Artificial Intelligence & Policy in India, Volume 1 (2020) by Indian Society of Artificial Intelligence & Law

Abhivardhan & Baldeep Singh Gill, Editors.

Artificial Intelligence and Policy in India, Volume 1(2020)

Editors: Abhivardhan, Baldeep Singh Gill. All rights reserved. No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the publisher and the authors of the respective manuscripts published as papers, except in the case of brief quotations embodied in critical reviews and certain other noncommercial uses permitted by copyright law. For permission requests, write to the publisher, addressed â&#x20AC;&#x153;Attention: Permissions Coordinator,â&#x20AC;? at the address below. ISBN: 979-86-627039-4-0 (Paperback via Amazon); 978-81-947131-0-4 (Online) Printed and distributed online by Indian Society of Artificial Intelligence and Law in the Republic of India. First edition, Volume 1, 2020. Price (Online): 300 INR Price (Paperback): 10 USD Indian Society of Artificial Intelligence and Law, 8/12, Patrika Marg, Civil Lines, Prayagraj, Uttar Pradesh, India - 211001 The publishing rights of the papers published in the book are reserved with the respective authors of the papers and the publisher of the book. The copyright of all the papers are reserved with the authors of the respective papers. For the purpose of citation, please follow the format for the list of references as follows: 2020. Artificial Intelligence and Policy in India. Prayagraj: Indian Society of Artificial Intelligence and Law, 2020. 9798662703940, 978-81-947131-0-4. You can also cite the book through citethisforme.com (recommended). For Online Correspondence purposes, please mail us at: editorial@isail.in | executive@isail.in For Physical Correspondence purposes, please send us letters at: 8/12, Patrika Marg, Civil Lines, Allahabad, Uttar Pradesh, India - 211001

Artificial Intelligence and Policy in India

Preface Artificial Intelligence and Policy in India is a special edited book published by the Research Directorate of Indian Society of Artificial Intelligence & Law. The book is a special coverage of policy papers, artificial intelligence and social sciences anthologies and briefs by our Research Members and in short, is a collection of novel ideas and propositions which researchers, editors and interns at ISAIL do believe in. We have also published some of the miscellaneous works on Artificial Intelligence and Law in this Volume. We believe the Discussion Papers are special proposals that have been accepted for the purpose of academic scholarship and open & rational discussion. The Recommendations for the Indian Strategy on AI and Law, 2020 (isail.in/strategy) have been contributed by Aastha Mittal, Koushik Doma Reddy, Rezbi Kaur, Saurojit Barua, Siddharth Jain, Kshitij Naik, Ruhi Tyagi, Trishla Parihar, Adetola Jesulayomi, Ritansha Lakshmi, Urvashi Arora, Dipali Khawle, Sarmad Ahmad, Manohar Samal, Paras Raj Maheshwari, Ananya Saraogi, Stuti Modi, Vedant Sinha, Vaishnavi Venkatesan, Sarthak Tripathi, Sameer Samal, Nayan Grover, Vasudha Tiwari, Ritam Khanna, Abhishikta Sengupta, Mriduptal Bhattacharyya and Nimrat Dhillon. I would like to express my deepest gratitude to Prof Suman Kalani, our Chief Research Officer, Abhivardhan, Abhishrut Singh, Prof Amrit Subhadarsi, Ankur Pandey, Shuvam Bhattacharya and the Secretariat for their support in making the editorial process possible.

Baldeep Singh Gill Chief Experience Officer Indian Society of Artificial Intelligence and Law.

Artificial Intelligence and Policy in India, Volume 1(2020)

Index Indian Strategy on AI and Law, 2020 Series. 1. February-March 2020 Recommendations Baldeep Singh Gill & Abhishrut Singh, Editors 2. April-June 2020 Recommendations Abhivardhan, Editor 3. Artificial Intelligence and Constitutionalism: Satiability and Extensivity of AI as a Legal Personality Sarmad Ahmad & Baldeep Singh Gill 4. Automated Killer: the treads around the soft earth of legality and Artificial intelligence Vedant Sinha 5. Artificial Intelligence and Juridical Considerations: Legal and Administrative Underpinnings in India Mridutpal Bhattacharyya 6. The Legal Viability of Patenting Ritam Khanna 7. Jurisprudential Modalities of Data Protection Regime and Privacy via AI in India Vaishnavi Venkatesan & Nayan Grover 8. Artificial Intelligence in Corporate Transactions: AI, Automated Mergers & Acquisition and Corporate Ethics Stuti Modi 9. Machine Learning and its Privacy Implications in India: Analysis of the Logistical Imperative in Data Protection and Jurisprudence Sameer Samal

Artificial Intelligence and Policy in India

Discussion Papers and Research Articles. 10. On the Entrepreneurial-Employability Ecosystem & Technology Preparedness due to COVID19 in India: Seminal Analysis & Proposed Solutions via the AI Ecosystem Abhivardhan, Suman Kalani, Ankur Pandey, Baldeep Singh Gill, Kshitij Naik, Manohar Samal, Ritansha Lakshmi, Adetola Jesulayomi & Ruhi Tyagi 11. AI and its Tortious Liability Sadaf Fahim

Artificial Intelligence and Policy in India, Volume 1(2020)

Indian Strategy on AI and Law, 2020 Series The Research Project by ISAIL, which proposes special and helpful recommendations and legal solutions on the dynamic and complex challenges related to the legal and ethical implementation of artificial intelligence and law in India. isail.in/strategy

Artificial Intelligence and Policy in India

February-March Recommendations 2020 Baldeep Singh Gill1 and Abhishrut Singh2, Editors 2

1 Chief Experience Officer Junior Associate Editor, Indian Journal of Artificial Intelligence and Law baldeep@isail.in

Artificial Intelligence and Constitutionalism: Legal and Administrative Challenges

1.1

Administrative challenges faced by Artificial Intelligence

• The R&D costs of AI are bulky and intergovernmental collaboration with relevant tech diplomacy can make India's constitutional governance system robust and replenishable. Other countries like Japan, UK and China have proposed to collaborate among industries and government entities in India; • The European Union has proposed public – private partnership incentives, particularly in affairs related to robotics and big data; • Democratized Artificial Intelligence in Technological Interactions • Open data: India is a densely populated country. The startups need strong and open data to build their prototypes. Though India has a national data and accessibility policy, i.e., the NCSS, the policy considerations are not competent enough as countries like the US, the UK and member-states of the EU have much availability of open database access for AIrelated research. Thus startups have to face many obstacles in innovation; • To overcome the problems faced by startups, the government can provide access to relevant data required for AI Ethics research through a database support policy. The Government can also provide access to data through open data platforms which provide range of data collected by various ministries. The State of Telangana has already developed its own database policy. Other states should also take some initiative if they can do so. Further state and central government agencies need to actively pursue and achieve a National Data and Accessibility Policy; • The data possessed by certain Private Sector Units (PSUs) may be accessed for rendering innovative solutions for AI research. To protect the interests of the PSUs, the data provided to researchers and developers should be encrypted using federated learning. Furthermore, the Union Government should promote data sandboxes for innovation and privacy protection; • The NITI Aayog has proposed to create traceability in the AI infrastructure using access control based on local and international regulations to entail robust price discovery mechanisms in data research. It will bring parity but would limit the use of private sector data and should not be replaced by open government data; • Re-thinking intellectual property regimes • The NITI Aayog pointed out in its reports that that Section 3(k), the Patents Act, 1970 (India) exempts artificial intelligence from the definition of a patent. Furthermore, any products or bi-products derived from algorithmic activity would face controversy over patent

Artificial Intelligence and Policy in India, Volume 1(2020)

issues. We believe that the NITI Aayog must consider the issue whether the patenting of products and biproducts of algorithms is possible and that whether the due extent of patenting of the bi-products of algorithmic activity therefore should also be answered. The question of problems faced by smaller companies for using or developing artificial intelligence regarding the issue when the products and bi-products of algorithmic data are patented is yet to be answered; 1.2

National infrastructure to support domestic development: Requires capacity for storage and processing

• There is a need to build an infrastructure of data storage for effective storage capacity, input/output operations per second and to gauge the ability to process massive amount of data; • For improvement in the incumbent national administrative infrastructure confined to cyberspace, the Union Government must upgrade and optimise the effect of scale. Scalability must be prioritized and would require a high bandwidth, low latency and creatively manifested cyber architecture. Thus, it requires appropriate last minute data curation environment; • Conceptualization and Implementation • Government entities and large companies should promote accessibility and encourage innovation in artificial intelligence; For example: the State Government of Karnataka in collaboration with NASSCOM is in the process of setting up a centre of excellence for data science and artificial intelligence; 1.3

Legal challenges faced by artificial intelligence

• Privacy and security is of foremost importance. The public and private sectors have failed to protect the interest of the citizens in cyberspace. The debate on privacy in India is based on the issue that companies are using consumer data inappropriately and they are misusing large numbers of consumer data for competitive advantages for perfecting their algorithms. Thus it may violate the Articles 14, 15, 19 & 21 of the Constitution of India, 1950 though it does not affect consumers negatively in a physical sense of understanding. Nevertheless, there are ethical implications and these implications must be adequately adjudged; • Internal and external auditing related to AI can be a reasonable mechanism towards creating transparency in the processes and results of AI-based solutions as they are implemented; • There are different forms and levels of transparency which include factors like criticality of function, potential direct and indirect harm, sensitivity of data involved, actor using the solution; • It will be important for India to define standards around human-machine interactions including the level of transparency that will be required; 1.4

Enacting a National Data Protection Law and Privacy law based on International Standards to Police AI

• In India, data are available to both public and private sectors which in turn may infringe the privacy of the citizens and persons under Indian Law. There are no proper guidelines to

Artificial Intelligence and Policy in India

regulate the privacy of the citizens. India needs to adopt an AI Privacy and Data Protection Law (or measure) which in turn may pass the following test we propose: ─ How algorithms are being trained? ─ How traditional data categories (PI vs. SPDI - meta data vs. content data etc.) need to be revisited in light of artificial intelligence? ─ How can a privacy legislation be applied to autonomous decision making? 1.5

Establishing sectoral regulatory framework

• The Central Government needs to set up a regulatory framework for public and private sector unit; • Stringent competition law: The use of competition law to curb data exclusionary or dataexploitative practices will first have to meet the threshold of establishing capacity for a firm to derive market power from its ability to sustain datasets unavailable to its competitors;

AI & Intellectual Property Law: Suggestions in Indian Context

• The next question which remains a confusion is whether IP protection can be claimed for an AI based software. In different countries, this question has different answers, however, in India for patenting AI, the Computer-related inventions (CRIs) guidelines can be followed. These guidelines have provisions on computer-based, algorithm-based and softwarebased innovations, inventions and discoveries, but many scholars provide that it can also be used for AI based inventions; • We could take a path similar to that of China which promises to become a world AI leader by 2030. China takes a three-tiered strategy: first, understanding and accepting the different functionalities of AI by this year of 2020. Second, make breakthroughs in the industry which would ultimately pave the way for them to become AI world-leaders by 2030, as they hope to be. Thirdly, to establish themselves at that pivotal position; • The AI ethics are not yet formalized and far from being legislated. When computers, machinery and systems are developed then the producers have to strike a balance between accuracy and fairness. Due to the lack of proper and formulated AI ethics, the producers and their superiors tend to bend towards the more profitable side which, here, is accuracy. Thus, sacrificing its fairness and bringing the possible risks of biases and unfair treatments and decisions. They are based on 3 components: lawfulness, compliance with laws, and regulations of the land which should be technologically and socially robust; • This would help maximize the benefits and minimizes the risks. There is a need for a social and technological- promoting safe atmosphere. This was a suggestion in the ISA 2019. What is needed in this year is its implementation and resultant outputs. Lawfulness could be ensured by checking whether the current IPR and AI laws go hand-in-hand with the laws of the land as they should not contrast the basic rules of the nation. Though these are new regulations, they will always remain inferior to the established laws; • A possible solution to this would be to develop an ‘AI Transparency Bill’ which would call for the need to decrease algorithmic biases and making the positions of those disfavoured even worse and this bill be enacted so as to ensure the protection of the biased subjects and the subjugation of those who bias and use this technology maliciously to their advantage.

Artificial Intelligence and Policy in India, Volume 1(2020)

This should also be open to public scrutiny. It is time to ensure that the differences which already exist to not further deepen with this drawback.

AI & Constitutional Law: E-Courts & AI in India

• Automation in Document Reviewing with the help of AI has extensive benefits for the Judicial System, automating the document reviewing process will help the Judges to selectively read only the information that is relevant on that particular date for that particular case, which will, in turn, reduce the time taken by the Judge in reading the whole file and Irrelevant material so that the Judge can concentrate on matters that are of much more importance. Such solutions can be developed for the Judges that help them understand at an instant whether a Judgement cited by a counsel has been overruled or is relevant to the present case • The only solution to help reduce the pendency of cases in the courts is to make them technologically advanced and move ahead of the traditional way of delivering justice, this could be achieved by developing solutions to keep a log of the proceedings in the courtroom and the decisions taken Judge Similar to the Traditional 'summary of proceedings' which are done manually, but one that will include even more details such as the time taken by both the parties in a Particular Case, the number of cases closed by the judge on a daily, weekly and Monthly biases, such solutions will help the Government as well as Public keep up to date about the Cases and will also decrease corruption. • AI-powered tools that will keep a log of Court Proceedings will also help judges to decide whether to post a particular Case for hearing on a particular date and decide if they will have the time to hear that case on a particular day according to the predictions made by the tool based on the average time taken to be heard by each case posted on that particular date, which will, in turn, increase the Judges efficiency in hearing and deciding cases. • Such Tools will help the Government Analyze the Courts performance and if it is not performing as expected to find solutions to increase its Performance. • The major reason for the pendency of Cases is the number of cases that are filed in the District Court, High Court and the Supreme Court is increasing each year by developing 'Predictive Justice' instruments. This Could be achieved by AI-Powered online tools for dispute resolution that help give the best solution for the Case by analysing the Case by itself and comparing it with the database available with it, this will not only reduce the number of cases filed in the courts, and also make Justice more approachable and inexpensive for the conmen people. • There are various instances when the Court has to hire a translator for the Case at documents are in a different language or the witness that appear before the court speak a different language this could increase the Cost in that particular matter and appointing a translator in the court is a lengthy process this could be totally eliminated by providing the court with Natural Language Processing solutions that will take out any human intervention at all. • Humans, at times we tend to have a bias towards a particular party and Judges are humans as well. Bias is a major cause of concern for Arbitration Cases in India and all around the world whereas Arbitration is the most sought after form of alternative dispute resolution. AI can be used in Arbitration matters which can help predict or come up with the fairest solution for the particular matter even if we don't rely totally on AI-powered Tools in Arbitration they will help the Arbitrator decide what would be the best possible solution.

Artificial Intelligence and Policy in India

• A central Database can be developed by the Government of all its citizens which may include all their identification details Like Adhar, PAN, Driving Licence in one database as well as details regarding their birth residence their Criminal Record. Such Database will help Courts much more efficiently on Cases before it as the Tools that they have will access this database and suggest probabilities regarding the particular case. Developing and maintaining a Central Database can be revolutionary in Legal Administration as well as Law Enforcement.

AI & Intellectual Property Law: AI, Creativity & Innovation Ethics

• The world intellectual property organization (WIPO), has performed various surveys and country-based analysis on the impact of copyright on AI and its impact on the economy. It has been clearly observed that providing patents to non – human entities in the field of artificial intelligence can lead to a boost in the economy and give an effective platform for more inventors to widen their horizons. As of now, India does not provide patents to the inventor of any AI software. This legal notion could be reconsidered in order to make sure that no individual or novel cases are left without any clarity. • Another key aspect to understand is whether the Indian legal system is providing a proper layout in order to deal with the complexities of both the industries. There is a need to provide a proper judgment on ‘computer-generated works’ and its role in authorship. Countries like Spain and Germany have clear and distinct laws on AI and Intellectual property rights which help them to minimalize their interlinked problems and then work for attaining a legal and just solution. • Countries like Japan, which is considered as one of the fastest moving economies has been applying the strategy for granting Artificial Intelligence programmer copyright for the concerned AI software which will increase or has a positive impact on the country’s economy. This strategy can also be used by India, in order to gain a positive result in the economy. • A proper standard must be laid down by the legislation in the context of eligibility of artificial intelligence software or computer-generated work in order to provide patent for the same. This will ease down the processing of courts and judgments and will help in resolving issues as fast as possible. • The use of the Abstraction-Filtration-Comparison test can be implemented in order to ensure that the infringement of the copyright has not occurred. Here the experts discuss two kinds of copying: literal copying and non-literal copying. They use certain tools in order to identify the codes which have been copied. If it is unable to find out through literal copying, then experts shift to non-literal copying. Here they observe the non-literal parts of the concerned software like the way it feels or its structure and sequence. This is known as the Abstraction-Filtration-Comparison test.

AI & Constitutional Law: Privacy Jurisprudence and Modalities of Artificial Intelligence

• The government must encourage public to trust and have confidence in AI technologies and at the same time protect civil liberties, privacy, and human values in their application to fully utilize the benefits of AI for citizens. AI should also be formulated and operated to keep in mind with ideals of human dignity, their rights, and cultural diversity of our country.

Artificial Intelligence and Policy in India, Volume 1(2020)

• Big data and predictive analytics are the keystones for analysing multiple types of data and its sources to predict and can be used to stop crimes before happening. • Integrate CCTV footage with social media applications and data collected in the control room. • Anticipate criminal intent through the use of sophisticated technologies and data-mining. • Setting up cyber-crime labs in the state to enable speedy registration and enable forensic investigation. • Improved and sophisticated data collection techniques are a mandatory prerequisite for running predictive policing programmes. • Manufacturers and operators of AI should be responsible and liable. • There should be a codified norm for AI to protect citizens rights and keep a check on the evil use of AI. • India should focus on Shared Benefit (benefit and empower as many people as possible). Shared Prosperity (The profit gained by AI should be shared, to benefit all of the people of our country) and Security of our country (recognising the need for AI to be safe and acknowledge their accountability) • Increasing interest in AI can be seen from the last few years. It is recommended that there should be a need for Data science courses focusing on the core element of AI development and Ethics. • The right to receive education or access information on new technologies and their ethical implications will facilitate that everyone understands risks and opportunities and is empowered to participate in decisional processes that crucially shape our future. • The government needs to ensure that the software developer in the private sector abides by the constitutional standards of due process, in the absence of a legal framework, shareholders strive towards carrying out measures that would defend them from unforeseen penalties and liabilities that may arise in the course of use and implementation of AI technology. • There is a requirement to amend the IPR law and incorporate the status of AI-generated work under it. Till now IP right is the incentive for the creators who create creative, original and useful work, IP right comes in various category like, Copyright, Patent, Industrial design etc. which was so far only eminent in the human domain. Since AI has started now creating music, article, create painting, design software etc. These activities meet almost all the criteria to get AI their IP right except the lacking legal personality. • Copyright has been approved only to natural or legal persons and any machine or tool used for creating any creative work are only considered as a mere tool and thus have not been granted any copyright in the programs name. Recommendation to amend the Copyright Act 1957, to include AI-related works as a separate category or to give AI recognition to be an author • The need is for the organizations to understand - Is the law allowing the organizations to collect, store and process the data. What is the responsibility of the entity storing the data and the laws about whether or what kind of data can be collected, what should be stored and how the ownership of the data should be established? • Data ownership: As devices are continuously exchanging the data between themselves and tons of data is getting stored in different places, there are many stakeholders and partners involved in this. The clarity around ownership of data needs to be established and looked into very carefully. • The hegemony of the human race may soon decrease to a certain level with the rise of dependency on AI technologies. To prevent unwanted changes humans must have the ability

Artificial Intelligence and Policy in India

to foresee law which must have the power to check any evil use of AI or to ensure that there must not be any negligence for a quick profit. AI is growing multi-fold technology and we do not know all the advantages or danger associated with it. Therefore, it is very important to have a double-layered protection model: first- technology regulators; and second- laws to regulate AI actions as well as for responsibility and liability of errors done by AI. As AI has the potential to impact the public at large so it has to be regulated and applied reasonably. The benefits are not discriminatory which is benefiting one particular section of people or causing harm to another section of people in the society.

Artificial Intelligence and Policy in India, Volume 1(2020)

April-June Recommendations 2020 Abhivardhan1, Editor 1

Chairperson & Managing Trustee abhivardhan@isail.in

AI & Constitutional Law: Legal Personhood of AI

• AI is once again a centre of discussion amongst the many technological advancements of the 21st century. While discussions, talks and developments of sentient machines have occurred since the inception of the Turing Test and the Dartmouth conference, research and development into this field that happens now is unprecedented, and large scale adoption and incorporation of AI into various institutions from company boards, organisations and firms to schools, parliaments, hospitals, etc is inevitable as the technology becomes democratized. • A human centric approach to defining AI, is that it is a variety of computer programs that are capable of assessing, performing and solving tasks that would require human like intelligence to execute. • The method of approach towards creating such programs is essentially developing them to mimic their human counter parts. Much like humans, AI is capable of operating through various degrees of autonomy, taking upon decisions to be executed independently, and are capable of “reflecting upon their past actions” by understanding errors in their execution and re writing their program independently to ensure that the task is executed effectively in the future when it does arise again. This process is referred to as Machine Learning, and allows AI to not only re write its own code but also enables it to take upon new skills and methods of task execution. • Machine learning is present as a trait across all the known forms of AI whether existent or hypothetical. The various kinds of AI are categorised as Artificial Narrow Intelligence (ANI/Weak AI), Artificial General Intelligence (AGI/Strong AI) and Artificial Super Intelligence. ANI is the only known and manifested form of AI today, and is essentially any AI that is strong at one task only, wherein its strength in executing that specific task surpasses human capability AGI is a speculated form of AI, which can operate, function and perform tasks as well as, or even better than their human counter parts. • ASI is also a speculated form of AI, wherein it is proposed that the functioning and operations of this AI will surpass the understanding and comprehension of even the most gifted of humans, across all fields of cognitive thought AGI and ASI exist only as hypotheticals, whereas ANI and its associated research and development constitutes the majority of the field of AI today. • It is through understanding Machine Learning, the capability of a machine to “Meta learn”, wherein a vacuum is observed with regards to the responsibility and accountability of any action executed independently by an AI If an AI teaches itself by re writing its own code, and executes a new action which results in an error inflicting damage to life or property, or even poses as a threat, who eventually is responsible? As the research and development towards AI grows and flourishes, making AI more autonomous and independent to the extent

Artificial Intelligence and Policy in India

•

• • •

•

where we lean towards the creation, use and application of AGI in various fields, holding the creators liable for the actions of autonomous AI will eventually fall short if not done effectively. One of the methods to circumvent and bridge the responsibility and accountability gap effectively is the legal recognition of AI, through the imposition of a legal personality upon it. Legal personhood is a concept imposed upon an animate or inanimate being in order to have the legal system around it recognise it as a legal entity. Once recognized, the legal entity is essentially holds a bundle of rights and responsibilities enforced in relation to its surrounding legal system, and can essentially be held liable for not adhering to its responsibilities and claim an infringement of rights against another entity. While the distinction between legal person and natural person remains intact, the granting of legal personhood to AI is often approached with concern or uneasiness. However, a look a jurisprudential history of the world and understanding the variety of legal systems that once existed and are present today, shines light on the fact that legal personhood is a flexible and variable aspect of any legal system dynamic enough to accommodate any legal subject as per the need of the society (the collective of legal subjects) governed by that system. The concept is also not static in its traits and characteristics, as the concept of legal personhood has always evolved and currently exists in many shades to accommodate a variety of legal subjects, such as natural persons, corporations, governmental organisations, etc. Human slaves under Roman law during the heights of the Roman Empire were not considered as natural persons or legal subjects, but rather as legal objects property that could be bought or sold by Roman masters. Similarly, slaves were also treated as property in the United States of America until the passage of the 13th U S Constitutional Amendment in 1865. Ironically, the abovementioned examples highlight not only the treatment of human beings as legal objects who would be considered natural persons in the 21st century, but also lacks the distinction between the concepts of legal subjects and legal objects. Roman slaves, being property themselves, were also granted property rights to a degree allowed for by their masters, and American slaves were almost always held responsible for their criminal acts so as to have their masters excluded from criminal liability. With context to the Indian legal system, the application of legal personhood to AI does not seem like a farfetched idea. Although there is nothing explicitly codified within the provisions of the Constitution of India, 1950 that concerns legal personality, several precedents identify the Indian recognition of the legal personalities of several animate and inanimate entities. The honorable High Court of Punjab and Haryana, Chandigarh, emphasised in the case of Karnail Singh v State of Haryana that the entire animal kingdom including avian and aquatic species has a “distinct legal persona with corresponding rights, duties, and liabilities of a living person”. It is further emphasised in Animal Welfare Board v Nagaraja that animals are entitled to fair treatment and dignity under Article 21 of the Constitution Lastly, inanimate beings, such as naturally occurring water bodies have been given some form of recognition as well, the example of this being the Whanganui river in New Zealand. The legal recognition of AI therefore is very foreseeable. However, much like the ever-changing concept of legal personhood itself, it needs its own flexibilities and types of recognition. This distinction could be made on the basis of the strength of the AI itself; having ANI legally recognized as legal agents, and recognising a potential AGI as a legal person. A legal

Artificial Intelligence and Policy in India, Volume 1(2020)

person bears rights and responsibilities to be enforced in accordance with their actions and behaviors that are conducted out of their own will. A legal agent however, acts within a specific requirement as prescribed by the principle entity on behalf of whom the agent functions. • This classification is justified, as although ANI can execute decisions through its own autonomy, it requires a degree of human interference in order to function effectively; be it the input of data, training of the algorithm etc. Such is the case that is observed with any example of AI that exists today, all of which are predominantly ANI. • Through this distinction and the recognition of ANI as a legal agent, the responsibility and accountability of an ANI’s actions can then be imposed upon an existing legal person that is involved in its operations and functioning. As legal agents, liabilities arising out of an AI's act shall be imposed upon the principal. The role of the principal herein, is circumstantial and dependent on the situation. This can either be: a. the legal entity that manufactures the AI, b. the developer/s responsible for the algorithmic error, c. Or even the end user of the AI, if the AI is used outside its intended situation. (example of such being numerous instances of Tesla accidents wherein the driver had activated the self-driving feature outside its intended use on a highway); • It is also recognised that although still considered as ANI, there are a variety of AI that exist, operate and function as a collective of multiple ANI algorithms working together to generate an outcome. This is because in order to respond to the specific task they are to tackle, one algorithm may require specific and processed data that can only be delivered by another algorithm. This creates a chain of processing, wherein the output of one algorithm is utilised as the input for the next in the chain. • Such clusters or collections of ANI algorithms functioning together may be classified as Artificial Adept Intelligence (AAI), and can be considered stronger than ANI but weaker than AGI, hence creating a category right in between of those two classifications. AAI can hence be defined as an AI that is a group of various ANI algorithms working towards one specific function, such as driving. • The objective of creating such a classification is to further extend the imposition of liability; for if an AAI maybe responsible for any error, liability can be imposed upon the developer/ developing team specific to algorithm that committed an error. For example: if one of the many the visual-centric ANI of a self-driving car is at fault, then it imposes no liability on the developer responsible for the ANI that handles locomotive function of the vehicle. • With regards to the legal personhood of AGI, it is observed that no AGI exists in our present day and hence, within the current legal scenario. However the likelihood of an AGI is inevitable and can create disruption if appropriate measures aren’t in place to smoothen the adoption, incorporation and existence of the same. Hence, it is recommended that councils or bodies are established with the objective that they work towards AGI and its potential legal personality. These bodies ought to involve inter-disciplinary efforts towards the solving of questions posed by the legal personhood of AGI, and will have to incorporate risk assessment strategies to assess the consequences of the same. • Such questions may transcend legal, technical, ethical and philosophical boundaries, and hence will require cross-cultural effort aside from just inter-disciplinary effort, towards the creation of neo-legal theories, such as a list of criteria establishing AGI personhood, the requirements of artificial consciousness, or even a theory of penology for AI.

Artificial Intelligence and Policy in India

• Lastly, a legislation can be drafted, bringing the recognition of the extensions of legal recognition of AI into codification. Important and dynamic social changes can be emphasised and addressed effectively once they are codified and are bought to consideration with regards to their potential implications. Such a consideration is reflected in the motion passed by the European Parliament resolution of 16th February 2017 with recommendations to the Commission on Civil Law Rules on Robotics (59f, 2015/2013 (INL)) A8-0005/2017.

AI & Intellectual Property Law: AI, Creativity and Innovation Ethics

• The World Intellectual Property Rights Organization defines Copyright as the rights creators have over their literary and artistic work. These include Economic Rights and Moral Rights. While Economic Rights allow authors to gain financially by granting them the right to exclusively use or authorize to use their work, Moral Rights allow the author to prevent any unauthorized distortion of their work. Ever since the 1970s creators have used technology to create content in which the creative input was provided by the programmers and the technology was merely a tool, just like a camera. • However, with the advancement in the field of Artificial Intelligence and the ability of AI to take autonomous decisions, the creative input of the programmers has reduced significantly. Now we have AI programs creating content such as news articles, novels, art, music which involve minimal human intervention. • However, our Copyright laws have not evolved to take into consideration the authorship issues involved with content generated by AI programs. Can an AI program be given copyright over the work it generates? If not, then should the work be placed in the public domain free of copyright, or should the humans behind that AI program be given the copyright? • This research considers the above issues and recommends that we should have specific provisions for AI-generated works outside the framework of traditional copyright, synonymous with related or neighboring rights. The Copyright Act, 1957 can be amended to confer neighboring rights on humans behind the AI program, just as it has provisions for Broadcasting organizations and performers. 2.1

Possible Scenarios

• No human author, no copyright: Section 13 of the Copyright Act, 1957 states that Copyright shall subsist in ORIGINAL literary, dramatic, musical, and artistic works. The test of originality has shifted from the “doctrine of the sweat of the brow” to “modicum of creativity”, requiring the author to demonstrate certain minimum creativity to claim copyright over his work. • Tested on this principle, the AI seems to lack any creativity of its own. It is designed to execute certain instructions and that is all. The Court of Justice of the European Union in Infopaq decision held that original work must reflect the author’s personality, which AI clearly lacks. In the Naruto case US Court of Appeals for the Ninth Circuit held that under the US laws, only humans can hold a copyright and thus denied copyright to a monkey over his selfie. • The US Copyright Office further states that it “will register an original work of authorship, provided that the work was created by a human being.” The Federal Courts in Australia also

Artificial Intelligence and Policy in India, Volume 1(2020)

stated that a work generated by a computer without human intervention cannot be protected under copyright laws. • The requirement of human author is obvious in Copyright Act, 1957 itself as Section 22 states that copyright shall subsist until 60 years following the year in which the author dies. Thus, under the current laws, there is no provision for granting copyright to AI generated works. • However, such a position in law will bring all AI generated works in the public domain and the AI programmer will have no incentive to invest in such programs, inhibiting the advancement of AI technologies. Entrepreneurs entering into such AI ventures might choose to keep it a trade secret, keeping the know-how of such technologies away from the public view. Thus, amendment in current law is desired to provide some protection to content generated by AI and incentivize the investment in this field. 2.2

AI as an Author

• If copyright law is to be amended or interpreted so as to protect AI generated works, is it feasible that AI is recognized as an author of such works? The argument that AI lacks the wilful intention to impress its personality on the content it creates can be held to be not decisive because even minors and incapacitated persons can be authors. However, under the current laws, AI cannot be treated as a legal person. The EU Parliament’s proposal to grant specific legal status to AI as “e-persons” was met with heavy criticism. In the face of such legal uncertainty over the status of AI, authorship cannot be attributed under copyright laws. 2.3

Authorship to Humans behind AI

• Since the intellectual property laws are not amenable to non-human authors, a few jurisdictions amended their laws to allocate copyright to humans operating the AI program. This legal fiction of conferring authorship on a person who is not the author-in-fact has been well recognised in “Work For Hire” doctrine under which the employer is taken to be the holder of copyright of work made for hire. This allows the employer to reap economic rewards by exclusive use of the work and incentivises him to invest further in artistic works. A similar analogy has been applied in AI generated works in Hong Kong, New Zealand, UK, Ireland. • Section 9(3) of Copyright, Designs and Patents Act 1988 of United Kingdom states that: “In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken”. • However, there remains an ambiguity over who the law would consider to be the person making necessary arrangements for creation of the work. Should it be the programmer or the user of the AI? The judiciary has approached this question on a case by case basis. • In Express Newspaper case, the court held that the computer was being used as a tool just as a pen and hence the copyright should be granted to the user. In Nova Productions case, the Court of Appeal evaluated the authorship of a graphic work in a computer game and held that “the user’s input is not artistic in nature and he has contributed no skill or labour of an artistic kind”. The authorship of that graphic work was thus awarded to the programmer. Another issue with this approach is will the person who makes necessary arrangements

Artificial Intelligence and Policy in India

be criminally liable if the AI violates the copyright of any other work, even if he had no mens rea? • Neighbouring Rights for AI generated works: It is evident that while rationale for granting copyright was rewarding authorship, in case of AI-generated works the rationale is limited to economic incentive as the AI lacks a personality of its own. We must also consider that there remains an inherent gap between human and AI creativity, and the latter itself is a product of human creativity. Thus, it can be inferred that instead of copyright, the legal framework of Related or Neighbouring Rights will be more suitable for the protection of AI-generated works. Related Rights are currently given to Broadcasting Organisations and Performers under Chapter VIII of the Copyright Act, 1957. These are of shorter durations with the aim to protect the investment and provide economic incentives. A separate provision granting related rights to AI generated work can also resolve the contention around the criminal punishment by not attributing offences under Chapter XIII of the Copyright Act, 1957 to the AI programmers or users. Instead the provisions could provide for immediate destruction of such works created by AI which infringe copyright of other authors. • It is recommended that specific provisions granting neighbouring rights to AI-generated works be enacted after considering the existing and potential state of AI. The provisions must ensure adequate protection to the investment of AI-programmer or user so as to incentivize further advancement in AI technologies. This will ensure that our copyright law adapts well to the technological and economic realities.

AI, Ecology & Space Studies: Artificial Intelligence and Commercialization of Outer Space

• The vision of commercial space activity in India is not a newfound phenomenon and has already been introduced through the Space 2.0 phase which is currently dedicated to enable space entrepreneurs, small and medium scale enterprises to compete in the commercial space race which is worth $300 billion dollars. Evidence of the origin of commercial space activities can be traced to the year 1992, which was when Antrix Corporation Limited, a company owned by the Indian Government was established. The Pragyan Rover launched with Chandrayaan- 2 is one of the most successful artificial intelligence rovers launched by India and showcases the potential of artificial intelligence in space missions. • Some of the areas where artificial intelligence can contribute in enhancing space commercial activities in India are improved risk assessment of projects, progress in data collection, analysis, transmission, mapping and management, efficacious manufacture and development of space products such as spacecrafts, rockets, probes, rovers, space suits and telescopes, technology capacity building, efficient launch and landing, improvement in mission success rates, indulgence in commercial remote sensing, prolonged space travel, simulated training for astronauts, improved mission support systems, amelioration of services in India like geospatial positioning, internet and telecommunications and some long term goals such as asteroid mining and space tourism. • In order to ensure the successful application of artificial intelligence in commercial space activities, it is extremely vital that a central space law is passed. Such law would have to preliminarily stipulate the areas of commercial space in which private enterprises can contribute and the areas in which they are restricted, provide guidelines for jurisdiction over space objects and discoveries, envisage clear principles of liability and a penal structure

Artificial Intelligence and Policy in India, Volume 1(2020)

mechanism. It is indisputable that in the initial decades of the operation of such law it will not be possible to accommodate fully privatised space commercial activities and supervision will require to be strict in order to facilitate sustainable and orderly utilisation of commercial space. In view of the fact that space activities involve country responsibility, a critical effect on diplomatic and international relations and impact on the planet itself, it is pertinent that the penal mechanism inculcated within an Indian space law will have to be closely connected to Indian criminal jurisprudence and will have to create a “right in rem”. It is noteworthy that in the absence of these elements in a codified space law, basic activities will not occur smoothly and the application of artificial intelligence might not bring out the best results. • After achieving the first step of implementing a resilient, all- embracing and coherent central space law, public- private partnership would have to be embraced within the provisions of such law. Several applicable public private partnership models and a model concession agreement to cater the relationship between the public and private sector will have to be formalised. Although the Indian Space Research Organisation has already floated a tender in the year 2017, after the adoption of a central space law, the volume of operations will significantly rise and the present structure will then be rendered insufficient. For increased development and usage of artificial intelligence in these activities, a partnership with technology based, robotics and artificial intelligence development companies will have to increase. Incentive schemes have been one of the most successful methodologies to attract investment and partnerships in any sector in India which has included tax and duty waivers, partial and absolute, land allocation and government grants. Attraction of investors and constructive public private partnerships between artificial intelligence tech- companies and the Indian Space Research Organisation can lead to the positive development of enhancement in manufacture and innovation of space products such as spacecrafts, rockets, probes, rovers, space suits and telescopes that will significantly boost the duration of space travel and its activities, ensure advancements in remote sensing and other services . • The Indian Space Research Organisation has already employed artificial neural networks in mission support systems and the collection, analysis, transmission, mapping and management of data. It is pertinent that artificial intelligence is also used for the improvement of simulated training of astronauts, risk assessment and analysis, which can result in progress of the mission success rate of the targeted commercial space activity. Softwares using artificial intelligence algorithms such as Space Mission Architecture and Risk Analysis Tool (SMART) are already being used for conducting risk analysis and assessment. Furthermore, Visual Environment for Remote Virtual Exploration (VERVE) is one of the training simulation platforms for astronauts. Therefore, it is trite that capacity building in artificial intelligence technology forms the crux of how progress can be achieved in these activities. Bilateral treaties that emphasise upon import of artificial intelligence technology for commercial space activities can prove to be resourceful for capacity building. However, it is necessary that simultaneous indigenous development is also facilitated and catered so that dependency rates do not remain extremely high in the upcoming decades. • It is significant that law and policy- making is rethought for the purposes of commercial space activities. This is because the rise in such commercial activities in space will inevitably result in the development of new professions and the increase in space travellers that will not essentially be astronauts. The role of artificial intelligence will be extremely high as more virtual reality based simulations will be used for the rigorous training of such non- astronaut space travellers.

Artificial Intelligence and Policy in India

• Another key aspect which a central space law would have to emphasize on is the distinction of regulations between autonomous and manned missions. A host of delegated legislation that include procedures for registration, mission supervision and licensing. In order to ensure the development and increment of space exploration and sub- orbital activities, it is pertinent that the application of artificial intelligence is also explored in robotics to create autonomous space products such as autonomous rovers, landers and probes. Simplification of procedure in attaining reciprocal treatment of intellectual property rights of existing artificial intelligence based space products and a robust mechanism that permits the ownership of certain specified space objects on discovery by private entities is capable of magnifying the amount of autonomous commercial space activities from India. The magnification of commercial space activities in India will also result in the indulgence of private entities in constructing launch pads and therefore, in order to avoid the incoherent construction and development that affects Master Plans of urban and rural development has to be regulated. • Increased usage of the latest 3D printing tools and the benefits under the “Make In India Scheme” can significantly aid in the reduction of manufacture, operation and ultimately, the overall mission costs. Indian space missions are already popular for being cost- effective and is also considered as one of the best nations who is capable of efficiently launching nano and mini satellites. However, the central space law will have to address certain challenges to ensure that all types of commercial space activities are cost- effective and ecological. Since space law is not concretely codified in India till today, its formulation can mandate manufacturers to research, develop and utilise clean and sustainable technology to build space products that reduce prices. Application of Space Based Solar Power (SBSP), reusable space vehicles, better payload management, efficient Power Management and Distribution (PMAD) and energy storage systems are few of the clean technology methods that can be mandated as “basic standards” for Indian based space products. Moreover, a space regulatory agency will have to be established which not only will regulate the private sector in India but will also regulate exports of Indian manufactured space products to other nations of the Global South. Furthermore, the national agency can also be entrusted to provide training to other Global South nations and also, create guidelines for Indian partnerships with other nations to launch their products into outer space. • Presently, a legal framework for space tourism and asteroid mining may not be conducive since India may be quite a lot of decades away from indulging in such activity. However, at the same time a resilient framework that accommodates a strong support system and services and participation towards other nations of the Global North that are closer towards achieving this goal can certainly act as a catalyst to speed up the process in India. • It is not unreasonable to infer that a commercial space race may lead unsustainable activities that harm space objects and the whole planet itself and therefore, it is extremely vital to stringently and manifestly formulate and implement policies that will facilitate and promote sustainable commercial space activity and sustainable exploitation of space resources.

Artificial Intelligence and Policy in India, Volume 1(2020)

AI and Constitutional Law: Civilian and Military Environments & the Compliance of AI

• Clive Humby coined the phrase, that “Data is the new oil”, however Lt. Gen Jack Shanahan feels differently, that it is “a mineral ore, there’s a lot of crap. You have to filter out the impurities from the raw material to get the gold nuggets”. The enormous amounts of data flowing within the physical infrastructure can be mostly non-consequential at face value, however when purified, it holds value, thereby forming one of the stress point in usage of AI, the dependability on the quality of the dataset. Therefore, the methods to purify the data are as important as the AI, it is being fed, based on AI gullibility. • AI is very gullible, amid pattern recognition the minutest of impurities data can consequentially change the nature and the results of the AI, so much that the deviation due to outliers, missing values or can alter the projected results. A large set of clearly labelled, well-organized data points is what machine learning algorithms need to learn from before they can try making sense of raw data the humans haven’t cleaned up for them. • In a military setting, the AI must be robust. For a robust AI to work in a surveillance situation, for example, to differentiate between a Civilian and a militant, it has to first understand the difference between what is normal and abnormal behaviour for a civilian and to do the same, it has to train on a wide assortment of civilian data, to establish a range of normality and abnormality. However, privacy concerns do arise and the Constitutional Courts can deprive the access to data per the privacy concern, deprives the military of relevant datasets to utilise. There the balance between the private data of a person is of National security consequence or not, has to be demarcated or even if this operation has to be performed or not. Doing it provides the state legitimacy to claim sovereign immunity under Tort Law. • The lines get blurrier in a warzone, where the citizen is often not hostile, but not cooperative with the occupational forces as well, and the data-points extracted from a set of co-operating citizen mindful of the law is different from those civilians whose rights and its co-relatives are altered in a warzone. Thereby different standards for AI arise in both peaceful and warzone situations and even within this scenario, there exists a probability for Algorithmic Bias to exist. Therefore, AI may be trained in a normal situation might probably tag normal civilians in warlike situations as militants. • It is not only sufficient to be able to conclusively segregate militants from a normal citizen, in a peaceful or a warzone scenario. It is equally important for the AI to delineate the explanations for the outcome achieved and answer questions such as “why”, “how” of such classifications. • Explainability of the AI employed in the military should be a big factor under consideration, and such models such be employed that does not morph into a Blackbox. Any such solution that seeks to unravel the reasoning such as deep explanation or model induction only does as a predictive, not an affirmatory solution. • If the AI decides to tag an individual, then the right of authorisation to execute an action, in that case, should still reside with a Human operator as a safety valve against any mistake that the AI might commit. • If the human operator relies on the AI to execute a decision, then the liability in such a scenario might not rest upon AI, as the human operator had the chance to exercise due

Artificial Intelligence and Policy in India

diligence. However, it varies in terms of operations, therefore will AI-powered apparatus be used in offensive operations is still to be determined. • The paradigms of warfare are changing. US Northern Command head Gen. Terrence O’Shaughnessy says the key to winning tomorrow’s all-domain wars is predicting an adversary’s actions hours and even days in advance. Synthesis of AI with increased situational awareness is eventually increasing the efficacy of the military, such as $1500 swarm drone using datalink and controlled by AI can enable 1000 such drones to perform simultaneously, in future will perform the same task as a $100 million jet. A simulated exercise, involving a human opponent to a unit mixed with robot, humans and drones reveals that they can do the same job that would have been done by a force 3 times their size. AI is lowering the cost of war overall, if there exists a moral cost of putting the responsibility and the trigger to a computer program will be the same or not is a different question to answer.

Artificial Intelligence and Corporate Governance: AI, Automated Mergers & Acquisition and Corporate Ethics

• The global market for Mergers and Acquisition in 2018 was worth $4 Trillion and amounted to 50,855 deals. Surprisingly, in 2019 despite of worries about potential downturn of global economy due to geo-political issues like Brexit and US-China Trade tensions, a 1% increase in deal value was witnessed, with only 2% downfall in deal count. The first half of 2019 witnessed strong mega-deal activity in the US, which was balanced by the slower second half of the year. Relevantly, the trend was opposite in Europe and Asia where the year started off slowly and the second half picked up. However, in 2020 with the anticipated Recession caused by Coronavirus it is essential to bear in mind that every downturn produces its own winners and losers which depends upon the strategies adopted, which differentiates them. The winners strategize to using scale and scope Merger and Acquisition and divest proactively in order to reshape their portfolios. This can be evidenced from the India M&A Report, 2019 which was published by Bain & Company in collaboration with Confederation of Indian Industry (CCI), where they reiterated that 70% of growth in 2018 was because of Distress Deals, enabled through the Corporate Insolvency Resolution Process under the IBC. Deals enlarging the sectors in which the business functions rather than helping it scale up its existing activities will lead Merger and Acquisition activities this year due to persistence of dynamics driving growth of these deals in the previous two years. 5.1

Expected M&A Trends in India, 2020

• It is expected that growth of M&A deals in the country will be favourable in the year due to the commendable regulatory regime introduced earlier by the government. • Key reforms include reducing tax rates which in turn will incentivise manufacturing facilities either setting up or acquisition, thereby inviting direct or indirect foreign investment supplementing M&A deals. Further, expansion of M&A transactions is also expected out of CIRP Process under IBC and as entities contemplate expansion of its core/non-core vertical and horizontal business through buyouts, to maintain their competitiveness in the market. Moreover, divestment of group business lead by restructuring and reorganising in order to obliterate archaic ways of doing business is another factor contributing to a positive and

Artificial Intelligence and Policy in India, Volume 1(2020)

conclusive trend of M&A Transactions. Additionally, business initiatives facilitating foreign investments would not only contribute to growth of volume of M&A deals in the country but also value.

Imperative Requirement for AI in M&A. • The year 2019, witnessed unprecedented regulatory scrutiny of M&A Deals. There is a tremendous rise in complexity, time and resources to surpass the regulatory hurdle. Consequently, resulting in costing more both in terms of cost and time. Further, highest deal failures can be traced back to poor due diligence that did not identified critical issues. The traditional approach of solely relying on Spreadsheet to analyse data is quickly becoming relic of the past. There has been a major shift towards a more dynamic, integrated and analytical process, tool and technique. This contemporary technique can be used to deliver what seemed impossible in the past, a combination of big picture insights and a microscopic level of detailed analysis, all in less time, effort and with more precision. Adoption of AI still in pre-mature stage in Corporate Finance. • AI has the potential of taking M&A Analytics to the next level, under which it is currently focusing on smart automation and making automation tolls and processes smarter and efficient. The efforts that were traditionally time-consuming, labour intensive and required human judgements is now being replaced with M&A Analytical platforms currently using AI and machine learning to analyse massive amount of data. Deloitte lead M&A platform called iDeal can not only make most of the work done with little or no human involvement but also learns from its mistakes when humans make correction, becoming more reliable, updated and accurate. Even in India, MnA Genome adopts AI as a tool to facilitate decision making at the due diligence stage and can also be used at the post-merger stage to align culture of both organisations. • However, the actual deployment of AI is still at an early stage in Corporate Finance as compared to other professional services. One factor being complexity of large corporate transaction, which makes it difficult to replicate and standardize tasks that need human judgement, collaboration and experience. In M&A deals it is difficult to reduce to replicable process since each transaction has a specific nature. Activity being heavily focused on data in Corporate Financing, the issue is with the accuracy and reliability. Work is required to clean the data to use it for AI application. A massive shift will be seen when data will be cleaned at source, transferred and stored with accuracy. This will minimise the grey areas. For people who think they can work with nothing but clean data, it should also be borne in mind that it is wasteful waiting for perfect data since it is not always practicable or cost effective to clean it up. In most business context, data would be cleaned only when it mattered for reasons of cost, trust or safety. Proposed Integrated use of Humans & AI. • As the Discussion Paper on National Strategy for Artificial Intelligence by NITI Aayog has validly established, the deployment of AI in each sector is supposed to consider the incremental value the adoption of technology can provide to improve the pre-existing processes

Artificial Intelligence and Policy in India

within each sector. The primary aim is to enhance the process for efficiency and effectivity rather than aspiring to be a tool to replace human decision-making in its entirety. This paper proposes an integrated human and machine-learning process in order to identify, classify, prioritize, organise and highlight document, which must be disclosed for business combination agreements for higher efficacy and speed and lower cost than humans alone can. The paper further suggests that an automated model, particularly Reflexive Random Indexing, can add appreciable value to the M&A due diligence process by identifying and indicating clearly related indirect connection. 5.2

Proposal of AI deployment in various stages of M&A Deals Process

AI in Deal Origination. • For at least two decades, web-based match-making services have been around for buyers and sellers of small companies. For example, New-York based Axial Network, being a new generation online service provider, is using algorithms to recommend most relevant parties that the buyers and sellers can approach, by taking into account each buyer’s and investor’s realtime intent, along with strategic and financial interest on both sides of the deal. Further, Euan Cameron, UK AI lead at PwC, highlighted that AI could be used effectively to learn from past deals, by identifying success and failure factors and inform the preferred characteristics of future deals. An even more futuristic approach would be to analyse historical data by understanding actual outcome of deals and to link this knowledge to features of the target company. AI in Company Valuation. • Identifying comparable business and assets is a crucial area which will benefit from integration of appropriate automated analytical process backed and checked by skilled professionals. AI in Negotiation. • AI can also be successfully used in negotiation which is considered to be the most human part of the deal. Right data is required in order to help identify clauses which commonly cause issue. Additionally, historical data could be utilized to identify what the market standard is on certain terms.

AI in Due Diligence. • Prolonged expensive hours are required in viewing documents for M&A market. The due diligence process could undoubtedly be largely automated, leading to faster and cheaper transactions which have a better risk management. Reflexive Random Indexing deploys an efficient, elegant solution for the challenge of collecting, classifying, organizing, prioritizing

Artificial Intelligence and Policy in India, Volume 1(2020)

and highlighting the corporate documents in issue integrated with Human input on risk assessment. • It is further recommended for the success of the model to make it more iterative, continually accommodating, objective and responsive. This is because the agreements are continuously revised during negotiation. Additionally, the model would highlight the most relevant words in each document enabling faster manual review. The software could use different colours for different queries so that the users can evaluate multiple queries simultaneously. The user could then easily label documents as responsive or not, giving the model an opportunity to improve continuously. AI in Deal Completion. • The highly technical process of deal completion which in itself involves negotiations, can also be simplified by using AI. Learning and insights could be gathered from hundreds and thousands of previous contracts. The inherent potential of machines over humans in this context should be judiciously used. AI in Post-Transaction to Exit/Divestment. • The analytical tool underpinning AI can be used for value creation activities that the business might not have identified before, which have an extensive reach across industries and value chains. 5.3

Recommendations for Ethical and Legal Consideration

• With the deployment of AI in M&A transactions, issues of privacy and security, liability and accountability, oversight, evaluation, transparency, redressal, lack of due process causing bias and discrimination also need redressal. • The recommendations for the paper propose a rule-based system applied contextually in designing ethics, due process, fairness and transparency. Moreover, an Algorithmic Impact Assessment is recommended, whereby the onus lies on the authorities to deploy guidelines and procedures for evaluating the impact of AI-driven solutions. A combination of delegating controlled discretion to automated system and adoption of Constitutional Principles of proximity, proportionality and arbitrariness to assess the use of AI in governance should be undertaken. An appropriate and contextualized process needs to be developed by the government, when decision making is being carried out by AI along with ensuring transparency regarding the factors undertaken for such decision-making. Further, Public Private Partnership requires a cohesive and uniform framework for regulating the partnership which is entered between government and private sector. There is an imperative need for establishment of adequate Redressal Mechanism, which would foster accountability and would be accessible to all stakeholders. Interdisciplinary approach for furtherance of AI, which is attained by integration of technological progress along with economic, political, demographic, anthropological and legal aspects which account for fairness and due process. Additionally, the technology benefits must reach lowest common denomination, adhere to international obligations, Sustainable Development Goals and guarantee socio-economic rights. AI should be deployed in such a manner that the existing domestic and international human right standards along with commitment to the environment is honoured.

Artificial Intelligence and Policy in India

5.4

Conclusion

• Conclusively, the deployment of AI in carrying out analytics of M&A Deals would surely induce some challenges. However, not only such problems can be fixed but the merits of such adoption outweigh the demerits of it. Hence, the use of AI in M&A Transactions remains inevitable in the future.

AI and Constitutional Law: Artificial Intelligence and Privacy

6.1

Introduction

• In today’s world, Artificial Intelligence is at the forefront of technological development. From merely texts in pages of a science fiction book to authentic existence, Artificial Intelligence has come a long way. Things that could only be envisioned earlier have now become a staunch reality. Some perceive it as a threat while several apprehend it as a tool for social good however irrespective of the perception, it is an indisputable reality of today’s world. • Despite the plethora of information surrounding AI and envisioning its implementation for societal good, it raises numerous concerns. A highly debated issue is that of privacy considerations. Emergence of sophisticated AI in its multi-faceted domains, which are majorly data driven, have exacerbated privacy concerns. In this information era, privacy hinges on our ability to control how our data is being stored, modified, and exchanged between different parties. • The following are recommendations for steps to be taken in order to ensure better privacy for a human being in this AI driven era. A part of recommendations provided are for improving certain general areas of the data protection regime so that when they interact with Artificial Intelligence, better privacy is ensured. While other solutions are specific to the interaction of Artificial Intelligence with consumer data. 6.2

Recommendations

• India presently does not have a data protection law however the Personal Data Protection Bill, 2019 which envisages a comprehensive policy on data protection is in development and is pending approval, which was the backdrop of the Justice K.S.Puttaswamy (Retd) vs Union Of India judgement. India needs to adopt a comprehensive framework for the regulation of data which suits the best interests of the country whilst dealing with omnipresent problems that plague the current landscape of data privacy in India. It needs to provide rights to individuals with stringent measures prescribed in cases of contravention of the same. • It is imperative that the data protection law, which is to develop into the law of the land, should be free of bias. In its current form, the bill provides unrestricted access to government agencies for matters of ‘nation concern’ or in the interest of the sovereignty and integrity of India. These terms are vague and leave room for ambiguity to kick in thus paving way for misuse of the provisions and defeating the very purpose of the act. Situations wherein the government can bypass such checks and balances should be dealt with exhaustively. It is imperative that more accountability and transparency be attributed to Governmental organizations/ agencies while dealing with the subject matter.

Artificial Intelligence and Policy in India, Volume 1(2020)

• It is proposed that India needs a Central Data regulatory authority which will act as the sole body handling data, leading to an increase in centralization and transparency whilst dealing with sensitive personal information. Instead of information being scattered across multifaceted domains, it would pass through a designated authority which shall be the fiduciary holder of this data. The concept of data protection officers as listed in the GDPR can also be implemented in India by molding it as per requirements wherein the data protection officer would be responsible for overseeing a company’s data protection strategy and its implementation to ensure compliance with the country’s policies and framework. Keeping It Simple. • Privacy Policies drafted by companies should be in plain simple language. The privacy policy is drafted for consumers, who don’t prefer reading technical jargons or legalese, so use of them should be avoided as much as possible. The Privacy policies are lengthy in the first place and a further addition of Technical Jargon and Legalese makes it even more difficult for consumers to read and comprehend it. A study by New York Times included analyzing the readability of privacy policies of different companies by using a Lexile test developed by the education company Metametrics. The test measures a text’s complexity based on factors like sentence length and the difficulty of vocabulary. They analysed 150 privacy policies and except two or three all others were found to be much less readable and more complex than the college study texts. This rings alarm bells and makes consumers more vulnerable to a privacy breach since the privacy policy to which they consented was not completely understood by them. In India where most of the population with access to internet is not digitally literate or is at a premature stage of understanding these nuances, this becomes a much huge problem. Consumers should not need an understanding of complex methodologies and processes involved in data collection in order to understand where their data is being used. And So, it should be mandated by the Data Protection Law in India that companies should have their privacy policies in plain and simple language which could be evaluated by some method or tool like the one involved in the study done by The New York Times. Choosing Rights Over Consent. • Until now data protection frameworks have had a consent-based approach i.e. where companies could collect and use data of consumers in any way after acquiring their consent on a privacy policy through click wrap. But in today’s age when data terminologies are so complex and as mentioned above the privacy policies are not easily comprehensible for a common consumer. Thus, it is suggested that while framing data protection laws a Rights based approach should be followed where certain guidelines are laid out by law as to what extent and manner the data of a consumer, although obtained with consent, can be used. Such laws should limit the use and collection of data in a manner to protect the violation of data rights of consumers. Breaking of such laws should lead to penalisation of the corporation doing so. This would make corporations and organizations handling consumer data more accountable as to their usage of that data. Especially in a country like India where a larger part of the population which has access to technology is not much digitally literate and is trying to learn little by little every day. These types of consumers are most vulnerable to violation of Data and Privacy rights and so a rights-based approach becomes more necessary for a country like India.

Artificial Intelligence and Policy in India

Opt-in rather than Opt-out. • To discover a consumer’s consent status, there exists two systems for data collection – Optin and Opt-out. In an Opt-in system for data collection, data collection is turned off by default and the consumer has to expressly give his consent in order to start data collection whereas in an Opt-Out system the consent is already presumed to be given by default and the consumer has to retract it if he wants the data collection to stop. To ensure that the consent given by consumers is closer to real consent, we need to ensure that the companies follow a strict Opt-in policy of Data Collection. Although General Data Protection regulation imposed a strict Opt-In system, the issue is still heavily debated. The pro Opt-out system group argues the following three major concerns – (1) It is that the consumers are likely to prefer the initial option set for them and it is highly unlikely that they’ll change the option to allow their data being collected no matter how convenient and clear it is made. It is based on the presumption of general preference of status quo by consumers. (2) It is stated that opt-in requirements can pose significant transaction costs to data collectors previously uninhibited by consent gates. An opt-in regime imposes more pronounced legal costs, technology process management, and business operations costs than does a typical opt-out regime. (3) It degrades the user experience by introducing unwanted disruptions like asking for consent prior to the main content being introduced which annoys the consumer and makes them prone to leave. Now all these points are not well established. The world we live in where our data can be used in numerous adverse ways, it becomes really important that the consumers are well informed about the data collection and they agree to it prior to it being in effect. The above-mentioned concerns can be well handled by introducing a much simpler privacy policy as mentioned earlier and it is much more likely consumers won’t mind a little disruption if they know that it is for their own safety and security. The current proposed method for Indian Data collection Regime is also Opt-In for the companies except when it comes to the government which is also a big concern.

Opt-in and AI. • This Opt-in system might be followed in the usual data collection by websites, but it is not even remotely followed when it comes to data collection by Artificial Intelligence. Data collection by Artificial Intelligence at a lot of points is without consent and even unflagged. And as stated earlier that how much more deeply our data can be used with the help of an AI it becomes even more crucial. The personalised AI powered medical advice portals, the online retail sites and the AI Traffic Cameras don’t warn us and neither ask for our consent before analysing and storing our data beyond the scope of what we provide them for. Proper Cautions Necessary. • There must be proper warning at all places wherever there is a chance that the consumer’s data might be collected and stored by an Artificial Intelligence enabled system and when it is collected for synthesis by an AI. Now it is even more important for us to learn when an AI enabled system is collecting our data due to its much larger ability of doing so. And at many places it does so without warnings like when google maps use location information of hundreds of users to warn about traffic to its other users, license plates being captured by

Artificial Intelligence and Policy in India, Volume 1(2020)

plate readers, driving speeds and car information captured by AI traffic cameras, Tesla selfdriving car systems storing every detail of our drive, our face data being stored by airport security face recognition system etc. Now gathering all this information is necessary for AI to provide better services to us but a caveat every time it collects data can be provided so that we can ensure that it is only collected to provide better services to us and not for any other interests. In a world filled with Artificial Intelligence all around us there need to be red flags everywhere where our data is being collected by them. • To be at the forefront of the AI revolution, is imperative to take proactive steps since the dangers and problems associated with AI are not yet completely known. Consequently, it is vital to have a two-layered protection model: one- technological regulators; and two- laws to control AI actions as well as for accountability of errors. Technological regulators include introducing a multi-layered security and defense software to protect data and in turn business. Layered security strategies are reactions to today's cyber threat landscape. Rather than simply waiting for attacks to hit endpoints, layered security takes a holistic view of cyber defense, accounting for the multitude of vectors by which modern malware is delivered. Apart from this, laws which protect the rights of consumers and ensure transparency and accountability along with these technological regulators will be an advantage to ensure privacy concerts pertaining to AI. Right to be Absolutely Informed. • There should be a clear declaration made to the consumers with regards to third parties using their personal data. The law should mandate it for companies to disclose the list of third-party organizations to consumers who will be accessing and using their data and also clearly state in the disclosure as to what part of the consumer data is provided to which organization for what purpose. Another important part of being absolutely informed should be the Right for individuals to receive the reasoning underlying any automated processing of their data, and the consequences of such reasoning for their rights and interests. Machines Learning Safely. • Machine Learning Algorithms are often trained to predict outcomes based on certain sets of data. These outcomes can be related to a mathematical issue or social issue etc. In order to achieve high, reliable levels of accuracy in prediction, these systems rely on large sets of data to learn from. These data set often involve personal and sensitive information of people. The issue of data privacy is amplified with machine learning. This data is forever stored with machine learning systems. These systems are even capable of re-identifying personalized data based on only minimal information. There is also a lack of transparency in how this personal data might be processed. In order to meet the needs of the future it is very crucial to train these machine learning systems but keeping privacy protection in mind is also necessary. So certain guidelines can be made mandating certain practices to be followed while training machine learning which protect the privacy of the people whose data is being used in the training process. These practices could involve – Federated Learning, Differential Privacy, Homomorphic encryption etc as per the recommendations of the experts in the field • Transparency and accountability also form an important part of the training process because in numerous cases there is a manifestation of unwanted bias which is reproduced via people building it or through biased data which is fed to train the AI and Machine Learning

Artificial Intelligence and Policy in India

algorithms. In such a scenario, transparency and accountability are two major pillars which will prevent the structure of AI from being corrupted and ensure it remains free of bias. Right to Erasure. • Right to Erasure enables any data principal to have his personal data deleted from all databases on his request. It is an essential part of European Union’s General Data Protection Regulations. This right ensures that a consent is valid only till it is an active consent. If once a consent was given for an information to be collected, it does not mean that the information shall be withheld for lifetime. This also provides a person an opportunity to get any false information about him, present on the internet, to be removed permanently. It is also famously known as “Right to be Forgotten” worldwide, which is different from the Right to be Forgotten envisaged in Indian Data Protection Law that only allows users to stop their data being disclosed any further on their request. Right to Erasure is a big step towards strengthening a person’s privacy. Although it is a qualified right, but still a valid request of Right to Erasure goes a long way in protecting a Person’s previously collected and retained from being exploited any further.

Can AI forget? • Right to Erasure becomes complicated when the data has been fed to Artificial Intelligence and Machine Learning systems. The first issue this raises is of how to reclaim the data and its influence on the resulting output. Generally speaking, AI cannot be taught how to “forget” something the way a human can. Technical experts do suggest a solution i.e. that if the key that allows AI to access a particular data is removed then the AI won’t be able to access that particular data and thus Right to Erasure would be met. Now coming to the second and much bigger concern i.e. the impact on the performance of AI and Machine Learning algorithms of the data being deleted. Although one person’s data being deleted wouldn’t have much effect but if thousands of people exercise their Right to Erasure then it could lead to a mess in AI’s performance. And so, it is recommended that the Right to Erasure be never made an absolute right so that if the data is crucial for the performance of AI, the right can be denied. Governing Privacy in Time of Crisis. • There is a need for certain legislative regulations in the data protection law that can govern the realm of privacy in crisis situations such as occurrence of war or outbreak of a pandemic. These regulations should mainly focus on regulating the actions of government during the crises. It is often seen that in the time of crises the government takes an undue advantage and under the garb of controlling the situation, it breaches the privacy of its citizens to a far extent. For example currently in India, the name of controlling the corona virus outbreak the government has rigorously increased surveillance over the general population. They are using drones with facial recognition systems in the background in order to track and trace citizens. The Aarogya Setu app introduced by the government and now being imposed on citizens through various ways has a flawed privacy policy. This app collects highly sensitive data related to health and whereabouts of the citizen and there is no clear mention of the purpose of usage in the privacy policy and also there are no clauses in the privacy policy

Artificial Intelligence and Policy in India, Volume 1(2020)

which limit the retention of data until a certain period of time. It is obvious that the government needs to be provided a bit of loose hand in such situations as health and life always trump over privacy but that loose hand should have some limits. Thus, the extended lineage provided to the government as to violation of and restrictions on the government to ensure the protection of privacy of the citizens in the time of a crisis should be clearly defined beforehand in the data protection legislation of the country. • Awareness is one of the most important tools to combat AI considerations and privacy aspects. India’s low literacy levels acts as a major barrier to the common people understanding privacy laws and resolving complexities, conducting, and inculcating public consultation goes a long way. If people are aware of the rights that vested in them, data can be circulated and regulated in a controlled manner where the individual is provided with ‘real’ options to regulate data in the manner they feel right, along the lines of GDPR. 6.3

Conclusions

• AI is ubiquitous in our everyday lives. It is the new ‘normal’ and is making a difference in numerous fields. It is defining our lives in new innovative ways every day. Its impact on society cannot be negated in any manner whatsoever. There exist numerous AI's that aid essential areas such as law enforcement to provide novel solutions to omnipresent problems, gradually annihilating the gap between already diminishing human and AI functions, amongst performing other functions for the overall betterment of society. However, with increase in involvement of AI in our lives, increases the exposure of our information to this world which leads to increase in threat to our privacy. The matter pertaining to privacy is a subject of great concern which needs to be addressed appropriately by way of a robust legislative framework covering all aspects mentioned above.

AI & Constitutional Law: AI, Criminology & Judicial Governance

• The creation of a project team of researchers for the sole purpose of a project of putting AI into the rounds in terms of the Justice system. In collaborations with technology giants, & law institutes. This team can comprise of three divisions: ─ Fillers – Engineers assigned to design, create & improve the “Reservoir AI”, with help of lawyers to decide which precedents/news/amendments/statutes to specifically use in order to avoid any chance of biasing of the AI. ─ Profilers – Criminologists, lawyers, psychologists & engineers dedicated to the creation a “Profiling AI”. Where, the psychologists shall be constantly evaluating the members of the profilers to identify any biases that might end up getting programmed into the system. The criminologists & lawyers shall be constantly researching on newer methods, developments in the field of social profiling. The engineers shall be responsible for creation & improvements of the said system. ─ Jury – Lawyers & criminologists & engineers whose objectives shall be the creation & required upgrading & updating of the “Advisory AI”. While, the lawyers shall look into the ethics part of the creation, the criminologists with the help of the engineers shall continue looking for biases in the system. • The development of a “Reservoir AI” to be used in the judicial system. This AI shall be constantly fed with precedents, news, amendments, and statutes. The only object of this AI

Artificial Intelligence and Policy in India

•

• •

•

should be to learn everything there is to learn about law, & to furnish the judge with whatever case law/law/provision/incident/observation that is relevant to the case at hand. The development of a “Profiling AI” to be used for criminal profiling. This AI shall be fed with principles of criminology & should be able to understand the concept; this AI should be able to compile a criminal profile of an individual based on user defined input. Such as name of offender, offense, age, sex, place of birth, vehicle type if any, medical records, previous criminal affiliations & allegations, etc. This AI shall then formulate a criminal profile of this individual based on the its understanding of the principles of criminal profiling, & facts surrounding the crime as well as evidence found at crime scene. This shall be further supplemented with observations by forensic science experts. The objective of this AI would be to use all the observations that have been made surrounding the crime to establish a criminal profile. This AI shall also access the geographical location of the accused & the surroundings of the accused to detect if the environment surrounding the accused is deviant, or delinquent. The development of the “Advisory AI” that shall take the deductions made by the profiling AI into mind, & shall further take the information given by the “Reservoir AI” into mind & shall formulate an advisory opinion of the case. I.e. The AI shall give a decision by itself, as an advisory opinion to the concerned judge. The formation of a review board to look into any appeals made against the decisions of a judge based on the Judge passing any sentence or decree contrary to the advice of the AI. Whereby the judge shall be liable to submit a report as to his actions. If unsatisfactory, an investigatory committee would be asked to launch an investigation against the conduct of the concerned judge. All of this shall be done in confidentiality, while the person appealing shall be awarded a protection program, so that the person shall be free of threats. Companies can be contracted to provide AI support to the law enforcement departments, which shall be of much help to the AIs in the judiciary. So, those certain facts can be understood & gotten by the AI firsthand. Such as, digital forensics, ballistic reports, gunshot detection reports, objects recognized, faces recognized, DNA analyses reports, etc. This shall further help the “Profiling AI” to compile the profile. Spreading of awareness about the field of AI, Law & criminology, so that more researchers come into this field, encouraging research institutions, premier law schools & societies to indulge in this field so that more can be achieved. The application of Artificial Intelligence to the justice system is still in its nascent stage in India. This was made evident when as recently as January 2020, Chief Justice of India S.A. Bobde spoke of the possibility of developing an Artificial Intelligence system for courts in order to address the issue of pendency of cases and mitigate delays in the delivery of justice. Several successful utilizations – such as that of the AI enabled video analytics platform ‘JARVIS’ (Joint AI Research for Video Instances and Streams) in Uttar Pradesh prisons and the AI system ‘PAIS’ (Police Artificial Intelligence System) which provides a database having records of prison inmates across the state of Punjab – highlight the potential of Artificial Intelligence for use in the criminal justice system. However, the objective in all such cases has been primarily to monitor inmates and their possible violent acts, criminal operations or prison breaches, or facially identify criminals. Artificial Intelligence systems can also play a significant role in ensuring the enforcement of the rights available to defendants, prisoners and undertrials, by way of automating routine tasks as well as supplementing human discretion with their wide range of computational skills, applications which have been largely unexplored in our country. Only the state of

•

• •

•

Artificial Intelligence and Policy in India, Volume 1(2020)

Odisha has invited bids for suitable AI applications to detect procedural mistakes made by prison authorities, and as late as December 2019. A nationwide following of suit would be a monumental step towards ensuring the prevention of violation of prisoners’ rights and the provision of an adequate rehabilitative environment. Defendants are often subject to bias during sentencing, a phenomenon which most notably affects individuals from disadvantaged communities. The requirement is the achievement of objectivity and neutrality in decision making, a feat which is sometimes difficult for humans to achieve, but can be sought to be tackled by Artificial Intelligence. AI algorithms should be employed to perform crime forecasting and risk and recidivism assessments on alleged criminals, which ascertains their likelihood of committing a crime in the future. This would enable judges to rely on the analytical and computational skills of AI as well as their own rationale and intuition, which is intrinsic to human nature. Adopting the same as a statutory responsibility should be considered. Arguably, such algorithms are also tainted with human bias, and are simply the mathematical articulations of such inherent bias. This is due to the fact that they rely on historical crime data and will therefore reinforce disparities and perpetuate the biases. However, the human aspect is an inalienable component of the criminal justice system and the use of predictive algorithms may be scrutinized, and the bias thus accounted for, in a way that the discretionary powers of a judge cannot, making the process more just and fair. Algorithms that don’t consider factors such as gender and locality of residence should be used to eliminate bias related to these aspects. They should be sourced from developers who ensure that the creation process is inclusive of individuals from different communities, genders and localities, and even those who have themselves had encounters with the criminal justice system. Their feedback would guarantee that diverse perspectives are considered while designing the algorithm, and that variables utilized as inputs are not related to factors such as the socioeconomic status of the defendant. Equality assessments are required to be formalized and performed before such technology is made available in the public sector. Incorporating the algorithm computed result as a baseline into the various factors used to determine the judgement can mitigate the unconscious pervasion of other human aspects as well, such as the effect of fatigue. A committee of experts should be constituted to establish a standard procedure for monitoring and scrutinizing the outcomes of algorithm use, as well as the data it is trained on, and to ensure that the process is scientifically tested. Moreover, this body should be responsible for making sure that the initial procurement of an algorithm is met with a series of tests to guarantee that its process is subject to the control required to grant the appropriate consideration to the rights of the accused. Often there arises opacity and unaccountability regarding the predictive algorithms, as their owners do not elaborate on the working of said algorithms or make the code publicly available, since it is proprietary information. To prevent this black box, AI software should be sourced only after satisfactory compliance is achieved by third party vendors to certain strict guidelines, namely, those relating to the validity and source of the input data. An attempt should be made to build public understanding of the method of computation of the scores. Although complete transparency of procedure is difficult to achieve, this can ensure the opportunity to ascertain the reliability of the code. Companies independent of the prison sector should play the key role in developing these systems. As mentioned by the CJI, it is imperative that the pressing issue of pendency of cases is dealt with. Courts and judges are ill equipped to deal with the sheer quantum of cases assigned

Artificial Intelligence and Policy in India

•

• •

to them. A closely related concern is that of the undertrials in prisons, who perhaps were not granted bail or could not afford it, and whose cases have not been heard. According to the prison statistics of the National Crimes Record Bureau released in 2019, it was found that more than 67 per cent of the prison inmates in India are undertrials who are yet to be convicted. Thus, even before their crime is proved, such individuals are made to repent for it, by the various impacts that imprisonment could have on their lives. Needless to say, the consequences would be the greatest for the economically weaker sections. Considering that it is unjust for undertrials to be made to remain in prison for extended periods due to the overburdened judicial system, certain advanced measures are required to be taken to ensure that cases are heard within a reasonable period of time. Artificial Intelligence should play a role in hastening human deliberations. AI systems are capable of processing large volumes of information, related to the facts of a case, laws, precedents, and so on, much more efficiently than their human counterparts. Having an AI computed report to form the blueprint for a judgement would expedite the process, and thus enable judges to hear a greater quantum of cases. Image and video analysis for cases tend to require numerous personnel who possess knowledge of the relevant subject matter. Due to the large amount of information involved, this process consequently poses the risk of human error. The use of AI systems for this purpose addresses both these concerns. The aforementioned risk analysis algorithms should also be used to determine the individuals to whom bail should be granted. If it can determine that an offender is not likely to engage in activities of a criminal nature if not kept in detention before their trial, due to a low risk score, this should increase the possibility of bail being granted to him or her. This would ensure a reduction in the population of prisons, without causing adverse effects to public safety. This has been found to be increasingly necessary as the nationwide occupancy rate of jails at the end of 2017 was found by the National Crimes Record Bureau to be a shocking 115.1 per cent, as 4.33 lakh inmates were occupying a meagre 1,400 jails. Another issue rampant in Indian prisons is that of police brutality. The treatment meted out to prisoners is often reflective of a procedure that is less than just and reasonable, and there is an absence of a system to closely monitor this police procedure. An AI enabled platform should be put in place to serve this purpose. As previously mentioned, Uttar Pradesh has employed the use of the Staqu developed AI enabled video analytics platform ‘JARVIS’, in as many as 70 prisons throughout the state. This platform analyses the activities in these prisons through the feeds provided by surveillance cameras installed therein. It has so far been trained to flag only violent acts and prison breaches of inmates. The platform should also be used to flag acts of police brutality and violence to shed light on such excesses by the authorities, that are usually unreported and hence overlooked, and should subsequently be employed in all prisons across the country. Furthermore, the risk assessment tests performed on the defendant should also be used for prison authorities, to predict and determine which of them are likely to breach procedure. AI can also assist defendants in proving their innocence from a scientific perspective. The use of AI technology to aid in complicated cases of forensic DNA tests or mixture interpretation, in cases involving numerous perpetrators, should be explored in the criminal justice system in India. Human analysis may sometimes fall short in the examination of complex evidence and data, if used alone, and the employment of AI systems to supplement it would provide greater clarity for the purpose of the trial.

Artificial Intelligence and Policy in India, Volume 1(2020)

• At this juncture, it may not be pertinent to devise a framework for the application of “mindreading” AI technology to the criminal justice system, due to the plethora of allied ethical concerns. However, such a system holds immense potential to improve risk analysis by making predictions of criminal behaviour, as well as to allow wrongly accused individuals an opportunity to prove their innocence. • Artificial Intelligence has evolved in ways that, a few years ago, could not be imagined, and has had a ground-breaking influence as a technology on all aspects of human life. It is essential for Artificial Intelligence to become increasingly integrated into the legal system, specifically in the criminal justice system, as it would prove to be of great assistance in ensuring that the rights of defendants, prisoners and undertrials are enforced, and thus revolutionize the functioning of the system. In furtherance of this aim, a comprehensive legal framework is required to be formulated and implemented, to ensure regulation and oversight of the shortcomings inherent in Artificial Intelligence algorithms and to make certain that they are rendered appropriate.

AI & Constitutional Law: Machine Learning and Privacy Mechanisms in India

• Democratic establishments throughout the world have incorporated privacy, including informational and data privacy, as an inviolable facet of the right to a dignified life. Thus, to safeguard the said right, it is imperative to establish a legal framework to protect personal data throughout all industries. With the advent of advance technologies, such as Machine Learning, the necessity to protect personal data has increased significantly. In furtherance of the same objective, the paper analyses involvement of personal data at various stages of Machine Learning and recommends protective measures. Although central legislation is required to govern data privacy in general, it is also imperative that specific industries implement data privacy policy measures targeting the protection of data generated and collected by the industry. Therefore, a policy framework consisting of central legislation and industryspecific policy is recommended. • The following policy measures are recommended in the central legislation: ─ Granting absolute data ownership to data principals for all categories of personal data. ─ Enforcing liability over Machine Learning developers by establishing data principal-data fiduciary relationship. • The following policy measures are recommended in industry-specific policy measures: ─ Enforcing compulsory data anonymization in datasets. ─ Mandatory usage of specific learning methods for certain industries. ─ Regulate data purchase from private entities. ─ Sectoral data localisation obligations. Data Ownership. • Individuals to be granted property rights overall categories of their personal data. Majority of the industries do not provide property rights over personal data to individuals. However, certain industries, such as the healthcare industry, provide partial ownership. The data in medical records is owned by the patient but the medium of transmission and storage is owned by the healthcare provider. In such situations, although the data is owned by the

Artificial Intelligence and Policy in India

data principal, he/she does not have any control over the medium of storage and transmission. At this juncture, regardless of central legislation that grants property rights over personal data, the intervention of industry-specific policy measure to ensure uniform rules for collection, transmission, dissemination and storage is necessary. Granting property rights over personal data to individuals would ensure a strict consent-based usage of such data as individuals will exercise absolute control over the generation, transmission and secondary usage of their personal data. Compulsory Anonymization of Personal Data. • Anonymization methods such as K-anonymity, L-diversity and T-closeness are commonly used to mask or remove personally identifiable information and sensitive column data from ML training datasets. These anonymization techniques are currently voluntary in practice and India does not have any policy measure to enforce such safeguard measures. Industries such as Agriculture do not generate personal data and thus require minimal data protection. However, Financial and Healthcare industry generates sensitive and critical personal data, and thus require stringent anonymization measures. The Government of India also offers certain open-access datasets containing data from various Ministries to enable innovation and development in India. Therefore, even if explicit consent is obtained from the data principal, it is essential for the aforementioned industry-specific data governance policy to enforce such anonymization techniques before personal data is used in ML training datasets. Mandatory usage of specific learning methods for certain industries. • The Draft Personal Data Protection Bill, 2019 has categorized personal data and has proposed general regulatory measures for the specific category. The proposed regulatory provisions are indeed promising but will fail to achieve its objectives if data is used for secondary purposes, such as Machine Learning training datasets. Therefore, specific learning methods have to be introduced for certain industries. Machine Learning algorithms’ training on clinical data for Healthcare development should be mandatorily performed using ‘Federated Learning’ likewise training on financial data for Investment and Banking development should be mandatorily performed using ‘Shared Machine Learning’. The aforementioned learning technologies, although voluntary, have been already implemented in certain overseas organisations. The industry-specific data regulatory policies for the Healthcare Industry and Investment and Banking industry should impose the mandatory usage of such training methods. Regulate Dataset Purchase from Private Sector. • The Government of India offers certain open-access datasets containing data from various Ministries to enable innovation and development in India. However, it does not consist of robust and comprehensive datasets across various sectors and fields, thereby creating a shortage of quality intelligent data. The quality of the dataset affects the performance accuracy of the Machine Learning program. Considering the lack of any legislative policy for data protection in India, majority of the data is owned and controlled by the private sector, and thus Machine Learning developers turn to private sector industries for datasets. However,

Artificial Intelligence and Policy in India, Volume 1(2020)

due to the high costs of such private-sector datasets, small-scale innovators and startups face a serious obstacle. Therefore, a regulatory framework has to be established that regulates the pricing and quality of the datasets to ensure a fairground for innovation in India for both large-scale innovators and startups. Sectoral Data Localization Obligations. • Considering the multitude of opportunities emerging from Machine Learning technology, the correct rationale for a developing country like India would be to analyze data localization measures. Industry-specific data localization measures for industries that generate sensitive or crucial personal data would ensure local storage of such data, increased accessibility to government institutions and growth in innovation. Local storage of data would provide the government with better access and control over citizen’s personal data. The rate of accessibility of data for quality machine learning datasets will likewise improve. Industry-specific data localization obligations are more favorable than absolute data localization measures overall industries as an absolute measure would cause a negative economic impact.

Artificial Intelligence and Intellectual Property: Collateral Parallelism

• AI is the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings. Computer systems began to mimic our perception and cognitive ability, our ability to create; these abilities are those for whose guidance and protection the laws and the law makers were primarily made. AI machines have several such qualities that make them stand equal to human beings. • AI can gather data and feedback and process those to improve their ability and create new and magnificent things that human can or even cannot without almost any human interference. These machines have the ability to learn and to take decisions based on that. Now the question which comes into light is when an AI system creates something on its own without any human interference will that work be made a patent? Will that system get the credit for that innovation or the owner of that system will be granted patenting right? Or the question which actually is getting discussed is whether such work should be granted patenting rights, like granted in the case of human being? • Current patent laws treat AI software inventions essentially as logical algorithms implemented on the computer. On inventorship, patent law states that someone (usually a natural person) who merely applies the logic to make something workable cannot be an inventor. So far machines were ‘that someone’, hence they were not a possible inventor under the law • According to WIPO, Intellectual Property is – to the unique, value adding creations of the Human intellect that results from human ingenuity, creativity and inventiveness. And what IP Laws do is to confer property like rights on these inventions or creativity. Simply speaking by assigning property rights on the product of our intellect laws give us control or exclusivity over that particular property or product. IPR can be said to be working as the root of the innovations. • The office of technological assessment in 1986 said that artificial intelligence should be considered legitimate co-authors. Critics has said that machine follow a routine as set by humans thus it cannot be considered as something worthy of patenting. It was suggested

Artificial Intelligence and Policy in India

•

that copyright production should not be granted unless granted to the creator, i.e., human beings on behalf. In the matter of patentability, the three important criteria to be kept in mind are novelty, inventive step and industrial use. But in the case of AI it is considered that novelty is not present, as some prior data has to be entered in the device or machine to make it work accordingly. A machine cannot develop the data in its own, thus it cannot be totally different of the prior art. Many inventions produced by AI are generally driven by Deep Neural Networks (DNN) and are heuristic in their behaviour. In such cases, we can focus on the end-result obtained from the process and not on the process itself. If the endresult meets the criteria set forth as ‘sufficient to imbue a human or natural person’ with an inventor status, then consequently the machine (or AI system) could also be imbued with the same status. Currently we have AI created music and art work. Example of such is e-David who is a robot and has done commendable work in the art field. It creates the portraits which never primarily existed by analyzing and observing the features like we human do. University of London Press Vs. University Tutorial Press said “the word original does not in this connection mean that the work must be expression of original or inventive thought. But that the work must not be copied from another work but should originate from the author.” In US to qualify as a work of authorship a work must be created by a human being. In the case of Naruto Vs. Slater where Naruto a monkey took some photographs, the US Court said that being an animal, i.e. non-human, he does not have standing in the court and cannot sue for copyright infringement. In 2010 US Supreme Court denied patenting to the programs because what they perform is mechanical rather than inventive. Similarly in Nigeria AI systems are said to lack legal personality and cannot be authors. Now in India there are no guidelines for AI related inventions but computer related work has been discussed and appreciated by making laws for the same timely. UK has expanded the scope of copyright protected work to expressly include the computergenerated work. The author of such computer-generated work according to section 178 of UK Copyrights, Designs and Patent Act, is deemed to be the person by whom necessary arrangements for the creation of the work are undertaken. Somewhat same idea has been perceived in Europe where European Patent Office has laid the guidelines where list of exclusions has been made. One of the ingredients of that list is Mathematical methods and programmes by computers. EPO has also made guidelines specifically for AI related inventions. It clearly states that when an AI classification method serves a technical purpose, the steps to generate the training set and train the classifier may also contribute to the technical character of the invention, if its support achieves the technical purpose Existence of AI should be recognized and such organisations should be made to deal solely with the recognition of AI and the laws to be made for them. Detailed and logical guidelines for patenting and copyright of the products of AI should be made at international level and implemented uniformly. We must acknowledge that AI systems demand a reconsideration of the extant IP laws. Intelligent IP should managed be in various ways like Data Privacy where high quality and accurate data can be accessed, by enabling the IP systems and tools with AI-based solutions, by empowering people to realize the benefit of AI in the IP domain an etc. IP management can greatly benefit in using AI during patent search and prosecution phase. Similarly, there could be more AI benefits that are not yet realized in IP. But what required for that to achieve is that as there is no sustainability currently in the field several steps should be considered for that. There should be uniformity across jurisdictions, the

Artificial Intelligence and Policy in India, Volume 1(2020)

system where a work is protected by laws in UK and is not in Nigeria is vague and should be made more precise. A multi-trading platform needs to agree on a position. IP rights for AI invention and creativity should be considered side by side criminal and civil liability for such rights. We should also consider that giving copyrights or authorship of an AI generated work to a human will lead to those inventions getting exploited by those who haven’t even created them. As the upcoming lawyers it becomes our duty to adapt such great changes and change the law according to the needs of this generation. The speciality of the lawyers is that we acknowledge past and consider future while deciding the present. While the scope of IP management automation and using AI tools is colossal, it is just a matter of time when IP management will become fully automated and self-driven.

AI & Intellectual Property Law: Economic Viability & Ownership Pertinent to Patent Law

• The working paper by NITI Aayog is appreciative of the efforts of China and emphasizes on pulling India to the International competency but it understood that it could not achieve levels of AI proscribed by China. In comparison to international standard, it is noted that China’s AI policy is similar to that of OECD AI’s principle- setting a well-known international standard of ethic on AI that it has to keep to make itself saleable to the global companies along with protecting the future jobs of humankind which is a risk that India might be overlooking. The point that is missing in NITI Aayog’s paper is that it advises the government to collaborate with different stakeholders with a different set of data whereas; this step lacks the transparency that the OECD principles expect. • Inclusion of Defence Sector Impact of AI: The Defence sector in India is to be technologically improved by the way of inclusion of AI. The Defence of the country is usually under the attack during many military operations and that leads to a loss of human life and manpower which is required in the need of the disaster and other key military keepings. This loss of live can be reduced by the way of the inclusion of the AI and development of bots that can replace the man-power in the warfare. It can also be used likewise to reach out to the sensitive warzones and provide a relief to the victims of the strikes. The pre and post war consolidations involve the usage of the warfare machinery and it has lead to develop a more nuanced structure to detect the potential targets and survey the affected areas of warzones • Further it is suggested that the AI can be a basis of the change in the process of the defence procurement. The defence procurement is one of the tasks that require a lot of effort of analyzing and crunching numbers to analyze the right option. It is an important task and has to be conducted with the accuracy and efficiency. The Soft AI can be a viable solution to make the analysis of the prototypes of ammunitions accurate and quick. Since the government has made a decision of privatizing the areas of the defence the tender bidding process and thereafter the processing of application can be conducted with the help of AI. • On the Government Sphere, the government laid out a Defence IPR policy for Developed software products in India by Ministry of Defence has set a new challenge in the AI start-up in India in the field of Defence. With a great amount of paper work to setup and a huge investment in the terms of procuring the rights from each circuit board to every software that potential will be acquired for its services in the field. The Categories of Owners of those

Artificial Intelligence and Policy in India

software can vary from the ‘deliverables’ meaning the developers who develop the software, to the owners of the hefty computer programs that make the government pay the pockets on the mere purchasing of these rights. • The Solution that is potent, is the joint ownership of the IPR along with many software developers. Further solution that seems feasible is to expand the make in India campaign of government to setup places of startup that work with loyalty to develop the sustainable technology by allocating the funds needed to develop an indigenous software and technology in India. The Government can also employee the US patent law strategy which flexible for both government and the start-ups. The MoD has to make sure that the there is a development under the guidance of the National Defence Research Labs so that the future of the AI can find its find its feet in India without much legal impediments. • Inclusion of the Judiciary as a Stake Holders of the Artificial Intelligence: The lacuna of the paper that the government did not account in the for Judiciary System and their travesty with the AI systems that are currently in place. The actions of the government indicated in the NITI Aayog report is based on the ground work approach. It wants to imply the use of the AI either in a form that revolutionaries the system of an industry or either it is worked upon at the younger population that is at academic places. However, the same is neglecting of the issue of the older demographics as the stakeholders. India has almost 19 lakh judges per 10 lakh people and with such increasing responsibility over the judiciary it is imperative that the ISA policy actions should have been enabled to aid in easing the burden of the judges. • A suggestive solution to the same is the inclusion of the processing of the errors that come across the filings that happen in an actual court. The process of filing involves the checking of various formalities in order to admit a plea in the court. The irregularities in these process are looked upon by the humans deployed for such jobs. Furthermore, there are typist who are accountable to write down the judgements with great precision and accuracy and sometimes they too are prone to the commit a fault in such practices. With the introduction of the deep machine and speech-to-text and text-to-speech it will be convenient for the judges for keeping the data secured and confidential as well as complete their pendency of cases with quick and accurate help.

AI & Intellectual Property Law: Ethical Stimulus of AI as a Legal Personality

• Technology has swiftly taken over almost every aspect around us and across the globe in the past few years. There was a time, not that long ago, when mobile phones were considered to be a luxury item and internet was a rarity, but with the advent of advancements in technology, the entire world wide web can be accessed with our fingertips. And now, one of the most contentious issues that has come into light is that of artificial intelligence. • There are many nuances to the proper definition of “Artificial Intelligence” across the board. According to Wikipedia, the all-knowing guide of the common folks, “artificial intelligence (AI) is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and animals”. The existing juxtaposition of the machine with the human mind is clearly evident here.

Artificial Intelligence and Policy in India, Volume 1(2020)

• As we are exploring the newer terrain of artificial intelligence, it becomes profoundly important for us to look into the legal and ethical facets in relation to it. Technology is changing at a rapid speed and it brings about changes in its surroundings as well, so naturally, the next step would be to ensure that the laws pertaining to the aforementioned are revamped too. • One area of law to be considered in consonance with artificial intelligence is that of intellectual property (IP). World Intellectual Property Organization (WIPO) defines IP as “creations of the mind, such as inventions; literary and artistic works; designs; and symbols, names and images used in commerce”. The prime focus is on “striking the right balance between the interests of innovators and the wider public interest, in order to foster an environment in which creativity and innovation can flourish”. • Alan Turing, an English mathematician and computer scientist, developed the “Turing Test” to define artificial intelligence as a way of making computers capable of thinking like a human being. Another school of thought in relation to defining AI is about pursuing four goals: ─ systems that think like humans, ─ systems that think rationally, ─ systems that act like humans, and ─ systems that act rationally. • As we delve deeper into the elements of AI, we see that the approach of “modelling humans” becoming more discernible. Intellectual Property law is all about protecting the creator’s innovations and ideas. So, the question here - the “elephant in the room”, so to speak - is if it matters whether that creator is a human being or someone pretending to function and think like a human being. • Intellectual Property Rights encourage development of novel, unique, and useful ideas and works. Presently, there is no international law that provides the protection of IPR to byproducts and creations that exist courtesy of artificial intelligence. One of the main contentions against the case for AI is the authenticity of what it creates and if it can be called inherently “new”. • Generally, humans provide data to AI, which it takes forward, and with the help of perfectly curated algorithms brings about its final result. This ends up questioning the validity of AI’s supposedly “original” work. Ada Lovelace had reservations about the same and pointed out that, “The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform.” • On the other end, the evolution in technology and AI has provided us with numerous examples where AI has stood its ground and originated ideas that are seemingly novel. Considering the fact that we have, indubitably, come very far and have achieved countless milestones in the field of technology since Lady Lovelace’s time, the above statement becomes suspicious. • Robot Rights is an intriguing and growing concept, where artificially intelligent entities are looking for rights, responsibilities and moral obligations similar to that of a human being. If the robots are granted human rights in the near future and proper legislation is provided for the same, they will also be bound to fall in the ambit of intellectual property law and granting them IP rights will become a reality. • Ethically and morally this does raise the question whether AI actually deserves these rights or not. If we take a look at the global level, there is an abundance of actual human beings who are fighting for basic human rights, so providing those human rights to artificial beings

Artificial Intelligence and Policy in India

•

shows the indifference for suffering of human lives and the social implications arising out of it are presumably outrageous. The notion behind protecting the rights of AI is sincere and genuine, but considering the discourse surrounding the proposition of AI rights v/s human rights, we feel there is still a long road ahead. There are many legal and ethical barriers to cross before reaching the finish line. Keeping this in mind, we cannot forget that our society is dynamic and constantly changing at a very fast pace, thus, law can never remain static. There is a need for action to be taken for the protection of AI generated works because if the work is innovative and helpful to the public, it should not go to waste and it must bring about something of value. It is recommended that the concerns surrounding IP rights for AI can be sought out by providing a regulatory framework separate from the current IP regime. Artificially intelligent entities differ from human beings in a multitude of ways, so instead of confining them to human rights, a separate legislation governing their own rights is the need of the hour. The human factor from AI cannot be separated because humans work to create AI and then humans work with AI, so there need to be measures ensuring the protection of developments that arise as a result of this collaboration. There must be steps to ensure that AI keeps working in a transparent and accountable manner, without any algorithm bias to keep it from misleading. We have to realise that machines cannot do everything alone, but in recent times, neither can humans.

11.1

Conclusion

• The worries surrounding AI must be done away with by formulating proper regulations concerning the rights and responsibilities of AI, which should not necessarily be similar to human rights but, at least, be compatible with them. The goal for working for the public good unites humans and AI and there needs to an understanding of trust between the two. There are many ethical and moral complications that must be scrutinized thoroughly before coming up with a legislation on paper. AI will keep on expanding economically in the coming years, so it must receive credit where its due. An international regulatory framework which overlooks issues pertaining to AI will help to sought out the above-mentioned concerns.

Artificial Intelligence and Policy in India, Volume 1(2020)

Artificial Intelligence and Constitutionalism: Satiability and Extensivity of AI as a Legal Personality Sarmad Ahmad1 and Baldeep Singh Gill2 Chief Managing Editor, The Indian Learning Chief Experience Officer, Indian Society of Artificial Intelligence and Law baldeep@isail.in 1

Introduction

Humankind has evolved on its own distinct path in comparison to other species; a path that exercises intellectual effort, resulting in the creation, manifestation and adoption of novelties that simplify, amplify or elevate day-to-day existence. A retrospective analysis of various centuries of human existence emphasizes this observation; by how the adoption of many technologies have eased life, and now are necessary for the sustenance of it.The 21st century, while saturated with the development and incorporation of various technologies, offers the potential of one novel technology that has exhibited signs of revolutionizing everything else; AI. The interest and curiosity around AI has reincarnated back from its early days of research and development in the 1950’s, and is now receiving extensive attention in regards of research, development, investments and economic incorporation in next decade. What distinguishes AI from other technologies, is that it offers the creation, reproduction and manufacturing of intelligence; the ability to learn, understand, and apply that understanding towards achieving a desired outcome.This has led to the creation of Autonomous Decision Making Systems (ADMS), technological systems or ‘Robots’ for the lack of a better word, that can take decisions in situations/tasksthat they are adept at with very minimal or no human involvement.Examples of many ADMS are prominent in various fields; some already exhibiting better performance than human beings at the task at hand. While current examples of ADMS showcase Artificial “Narrow” Intelligence (ANI), the capability and research into Artificial “General” Intelligence (AGI) is already underway; i.e, ADMS that can showcase and exhibit intelligence in more than one field, on par with human capability. Analogically, how the industrial revolution harnessed the capabilities of physical power, the use of AI introduces the capability to harness cognitive power. Directly, attempts towards the regulation of AI requires an understanding of Machine Learning (ML). ML is one of the subsets of the science of AI that allows machines to essentially Meta-learn; learning how to learn. Its advent poses the following situation; an instance where minimal input is provided to a machine by a human, which then continues to learn, adapt and improve on its own (SAS Institute Inc.).It is through understanding ML, the capability of a machine to “Meta-learn”, wherein a vacuum is observed with regards to the responsibility and accountability of any action executed independently by an AI. If an AI teaches itself by rewriting its own code, and executes a new action which results in an error inflicting damage to life or property, or even poses as a threat, who eventually is responsible? As the research and development towards AI grows and flourishes, making AI more autonomous and independent to the extent where we lean towards the creation, use and application of AGI in various fields,

Artificial Intelligence and Policy in India

holding the creators liable for the actions of autonomous AI will eventually fall short if not done effectively. The prospect of AI does not only bring the certainty of growth, development and progress, but it also carries with itself ambiguity of its nature of being. Increasing incorporation of ADMS/AI into our daily lives aside from extensive research and development highlights that a fully capable, autonomous and independent AGI is not too far away from sight. The manifestation of such a technology will eventually exhibita need for a system of accountability and responsibility. Hence, keeping in consideration the lack of such a governance over AI, the question of granting them the status of a legal entity has to be entertained, and the granting of this status could dramatically remodel the legal system as we know it today. 1.1

Objectives and Initial Theses

The objective of this research is to analyze the possibility of granting AI a Legal Personality, chart the extent of this personality in relation with the legal system around it, attempt to evaluate the implications of this possibility upon the law and assess the constitutionality of this probability with exclusive reference to the Constitution of India, 1950, while providing recommendations for the same.

Tracing the capabilities and capacities of AI

2.1

AI Classifications on basis of its Strength

The notion of an omniscient artificial intelligence that is malicious in its functioning is often either idealized or feared, and experienced through the mediums of Hollywood movies or science-fiction literature. Present day reality however, couldn’t be further from such a perspective, for what is often under-represented about the actuality of Artificial Intelligence is that its research and development is far from such a reality (Krishna). Arguably, what drives the research and development towards AI is the potential it promises, rather than what it delivers today. This has allowed speculation to flourish, and in turn has provided for a blueprint that shines light on what we are to expect in the near future. AI can be broadly categorized into three widely acknowledged categories; Artificial Narrow Intelligence (ANI/ Weak AI), Artificial General Intelligence (AGI/ Strong AI), and Artificial Super Intelligence (ASI) (Fourtané). ANI is a narrow form of AI, and adequately titled as such because it is only trained and capable of performing one task exclusively. It has to be noted that in its execution of its one exclusive task, it exhibits better performance than the average human. ANI is currently the only existent form of AI, and executes a variety of functions in the undercurrent of the post-modern world. Examples of ANI algorithms include facial recognition AI used in surveillance and social media picture-tagging, natural language processing (NLP) used in assistant applications like Amazon’s Alexa, Google’s voice assistant and Apple’s Siri, or depth perception algorithms utilized in selfdriving cars that measure and perceive the distance between the vehicle and other vehicles in its environment. While it is known that AI is a software that is capable of executing tasks that would usually require human intelligence, the objective behind developing ANI is to create algorithms that specifically handle abilities and skills that are either natural and instinctive, or learnt and mastered through repetitive practice by a human being. Abilities that are learnt and mastered through repetitive practice vary and differ between human beings across all walks of life and

Artificial Intelligence and Policy in India, Volume 1(2020)

age groups, and can be distinguished with cultural, religious, traditional, social, economic and political boundaries. E.g. A human being may compose music, manifest a form of art, organize research information, cook a specific dish, perform a surgery, and may draft theories of law, politics or economics; through repetitive and deliberate practice. While the intricacies of these abilities may differ due to the abovementioned distinctions, there still exists the undeniable set of natural, instinctive abilities possessed by humankind as a whole; all humans can recognize faces, perceive depth in their surrounding spaces, empathize, and process and produce their regional languages, or an additional few. It is in attempt to recreate such capabilities that has led to research and development of ANI algorithms that can perform singular, exclusive tasks through its trained capacity. AGI, which exists today only as a hypothetical concept, is essentially AI that can engage itself in tasks a human could execute generally, either on par with a human being or better than one. While an ANI algorithm can execute one task better and more efficiently than the combined efforts of a hundred human beings, it ceases to be a performer in the first place in comparison to the wide variety of tasks the five-year old human child can engage themselves into. While the concept may seem far-fetched, understanding how AGI functioning could differ from ANI functioning is necessary. While an ANI requires massive volumes of training data to be wellversed with the task at hand and execute it, an AGI could be a relatively fast learner in comparison as it could use its existing processed data to learn tasks than overlap with one another, eventually creating a chain of associative-based learning that is natural to human beings (Joshi). This could be illustrated with the following example; A NLP algorithm (exclusively ANI) will have to take in massive volumes of data of the English language including its associated rules of grammar, punctuation, enunciation, semantics and syntax, in order to effectively process English. However, the same NLP algorithm which could be one of the many algorithms in an AGI, could utilize associative learning to identify the overlap of linguistic rules in English, French and other similar languages, and accordingly learn those languages as well without needing exclusive datasets for those languages, all on its own. ASI, which is the last of the three categories is also a hypothetical like AGI. This refers to an AI whose functioning, operations and methods of reasoning surpass the interpretation, understanding and thinking of human beings (Asokan). Many AI developers, futurists and technology enthusiasts refer to this as an event characterized by massive chains of “intelligence explosions”, collectively referring to it as “Singularity” (Tzezana). 2.2

The Applications of AI today

Alongside ML that has been mentioned above, Deep Learning (DL) is a novel form of machine learning that utilizes neural networks (strings of algorithms that transmit data to one another in a manner that mimics the transfer of information in the neural connections of the human brain) to speed up the processing of information, which in turn trains the AI faster (Perez). Why ML and DL are being researched, funded and developed is quite straightforward; it is being utilized to address the novel problems and situations of the 21st century, the kinds of novel events that have been observed exclusively in the age of information technology. AI research and development has been recurring since the 1950’s ever since the Turing test was established.Some may even argue that AI research dates far before the term “AI” itself was coined; a time where the idea of an artificially sentient being was referred to as “Automata” (Schuchmann). Regardless, these periods of AI research always led to an AI winter; periods of time where anything related to AI was dismissed, solely because there existed a hype over the subject that

Artificial Intelligence and Policy in India

was not fulfilled due to the overpromises of developers, high end-user expectations and lack of the necessary and adequate infrastructure. The age of information technology surpasses all of these obstacles and provides atop of the requirements, a new weapon in the arsenal of AI development; Big Data. The vastness of unstructured, random and vastly correlated data created by the globe gives rise to two factors between AI and Big Data that generates an inorganic symbiotic relationship between the two (Casey); wherein AI requires massive volumes of data to be trained in its task, and Big Data needs to be assessed and analyzed by efficient and reliable AI (as they are immune to cognitive dissonance and mental fog unlike humans, and don’t require washroom breaks or paid leave) in order to make comprehensible judgments out of it. Any field, sector, discipline, profession or platform that involves patterns and repetition of behavior or directly or indirectly involves the analysis of information, has room to accommodate AI. Given that very few sectors exist without these two characteristics, AI’s application is foreseeable and inevitable almost everywhere. AI is being utilized across a variety of existing disciplines and fields today, be it for research, legal consultancy (BNH.AI), playing a game of Go, medical examination and diagnosis, and outer space research and development (AI SpaceFactory). Furthermore, AI has created fields and disciplines through its research and development; the Not Company is a Chilean based company that has made its own niche in the vegan-food market using AI. It utilizes algorithms to breakdown popular foods to the molecular level and identify how it can recreate those foods on a molecular level using plant-based ingredients (NotCo). Contemporary developments are being made in the present as well. At the time of drafting this research amidst the COVID-19 pandemic, AI still makes massive strides; either by facilitating the accurate diagnosis of the disease (Six) or by proving its efficiency in replacing humans in a day and age of social distancing protocols and nation-wide lockdowns (GAONA). Nationstates have already established strategies in place for smooth responsible mitigation towards AI adoption, which makes the transition inevitable (BAAI).

Tracing precedents: A critical analysis of Legal Personhood

3.1

The Concept of Legal Personhood

In order for an entity to integrate into a legal system, it must be first considered a legal person. A legal person, as widely known, is any recognized entity endowed with rights, duties and capacities. Various municipal systems of law refer to this concept in a variety of manners, but all of these theories boil down to the same rudimentary acknowledgment, that all legal persons are inherently legal subjects; as they are subject to the law. Entities that are subjected to the law, involve themselves with legal objects or instruments that impose rights or duties upon them; i.e. a contract, a property, or an intellectual property. Through the exercising of the capacity endowed by the legal system that recognizes it, the entity can choose to enforce its rights and enact on its duties. In contrast however, the lack of exercising this capacity can lead to the non-performance of these duties, and an over-exercising of this capacity beyond its legal recognition can lead to the infringement of another entity’s rights; both of which can be addressed by the pre-stipulated mechanisms of the legal system in which they are recognized. Simplified, a legal person is a bundle of a rights and duties, and can refer to natural persons (human beings) or juristic persons (such as corporations). Consequently, a legal person has the duty to obey the law, while enjoying the benefits of protection upon rights and privileges accorded to them (Mahajan, 2016). Many common law and civil law systems don’t have a codified

Artificial Intelligence and Policy in India, Volume 1(2020)

definition of the term legal person, but rather harbor an implicit understanding of what it means for an entity to bear legal capacity recognized within its framework. However, what is often overlooked in the granting of a legal personality is the associated acknowledgment and recognition of an entity within the legal system and directly, its trait of being identified in relationship with it by other legal systems. Such a phenomena has underlying foundational theories in many other disciplines of cognitive thought such as the philosophical school of structuralism or the psychological theories of associative learning; but is commonly acknowledged and referred to in legal nomenclature, as nationality. While legal personality offers recognition within a legal system, its nationality expands the scope of its recognition across national borders by attaching it to a recognized entity responsible to enact its governing system of law; the sovereign state. States themselves are recognized personalities subject to the body of Public International Law (Shaw, 2016). Through this “domino effect of recognition”, a legal person of one state may have their rights enforceable in another state, and can also be held liable in them (Warren, et al., 1938, 1941). This phenomena expands the recognition of an entity, and directly, its attached rights and duties. 3.2

Application and Impact upon AI

In context of a potential AI personality, recognition would include ADMS and AI into an existing international integrated system of legal recognition of various entities. In our current tech-reliant civilization which is observably headed towards being tech-centric, this recognition could pose itself to be of immense utility. This is because it is well understood that legal personality is a flexible and changeable aspect of the legal system, and it is well recognized that legal subjects in a legal system are not exactly identical in their natures, capacity, rights or duties, depending on the need of the community (Genderen, 2018). This is recognized in International Law as well, as the scope of a legal personality is measured by the need of society under different circumstances (Genderen, 2018). 3.3

Precedents of the dynamism of Legal Personhood

One of the many obstacles often dragged into this discussion is the element of consciousness, or the lack thereof. It is commonly agreed by academics of various disciplines that machines of the present age don’t exhibit consciousness (Hildt) (which doesn’t rule out the possibility of them developing it). The question nonetheless, “does Consciousness matter?” does encapsulates one of the biggest dilemmas in the field of AI ethics. While the question could never meet resolution in other disciplines, a simple historical retrospection into the precedents of law will provide a straightforward answer from the discipline; consciousness is not necessary. The ethical dilemma is still entertained nonetheless by scholars that are opposed to the granting of AI legal personality under the notion that extending the class of legal persons can come at the expense of the interests of those already within it (Of, For, and By the People: The Legal Lacuna of Synthetic Persons, 2017). Times of dynamic legal change have always put forward questions and thoughts for consideration. Historical retrospection and precedents fill in the spaces of confusion here as well, as circumstantial notions continuously evolve to highlight fresh issues arising out of various places, time periods and cultures. Human slaves in the Roman Empire and in the centuries following it, were never considered human beings, nor were given the rights that were granted to their masters that were higher on the social hierarchy (Genderen, 2018). They however still had the possibility of owning an amount of property that was specified by their masters. This

Artificial Intelligence and Policy in India

highlights a peculiar instance of conceptual merging; wherein the status of a legal subject and a legal object could be imposed upon the same entity. In the United States of America prior to the 13th U.S Constitutional Amendment of 1865, slaves were punished for criminal acts instead of the liability being imposed vicariously upon their slaver masters. Abolitionist William Goodell in 1853 observed that a slave “became a person” whenever they were to be punished, were under control of the law but couldn’t perceive it to be friendly towards them, and couldn’t testify against white Americans (Slaves and the Rules of Evidence in Criminal Trials- Symposium on the Law of Slavery: Criminal and, 1993). Arguably, legislators, enforcers and defenders of slavery law in the United States did understand that granting slaves their rights came at the expense of nullifying their draconian ideals, and approached the legal recognition of slaves by “cherry-picking” whatever could facilitate their biased perspectives. This directly justified the atrocities committed against an entire demographic that would be considered natural persons and be protected against racial discrimination in the 21st century. With reference to the Indian legal system and regards to animals, the honourable High Court of Punjab and Haryana, Chandigarh, emphasised in the case of Karnail Singh v. State of Haryana, that the entire animal kingdom including avian and aquatic species has a “distinct legal persona with corresponding rights, duties, and liabilities of a living person” (SHARMA, 2019). It was further emphasised in Animal Welfare Board v. Nagaraja that animals are entitled to fair treatment and dignity under Article 21 of the Constitution of India, 1950 (K.S. Radhakrishnan, 2014). While the abovementioned illustrated how legal recognition was granted to a variety of sentient beings and was denied even to humans beings; the species that highlighted questions of consciousness in the first place, various examples exist that highlight the granting of legal personhood to inanimate entities. The Whanganui River in New Zealand was granted legal personhood in order to facilitate its preservation (WARNE). In a UK case (LJ, 1991), a temple that was recognized as a legal person in India (being physically present there) could assert rights and make claims under English Law. The courts recognized this in light of the principle of comity of nations. Furthermore, Japanese history documents a curious event of a statute being arrested as a suspect in a case of theft, by 18th century jurist and Judge Ooka Tadasuke (Muzachan). The statute was of Jizo, a Japanese deity and Buddhist monk. It was brought to court tied as it was the only witness to the theft of an article of clothing owned by a traveling merchant that rested under the statue’s shade. While the arrest was only a means for Judge Tadasuke to solve the case and identify the real culprit, the undercurrent of imposed persona ficta with its associated rights and duties and the judge’s wit can’t be denied; forJizo is the protector of travelers in Japanese medieval Buddhism (Chavez). 3.4 Closing Remarks on Legal Personhood of AI With the abovementioned, it is emphasized that the theory of legal personhood is as dynamic is it is required by the society around the entity in question. It is a dynamic legal fiction, and often is the key element necessary for the suspension and sustenance of many theories of law, including the upholding of establishments that are often overlooked such as states and corporations (Criteria for Recognition of AI as a Legal Person, 2019). The element of legal fiction is highlighted in jurist Naffine’s theory of the “Cheshire Cat”, named after the character in Lewis Carroll’s enduring tale Alice in Wonderland. The Cheshire cat theory emphasizes that the legal person exists only as an abstract capacity to function in law without any ethical or moral

Artificial Intelligence and Policy in India, Volume 1(2020)

constraints, and such capacity is endowed by the law because it is convenient for the law to have such a creation (Who are Law's Persons? From Cheshire Cats to Responsible Subjects, 2003). The legal recognition of AI therefore is very foreseeable. However, much like the everchanging concept of legal personhood itself, it needs its own flexibilities and types of recognition. This distinction could be made on the basis of the strength of the AI itself; having ANI legally recognized as legal agents, and recognizing a potential AGI as a legal person. This is discussed in the next section.

Tracing the points of confluence: AI Law for the Artificially Intelligent?

4.1 Short Term fixes: Addressing the AI Black Box through Legal Recognition The legal recognition of AI and ADMS will undoubtedly lead to dynamic new changes and expand the frontiers of a legal system, internally and externally. This will bring AI into the chain of accountable entities that could be held responsible for any infringement of life and property. AI is still not fully autonomous, independent and conscious which would make it unable to comprehend and understand the scale or impact of any action that maybe of legal concern, but recognition would still address the legal vacuum that rests with regards to the accountability and responsibility of those actions, which is discussed ahead. Prior to that aspect, legal recognition would also provide temporary regulation against the AI Black Box phenomena (maize). The AI Black Box is the label given to the inability to understand an algorithmâ&#x20AC;&#x2122;s reasoning in processing its training data, what data is it utilizing amongst the variety of training data presented and how it is interpreting that data towards generating an outcome. The AI black box is known for being ambiguous to its developers as well, and can manifest itself in various algorithms that are involved in the day to day activities of individuals carrying out their personal lives, such as loan application processing algorithms, education program assessments, employee recruitment assessments, etc. The black box can become a concern when it propagates biases against various groups of society that would raise a legal concern, such as Amazonâ&#x20AC;&#x2122;s use of an algorithm (now revoked) that processed recruitment applications which discriminated against women (Dastin). Questions of information security are raised as well if the algorithms are made more transparent, as this would require independent AI to assess these processes and have software codes of the principle AI and the interpreting AI to be made available to the public and in turn, have it exposed to malicious intentions as well (Burt). Till productive strides can be made in AI transparency, the black box will have to be assumed as a trait highlighting AIâ&#x20AC;&#x2122;s autonomy, and ought to be legally recognized in the chain of decision-making executed by various institutions and establishments. The Committee of Ministers of the Council of Europe very recently has acknowledged the black box phenomena in a recommendations report that highlights the potential impact algorithmic systems pose upon the human rights of European citizens (Recommendation CM/Rec(2020)1 of the Committee of Ministers to member States, 2020).

Artificial Intelligence and Policy in India

4.2 The Distinctions of Legal Recognition: Classifying AI The legal recognition of AI would not be straightforward in its implementation, but would require a detailed and intricate strategy in response to the complexity that the problem poses. Herein, it is proposed that any AI that is in existence today (ANI) be recognized as legal agents and any theories of artificial personality be reserved for the potential of AGI, which would be recognized as legal persons. The distinction is made on a very simple observation made on the strength classifications of AI. A legal agent acts within a specific requirement as prescribed by the principle entity on behalf of whom the agent functions. A legal person however bears rights and responsibilities to be enforced in accordance with their actions and behaviors that are conducted out of their own will. Although ANI can execute decisions through its own autonomy, it requires a degree of human involvement in order to function effectively; be it the input of data, training of the algorithm etc. Liability “Neural Networks”: ANI as Agents and the extension of Liability. Through this distinction and the recognition of ANI as a legal agent, the responsibility and accountability of an ANI’s actions can then be imposed upon an existing legal person that is involved in its operations and functioning. This will create, for the lack of a better metaphor, a “neural network” of new liabilities, as the number of individuals involved in the development and use of an ANI are numerous vary across different AI. As legal agents, liabilities arising out of an AI's act shall be imposed upon the principle. The principle however would be circumstantial and dependent on the situation, and its subjective causes and actions. This can either be: 1. The legal entity that manufactures the AI, 2. The developer/s responsible for the algorithmic error, 3. Or even the end user of the AI, if the AI is used outside its intended situation. The Criminal Liability of AI is further discussed by Gabriel Hallevy in (The Criminal Liability of Artificial Intelligence - from Science Fiction to Legal Social, 2016). He proposes three models (perpetrator-by-another; natural-probable consequence; or direct liability) for making the AI systems liable for the offences. The perpetrator-by-another and natural-probable model can be used to decide the liability of ANI whereby the programmer or end user can be held liable for any consequence or offence committed by the ANI. The end user of the AI will have to be involved into the chain of ANI liability as well, as there exist a variety of hypotheticals wherein a user has used an AI outside its intended environment of application which has resulted in the action of legal concern. This has to be done to fully address the novelty of this “complex distributed responsibility”, which will definitely lead to changes in the established bodies of civil law, such as contract, business and investment law around the world (Vital, Sophia, and Co.—The Quest for the Legal Personhood of Robots, 2018). Furthermore, the pioneering work in the field of AI broadens the definition of AI. Myriads of ANI available till date are equivalent to and should be categorised under computer software. Thus, such ANI should be considered as a service since there exists no transfer of ownership rather the legal entity or corporation is being paid for the delivering certain service or software or ANI. Therefore, the entities developing ANI should be held liable for its breach or offence.

Artificial Intelligence and Policy in India, Volume 1(2020)

4.3 Reinforcing the law for the arrival of AGI Some scholars argue that the hypothesis of a conscious AI that would require a legal personality should be dismissed for the foreseeable future (Vital, Sophia, and Co.â&#x20AC;&#x201D;The Quest for the Legal Personhood of Robots, 2018), but that doesnâ&#x20AC;&#x2122;t nullify the need for dialogue and legal preparedness for the birth of such a technology (Koch). Such actions will however require inter-disciplinarian effort. In order to identify and effectively grant an AGI legal personality, it should meet a standard set of criteria in order to ensure that such recognition is justified. Various scholars have a variety of approaches to creating such a list of criteria, but the two most holistic approaches are identified below. F. Patrick Hubbard in his article ('Do Androids Dream?': Personhood and Intelligent Artifacts, 2010) states that a computer system (AI in context of this research) should be granted right to legal personality if it meets the following three criteria: 1. It has an ability to interact with its environment and to engage in complex thought and communication, 2. A sense of being a self with a concern for achieving its plan for its life, 3. The ability to live in a community with other persons based on at least mutual self-interest. Jurist Jacob Turner identifies here that the second point loosely resembles the idea of consciousness and if legal personality were to be approached pragmatically, then it would not be necessary to establish the second criteria (Turner, 2019). Robert van den Hoven van Genderen however, addresses this criteria list with far more intricacy and states that in order for an AI to be granted legal personality, it must: 1. be necessary in a human society and have socio-economic relevance as to acquire legal certification, 2. pass a test of determination of autonomous intelligence, 3. have sufficient social intelligence, 4. be able to respond to changing circumstances; possess adaptive intelligence, 5. be accepted by other legal persons by creating trust and reliance which would facilitate integration into economic, legal and social structures, 6. be registered in a public register of robots which have specified legal competences for specified roles and tasks (Genderen, 2018). 4.4 Recommendations Establishment of Bodies and Institutions addressing questions of AI. Bodies ought to be established with the objective that they work towards AGI and its potential legal personhood. These bodies ought to involve inter-disciplinarian efforts towards the solving of questions posed by the legal personhood of AGI. Such questions may transcend legal, technical, ethical and philosophical boundaries, and hence will require cross-cultural effort towards the creation of neo-legal theories, such as a theory of penology for AI. Lastly, these institutions ought to be public in its structures, involving states, public and private entities in its operations in the realm of AI ethics and ought to strive towards the manifestation of new rules and principles that would be necessary to navigate the realm of the tech-centric society that is incoming.

Artificial Intelligence and Policy in India

The Importance and necessity of codification. Codification as a due process of the law is a necessary element of legal systems that has evolved along aside the development of these systems themselves. Codification serves a variety of purposes when it comes to contemporary developments being faced by the law. These include an ease of enforcement of the law, efficacy and efficiency of governmental and administrative functions, and ease of access to the law for the public and concerned parties. All of these factors can facilitate the creation of an effective body of law governing AI, which ought to be considered for the long term harmonization of AI into various social spheres. This can be enacted through the act of effective legislations that legally recognizes AI as an artificial person and as a factor in the development and function of future societies. Lifespan of AGI. A prolonged discussion has been put forward for granting the AGI a legal personhood. Aside from ANI, as AGI is believed to have a human like consciousness and cognitive thinking, it should be given a life span. The unforeseeable future brings numerous possibilities; as predicted by many scholars, AGI should be able to replicate and multiply itself. To prevent a dystopian population/production outbreak of AGIs, it should be assigned a precise life span similar to humans.

The dawn of a new age: Conclusion

Philosopher and the founder of modern political philosophy Thomas Hobbes in his book Leviathan described what it meant to be a person, as: “He whose words or actions are considered, either as his own or as representing the words or actions of another man, or of any other thing to whom they are attributed, whether truly or by fiction. When they are considered as his own, then he is called a natural person: and when they are considered as representing the words or actions of another, then he is a feigned or artificial person” (Hobbes, 1909–14). He then proceeded to describe the origins of the word person, coming from the Latin word ‘persona” and the Greek word “prosperon”, which refers to a mask used in theatres. In modern days, often revisiting the literature preserved in global history provides for effective guidance and direction towards the solving of problems, regardless of how specific that problem is to its time period of inception. The 21st century is already underway towards large scale AI incorporation for any function it may be deemed necessary. Discussions and their subsequent actions have to take place accordingly, and the prosperon must be utilized effectively.

References 1. AI SpaceFactory. AI SpaceFactory Homepage. AI SpaceFactory. [Online] AI SpaceFactory. [Cited: May 09, 2020.] https://www.aispacefactory.com/.

Artificial Intelligence and Policy in India, Volume 1(2020)

2. Asokan, Akshaya. Who is afraid of Artificial Super Intelligence? Analytics India Magazine. [Online] Analytics India Magazine Pvt Ltd. [Cited: May 08, 2020.] https://analyticsindiamag.com/whosafraid-of-artificial-super-intelligence/. 3. BAAI. Beijing AI Principles. BAAI. [Online] BAAI. [Cited: May 10, 2020.] https://www.baai.ac.cn/news/beijing-ai-principles-en.html. 4. BNH.AI. BNH.AI Homepage. BNH.AI. [Online] BNH.AI. [Cited: May 09, 2020.] bnh.ai. 5. Burt, Andrew. The AI Transparency Paradox. Harvard Business Review. [Online] Harvard Business School Publishing. [Cited: May 13, 2020.] https://hbr.org/2019/12/the-ai-transparency-paradox. 6. Casey, Kevin. How Big Data and AI work together. The Enterprisers Project. [Online] Red Hat, Inc. [Cited: May 09, 2020.] https://enterprisersproject.com/article/2019/10/how-big-data-and-ai-worktogether. 7. Chavez, Amy. A guide to Jizo, guardian of travelers and the weak. the japan times. [Online] THE JAPAN TIMES LTD. [Cited: May 12, 2020.] https://www.japantimes.co.jp/community/2012/03/31/our-lives/a-guide-to-jizo-guardian-of-travelers-and-the-weak/#.XvJbCygzZRY. 8. Criteria for Recognition of AI as a Legal Person. Roman Dremliuga, Pavel Kuznetcov, Alexey Mamychev. 2019. 3, s.l. : Canadian Center of Science and Education, 2019, Vol. 12. 9. Dastin, Jeffrey. Insight - Amazon scraps secret AI recruiting tool that showed bias against women. Thomson Reuters. [Online] Reuters. [Cited: May 13, 2020.] https://in.reuters.com/article/amazoncom-jobs-automation/insight-amazon-scraps-secret-airecruiting-tool-that-showed-bias-against-womenidINKCN1MK0AH. 10. 'Do Androids Dream?': Personhood and Intelligent Artifacts. Hubbard, F. Patrick. 2010. Philadelphia : Temple Law Review, 2010, Vol. 83. 11. Fourtané, Susan. The Three Types of Artificial Intelligence: Understanding AI. Interesting Engineering. [Online] Interesting Engineering, Inc. [Cited: May 08, 2020.] https://interestingengineering.com/the-three-types-of-artificial-intelligence-understanding-ai. 12. GAONA, J. MAURICIO. COVID-19 will accelerate AI's replacement of humans as factor of production. The Hill. [Online] CAPITOL HILL PUBLISHING CORP. [Cited: May 14, 2020.] https://thehill.com/opinion/technology/492492-covid-19-will-accelerate-ais-replacement-of-humans-as-factor-of-production. 13. Genderen, R. van den Hoven van. 2018. Do We Need New Legal Personhood in the Age of Robots and AI? [book auth.] Mark Fenwick, Nikolaus Forgó Marcelo Corrales. Robotics, AI and the Future. Singapore : Springer Publishers, 2018. 14. —. 2018. Do We Need New Legal Personhood in the Age of Robots and AI? [book auth.] Mark Fenwick, Nikolaus Forgó Marcelo Corrales. Robotics, AI and the Future. Singapore : Springer Publishers, 2018. 15. Hildt, Elisabeth. Artificial Intelligence: Does Consciousness Matter? frontiers in Psychology. [Online] Frontiers' social media. 16. Hobbes, Thomas. 1909–14. Of Man, Being the First Part of Leviathan. New York : P.F. Collier & Son, 1909–14. 17. Joshi, Naveen. How far are we from achieving artificial intelligence? Forbes. [Online] Forbes Media LLC. [Cited: May 08, 2020.] https://www.forbes.com/sites/cognitiveworld/2019/06/10/how-farare-we-from-achieving-artificial-general-intelligence/#571b82536dc4. 18. K.S. Radhakrishnan, J. 2014. Animal Welfare Board of India v. Nagaraja & Ors. CIVIL APPEAL NO. 5387 OF 2014, New Delhi : Supreme Court of India, 2014. 19. Koch, Christof. Will Machines Ever Become Conscious? Scientific American. [Online] SCIENTIFIC AMERICAN, A DIVISION OF SPRINGER NATURE AMERICA, INC. [Cited: May 13, 2020.] https://www.scientificamerican.com/article/will-machines-ever-become-conscious/. 20. Krishna, Sreekar. Still in their infancy, AI algorithms need parenting. QUARTZ. [Online] Quartz Media, Inc. [Cited: May 07, 2020.] https://qz.com/work/1298462/ai-is-still-in-its-infancy-and-itneeds-parents-to-avoid-bias/. 21. LJ, Purchas. 1991. BUMPER DEVELOPMENT CORPORATION LTD V COMMISSIONER OF POLICE OF THE METROPOLIS. [1991] 1 WLR 1362, 1991.

Artificial Intelligence and Policy in India

22. Machine Learning: What it is and Why it matters. SAS. [Online] [Cited: May 07, 2020.] https://www.sas.com/en_in/insights/analytics/machine-learning.html. 23. Mahajan, V.D. 2016. Jurisprudence and Legal Theory. Lucknow : Eastern Book Company, 2016. 24. maize. WHAT IS THE AI BLACK BOX PROBLEM? maize. [Online] [Cited: May 13, 2020.] https://www.maize.io/en/content/what-is-ai-s-black-box-problem. 25. Muza-chan. Shibarare Jizo, the rope-tied statue. Muza-chan's Gate to Japan. [Online] [Cited: May 12, 2020.] https://muza-chan.net/japan/index.php/blog/shibarare-jizo-the-rope-tied-statue. 26. NotCo. NotCo Homepage. NotCo. [Online] NotCo. [Cited: May 09, 2020.] http://www.notco.com/. 27. Of, For, and By the People: The Legal Lacuna of Synthetic Persons. Joanna J Bryson, Mihailis E Diamantis, Thomas D Grant. 2017. 3, s.l. : Artificial Intelligence and Law, 2017, Vol. 25. 28. Perez, Carlos E. Deep Learning: The Unreasonable Effectiveness of Randomness. Medium. [Online] [Cited: May 09, 2020.] https://medium.com/intuitionmachine/deep-learning-the-unreasonable-effectiveness-of-randomness-14d5aef13f87. 29. Recommendation CM/Rec(2020)1 of the Committee of Ministers to member States. Europe, Council of. 2020. s.l. : Council of Europe, 2020. 30. SAS Institute Inc. Machine Learning: What it is and Why it matters. SAS. [Online] SAS Institute Inc. [Cited: 07 May 2020.] https://www.sas.com/en_in/insights/analytics/machine-learning.html. 31. Schuchmann, Sebastian. History of the first AI Winter. Medium. [Online] [Cited: May 09, 2020.] https://towardsdatascience.com/history-of-the-first-ai-winter-6f8c2186f80b. 32. SHARMA, HON'BLE MR. JUSTICE RAJIV. 2019. Karnail Singh and others v. State of Haryana. CRR-533-2013, Chandigarh : HIGH COURT OF PUNJAB & HARYANA, 2019. 33. Shaw, Malcom N. 2016. International Law. Cambridge : Cambridge University Press, 2016. 34. Six, Ory. Diagnosing COVID-19 using AI-based medical image analyses. Quantib. [Online] Quantib B.V. [Cited: May 14, 2020.] https://www.quantib.com/blog/diagnosing-covid-19-using-ai-basedmedical-image-analyses. 35. Slaves and the Rules of Evidence in Criminal Trials- Symposium on the Law of Slavery: Criminal and. Morris, Thomas D. 1993. 3, Chicago : Chicago-Kent Law Review, 1993, Vol. 68. 36. The Criminal Liability of Artificial Intelligence - from Science Fiction to Legal Social. Hallevy, Gabriel. 2016. 2, Ohio : Akron Intellectual Property Journal, 2016, Vol. 4. 37. Turner, Jacob. 2019. Robot Rules Regulating Artificial Intelligence. Cham : Springer Nature Switzerland AG, 2019. 38. Tzezana, Roey. Singularility: Explain it to me like I’m five years old. Futurism. [Online] [Cited: May 08, 2020.] https://futurism.com/singularity-explain-it-to-me-like-im-5-years-old. 39. Vital, Sophia, and Co.—The Quest for the Legal Personhood of Robots. Pagallo, Ugo. 2018. s.l. : MDPI AG, 2018. 40. —. Pagallo, Ugo. 2018. s.l. : MDPI AG, 2018. 41. WARNE, KENNEDY. A VOICE FOR NATURE. National Geographic. [Online] National Geographic Partners, LLC. [Cited: May 12, 2020.] https://www.nationalgeographic.com/culture/2019/04/maori-river-in-new-zealand-is-a-legal-person/. 42. Warren, C. (United States), Greenshields, R.A.E. (Canada) and Hostie, J.F. (Belgium). 1938, 1941. Trail Smelter Case (United States v Canada). Washington/Ottawa : Arbitrational Tribunal, 1938, 1941. 43. Who are Law's Persons? From Cheshire Cats to Responsible Subjects. Naffine, Ngaire. 2003. 3, s.l. : Modern Law Review, 2003, Vol. 66.

Artificial Intelligence and Policy in India, Volume 1(2020)

Automated Killer: the treads around the soft earth of legality and Artificial intelligence Vedant Sinha1 1

Honorary Research Member, Indian Society of Artificial Intelligence and Law sinha.vedant18@gmail.com

The emerging paradigms of AI in the next-generation battlefield

Zeroth law (Asimove, 2001): Asimov’s principle may remind us of the basic reason for the conception of the autonomous system is to benefit the existence of humanity and not further provide the tools of violence. Therefore, matching up with the intent of the human being, AI and robots are not directly involving themselves within the conflict, providing more support roles, computational authority, and data fusion through sensory vestiges on the battlefield to augment the ability of the conventional forces. The new fight is being fought out on the digital realm before blood spills and has resulted in many countries augmenting their data analytical methods and abilities through the induction of AI in the ORBAT and analytical teams.1 Clive Humby coined the phrase, that “Data is the new oil”, however Lt. Gen Jack Shanahan feels differently, that it is “a mineral ore, there’s a lot of crap. You have to filter out the impurities from the raw material to get the gold nuggets”. The enormous amounts of data flowing within the physical infrastructure can be mostly non-consequential at face value, however when purified, it holds value, thereby forming one of the stress point in usage of AI, the dependability on the quality of the dataset. Therefore, the methods to purify the data are as important as the AI, it is being fed, based on AI gullibility. The gullibility of the AI makes it sensitive to the minutest of changes, amid pattern recognition the minutest of impurities data can consequentially change the nature and the results of the AI, so much that the deviation due to outliers, missing values or can alter the projected results to a great effect. A large set of clearly labelled, well-organized data points is what machine learning algorithms require to understand from the data, once the data set is manually cleaned and provided to them. Such a set has to be manually vetted through human diligence, and such vetting alongside it raises a whole slew of issues ranging from privacy to sovereign liability of the states and its privilege in accessing it. However, the primary issue to focus around it is the focus upon the analytical value that the nation begets through such data. For example: The USA was involved in a lot of Counterterrorism operations through drone bombing over Pakistan and Afghanistan, they used to vet the information received through a Video feed, mobile towers and like sources of metadata, which were then, segregated and filtered out based on the data obtained from such filter were further used to mark the threat level of the target being observed and marked for elimination after specific vetting. The program despite undergoing human vetting still yielded a high amount of Civilian casualties against the actual military 1

The orbat across various armed forces are undergoing severe changes with respect to the integration of the AI, explicit introduction into direct confrontation in the front lines might not be the goal for the majority of the armed forces, but with emerging technologies of swarm drones, there will be renewed efforts upon introducing as soon as possible.

Artificial Intelligence and Policy in India

goals achieved. (Kalvapalle, 2018) Regarding the applicability of international law in this aspect, The legality of the drone bombing has been under the grey area. Furthermore, such hostility marking often resulted in thousands of innocent civilians being killed. The lines get blurrier in a warzone, where the citizen is often not hostile, but not cooperative with the occupational forces as well2, and the data-points extracted from a set of co-operating citizen mindful of the law is different from those civilians whose rights and its co-relatives are altered in a warzone. Thereby different standards for AI arise in both peaceful and warzone situations and even within this scenario, there exists a probability for Algorithmic Bias to exist. Therefore, AI may be trained in a normal situation might probably tag normal civilians in warlike situations as militants. In such a military setting, the AI must be robust.3 For a robust AI to work in a surveillance situation, for example, to differentiate a Civilian amongst an occupied population, it has to first understand the difference between what is normal and abnormal behaviour for a civilian and to do the same, it has to train on a wide assortment of civilian data, to establish a range of normality and abnormality. However, privacy concerns do arise and the Constitutional Courts can deprive the access to data per the privacy concern, deprives the military of relevant datasets to utilize. There the balance between the private data of a person is of National security consequence or not, has to be demarcated or even if this operation has to be performed or not. Doing it provides the state, the legitimacy to claim sovereign immunity under Tort Law. (UN, 1996) Moreover balancing the question of a matter of national importance and International Law, there is a very fine thread for the military to balance.4 Indeed, the military cannot disclose elements for reasons of national importance, however, the presence of state-sponsored legislation, issuing guidelines for the presence of ethics in AI becomes a requirement of the time. Within the ambit of the Additional protocols I and II to the Geneva conventions (ICRC, Additional protocols) adopted to deal with the concerned areas of international armed conflict and the civil wars respectively, were a layer of protection provided to the civilians as they are the most affected ones in armed conflict and provided as principles of conduct for the military. Quantification of such principles and evaluation against the set goals set out in the principles of international humanitarian principles is a task that requires a translative form of the principles of IHL, whose conversion into the purported principles that can be tested against the results propounded by an AI.

Integrating the principles of Distinction

The principle of distinction between civilians and combatants was first outlined in the St. Petersburg Declaration, which states that “the only legitimate object which States should endeavor to accomplish during war is to weaken the military forces of the enemy”. (ICRC, 2005) Article 13(2) of Additional Protocol II to the Geneva conventions (OHCHR), prohibits making the civilian and individuals as the object of attack. The directing attacks against 2

This can be clearly inferred from the Kashmir in India, whilst there exists some portion of demography that considers Indian forces as occupational, however they cannot be termed as militants. 3 An AI can be said to achieve the title of ‘robust’ when it can fuse information from various sources and can produce derivative conclusion faster, is self improving. 4 Such issues form the core of the AI within the ambit of Constitutionalism, state control and data access issues.

Artificial Intelligence and Policy in India, Volume 1(2020)

civilians is frowned against and such statute is also contained in Amended Protocol II to the Convention on Certain Conventional Weapons. (UN, 1996). It is laid out within the Protocol III to the Convention on Certain Conventional Weapons, which has been made applicable in non-international armed conflicts according to an amendment of Article 1 of the Convention adopted by consensus in 2001. The Ottawa Convention which bans anti-personnel landmines states inter alia, on “the principle that a distinction must be made between civilians and combatants”. However, the underlying principles upon whose distinction is possible varies politically for different states, it is an apparent danger that the chances of AI faltering upon the principles of distinction, as many AI applications have displayed propensities to be sensitive to change the conclusions with the smallest amounts of the impurities within the dataset.5 It is not only sufficient to be able to conclusively segregate militants from a normal citizen, in a peaceful or a warzone scenario. It is equally important for the AI to delineate the explanations for the outcome achieved and answer questions such as “why”, “how” of such classifications. The specific rules where the principle of distinction is, set out concerns Article 48 and 52 of Additional Protocol 1 to the Geneva Conventions. Insofar defining and differentiating combatant and a military object that can be lawfully attacked. Any direct attack against a civilian or civilian object is a violation of IHL, operating of lethal force upon the Civilian and/or civilians’ objects are war crimes. Any weapon which is incapable of distinguishing between civilians/civilian objects and fighters/military objects is also prohibited under IHL. Thereby, algorithm operated weapons that cannot follow the distinction rules are thereby banned under IHL. Integration of the ethics into AI can be as successful as the bias and the data collection procedure of the individual, thereby it is imperative to delineate the principles and the subject held to be desirous of such amalgamation. 2.1

Principles for integration

The viability of such programs can be committed as within the arena of a smart contract, as there are startups and existing technology that propose the conversion of a set of human-understandable principles that form the core of an E-Contract, such that an E-contract once executed is irreversible in nature and cannot be altered ex-post the agreement has been finalised. Smart contracts involve the basic idea that contractual clauses are embeddable into the hardware and software, to place the execution of the contract out of the control of the parties and thus dissolve trust-related friction observed in conventional contracts, in implementing it upon contingent fulfilment of the principle of the humanitarian principles, such implementation can be beneficial not only upon the military but in other horizontals as well. The action will require the implementation of Turing complete platforms and such conversion has to perspire from the bare conversion of a set of understandable set of principles into a dichotomous outline understandable by the machine as well. The classification between a civilian in conflict zones needs to be set upon the standards and norms of international humanitarian law and the principles of the least harm to the civilians possible and such applicability have to be imposed upon the concerned AI as well.

At the end of the day, it is upon the military to delimit the range of behaviour that can be construed as a civilian or a military behaviour.

Artificial Intelligence and Policy in India

Explanability of the AI

XAI employed in the military should be a big factor under consideration, and such models such be employed that does not morph into a Blackbox. Any such solution that seeks to unravel the reasoning such as deep explanation or model induction only does as a predictive, not an affirmatory solution. US army went ahead with the establishment of the project maven, i.e. an algorithmic warfare cross-functional team to specifically hunt down the Islamic militants in Iraq and Syria by synthesizing surveillance videos and audio-visual metadata into actionable intelligence vetted by intelligence expert. The actionable intelligence so provided by the algorithms answers the question in a probabilistic answer, without getting into the variable at play and explaining such results.6 The way for providing such an explanation is through Deep explanation: an inversion of deep learning, wherein machine learning is using to examine and analyse the explanation, unpack the procedures of a BlackBox or by adopting such ML algorithms that are more explainable than others. The principle of distinction underpinning many rules of IHL is that only fighters may be directly targeted. This is a necessary compromise that IHL provides to protect civilians in armed conflict. Without the principle of distinction, they would be no limitation on the methods of warfare. 3.1

Integrating the principles of proportionality

The principle of proportionality limits upon the military, that the harm so inflicted upon the civilian should be kept to a bare minimum and should be insofar proportionate to the factor that warrants such military response contingent upon the military advantage so accrued.7 The article where proportionality is most prevalent is in Article 51(5) (b) of Additional Protocols-I concerning the conduct of hostilities which prohibits attacks when the civilian harm would be excessive about the military advantage sought. This is an area of hostilities where we often hear the term â&#x20AC;&#x2DC;collateral damageâ&#x20AC;&#x2122;. The principle cannot be applied to override specific protections or create exceptions to rules where the text itself does not provide for one. The principle of proportionality itself is to be found within the rules of IHL themselves. Proportionality is only applied when a strike is made against a lawful military target. The inter-relation between the principle of proportionality and the principle of distinction becomes opaque when talking in terms of the AI, contingent upon the principles and the truths set apart for the working of the AI as done so by the programmer, it should be a principle of additional vigilance as co-relativity to co-existence is not a definitive marker of impending or even a shared version of ideology. Moreso, it boils down to the classification and there stands a possibility that a concentration of civilians wrongfully attributed to that of a congregation of militants which will lead to disastrous results. 3.2

Principle of military necessity

Military necessity permits armed forces to engage in conduct of destruction and violence so necessary to obtain the military objectives. The concept of military necessity acknowledges that 6

Such marking of the civilian as a militant or a person of actionable interest should not be contingent upon deterministic design conferred upon the algorithm, as the final judgement should lie with the human operator.

Artificial Intelligence and Policy in India, Volume 1(2020)

under the laws of war, winning the war or battle is a legitimate consideration. It is important to note that the notion itself is to be found within the rules of IHL. For example, Article 52 of Addition Protocol I lists those objects that can be subject to lawful attacks. The notion cannot be applied to override specific protections or create exceptions to rules where the text itself does not provide for one. Within the doctrine of the military necessity underlies jus ad Bellum and Jus in Bello, which restricts the circumstances of resorting to force for a state and how hostilities are conducted where such restriction of the force fails respectively. However, the adulation to the conventions does not explicitly legislate the avenue of the AI. Such an exercise can be interpreted by parties as the addition of an additional layer of obfuscation and it requires the strengthening of international law and its integration, coherence with the Municipal law, with live dangers of this method of transposing humanitarian principles to be interpreted by the individual as an adage, and an extra layer of obfuscation to the clarity behind the operation of the AI, reducing its importance to that of a standard form contract.8 3.3

Liability in case of AI committed war crimes

If the AI decides to tag an individual, then the right of authorization to execute an action, in that case, should still reside with a Human operator as a safety valve against any mistake that the AI might commit. If the human operator relies on the AI to execute a decision, then the liability in such a scenario might not rest upon AI, as the human operator had the chance to exercise due diligence. However, it varies in terms of operations, therefore will AI-powered apparatus be used in offensive operations is still to be determined. The principle of absolute liability should be applied within the realm of AI-powered weapons, liability will fall upon the user despite the range of the casualty alongside the liability of the programmer in his/her laxity in or for some default happening which results in mistargeting and civilian casualties. Tort law has to be used to discuss the relationship between the explainability and the liability in the areas where the actors have to indeed use the model to avoid the liability if the user intends to avoid liability if the model consistently outperforms humans, its conjunction with the international humanitarian law for dictating the interaction with the actions of the military has to have an overarching effect on the conduct of the military-related to the deployment of the AI in the battlefield.9 Such an assignment and the AI should be cognisant and must conform to the protocols and the laws set out in the Geneva conventions and the additional protocols to the Geneva conventions.

In absence of a clear delineated principle that connects the AI with the international humanitarian principles, it is open to interpretation to the wider public as to how they can interpret the principles of IHL and the application of AI to the same. 9 Tort law is the most viable component for the regulation of AI in both military and civilian application, it requires answering the question of loci of responsibility within the action so committed that infringes the civil rights of the other. Since, it clearly outlines rather the hohfeldian duty and right corelative between in rem and in personam, thereby in authorâ&#x20AC;&#x2122;s opinion, tort is applicable to the aforesaid scenario.

Artificial Intelligence and Policy in India

Conclusions

The paradigms of warfare are changing. US Northern Command head Gen. Terrence O’Shaughnessy says “the key to winning tomorrow’s all-domain wars is predicting an adversary’s actions hours and even days in advance.” (HITCHENS, 2020) Synthesis of AI with increased situational awareness is eventually increasing the efficacy of the military, such as $1500 swarm drone using data link and controlled by AI can enable 1000 such drones to perform simultaneously, in future will perform the same task as a $100 million jet. A simulated exercise, involving a human opponent to a unit mixed with robot, humans and drones reveals that they can do the same job that would have been done by a force 3 times their size. AI is lowering the cost of war overall, if there exists a moral cost of putting the responsibility and the trigger to a computer program will be the same or not is a different question to answer.

References 1. Asimove, Issac. 2001. Isaac Asimov's "Three Laws of Robotics". Auburn. [Online] auburn university, 2001. [Cited: June 24, 2020.] http://webhome.auburn.edu/~vestmon/robotics.html. 2. Diakonia. Basic principles of IHL. DIAKONIA. [Online] [Cited: June 24, 2020.] /www.diakonia.se/en/ihl/the-law/international-humanitarian-law-1/introduction-to-ihl/principles-of-international-law/.. 3. HITCHENS, THERESA. 2020. Breaking Defense. [Online] May 5, 2020. [Cited: June 24, 2020.] https://breakingdefense.com/2020/05/the-key-to-all-domain-warfare-is-predictive-analysis-gen-oshaughnessy/. 4. ICRC. Additional protocols. Protocols additional to the geneva convetions of 12 august 1949. Geneva : ICRC, Additional protocols. 5. ICRC. 2005. The principle of distinction between civilians and ocmbatants. IHL database. [Online] ICRC, 2005. [Cited: June 24, 2020.] https://ihl-databases.icrc.org/customaryihl/eng/docs/v1_rul_rule1.. 6. Kalvapalle, Rahul. 2018. What is Project Maven? The Pentagon AI project Google employees want out of. Global News. [Online] April 5, 2018. [Cited: June 24, 2020.] https://globalnews.ca/news/4125382/google-pentagon-ai-project-maven/. 7. OHCHR. Protocol II additional to the geneva conventions. OHCHR. [Online] [Cited: June 24, 2020.] https://www.ohchr.org/EN/ProfessionalInterest/Pages/ProtocolII.aspx . 8. UN. 1996. CCW amended protocol II. United nations geneva . [Online] may 3, 1996. [Cited: june 24, 2020.] https://www.unog.ch/80256EE600585943/(httpPages)/7E15EA4DFB2D6575C1257EEB0035F3 42?OpenDocument.

Artificial Intelligence and Policy in India, Volume 1(2020)

Artificial Intelligence and Juridical Considerations: Legal and Administrative Underpinnings in India Mridutpal Bhattacharyya1 1

Honorary Research Member, Indian Society of Artificial Intelligence and Law mridutpal.bhattacharyya@gmail.com

Introduction

“I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.” ― Alan Turing, Computing machinery and intelligence (Artificial Intelligence, 2020). The two ancient questions, perhaps the ones demanding the curiosity of every thinker around the globe, demanding research & demanding knowledge & education, the questions – where do we come from & where are we headed, are something unanswerable as of now. Despite there being widely accepted & acclaimed theories such as the coming of humans into existence via the initial primordial soup that had enveloped the Earth after the big bang, which gave birth to the first unicellular organisms, as consequence of repeated natural forces standing in collision to the soup, the theory of evolution propagated by Charles Darwin – the evolution of homosapiens from apes, there have been disputes as to the credibility of these theories & many others, since when there is thesis there has to be antithesis & then synthesis since that is the nature of humanity itself. There have been contradictions to these & other theories answering where we come from. But, there has seldom been clear suggestions & insights as to where are we as a collective organism, the human existence is headed. The future of humanity is uncertain, whether the Earth shall be destroyed & humans shall become intergalactic colonists to set off to far planets to live, or whether humans would infuse themselves with mechanical parts, or become immortals through spiritual methods is up for debate. But, there’s no denying the fact that whatever happens, no one can predict with a hundred percent accuracy, & whatever is to come shall be harbingered by the future itself. Development has to come without cessation in all fields, in order for humanity to be able to prosper & not merely exist. Artificial Intelligence, its advent, its inception, its development has forever been cause for debate & heated arguments & even exhibitions of violence, but a simple fact cannot be ignored that the field has been improving, evolving at a rapid pace from time to time, although there have existed moments of brief pauses in the history of development in this field. Although still in its adolescence AI has high applicability in multitudes of fields. Be it agriculture, industrial planning, commercial relations, or any other field. But, there comes a time when bold steps are needed, there comes a time when a leap of faith has to be made, & the time is nigh. The time has come to make a gallant maneuver, incorporating AI into the procedures of the law enforcement & the Judiciary. Bearing in mind the possibilities of ridding the Indian society of crime & striking at its very roots, bearing in mind the corruption in the Indian law enforcement & judicial machinery, this paper shall go into details on the criminological & social aspects of crime, & how the crime, delinquency &

Artificial Intelligence and Policy in India

deviance can be inhibited from happening in the first place, & suggestion of methods & reforms to curb the corruption faced by the Indian citizenry while seeking approach & remedy from the judiciary against wrongs done to them, & the rewarding of speedy justice to the Indian citizenry as per their Constitutional right, this paper shall further suggest methods to not only inhibit crime but also effectively apprehending criminals aided by the coming of age blessing of artificial intelligence.

Crime & Criminology: an understanding

In order to understand the work that needs to be done to inhibit crimes & apprehend criminals, a basic need – perhaps the most elemental & fundamental pre-requisite that has to be checked out is the knowledge & the understanding of what crime is. In order to do just that, a trip is necessary through the teachings of Edwin H. Sutherland, whose book “principles of criminology” is one of the most authoritative documents on the subject. Criminology is regarded by him as a “Body of knowledge regarding crime as a social phenomenon.” (Sutherland, 1924) This study, includes the studying of the processes of making of the law, the breaking of the law & the reaction towards the breakage of the law. These three processes serve as three aspects of one singular, unified sequence of interactions. In his words, “Certain acts which are regarded as undesirable are defined by the political society as crimes. In spite of this definition some people persist in the behavior and thus commit crimes; the political society reacts by punishment or other treatment, or by prevention. This sequence of interaction is the object matter of criminology.” (Sutherland, 1924) Hence, it can be said that crime is any undesirable act by an individual of a state against the state or another individual which can be deemed to be undesirable in the eyes of the society & political authority at large. While, criminology is the study of this exact phenomenon, along with the study of what causes the crime which is called criminal etiology, & the punishments rendered out to the person convicted of such a crime or penology. Any behavior which to the political society at large seems undesirable is considered to be criminal behavior.

Explanation to Criminal Behavior

Criminal behavior can be explained in two ways: The scientific explanation – the scientific explanation takes an approach from an analytical perspective in terms of the factors operating at the time of the occurrence of the event exhibiting criminal behavior, or in terms of the factors which were at play in the instance of such behavior in the past (Sutherland, 1924). In the first case, the explanation happens to be mechanistic, & in the latter historical, or genetic. Persons inclined towards higher analytical prowess such as physical & biological scientists take the first procedure to be more desirable & superior to the latter method due to it being easier to measure. Nevertheless, such methodical attempts have been unsuccessful owing to attempt in separating personal & social pathologies. The work from this perspective has led the belief that the acts of crime occur owing to the person-situation complex & the history of the person as well & no other mechanistic reason (Sutherland, 1924). I.e. the opportunistic criminal jumps at an opportunity only when that situation is

Artificial Intelligence and Policy in India, Volume 1(2020)

important to that individual. E.g. a person robs a poorly protected bank to minimize the chances of getting inhibited or apprehended. But the moment there is a lack in security of a highly protected bank, the gaze of the criminal shifts to this bank instead of the previous unprotected bank. Or, it might so happen that this particular criminal direly needs huge amounts of money as soon as possible, in which case this criminal will take a greater risk to rob the better protected bank housing higher amounts of money, despite there being greater chances of this criminal getting inhibited or apprehended. Genetic explanation of criminal behavior – This philosophy deals with the process that via which a man goes down the path of crime. The postulates are (Sutherland, 1924) – • Criminal behavior is learned (Sutherland, 1924) – this theory suggests that criminal behavior is merely learned & not inherited, since a person is not known to inherit knowledge. I.e. a person is much less likely to make a mechanical invention unless the person has had prior training in the field of mechanics (Sutherland, 1924). The same way, a person does not indulge in criminal behavior unless he or she experiences it first hand through some external interface. • Criminal behavior is learned from other persons while communicating (Sutherland, 1924) – this theory suggests that the transfer of information from one person to another also transfers ones delinquency to another (Sutherland, 1924). This might be verbal communication or gestures. E.g. The sharing of ideas to giving insight to a person on the works of factory owners which are derogatory to the rights of the laborers may cause this person to talk to the laborers & incite them to start a riot. • The principle part of learning of criminal behavior happens in intimate groups (Sutherland, 1924) - this theory suggests contrary to the popular notion that movies, video games, newspapers causes criminal behaviors that, criminal behavior is picked up on in close groups (Sutherland, 1924). E.g. A group of boys where one boy starts abusing drugs & the other boys follow in his tracks. The learning of criminal behavior includes (Sutherland, 1924) – • Very complicated or simple methods of committing a crime (Sutherland, 1924). – Imparting of knowledge among the communicating group of ways to accomplish the task at hand. • Specific directions of motives, attitudes, rationalizations, drive (Sutherland, 1924). – Making one believe in the motive behind the crime, the drive behind it, the attitude that needs to be utilized to accomplish the task, & the rationalization of the task. • The specific directions of motives are learnt by the definitions of the legal codes being deemed favorable or unfavorable (Sutherland, 1924) - the surrounding environment of a person plays a huge role in shaping up the person. I.e. If a person is surrounded by family members or other persons of the society say friends, who insult, disrespect, & hate the provisions of the law & think that they are not worth adhering to, this person shall also have imprinted in his mind those exact same ideals, consequently giving rise to criminal behavior. While, if a person is surrounded by people who love, respect & adhere to the law at all times, this person shall have those very values imprinted onto his mind, in which case this person shall be sculpted into a law-abiding citizen. A person becomes delinquent owing to more provisions of the law which are favorable to criminal behavior rather than more provisions of the law which are unfavorable to criminal behavior (Sutherland, 1924) - The principle of differential association (Sutherland, 1924), this theory

Artificial Intelligence and Policy in India

suggests that a person becomes a criminal not only when exposed to & in contact with criminal patterns, but also when not in contact or at exposure to anti-criminal patterns. Every person assimilates, adapts to the surrounding culture unless a contrary culture is present (Sutherland, 1924). Differential association may vary in frequency, duration, priority, & intensity (Sutherland, 1924) – This theory suggests that the results of differential association vary on the basis of frequency, duration, priority, intensity of the association. I.e. the frequency of the association & the duration of the association definitely play a part into how much delinquent information or motives are to be imprinted upon the mind of an individual. While how much priority this individual designates to the intimate group he or she mingles in, & how much priority is assigned to what information is being transferred in that group also plays a part in determining the result of the differential association. Intensity however, cannot be so clearly defined but can be stated to be the emotional reaction that is incurred by a person when in contact to or at exposure to delinquent patterns. The process of learning criminal behavior is similar to the process of learning anything else (Sutherland, 1924). – This theory suggests that the process of learning any delinquent behavior in an intimate group through criminal patterns is the same as the process of learning anything else. I.e. The learning of criminal behavior is not restricted only to the act of imitation. E.g. a person seduced into a criminal act learns the criminal act through association but this process cannot be said to be a simple procedure of imitation, since the crime of this individual takes a life of its own as soon as the person decides to serve as an accomplice to the crime, & starts planning or executing new methods to accomplish the criminal task at hand. Although criminal behavior is the result of certain needs & values, those needs & values cannot be considered to be the explanations of the criminal behavior. Since, those needs & values are also existent in non-criminal behavior (Sutherland, 1924). – I.e. A labor works for money while a thief steals for money, both are driven by the same physiological needs but the need for money causes the labor to do honest work while the same need causes the thief to steal, & hence the physiological needs of clothing, food & shelter or any need for instance for which an honest man would work cannot be considered to be an explanation for delinquent behavior.

The Social Aspects to Crimes

The criminal law is conventionally defined as set of rules of human conduct, which have been drawn up by political authorities (Sutherland, 1924). The characteristics of these rules therefore are polytonality, specificity, uniformity & penal sanction (Sutherland, 1924). It is that criminal law that sets ground-rules as to what is permissible, & what is liable of punitive actions. Deviance – any behavior straying from the accepted social norms put in place by a political society & the authorities(Google, 2020), delinquency – minor crimes or misdemeanors especially those committed by younger people(Google, 2020) & crime has been doing the rounds for centuries, has been prevalent in society since civilization itself. From a sociological perspective, great scholars of sociology have put forth their brilliance in theories propagated by them & studies suggested. Inevitably, the name of Emile Durkheim – a French sociologist comes up who did groundbreaking work in studying the society as a whole, & set a new benchmark for sociologists to come.

Artificial Intelligence and Policy in India, Volume 1(2020)

Such a framework used to understand the functioning of society called the structural-functional theory was worked upon by this brilliant man, whereby the common colloquial term called “Durkheim’s Deviance” came into place. This simply deals with the causes of deviance in the society at large, & its functions. The structural – functional theory did not see the individual alone but understood the individual to be a part of the society at large, & worked upon the society’s collective hive-mind. The basic idea is that the society is a complex unit, comprised of many interrelated parts we see as individuals10. • Clarification of right & wrong – Reactions & responses to deviant behavior helps society as a collection of individuals understand that it is wrong (Sparknotes, 2020). E.g. A student cheating on a test, getting caught & consequentially failing the test causes other students to bear in mind from the next time that cheating is wrong & prohibited. • Unification of individuals in society – Acts of deviance, especially terror causes the society to experience a collective emotion of unity. E.g. Following the attack of September 11th in the U.S.A, there was a surge of patriotism all over the world, & people were united against terrorists. • Promoting social change – Acts of deviance by individuals can cause the dominant majority of the society to avail alternative norms. E.g. the act of deviance committed by Rosa Parks in the Montgomery incident in 1955 caused the U.S. Supreme Court to declare that segregation of people in public transport is unconstitutional. • The function of the deviant in society is significant, & perhaps delinquent behavior cannot be done away with totally, but measures can be made to limit the behavior to a minimalized extent. • The same way came into force the Strain – Theory, developed by sociologist Robert Merton – which suggested that at times people try to fulfill their aspirations through legitimate & institutionalized means but they fail owing to one thing or the other. Their road ahead is blocked, & when these people are inhibited from being able to achieve these goals through legitimate, accepted means, they experience frustration or strain, which can lead to deviance. Robert Merton also suggested that, when a person is unable to attain his or her goals through institutionalized means, he or she feels “anomie” or a feeling of being isolated or a feeling of disconnection from the society. Hence, it can be inferred that the structure of the society causes a lack of legitimate opportunities to the citizenry, & the citizenry experiences anomie, & ends up wanting to take the path of deviant or delinquent or even criminal behavior to achieve their goals, one reason is the fact that financial constraints cause multitudes of children to miss out on getting higher education or even any education at all, for these people – legitimate options are limited & perhaps even absent at times. They do not have any other option except to tumble down a dark alleyway of crime. There are many other reasons that hold a developing country back from eradicating crime once & for all. These theories have all aimed to prove one single thing, i.e. disorder in the society, disorder caused by the government, imbalances in harmony, all are followed by anarchism from the In Durkheim’s works he inferred the functions of deviance in a society; he suggested that deviance is a necessary evil in any society owing to its contributions in the social order as follows Affirmations of cultural norms – The punishment dealt out to criminals, reinstates the knowledge of the society of the punishment that would be dealt out to be them if they decided to dabble in crime. i.e. Unaccepted & condemned behavior is prohibited no matter what the circumstances are. E.g. sending a thief to prison reaffirms our cultural value that thievery is wrong. Just as the existence of god would come into question if the existence of devil was in question. 10

Artificial Intelligence and Policy in India

fundamental units of the society – individuals in the form of deviance, or delinquency, & that is a fact which cannot be refuted. Such an imbalance & perhaps the biggest imbalance in harmony, prosperity, & happiness caused to the society is the parasitic nature of corruption which is eating away at our motherlands bosom for decades. This same evil is the reason behind the diminishing faith of the citizenry in the remedial powers of the judiciary, & thus the doors to the halls of justice are a far fetch for the citizenry, owing to the image of delay & injustice owing to corruption that has imprinted itself upon the minds of the citizenry.

Corruption in the Judiciary

It is known to all that there is corruption at every rung of the ladder of hierarchy, in every nook & cranny of the governmental organizations as well as private organizations, waiting for its prey – the unsuspecting citizen of India. Nevertheless, the corruption prevalent in the lower levels of the hierarchy is insignificant in relation to the corruption that is prevalent in the highest level, which is the root of all corruption, the kingdoms of corrupt kings is corrupt as well. Aiming at the apex of the food chain, the corruption of the higher authorities comes to sight. The Constitution is a charter is a sacred written document of the rights of the citizens, but the guarantors of those rights are Courts of Justice, & when the citizens plead for alleviation of their grievances from these guarantors while knowing the corruption that is prevalent, it’s inevitable that their faith shall tumble & fall like a house of cards,& thus citizens refrain from approaching the Courts & at times even prefer to take the law into their own hands, & suffer prosecution rather than approach the Courts & be bereft of their rights for ever lengthening amounts of time, & incur expenses at an enormous rate & of an huge volume, while continuing to be suffering not only from the situation that made them approach the Court in the first place but also from the corruption prevalent at every stage of the process. Here follow a few instances by names that are evidence enough to understand & fathom the loss of faith in the Indian judiciary & law enforcement that the society by large is experiencing: • Justice S.P Sinha – was impeached & was the first Judge to be impeached in independent India, in the year 1949. He was impeached on the grounds of improper exercise of judicial functions, the result of which was the lowering of the dignity of his Office (Research Paper on Corruption in Indian Judiciary-National and International perspective). Five charges were framed against whom by C. Rajagopalachari the then Governor General of India himself. His decisions on these five cases were concluded to be un-judicial & based on ultrajudicial considerations which led the belief that it was a clear case of corruption (Plot4plot, 2020). • Chief Justice K. Veeraswami (Madras High Court) – In 1979, CBI was allowed to file a case against Veeraswami of disproportionate wealth/income. The Madras High Court Judge was charged under the Prevention of Corruption Act, 1946. To which the Chief Justice of India later remarked in a case that the sanction of the CJI is needed for a F.I.R or case to be filed against a sitting Judge (Research Paper on Corruption in Indian Judiciary-National and International perspective). • M.L. Singh, Mehtab Singh Gill, Amarbir Singh (Judges of Punjab & Haryana High Court) – were convicted of having approached R.P Sidhu the chief of PPSC, to make sure that their daughters & other kin top the examination. Mehtab Gill & Amarbir Singh resigned, & M.L Singh continued despite no work was allotted to him following 2 committee

Artificial Intelligence and Policy in India, Volume 1(2020)

investigations (Research Paper on Corruption in Indian Judiciary-National and International perspective). • Ranganathan Mishra, K.N Singh, M.H Kaniya, L.M Sharma, M.N Venkatachalliah, A.M Ahmadi, J.S Verma, M.M Punchhi, A.S Anand, S.P Bharucha, B.N Kirpal, G.B Petnaik, Rajendra Babu, R.C Lahoti, V.N Khare, & Y.K Sabharwal. (Judges) – The former Law Minister – Shanti Bhushan enclosed in a cover the names of 16 chief justices & handed it over to the Court, virtually demanding it to be read. Out of these 16, 8 were deemed to be definitely corrupt. Shanti Bhushan was the advocate, behind the setting aside of Indira Gandhi’s election in the year 1975 (Research Paper on Corruption in Indian Judiciary-National and International perspective). • Soumitra Sen (Calcutta High Court Judge) – In 2011, became the first Indian Judge to be impeached by the Rajya Sabha, owing to charges of misappropriation of funds. He was alleged to have appropriated Rs. 32 Lakhs, as a Court appointed receiver in 1993, in a suit between Steel Authority of India (SAIL) & Shipping Corporation of India over the supply of fire bricks. The committee investigating this matter reported that the charges were proved, & Sen had kept the money in his personal account, which he later returned back in 2006, after an order by the High Court (Research Paper on Corruption in Indian Judiciary-National and International perspective). These instances, are enough to substantiate the notion of lack of faith in the Indian judiciary & moreover as to be proven later the need for Artificial intelligence to be incorporated into the judicial machinery. 5.1

Criminal Profiling – what it means

Crime, deviance, delinquency demand secrecy & it is this secrecy whose veil cannot be pierced by conventional ways of law enforcement & psychiatric compilations such as psychiatric profiling, thus need arises to bring forth the hound - the procedure that can be efficiently utilized to apprehend a perpetrator, the method of - Criminal profiling. The age-old method of criminal profiling has been defined as a technique utilizing which, many characteristics of a perpetrator or perpetrators can be predicted based on the way the crime was committed, i.e. the behavior exhibited in the commission of the crime. Criminal profiling in itself is conceptually ancient & indicates the fascination that humans possess when it comes to assessment & prediction of criminal behavior. It is a commonplace mistake on part of the unaware citizenry to consider criminal profiling to be a biased procedure against individuals who have an outward appearance that is in popular opinion considered to be significantly expressing a person’s affection to crime or high chances of having committed a crime. But, that is a mere fallacy, not a myth but a hoax. Since, the art of criminal profiling does not relate to an individual or the accused, but is the examination of a crime to interpret the behaviors evident in the commission of that crime (Kocsis, 2006). This behavior when analyzed emanates a description of the individual who is the most likely to have committed the crime (Kocsis, 2006), irrespective of whether it’s the accused person or not. Nevertheless, the factor that behavior is to some extent a reflection of a person’s personality has persisted as a corner-stone of modern day science of psychology (Kocsis, 2006). Interestingly, the earliest applications of the symbiosis of behavior & personality are exhibited in the form of a criminal investigation in England in the Victorian era. Following a series of murders, dubbed the first serial killings of the modern world. The perpetrator came to be

Artificial Intelligence and Policy in India

popularly known as Jack the Ripper, owing to his very own suggestion of this name to the local newspapers. The London’s Criminal Investigations Division in 1888 sought the assistance of a physician known as Dr. Thomas Bond. Dr. Bond was consulted to examine the available evidence concerning the murders. The ripper has been said to have committed eleven murders, but has been formally linked to only five, known as the “canonical five”. Notable components of Dr. Bond’s report included evaluations of the behaviors exhibited during the course of action. Whereby It was mentioned clearly by him that the five murders were committed by the same person, in the first four – the throat had been slit from left to right, in the last case however, due to the extensive number of mutilations the direction of the cut could not be discerned, & it was opined that the woman was lying down when murdered (Kocsis, 2006). These details were of a forensic nature however. Also, encompassed were other probable characteristics of the perpetrator. Contrary to the popular opinion of the Jack the Ripper having anatomical knowledge, Dr. Bond opined that the perpetrator did not have any anatomical knowledge be it human, or animal. I.e. The perpetrator was neither a medical practitioner, nor a butcher. It was also opined by him that the perpetrator suffered from periodical attacks of homicidal & erotic mania (Kocsis, 2006), & it was suggested that the external appearance of the perpetrator was quite likely to be that of an inoffensive looking man, probably middle aged & neatly & respectably dressed. He would be solitary & eccentric in his habits, & possibly would be a man without a regular occupation & living on a small income or pension (Kocsis, 2006). However, the case of the Whitechapel murderer remains unsolved & therefore prevents any evaluation of the accuracy of Dr. Bond’s analysis. Nevertheless, this analysis is an early historical example highlighting the fundamental concept of criminal profiling & the core tenets of what typically comprises a criminal profile (Kocsis, 2006). Another instance presents itself when Dr. Walter Langer was contacted by the US Office of Strategic Services (predecessor to the contemporary CIA) to conduct a psychological evaluation of Adolf Hitler (Kocsis, 2006). Dr. Langer discovered that Hitler’s likeliest reaction if confronted by defeat would be to commit suicide rather than face the humiliation of possible capture & trial for his actions. In lieu of Hitler’s ultimate fate, Dr. Langer’s prediction proved insightful (Kocsis, 2006). These instances however can be argued against owing to the lack of evidence as to whether the predictions even were accurate or not. However, the progenitor of contemporary criminal profiling, is the work of Dr. James Brussel in 1950s, Brussel was a psychiatrist whose skills in evaluation of crimes led his being consulted on a number of infamous cases. The most wellknown amongst these were his involvement in the investigation of bombings that plagued the city of New York & were termed as the work of the “Mad Bomber of New York” (Kocsis, 2006). Having considered the available case material in relation to the bombings, Dr. Brussels constructed a profile that identified numerous characteristics that were subsequently found to match the attributes of the bomber George Metesky. Most noteworthy among his predictions were Brussel’s conclusion that the bomber would be a very neat person. This, according to him was manifested in a number of behavioral attributes surrounding the crimes, such as the apparent care the bomber demonstrated in drafting letters to the authorities when seeking to communicate & explain his actions. He also opined that the perpetrator would be a neatly presented individual with an affinity towards blue double-breasted suits that would be probably worn with the jacket fastened. Following the apprehending of Metesky, he was allowed to change out of his pyjamas before being taken into custody, & he changed into a blue double –breasted suit, which he wore with all the buttons fastened (Kocsis, 2006).

5.2

Artificial Intelligence and Policy in India, Volume 1(2020)

Biographical Structure of A Criminal Profile

Elaborating on the exact objectives & structure of criminal profiles is not straightforward at all. Partially owing to the differing disciplinary perspectives surrounding the technique & partially because of the ever-developing & diversifying conceptions of the practice. Putting aside the limitations caused by variations, an underlying consensus can be discerned as to the contents of what categorical characteristics are to be present in a criminal profile, which is biographical in nature (Kocsis, 2006). The characteristics are as follows – • Likely demographics (age, gender, etc). • Legal history, including any antecedence (I.e. History of prior criminal offences/ convictions). • Vocational background (I.e. The work the offender is likely to be engaged in, if any). • Family Characteristics (I.e. the likely background of the offender’s family). • Habits & Social Interests (sports, hobbies, or other interests that the offender may have). • Mode of transportation (Type of vehicle, if owned). • Various personality traits (demeanor, appearance, etc). (Kocsis, 2006) • Along with such biographical characteristics, at times approximate locations of an offender’s residence, as well as other geographical characteristics are also included (Kocsis, 2006).

ARTIFICIAL INTELLIGENCE – what’s all the uproar about? “Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an ‘intelligence explosion,’ and the intelligence of man would be left far behind. Thus the first ultraintelligent machine is the last invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control.” Irving Good (Tegmark, 2017). Artificial intelligence or machine intelligence refers to thinking capabilities & intelligence demonstrated by machines in contrast to the human intelligence while ultra-intelligence refers to super intelligent machines. In order to further delve into the workings & discussions on artificial intelligence a few concepts should be understood first to better grasp the arguments. Such a question comes up, referring to what the meaning of life even is. We know that all things are made up of atoms congregating into molecules comprising matter. In a broader aspect, matter is not what is replicated to form life or anything else, but information made of bits is. This replicated information shows the placement of atoms. When a bacterium creates a copy of its DNA, no new matter is made, but the same placement of atoms or the same information is replicated into a newer copy. In other words, we can think of life as a self-replicating information processing system whose information (software) defines both its behavior & its blueprint (hardware) (Tegmark, 2017). Hence, three stages of life can be said to be existent – biological, cultural & technological. While the biological evolution can be understood as the evolution of apes from unicellular organisms, cultural evolution can be said to be the evolution of modern day humans from apes, while technological evolution can be said to be the evolution of narrow artificial

Artificial Intelligence and Policy in India

intelligence systems from ancient computer systems (Tegmark, 2017). Yet, despite the newest innovations, all life forms that we know of remain biologically limited by their hardware. In relation to the arguments & controversies of whether an Artificial General Intelligence is possible, two categories of ideologies are existent – • Digital utopianism – the ideology that believes in the next step of cosmological evolution being the age of AI & consequently AGI, & believes that it will be beneficial to humanity (Tegmark, 2017). • Techno-skepticism – the ideology that suggests that building an AGI is so hard & so demanding that it’s hundreds of years afar if at all possible, & thus no need for worrying about it is present as of now (Tegmark, 2017). As to the other people who oppose the advent of AI in the first place they can be simply replied to by saying that – being threatened by a machine whose goals are misaligned with ours, suggests that its these goals of the machines that threaten us & not the fact that whether the machine is self-conscious or experiences a sense of purpose (Tegmark, 2017) In relation to the latest developments, neural networks have now succeeded in transforming both the biological & artificial intelligence, & have recently initiated its dominance in the subfield of AI known as machine learning (Tegmark, 2017). Currently the most well-known model of neural networks represents the state of each neuron by a single number; each neuron updates its state at regular time steps by simply averaging together the inputs from all connected neurons, weighting them by the synaptic strengths, optionally adding a constant, & then applying an activation function to compute the next step (Tegmark, 2017).

5.3

Successes of Deep Learning Procedures

In 1997, IBM’s Deep Blue managed to outplay the then world chess champion Gary Kasparov (Tegmark, 2017), IBM’s Watson dethroned the human world champion in the quiz show jeopardy (Tegmark, 2017). In 2015, Google DeepMind released an AI system using deep learning that was able to master dozens of computer games like a kid would – with no instructions whatsoever – and it soon was able to play better than any human (Tegmark, 2017). However these AIs depended more on custom coding than learning, In 2016, the same company built AlphaGo, a Go- playing computer system that used deep learning to evaluate the strengths of different board positions & defeated the world’s best player (Tegmark, 2017). In 2014, a team at Google led by Ilya Sutskever made an AI that could recognize images by using neural networks (Tegmark, 2017). Dares & Prospects. Everything we love about our existence & today’s life is the product of human intelligence, being able to amplify it with the power of artificial intelligence, shall enable us to make life even better. Even modest progress in the field of AI might translate into major breakthroughs in the fields of science, arts & technology, while causing decreases in the amounts of disease, accidents, injustice, poverty, drudgery, etc. (Tegmark, 2017).

Artificial Intelligence and Policy in India, Volume 1(2020)

Hence, many important questions come up which we should work upon by furnishing ourselves the answers to. A few of those are – 1. How can we make future AI systems more robust than today’s, so that they do what we want without crashing, malfunctioning or getting hacked (Tegmark, 2017)? 2. How can we update our legal systems to make it more fair and efficient and to keep pace with the rapidly changing landscape (Tegmark, 2017)? 3. How can we make weapons smarter and less prone to killing innocent civilians without triggering an out-of-control arms race in lethal autonomous weapons (Tegmark, 2017)? 4. How can we grow our prosperity through automation without leaving people lacking income or purpose (Tegmark, 2017)?

The AI & Law Question

“Our intelligence is what makes us human, and AI is an extension of that quality.” – Yann LeCun (2019). It is only human for us, the dominant species on the planet Earth to make mistakes, & at times those mistakes end up making us suffer beyond measures, or worse still, those mistakes make others suffer. Humans are imperfect, and it is this imperfection that demands a better efficiency in the handling of matters of grave importance. Such as cases in the Courts of Law, & law enforcement procedures to inhibit crimes or apprehend criminals. Not only would the incorporation of AI into the judicial machinery make a perfectly efficient system of judicial governance & administration of justice but would also make for a judiciary system free from corruption. Recent studies have also shown that the feeding of a system with prisoner data can predict the probability of the prisoners returning to crime (Tegmark, 2017)

AI & Law Enforcement

AI can be put to use in a few sub-fields in law enforcement. Such as – 1. Public Safety Video and Image Analysis - Video and image analysis is commonplace nowadays in law enforcement communities in order to obtain information regarding people, objects, and actions to facilitate criminal investigations. However, the analysis is very laborintensive & requires personnel with subject matter expertise. The procedure is also prone to human error owing to the vast amount of informational traffic coming in continuously that needs to be monitored. AI can be used to overcome these difficulties with the help of facial recognition capabilities of the AI, while linking it with criminal databases to easily recognize any imminent threats anywhere at any time. The AI can easily learn & develop its own facial recognition algorithms after some time & can become an expert, while the required personnel for this job can be allocated other jobs (USING ARTIFICIAL INTELLIGENCE TO ADDRESS CRIMINAL JUSTICE NEEDS, 2019). 2. DNA analysis - AI can also benefit the law enforcement community in a scientific field from an evidence processing standpoint. In case of forensic DNA testing, this has had an impact on the criminal justice system over several years in the past. Biological evidence, such as blood, saliva, semen, and skin cells, are transferrable via contact with people and objects

Artificial Intelligence and Policy in India

during the commission of a crime. The advancement of DNA technology has increased the sensitivity of DNA analysis, allowing forensic scientists to detect and process low-level, degraded, or otherwise unviable DNA evidence. As a result of increased sensitivity, smaller amounts of DNA can now be detected, leading to the possibility of detecting DNA from multiple contributors, even if at very low levels. Hence, convolution or separation of identities of DNA contributed by multiple perpetrators poses a challenge to forensic experts. AI has the potential to address this challenge. DNA analyses produces large amounts of complex data in electronic format; these data contain patterns, which might be beyond the range of human analysis but may prove worthwhile with the increase in sensitivity of these DNA analysis machinery & increase in efficiency of AI, AI can overcome these challenges (USING ARTIFICIAL INTELLIGENCE TO ADDRESS CRIMINAL JUSTICE NEEDS, 2019) 3. Gunshot detection - The unearthing of pattern signatures in gunshot analyses is another area in which AI algorithms can be used. In order to identify several gunshot attributes & allocate several attributes to the gunshot in the forms of predictions such as caliber, ammo type, number of firearms present, type of firearm, etc. (USING ARTIFICIAL INTELLIGENCE TO ADDRESS CRIMINAL JUSTICE NEEDS, 2019). 4. Crime forecasting - Predictive analysis is a complex procedure utilizing large volumes of data to forecast and formulate potential outcomes. In criminal justice, the job mainly rests with the police, probation practitioners, and other professionals, who have to gain expertise over many years, & through continual processes of trial & error, putting to risk the lives of numerous civilians with each & every error. The work is time-consuming and subject to bias and error. With AI, volumes of information on law and legal precedence, social information, and media can be used to suggest rulings, identify criminal enterprises, and predict and reveal people who are at risk from criminal enterprises (USING ARTIFICIAL INTELLIGENCE TO ADDRESS CRIMINAL JUSTICE NEEDS, 2019).

AI, Criminal Profiling & the Judiciary

There are numerous instances of judicial systems that already employ artificial intelligence tools in criminal proceedings. The Indian judiciary stands only to benefit from the incorporation of AI into its systems. In order to pull back the faith of the citizenry of India into the Justice system. The faith that has been lost owing to the numerous cases against jurists, advocates, law enforcement officials & politicians. India herself stands to benefit from the incorporation of AI into the judicial machinery. While, ridding our motherland of corruption in the class A citizenry once & for all. Here follow a few ways AI can be incorporated into the judicial machinery – 1. The use of AI in bail decisions - New Jersey and California are two states which have shifted from cash bail and fixed bail schedules towards new risk assessment systems. In California, the underlying principle of this recent change in 2018 “is that a suspect will be evaluated on the basis of risk to public safety and the likelihood of not appearing in court, rather than on his or her ability to post a certain bail amount.” (Benoît Dupont, 2018) The aim of this new maneuver is to make sure that judges make pre-trial detention or release decisions based in part on empirical systems which are entirely analytical & determine whether a person has a high probability of fleeing or committing another crime & not just decide what’s to happen to the accused person based on merely the person’s capability to pay a certain amount of money to excuse himself from jail time. It’s high time that India starts the usage of such

Artificial Intelligence and Policy in India, Volume 1(2020)

systems in order to cause a better anti-criminalistics pattern which shall go down as a no tolerance policy to crime & therefore cause a drop in crime rates. 2. The use of AI in sentencing - Jurisdictions vary in how they determine sentencing decisions, but all of them tend to be based upon aspects such as the severity or classification of the crime, if the accused is a repeat offender of the crime or has committed any other crime in the past or any pre-existing guidelines suggesting certain punishments for certain crimes. It is humanely impossible for a judge to have all such details by heart, & no matter how much of information is provided to a judge to pass a decree on the situation, chances of errors being made are always present. Hence the usage of AI in sentencing can significantly improve efficiency, & any chance of injustice due to errors or mistakes or even corruption can be minimized (Benoît Dupont, 2018). A prominent example of AI being used for sentencing is the usage of COMPAS (Correctional Offender Management Profiling for Alternative Sanctions), however accusations of the software being racially biased were made in 2016 (Benoît Dupont, 2018). 3. The use of AI to compile a criminal profile of the accused – Feeding in the biographical requisites for a criminal profile & feeding in all available information about the accused, can make an AI be easily able to compile a criminal profile for the accused be it for the usage in a Court of Law or be it during the investigation of a crime. This profile can further be used by judges to rule on whether this accused is a particular threat to society & should be out on parole or not. This profile can be even used for the purpose of sentencing, where the forensic evidence & all evidences are fed into the system & the AI draws to web together & links the evidences while bearing in mind the criminal profile itself. Whereby, a perfect conviction or acquittal can be made, as efficiently as possible without causing an eternity of wait to the citizenry. 4. The use of AI to compile a geographical profile of the accused & deducing whether the accused is deviant – Feeding the system with the records of the accused, & the geographical location of the accused’s residence, the AI can easily discern whether the area of residence of the accused is prone to hostilities of a criminal nature, or deviancy or delinquency, & thus can give an insight as to the presence or absence of criminalist patterns around the accused. 5. The inference of guilt using AI – AI can be fed the records of the accused while, the criminal profile & geographical profile can be made by the AI, & the AI can itself identify the accused as guilty or not-guilty based on the amount of evidence provided. The AI has to be fed the principles of criminology & the AI should understand the method of criminal profiling perfectly & shall use a deep learning process using neural networks to ultimately synthesize whether the accused is guilty or not. AI – the harbinger of Justice . Artificial Intelligence can be used as an advisor to judges, acting only as an advisor & not a judge itself until the arrival of the AGI which can think, feel, and have instincts & impulses of a human. The doctrine of natural justice to be preserved & the doctrines of judicial governance to be preserved needs there to be a presence of the concept of “Justice, equity & good conscience” which is not possible as of now in an artificial intelligence, owing to its lack of humanity. Hence, AI can be used as a personal advisor to the judge, this shall facilitate speedy justice, since the judge shall not have to wait for day, months or even years to clearly understand the

Artificial Intelligence and Policy in India

situation & shall not have to rely upon the advocates completely to either prove the absence or presence of guilt, in order to grant an acquittal or conviction. This maneuver shall undoubtedly renew the faith of Indians in the Judiciary. However, in order to minimize any & all chances of corruption creeping into the judiciary despite there being present an AI, a review board should be put into place which shall go upon the decisions of a judge, shall review it whenever an appeal to a higher court is made, or whenever the advice of the AI although seemingly perfect is not accepted, & a ruling to the contrary is made. Following which, the review board shall ask the concerned judge to provide a report as to why the advices of the AI were not abided by, to which the judge shall be liable to answer. If the Judge answers satisfactorily, the review would be over if not, a formal investigation should be launched into the conduct of the said judge.

Conclusions and Recommendations

“It’s every man’s business to see justice done.” - Sir Arthur Conan Doyle (Crime, 2020) The age of the AI cometh, & there’s no stopping it, the question of an AGI is not something to be concerned about as of now, when todays AI can be effectively used for the right purposes, while continuing research in the field to bring forth an AGI someday. The blessing of AI can undoubtedly redeem the sinking ship of Indian justice, due to political or fiduciary injustice & inclination. The AI can be brought into the Indian judicial machinery as well as the Indian law enforcement community to take it by storm & leave behind the rubble & debris of an ancient epitome of injustice, while driving the citizenry of India forward into the flush of a new dawn of Justice where every man shall have his rights & have his rights respected. Where, delay, injustice, corruption, paperwork, & uselessness are not what come first to the mind of people when talking about Courts or thinking about approaching the courts to find alleviation to their grievances. Where, people are truly proud of what their country has to offer, where people are united in a nation where criminality is truly condemned, without corruption. Where lawyers are not looked upon as frauds or liars but the noblest of the noble. With certain few policies, enactments & organizations, the age of AI can come & come to stay. Justice can invariably prevail. Where people are satisfied, & do not experience anomie, where deviance, delinquency & crime have an all-time low owing to perfect decisions by the government, causing happiness for one & happiness for all. 1.

The creation of a project team of researchers for the sole purpose of a project of putting AI into the rounds in terms of the Justice system. In collaborations with technology giants, & law institutes. This team can comprise of three divisions. – ■ Fillers – Engineers assigned to design, create & improve the “Reservoir AI”, with help of lawyers to decide which precedents/news/amendments/statutes to specifically use in order to avoid any chance of biasing of the AI. ■ Profilers – Criminologists, lawyers, psychologists & engineers dedicated to the creation of a “Profiling AI”. Where, the psychologists shall be constantly evaluating the members of the profilers to identify any biases that might end up getting programmed into the system. The criminologists & lawyers shall be constantly researching on newer methods, developments in the field of social profiling. The engineers shall be responsible for creation & improvements of the said system.

Artificial Intelligence and Policy in India, Volume 1(2020)

■ Jury – Lawyers & criminologists & engineers whose objectives shall be the creation & required upgrading & updating of the “Advisory AI”. While, the lawyers shall look into the ethics part of the creation, the criminologists with the help of the engineers shall continue looking for biases in the system. 2.

The development of a “Reservoir AI” to be used in the judicial system. This AI shall be constantly fed with precedents, news, amendments, and statutes. The only object of this AI should be to learn everything there is to learn about law, & to furnish the judge with whatever case law/law/provision/incident/observation that is relevant to the case at hand. The development of a “Profiling AI” to be used for criminal profiling. This AI shall be fed with principles of criminology & should be able to understand the concept; this AI should be able to compile a criminal profile of an individual based on user defined input. Such as name of offender, offense, age, sex, place of birth, vehicle type if any, medical records, previous criminal affiliations & allegations, etc. This AI shall then formulate a criminal profile of this individual based on the its understanding of the principles of criminal profiling, & facts surrounding the crime as well as evidence found at crime scene. This shall be further supplemented with observations by forensic science experts. The objective of this AI would be to use all the observations that have been made surrounding the crime to establish a criminal profile. This AI shall also access the geographical location of the accused & the surroundings of the accused to detect if the environment surrounding the accused is deviant, or delinquent. The development of the “Advisory AI” that shall take the deductions made by the profiling AI into mind, & shall further take the information given by the “Reservoir AI” into mind & shall formulate an advisory opinion of the case. I.e. The AI shall give a decision by itself, as an advisory opinion to the concerned judge. The formation of a review board to look into any appeals made against the decisions of a judge based on the Judge passing any sentence or decree contrary to the advice of the AI. Whereby the judge shall be liable to submit a report as to his actions. If unsatisfactory, an investigatory committee would be asked to launch an investigation against the conduct of the concerned judge. All of this shall be done in confidentiality, while the person appealing shall be awarded a protection program, so that the person shall be free of threats. Companies can be contracted to provide AI support to the law enforcement departments, which shall be of much help to the AIs in the judiciary. So, those certain facts can be understood & gotten by the AI firsthand. Such as, digital forensics, ballistic reports, gunshot detection reports, objects recognized, faces recognized, DNA analyses reports, etc. This shall further help the “Profiling AI” to compile the profile. Spreading of awareness about the field of AI, Law & criminology, so that more researchers come into this field, encouraging research institutions, premier law schools & societies to indulge in this field so that more can be achieved.

References 1. Artificial Intelligence. Goodreads. [Online] [Cited: May 01, 2020.] https://www.goodreads.com/quotes/tag/artificial-intelligence.

Artificial Intelligence and Policy in India

2. BenoĂŽt Dupont, Yuan Stevens, Hannes Westermann, Michael Joyce. 2018. Artificial Intelligence in the Context of Criminal Justice . s.l. : Korean Institute of Criminology, 2018. 3. Corruption. Goodreads. [Online] [Cited: May 03, 2020.] https://www.goodreads.com/quotes/tag/corruption. 4. Kocsis, Richard N. 2006. Criminal Profiling Principles and Practice. Totowa, New Jersey : Humana Press Inc., 2006. 5. Plot4plot. Indian Corrupt Judges. [Online] [Cited: May 03, 2020.] http://indiancorruptjudges.com/Plot4Plot/008_07.htm. 6. Research Paper on Corruption in Indian Judiciary-National and International perspective. Anthala, Ram Hari. Chandigarh : Punjab University, Dept. of Law. 7. Sutherland, Edwin H. 1924. Principles of Criminology. Bloomington, Indiana : J. B. Lippincott Company, 1924. 8. Tegmark, Max. 2017. Life 3.0 - Being Human in the Age of Artificial Intelligence. New York : Alfred A. Knopf, 2017. 9. 2019. Top 10 Artificial Intelligence Quotes That Will Inspire You. DZone. [Online] May 07, 2019. [Cited: May 04, 2020.] https://dzone.com/articles/top-10-artificial-intelligence-quotes-that-will-in. 10. USING ARTIFICIAL INTELLIGENCE TO ADDRESS CRIMINAL JUSTICE NEEDS. Rigano, Christopher;. 2019. 280, s.l. : National Institute of Justice, 2019.

Artificial Intelligence and Policy in India, Volume 1(2020)

The Legal Viability of Patenting Ritam Khanna1 1

Honorary Research Member, Indian Society of Artificial Intelligence and Law ritamk3@gmail.com

Current Trend of AI- Patents in-and-out of the industry

The Global figures of the patent filings increased at a rate of 5.2% as per the reports of the World Intellectual Property Organization (WIPO) in 2019. (World Intellectual Property Organisation, 2019). Asia is ahead of all the continents in this trend. Developing countries like China and India are important players of this trend and account for more than 50% of the patent filings around the globe. In India, the developers of Artificial Intelligence (AI) contributed a share of 89 % of patents filed in 2019 (TATA and CII Report, 2019). With every minute addition of features to the Artificial Intelligence Machinery, a constant change of the technology and human innovation, legal implementation of the patenting is not effective enough. Furthermore, a 2018 tender (Rajasthan HC, 2018) posted by the website of Indian Patent Office (IPO) gives out a notice Inviting Expression of Interest (EOI) for making use of Artificial Intelligence (AI), Blockchain, Internet of Things (IoT) and other latest technologies in the Patent Processing system of IPO. Implying that the change in the industry is not only affecting the stakeholders outside but inside the industry. 1.1

The Economics Behind the Patenting Laws in India

Patents represent, like other IPs, a competitive resource and could confer a competitive advantage. Industrial Properties and patents act as catalysts in the business context. (R&D and Patenting by Firms in India in High and Medium High Technology , 2014)Industrial Property can influence the revenues generating potential and capability of enterprises and can affect business sustainability and business risk. The economics behind the patenting is a game of cost. (Kaufer, 1989)The imperfect market, like India, run by the monopolization of the products that are innovative, highly in-demand or whose demand can be created by the business minds. The incentive to make the innovations would be lost if the patents, which is the source of granting the monopoly to the innovators, is not given to the start-up innovators. It is a problem with the current AI ownership and patenting as humans can be the only innovators under the Patent Law. Furthermore, Patenting Innovation has an added economic edge. It creates a gap between marginal social benefit and marginal social cost. Although, considered a small price to pay for progress, but another bit of imperfection may not make all that difference to total social product for the big developers and companies. In third world countries, very rarely are patents included in licensing agreements. The patents continue to be held by the parent companies and subsidiaries may have their use, if the patenting firms choose to invest in the particular; country. Most licensing agreements with third world companies generally cover only the use of blueprints, drawings, managerial expertise and trademarks. The less industrialised the country, generally speaking, the smaller is the proportion of patent licences to total know-how agreements. From this point of view, India

Artificial Intelligence and Policy in India

belongs firmly to the group of poor third world countries (The Composition -of Licensing Fees and Arrangements as a Function of Economic Development of Technology Recipient Nations, 1980). It has been argued that if technology transfer is to be effected successfully and cheaply between two countries, the difference in their level of technological development must not exceed a certain threshold limit. (Delome, 2008) Artificial Intelligence technology has its impact on all the three economic sectors as per the NITI Aayogâ&#x20AC;&#x2122;s report 2018. (Aayog, 2018). A study by the Microsoft and IDC Asia indicating that the innovations are to increased by x2.2 times in India implicating the increase in productivity (Microsoft and IDC, 2019). The cost of the patents is highly dependent on the ownership, the level of innovation along with the actual cost put forward in bringing the output innovation. Legal implications have a drastic effect on cost. Thus, if the cost of filing and owning the patent is high, naturally the large-scale development of the innovation is not a certainty ( A Framework For The Economic Analysis Of Patents In Business Context, 2014). 1.2

The Issues of patenting the AI

The 2005 Amendment onwards the Indian Patent Act has redefined the three criteria of the patentability as the granting of the patent is based on the following first, Inventive step, second, novelty and third, the Usefulness ( Intellectual Property Rights And Innovation In Developing Countries: Evidence From India, 2008). It had shifted the burden of proof to the infringers. (Impact of patent policy changes on R&D expenditure by industries in India, 2017) Looking at the current scenario, it can be a loophole, as, unlike other tangible innovation. The innovations in deep learning do not have proper compartmentalization. Thus, the procedure of patenting is difficult for any innovator. On reading of Â§ 2(1) and (y) Â§ 2(1)(s) Patent Act, 1970, it observed that it is unable to accept the ownership of Deep Learning Machines as the deep learning machines tend to have stimulations similar to human cognitive skills. These kinds of machine making are intensive and employ different software and programmes that might be patented software programmes. But in a hypothetical, an innovation made by the deep learning machine-like AI machine operating on a medical procedure invents a new procedure of performing a surgery. This raises the following questions: Who is the owner of the Patent: the Machine or the owner of the machine? Section 6(1) of Indian Patents Law of 1970 categories any citizen into different innovators, inheritors of the innovators and the foreign patent filers to encompass the patentee of the claimant. It explicitly restricted the criteria of persons to the legal as well as natural persons including the government. The question of the AI being a human intelligent creation and thereof making a sub-invention in its operation or any other kind of infringement is disputable. Patents are exclusively meant for corporations or the person with a human ability to create innovations. Although, there is a thin line of difference between the future deep learning machines creations and the human-created inventions. Logically speaking, the AI innovations should be entitled to hold the patentee title of the innovations created by the nature of the human experiences combined with the proper knowledge to develop their foundations of any innovations. They will be as good as any human inventions. However, the legal loophole is that the definition of the patents applies to natural persons or corporations. Granting AI to the original makers or the legal entities would be a

Artificial Intelligence and Policy in India, Volume 1(2020)

problematic array of civil lawsuits seeking compensation for any development of infringements alike. The Companies thereof will argue that the end-users will be the liable party for such causes of the infringement, which is legally correct deposition as they comply with the definition of ‘assignee’ under the Patent Act in a broader sense. The other side of the corporations has led on to argue that the inventors should be the patent holder as economically it will be having a first-mover advantage which can make the innovations of the innovations safe and prosperous for a lifetime. There is a utilitarian perspective that the inventor ought to be allowed a right to the benefit of his invention for some certain time, as an encouragement to men to pursue ideas, which may produce utility. (Intellectual Property in New Technological Age , 2016)If more of the patents are allowed to the AI then they might lead onto diminish the incentive need of the humans ( Thinking About Thinking Machines: Implications of Machine Inventors for Patent Law, 2002) and a create overriding monopolies for the companies who own them by reaping interest on the inventions. Patentability of AI A dividing line between what is patentable and what is a mere extension of existing knowledge is also grounded on ‘human capabilities’ by comparing what the notional ‘person skilled in the art’ would have been able to discover without unusual effort and the additional step of human ingenuity made by the inventor. (Vertisky, 2018). Patent claims that is directed to abstract ideas (e.g. a mathematical algorithm), natural phenomena or laws of nature are not eligible for patent protection (1981). The Supreme Court of the United States explained that “they are the basic tools of scientific and technological work,” and that granting monopolies on those tools through patent rights might impede innovation (2017). In India, we have a similar position as in the US but artificial intelligence has been patented on the parameters judged based on their genuineness. The NITI Aayog’s AI paper explores the use of blockchain and AI in the governance of different departments of the Government of India and also, for enabling several socialist reforms. Due to this reason, the AI systems that are responsible for innovating ideas will be encroaching the territory of the “human genuiness” of the patents. On the bright side, the AI in the government sector can channelize the cost-cutting patenting for the sake of the socialist reforms or public welfare. In Alice decision of the United States, the AI which capable of mimicking the human mind cannot be considered for patenting as they are nothing but ordinary functions of humans (2014). United States Patent Office has affirmed the same and overturned the opportunity to make their laws AI inclusive (United States Patent and Trademark Office, 2020). In India Two AI software has been granted patents (TATA, 2020). However, the same is mere software that is aiding in the performance of the AI not consisting of AI invention to be claimed as a patent. The Software patent also creates the problem of ownership and thereof the monopolizing of the AI by a single unit in the market which will be contrary to the socialist reforms that the government is aiming through the NITI Aayog’s AI policy. Who is liable for the infringement of a patent by the AI machines? The Another aspect of the AI’s inability to be compatible with the patent laws in India is because as soon as any AI start-up or the company would file for the patent, a claim for infringement would stand. The AI are relying on their intellectual capabilities on the software or database system and cannot be considered innovations until the software have been recognised as innovative or unique.

Artificial Intelligence and Policy in India

The US patent law acknowledges, “Induced infringement”, as whoever actively induces infringement of a patent shall be liable as an infringer ( World Economic Forum). It means that the alleged inducer must have knowingly aided another’s a direct infringement of a patent. Through this doctrine, the AI cannot be held responsible but the innovator will be held liable for the effect of it being a helper to the cause. As per the European Parliament resolution came up with the recommendations to the Commission on Civil Law Rules on Robotics (hereinafter European Parliament Resolution), AI cannot be held liable per se for acts or omissions that cause damage to third parties (e.g. patent infringement). Instead, AI’s act would have to be traced back to a human agent, such as its manufacturer, operator, owner or user, if that agent could have foreseen and could have avoided AI’s harmful behaviour. Holding a product’s manufacturer liable for patent infringement is common practice in patent litigation (SUBBARAM, 2015) and it is considered suitable because the developers ultimately create the AI (that infringes the patent). From a socialist perspective, they are usually in a relatively better position to foresee the infringement than the end-users, and have likely derived economic value from the AI (e.g. selling AI to the end-users). However, the only concern is that if all the burden is shifted towards the maker then it will be subject to the criteria that whether or not the innovator had foreseen such infringement as it created the human mind replica, but is no tamer of it?

The Prospective of Patent Litigation and Artificial Intelligence

The Patent filings are sought to be revolutionized by the way of the introduction of the AI from the process of filing them, to the step of hearing of the claims and granting the patents to the patentee. Presently, the drafting of the techno-legal document, namely, the patent specification (which contains the details of the invention to be protected), filing and processing of the applications for patents. It is conducted by the help of the patent agents but with the intervention of the AI, the applications it might be analyzed by the AI for granting patents. The Patent Office is also supervising authority of the procedure of granting the patent and regulation of it thereof. It involves the process of passing on the information of the updates in the procedure of patents systems. However, it suffers from the defects of the unawareness on the part of the patent filers and the gives them the trouble of subscribing the copy of the journal of the same, making it a restrictive pool of the applicants which include informed agents and lawyers. This proves to be a problem for the Start-up AI innovators who want to patent their innovations and lack enough resources. According to Department of Industrial Policy & Promotion (DIPP), 6096 entities have been recognized as start-ups that are majorly focusing on Machine Learning for the category of products, services and applications that they provide today (Machine Learning Startups To Watch In 2018, 2017). Start IP India initiative by AI start-ups had 2018 as their best funding year, with the investment of $99.5 billion which is approximately 10% of last year's total investment of $9.33 billion. ( Why Microsoft Leads The AI Patent Race Going Into 2019 , 2019) The effect of keeping the cost of the patent high with the complicated procedures and knotty lengths of patent papers creates a spurring cost for applications to protect their nascent inventions. European Patent Office latest report addressed the problem of the patents for the start-ups and balanced the option of interests of the small industries to legal compliance. A possibility of implementation of patents and copyright together in the inability of each of the laws to

Artificial Intelligence and Policy in India, Volume 1(2020)

protect them wholly was suggested as an option. Copyright can provide a collaborative basis for the low-end algorithm and on the other hand, the Patent Laws can protect high-end innovations (Patenting Artificial Intelligence Conference Summary, 2018). On a brighter side, the Indian Patenting Law firms and analyst have been planning to utilise the AI to its optimum (Financial Times, 2020). In some cases, technology can unlock value for client companies by finding patents that have been forgotten because they have moved out of a product line. The companies can then generate revenue by selling them. As well as saving time and money in the patent application process, AI technology is a powerful tool in preventing IP infringement. It is even possible that de-shelving of the patent application will lower the cost for the overburdened Judicial System in India.

Conclusion

The Indian Patent Act suffers from the inability to distinguish abstract innovations and other kinds of innovations. Since the AI is a technologically and advancing field in the upcoming years the patent laws are unable to make the incentive enough for the companies or start-ups to work with AI. Their innovations are at the risk of suffering from the being unprotected by the law and give due credit for the work that innovators undertake. The credibility of the innovators is reduced when they are subjected to the infringement cases on them. No AI can be build filled with unique innovations from the assembling of the parts to the performance. With a great amount of paperwork to set up and a huge investment in the terms of procuring the rights from each circuit board to every software, that potential will be acquired for its services in the field. On the Government Sphere, the government laid out a Defence IPR policy for Developed software products in India by Ministry of Defence has set a new challenge in the AI start-up in India in the field of Defence (Ministry of Defence). It poses the question that if the government faces the charges of infringement to the AI that is applicable for the societal welfare then can they be able to forego the criteria of true and first innovation to be maintained or rather they might spend millions of finances to acquire the rights of the same from another country? The same is a question that the judicial reviews by the court will be able to solve with their interpretation and with the proper updated Patent Acts which much wider definitions and patentee friendly procedures for applications.

Recommendations

The following are the recommendations that we would be looking for ensuring the AI and Patent compatibility soon: â&#x20AC;˘ Flexible AI in the Patent Registering Offices: It is suggested that patent agents can be replaced by a mix of Classic AI and Deep Learning, which is vested with the expert knowledge of set standards of assessing the three criteria for the further processing of the application. Using natural language processing to understand patent documents and to automatically classify them can be a viable option. Further, the obvious human errors or interventions can be reduced to maintain a fair and speedy process. However, the same is a bane for the employment and the human capital employed to ensure the compliance of the same is reaching the experience of the humans working at this job. It may also cause problems with the litigation process as it would be difficult to determine the negligence of the AI or Human.

Artificial Intelligence and Policy in India

• Determining the Proportionality principle for the AIs: Any tech litigation is eventually going to answer the question of proportionality of the tech system, and if there is a less restrictive alternative that is available. It is possible that in the AI infringement cases can be a road bumper for the development of the innovations in India and thus, there is a need that policymakers and the judicial system need to devise to prove how much infringement is a injurious to the right of the other party. A slight mark of innovation might be able to get a patent but it would affect the competition between the similar natures AI developers pushing the cost to a higher scale. • Human Ownership of AI liability: The ultimate goal of attributing Patenting to human authors is to continue to encourage the development of artificial intelligence. Artificial intelligence machines do not require financial incentives, unlike human programmers and they will be working to create an output- irrelevant to being innovative or redundant. Their uniqueness is only depended on what all ‘intelligence accessories’ the programmer can attribute to the AI machine. Without the programmer’s contribution, artificial intelligence devices would simply not be available for use by the public. Patent Act does not require a particular threshold of human control or input in the invention process for granting patent rights, but it frames the questions of ownership and patentability in terms of human creation. • The general definition of a “legal person,” which is a subject of legal rights and obligations is likely broad enough to encompass AI as long as AI’s role as an inventor is subject to legal rights and obligations. However, even if taking the inventors ownership to be considered, their liability should be proportional to the actual level of instructions given to the AI and of its degree of autonomy, so that the greater learning capability or autonomy, and responsibility of its trainer inventor can be established and regulated. • Shorter terms of Patents to the software innovations: It is also suggested that concerning the development of the AI and economic market of ease to work with, shorter patent terms to software patents can work well in favour of the patent litigation resolution. It can be a feasible option to balance the public policy of allowing the start-ups as well as allow the tech giants to reap the profits from the patent within a short investment period. • Multi-Level Model for distinguishing the AI for Patents: The Three-Level model of AI is similar to the lines of the WTO guidelines that categorises the AI into three parts (The impossible nightmare? IP in recession. Journal of Intellectual Property Law & Practice, 2008). The assertion behind this model is a that the AI revolution can be divided into the constituent part and can be summarised into 3 broad categories (1) advanced or neuroautonomous, (2) automated or semi-automated and (3) autonomous. According to a report published by Microsoft (2018), they can encompass all the possible AI innovations and are distinguished on the specific criteria similar to that of establishing the granting of patents: creativity, rational intelligence, unpredictability, innovations which are simulations of the human mind experiences (Impact of artificial intelligence on patent law. , 2019). A first criterion is a software-intensive form of AI categorisation, which involves the functions controlled by the humans like assemble, and replicate circuits designed by humans, process vast volumes of data accurately, efficiently and rapidly, well beyond the capacity of the human brain and traditional software algorithm etc. The second criteria of Semi-Automated AI are the function that involves less interaction with human control and it is further involves a sub-division of soft AI and Hard AI. The former functions likeability to make “decisions” within the environment for which it was designed and to further the task it was designed to accomplish, without human interaction and the latter involves the function of Designed to

• •

•

Artificial Intelligence and Policy in India, Volume 1(2020)

think creatively making intelligent human-like decisions having the competent cognitive ability. The third criteria is a deep learning machine that is depended on its system processing a merger of biological intelligence, it is the criteria of the distant future. If the Patents are granted in the same classifications then they can solve the problem of the ownership as well as a liability by adjudging the criteria of the reasonable foreseeability and the link of cause and effect similar to determining the tortious liability. ‘Turing test criteria’ for adjudging the innovation: Sir Alan Turing proposed a test called the ‘Turing test’ (Turing, 1950). The test called for the users to converse with a machine/human in a text-only format, and then suggest whether they believed they communicated with a human or a machine. This one of the near short term solution to the problem of the AI and patent compatibility. It would be easier for the lawmakers to mark the distinguishing feature of the deep learning machine or a classic AI machinery and as a temporary test to prove the liability of infringement looking at the ability of the machine to comprehend consequences of actions. The creation of Insurance Funds or Contracts for the AI-related infringement: The infringement cases that can be arisen out of the creations that the innovators and start-ups engage in raises financial and legal contingencies. This creates the sense of protection to the incentive market of AI in India and allows the start-ups to work on their innovations by ensuring an insurance system with a fund to ensure that reparation can be made for damages that the innovation might bring upon itself or the creators. The AI giants can come into the sophisticated contracts that cover some of the foreseeable contingencies beforehand between the AI and end-users Lowering the cost of patents contributing to the social system: The problems that India as a welfare state will be handling is to accommodate to the image of the backyard of AI as well as using the innovation to yield public welfare like pharmaceutical patents of the medicines the AI patent cost can be lowered to provide the essential AI technology to economic sectors. Lowering the subject-matter patentability standard for AI inventions relating to areas deemed more socially beneficial, such as healthcare, the environment, criminal justice and education, might be one way to help balance-promoting innovation with mitigating ethical concerns.

References 1. World Economic Forum. Artificial Intelligence Collides With Patent Law. [Online] http://www3.weforum.org/docs/WEF_48540_WP_End_of_Innovation_Protecting_Patent_Law.pdf. 2. A Framework For The Economic Analysis Of Patents In Business Context. Giordani, S. 2014. 2014. 3. Intellectual Property Rights And Innovation In Developing Countries: Evidence From India. Sharma, A. Dutta and S. 2008. s.l. : Enterprisesurveys.org., 2008. 4. Thinking About Thinking Machines: Implications of Machine Inventors for Patent Law. Vertinsky, L. and Rice, T. 2002. 2, 2002, B.U. J. Sci. & Tech. L. , Vol. 8, pp. 574, 576-77. 5. Why Microsoft Leads The AI Patent Race Going Into 2019 . Columbus, L. 2019. 2019. 6. NITI Aayog. 2018. National Strategy on Artificial Intelligence. 2018. 7. 2014. Alice Corp. v. CLS Bank Int’l, . 134 S. Ct. 2347, 2355 , 2014.

Artificial Intelligence and Policy in India

8. Delome, j. 2008. The changing legal framework for technology. Transfer: some implications. North / south technology transfer: the adjustments ahead. 2008, p. 89 . 9. 1981. Diamond v. Diehr, . 450 U.S. 175 , 1981. 10. Financial Times. 2020. AI drives down cost and drudgery of routine patent work. 2020. 11. Impact of artificial intelligence on patent law. . Chimuka, G. 2019. 2019, p. 59. Towards a new analytical framework – [ theMulti-Level Model]. World Patent Information. 101926.. 12. Impact of patent policy changes on R&D expenditure by industries in India. R. Sharma, A. Paswan S. Ambrammal and M. Dhanora. 2017. 2, 2017, Journal of World Intellectual Property , Vol. 21, pp. 52-69. 13. Intellectual Property in New Technological Age . Lemly, Mark A. and Merges, Peter S. Menell and Robert P. 2016. s.l. : Copyrights, Trademarks and State IP Protection , 2016, Vol. 1. 14. Kaufer, E. 1989. The Economics of Patent System. 1989. 15. Machine Learning Startups To Watch In 2018. Columbus, L. 2017. s.l. : Forbes, 2017. 16. Microsoft and IDC, 2019. 2019. Microsoft – IDC Study: Artificial Intelligence to nearly double the rate of innovation in Asia Pacific by 2021. Microsoft Asia News Center. [Online] 2019. https://news.microsoft.com/apac/2019/02/20/microsoft-idc-study-artificial-intelligence-to-nearly-double-the-rate-of-innovation-in-asia-pacific-by-2021/. 17. Ministry of Defence, 2006. Defence Policy for Jointly Developed Software Products. s.l. : Ministry of Defence, India. D(IT) Div ID No. 4(14)/2006/D(IT). 18. Patenting Artificial Intelligence Conference Summary. 2018. 19. R&D and Patenting by Firms in India in High and Medium High Technology. Ambramal, S. and Sharma, R, 2014, Journal of Chinese Economic and Business Studies, Vol. 12, pp. 181-187. 20. Rajasthan HC, 2018. 2018. Expression Of Interest Document For Selecting Agency For Providing Solution For AI Based Video Analytics And Traffic Management. [online] Available at:. [Online] 2018. http://risl.rajasthan.gov.in/wp-content/uploads/2018/08/EOI_AI_based_video_analytics_traffic_mana. 21. 2017. Smart Sys. Innovations, LLC v. Chicago Transit Auth., . 873 F.3d 1364, 1378, s.l. : Federal Circuit , 2017. 22. Subbaram, N.R. 2015. Subbaram: patent practice and procedures . 2015. 23. TATA and CII Report, 2019. 2019. Understanding the Dynamics of Artificial Intelligence In Intellectual Property Rights. Confederation of Indian Industry . : TATA consultancy Services, 2019. 24. TATA, Consultancy. 2020. . Can Artificial Intelligence Software Be Patented In India? Patent Blog And Patent News For Legal Services. [Online] 2020. https://ttconsultants.com/blog/can-artificial-intelligence-software-be-patented-in-india/. 25. The Composition -of Licensing Fees and Arrangements as a Function of Economic Development of Technology Recipient Nations. Contractor, F J. 1980. 1980, Journal of Busines Studies . 26. The impossible nightmare? IP in recession. Journal of Intellectual Property Law & Practice. Phillips, J. 2008. 9, 2008, Vol. 3, pp. 535-535. 27. Turing, A. 1950. Computing machinery and intelligence. : Mind, LIX, 1950. 28. United States Patent and Trademark Office. 2020. Petition Decision: Inventorship Limited To Natural Persons. 2020. 29. Vertisky, L. 2018. Thinking Machines and Patent Law’. Barifield. Research handbook on the law of artificial intelligence. 2018, p. 498.

Artificial Intelligence and Policy in India, Volume 1(2020)

30. World Intellectual Property Organization. 2019. World Intellectual Property Indicator: Filing Patents, Trademarks and Industrial Design Reach Reacord Heights in 2019.: WIPO, 2019. https://www.wipo.int/pressroom/en/articles/2019/article_0012.html.

Artificial Intelligence and Policy in India

Jurisprudential Modalities of Data Protection Regime and Privacy via AI in India Vaishnavi Venkatesan1 and Nayan Grover2 12

Honorary Research Members, Indian Society of Artificial Intelligence and Law nayan.grover20@gmail.com

Introduction

There have been numerous attempts by academicians, jurists, legal theorists as well as philosophers to intellectualize the concept of privacy. However, the term has numerous connotations and encompasses multi-faceted aspects. Black’s Law Dictionary incudes various components of the term and defines it as “The right to be let alone; the right of a person to be free from unwarranted publicity; and right to live without unwarranted interference by the public in matters with which the public is not necessarily concerned.” (Dictionary) Term "right of privacy" is generic term encompassing various rights recognized to be inherent in concept of ordered liberty, and such right prevents governmental interference in intimate personal relationships or activities, freedoms of individual to make fundamental choices involving himself, his family, and his relationship with others. It is the right of an individual (or corporation) to withhold himself and his property from public scrutiny, if he so chooses. Considering the vast literature available on the subject, privacy can be categorized into six broad headings, which coherently come together to put forth the meaning of the highly complex term ‘privacy’. It includes: 1. The right to be let alone (Mathew, 1975). 2. limited access to the self- the ability to shield oneself from unwanted access by other. (Rao, 1964); 3. Secrecy (Cases, 1998); 4. Ability to exercise control over personal information about oneself (Lahoti, 2005) ; 5. The protection of one’s personality, individuality and dignity (Cases, 2009) and lastly 6. Control over one’s accesses to intimate relationship or aspects of life (Singh, 1997). In an attempt to define privacy, the above-mentioned aspects together provide for the meaning behind privacy as a whole and each of these facets have been evolved through privacy jurisprudence in India over time. To understand and enlist threats to data privacy, difficulties involved and their interplay with Artificial Intelligence which will be done in the latter part of this paper, it is imperative to primarily analyse the privacy jurisprudence and its evolution in the Indian context as well as a comparative analysis vis-à-vis jurisprudence of other countries having a strong privacy regime.

Literature Review

There exists a variety of literature on privacy concerns and data protection laws. While dealing with privacy concerns with respect to Artificial Intelligence, Vidushi Marda (2019) in her report talks about the Data Protection Bill which seeks to strengthen India’s privacy regime through numerous positive measures but also seems to undermine privacy with respect to a few considerations by providing worryingly broad exceptions for government data processing. Presently,

Artificial Intelligence and Policy in India, Volume 1(2020)

the bill allows government processing of personal data without consent if is shown to benecessary for “any function of Parliament or any State Legislature,”, or necessary for the exercise of the State “authorized by law for the provision of any service or benefit, which is vague and paves way for loopholes with can be misused. Although this paper deals with the problems associated with data privacy in depth and talks about numerous challenges, it poses a lot of questions but doesn’t provide a solution to combat such privacy related concerns. Pallavi Gupta in her paper (2019) talks data protection with regards to privacy and elaborates on the concept by stating that every AI interface relies entirely on the data which is being fed into its system. The author also poses the question of who is to be blamed for any action that an AI system undertakes. This directs focus on the problem of discrimination with respect to the data being fed in, which is closely related to data privacy. The paper talks about the contribution of the IT Act, which acts as a sole legislation touching upon the aspect of consent and data privacy remotely. Even though there exist some sections of the Information Technology Act which deal with the subject, there is a need for a more robust framework. It raises the requirement for comprehensive data protection legislation in India, on the lines of European Directive on Data Protection, UK Data Protection Act (1998).The author also recommends the two- layer protection model to regulate data usage with respect to AI. She states that it is of utmost importance to have a two-layered protection model: one- technological regulators; and two- laws to control AI actions as well as for accountability of errors. Arindrajit Basu and Pranav M.B in their article (2019) put forth their opinions on privacy and Artificial Intelligence by stating that existing AI principles do not adequately articulate how the legal rule can be applied to various scenarios by multiple organisations. According to them, there is no need of a new “Law of Artificial Intelligence” to regulate this space.While there is consensus on the above opinion and the authors in their article have dealt with the issue in depth but are lacking in delving into privacy jurisprudence and explaining the same. Sreekanth Mukku (2019) in his thesis has conducted an in-depth comparative analysis with respect to the convergence and divergence of privacy and data protection laws of the two countries. While dealing with NITI Aayog’s strategy to tackle AI, the author points out an important contradiction. On the one hand the document explains the privacy concerns and how personal data of the citizens is to be protected, on the other hand it lays out adoption of sophisticated surveillance systems and use of social media platforms to monitor people’s movement to maintain public safety. Both are inherently contradicting. Furthermore, the author elaborates on the lack of liability on the government’s part by stating that the document recognizes the privacy rights of consumers and need for regulation of capturing, processing and inappropriate use, discrimination and so on, but fails to underscore the role of the government in data governance and remedy and redressal mechanism if government is the offender and violated data privacy. There exists a variety of literature with regards to data privacy and AI which covers a broad spectrum of areas ranging from data protection laws, already existing jurisprudence, taking inspiration from other countries and suggestions involving adoption of a new legislation to combat this issue to name a few, however at present there subsists a pressing need for literature which deals with privacy jurisprudence as a concept in detail as well as its applicability in India. After such analysis, this jurisprudence is to be co-related to the current scenario of Artificial Intelligence to arrive at a practical solution which will be included in the form of recommendations. The current paper seeks to delve into the depths of privacy jurisprudence by

Artificial Intelligence and Policy in India

undertaking a comprehensive analysis of already existing laws, precedents, fundamental duties and rights to arrive at a logical as well as amicable solution to combat the challenges pertaining to data privacy and AI ethics in the current landscape.

Privacy Jurisprudence In India: An Evolution:

Privacy as a concept is closely related to personal liberty which is one of pillars of democracy and forms a part of the preamble. Despite the fact that right to privacy is not explicitly mentioned in the Indian Constitution and plays no part in the constitutional assembly debates, it is inherent in constitutional principles and can be carved out through judicial pronouncements over the years leading to privacy ultimately being avowed as a fundamental right in the landmark case of K.S. Puttaswamy v. Union of India. (Sikri, 2014) To understand privacy in today’s technology driven world with respect to data privacy and Artificial Intelligence, a chronological analysis of privacy jurisprudence as delivered in various cases over the years is crucial. In India, the right of privacy has been developed through judicial rulings, from the rights under Articles 19(1) (a) (1950)(i.e. the fundamental right to freedom of speech and expression) and 21 (i.e. the right to life and personal liberty) of the Constitution. (1950) The very first case which dealt with the issue was M.P. Sharma v. Satish Chandra (Jagannadhadas, 1954)in 1954. In that case, the Court while dealing with search and seizure held that privacy was something alien to the Indian Constitution. When the constitution makers did not deem fit to include privacy as a right akin to the one provided in the American Fourth Amendment, there existed no justification to import it, into a totally different fundamental right by some process of strained construction. Immediately following the MP Sharma case, the court in the case of Khadak Singh v State of UP (Ayyangar, 1963)held that "the right of privacy is not a guaranteed right under our Constitution and therefore the attempt to ascertain the movements of an individual which is merely a manner in which privacy is invaded is not an infringement of a fundamental right guaranteed” It can be seen that even though the court does find a right to privacy, refrains from framing it in that manner. A crucial part here is that of the minority opinion of Justice Subba Rao, who wrote, “Constitution does not expressly declare a right to privacy as a fundamental right, but the said right is an essential ingredient of personal liberty. We would, therefore, define the right of personal liberty in Article 21 as a right of an individual to be free from restrictions or encroachments on his person, whether those restrictions or encroachments are directly imposed or indirectly brought about by calculated measures. It so understood, all the acts of surveillance under Regulation 236 infringe the fundamental right of the petitioner under Article 21 of the Constitution” which is very relevant to the current jurisdiction on the subject. The next case in the landscape of privacy jurisprudence, which came to be dealt with about a decade after the Kharak Singh case proved to be an important one where the courts for the first time felt the need to recognize the right to privacy. In the case of Govind Singh v State of MP (Mathew, 1975),even though the facts were similar to the Kharak Singh case, the approach of the court whilst dealing with privacy changed. In this case, the petitioner Govind contested the validity of the MP Police Regulations 855 and 856, which relate to surveillance, including through domiciliary visits. The court in this case referred to American jurisprudence and constitutional principles which had undergone significant changes since the Kharak Singh case.

Artificial Intelligence and Policy in India, Volume 1(2020)

The Court in this case relied on Griswold v. Connecticut (Cases, 1965) and Roe v. Wade (Cases, 1973)(which dealt with right to abortion and privacy), both of which had found 'privacy' as an " interstitial or '' penumbraP ' right in the American Constitution - that is, not reducible to any one provision, but implicit in a number of separate provisions taken together. J Matthew in this case while recognizing the right to privacy held that, “The right to privacy in any event will necessarily have to go through a process of case-by-case development. Therefore, even assuming that the right to personal liberty, the right to move freely throughout the territory of India and the freedom of speech create an independent right of privacy as an emanation from them which one can characterize as a fundamental right, we do not think that the right is absolute.” The courts while dealing with the aspect of privacy moulded the concept with a view of according it legal recognition so that its breach could be remedied. Privacy was being given wider and wider field of operation including therein matters pertaining to health, personal communications, family, personal relations and a right to be free from harassment and molestation. Similarly, in the case of R. Rajagopal and Anr. v State of Tamil Nadu (Reddy, 1994) the court while recognizing that privacy as a ‘right to be left alone’, implicit under Article 21 of the Indian Constitution also laid down position will be different if a person voluntarily thrusts himself into such a situation. Subsequently in the case of People’s Union for Civil Liberties (PUCL) v Union of India (Reddi, 1997)the Hon’ble Supreme Court clearly held that:- “We have, therefore, no hesitation in holding that the right to privacy is a part of the right to''life `` and''personal liberty” enshrined under Article 21 of the Constitution. Once the facts in a given case constitute a right to privacy, Article 21 is attracted. The said right cannot be curtailed “except according to procedure established by law A pertinent point to note at this stage is that in the landscape of privacy jurisprudence in India, the court has often dealt with the concept in the offensive, i.e. the question of privacy comes into effect when it is infringed either by the state or by an individual, for example domiciliary visits, state surveillance, telephone tapping, revelation of contagious diseases to name a few. In the earlier cases the court relied on the philosophical aspect of privacy by correlating it with the right to dignity but with the evolution of privacy jurisprudence, the stance adopted by the courts changed. Contrary to evolutionary principles, in the recent landmark case dealing with privacy, K. S. Puttaswamy (Retd.) v Union of India (Sikri, 2014)the court regarded privacy as an inherent right which needs to be protected at the outset rather than when an offence infringing privacy has already been committed. The court recognized a positive as well as a negative right to privacy, the onus of which is placed on the government. The negative content acts as an embargo on the State from committing an intrusion upon the life and personal liberty of a citizen and its positive content imposes an obligation on the state to take all necessary measures to protect the privacy of the individual. Therefore, the constitutional protection of privacy may give rise to two inter-related protections i.e. (i) against the world at large, to be respected by all including State: right to choose what personal information is to be released into the public space (ii) against the State: as necessary concomitant of democratic values, limited government and limitation on power of State. Consequently, this landmark judgement while acting as a milestone in privacy jurisprudence has led to privacy being accorded as something which is much more than a mere common law right. It is now robust and sacrosanct, which has been held to be a fundamental right under Article 21 of the Indian Constitution. While dealing with the right to information privacy in

Artificial Intelligence and Policy in India

today’s technologically driven world, the Hon’ble Mr. Justice D.Y. Chandrachud concluded as under: “457. Informational privacy is a facet of the right to privacy. The dangers to privacy in an age of information can originate not only from the state but from non-state actors as well. We commend to the Union Government the need to examine and put into place a robust regime for data protection. The creation of such a regime requires a careful and sensitive balance between individual interests and legitimate concerns of the state. And thereby stated these to be policy considerations to be dealt with by the Union government while drafting a robust and comprehensive data protection framework.

Problems with the Artificial Intelligence

With increase in technological advancements every day the involvement of AI in our lives is becoming more integrated, in ways we are unable to fully understand yet. ‘AI is often used as an umbrella term to describe a collection of related techniques and technologies including machine learning, predictive analytics, deep learning, natural language processing and robotics. Real-life applications of AI technologies are already established in our everyday lives, although many people are not conscious of this. For example, being greeted by an automated voice on the other end of the phone, or being suggested a movie based on your preferences, are examples of mainstream AI technology’ (2018). Now there are increasing privacy concerns with every new technology introduced, then what makes AI such a big concern. The extent and much higher efficiency of AI creates more large and complex problems for sustenance of Information Privacy. For example – ‘the use of CCTV cameras in public spaces for surveillance is a relatively widespread practice and not considered to be unreasonably intrusive in modern society. However, combined with the use of facial recognition software, a network of cameras could be transformed into a tool that is much more privacy invasive.’ (2018) Use of such privacy invasive tool by an authoritarian regime like China can lead to far reaching human rights violation. Some usual privacy concerns that emerge when we introduce AI are 4.1

Blurred Lines on Sensitive Information

In today’s day and age every country’s data protection regime aims to protect and safeguard sensitive or personal information of user. Personal and sensitive information of a user is generally considered as the information that defines his unique personality and character, and that can be used to trace back to and identify the individual and thus is violative of an individual’s privacy rights. But when it comes to the use of AI it is hard to identify that what can or cannot be used to trace back to an individual. AI has the ability to analyse and interlink data in order to make predictions in various facets. It can recognize logic and patterns which a human mind will not be able to imagine. So, what may seem non–personal information at first, combined with other non-personal information, can be used by an AI to predict information that may be deemed as personal and can be used to identify the individual. While personal data is routinely (pseudo-)anonymised within datasets, AI can be employed to de-anonymise this data. (2020) Thus it becomes impossible to draw a clear line between sensitive and non-sensitive information.

4.2

Artificial Intelligence and Policy in India, Volume 1(2020)

Unclear Extent of Data Collection and Use

Data and information should be collected from an individual with his full knowledge and when it comes to personal information then along with knowledge, the necessity is must. But the techniques followed by AI systems to collect data is such that often times users are not even aware that their data is being collected for use. AI-driven consumer goods and autonomous systems are often fitted with sensors that produce and collect large amounts of data in their vicinity without the knowledge or consent of those around it. (2018) As the Internet of Things (IoT) pushes the network further into our physical environment and personal spaces, the scope of data created, collected and fed into AI systems stands to delve further into our personal lives. (2018) For Example - Technological developments in IoT devices, smartphones and web tracking means that the data being fed into AI systems is often not collected in a traditional transaction whereby people consciously provide their personal information to someone who is asking for it. (2017) Similar to the protocol followed in collection, the user should also be made fully aware about who all will use their data and till what extent it would be used. Now working and development of AI and machine learning systems is based on large chunks of data being fed to them. Most of the times the individuals whose data is used to train AI and Machine learning systems are unaware of this use of their data as usage of data for training of AI is not a reasonably predictable use. This Unclear extent of data collection and use is attack on one of the fundamental pillars on which information security laws around the world are based. 4.3

Undefined Purpose of Use

Providing an explanation of the purpose of collection (generally through a collection notice) is how most organisations adhere to the purpose specification principle. The ability of AI to extract meaning from data beyond what it was initially collected for presents a significant challenge to this principle. In some cases, organisations may not necessarily know ahead of time how the information will be used by AI in the future. (2018) 4.4

Transparency

One of the major principles on which the Data protection regime is based on is the transparency in the usage and processing of data. But in the case of Artificial Intelligence this has become nearly impossible. Now machine learning systems work on data and then develop their own logic which is many a times unknown, even to the developers if that system. Moreover, when it comes to deep learning, the problem becomes more complex. In deep learning systems, data is processed through different layers, output of each previous layer is input for the next one. Even if the underlying algorithm is transparent to the user, because the pattern of interactions is very complex and often uses clusters of factors that make no intuitive or theoretical sense. (MacCarthy, 2020) This is also known as the â&#x20AC;&#x2DC;black boxâ&#x20AC;&#x2122; problem with AI systems and is heavily criticized. 4.5

Discrimination

The AI and Machine Learning systems are trained using the data collected by humans. If the data fed into these systems contains bias, then it is likely that the outcomes of the AI system

Artificial Intelligence and Policy in India

will also have bias present in them. Now these systems might be used to perform crucial functions like crime predictions, allocation of funds, etc. Due to the bias present in the predictions some social groups, especially minorities, can suffer severe damage. For example, the predictivepolicing program, called Pred-Pol, had led to heavier policing of minority neighborhoods in Los Angeles. The program unfairly targeted Latino and black neighborhoods. (Miler, 2020) Another issue related to discrimination that arises is that accurate predictions may reveal sensitive attributes that could be used to discriminate. (2018) ‘Using machine learning methods, highly sensitive information can be inferred or predicted from non-sensitive forms of data. People’s emotional states e.g. confidence, nervousness, sadness, and tiredness, can be predicted from typing patterns on a computer keyboard.’ (Identifying emotional states using keystroke dynamics’, 2011) When sensitive personal data, such as information about health, sexuality, ethnicity, or political beliefs can be predicted from unrelated data (i.e. activity logs, phone metrics, location data or social media likes) such profiling poses significant challenges to privacy and may result in discrimination. (2018)

A Comparative Analysis of Indian Privacy Laws Vis-À-Vis Other Jurisdictions

India is yet to enact a comprehensive central legislation which deals solely with privacy considerations and data protection regulations. It is still in its nascent stage of development as compared to other highly regulated jurisdictions like that of European Unions’ (Hereinafter EU) GDPR, which encompasses a set of stringent measures for data protection and privacy. This piece seeks to draw a comparison between India’s Personal Data Protection Act, 2018 which is a draft bill pending approval with that of GDPR and US’s privacy regulations holistically. India through the Personal Data Protection Act, is moving away from harm-specific provisions, as adopted by the US which is more inclined towards providing solutions for privacy infringement and is rather in line towards the approach adopted by the EU in GDPR towards restrictive and preventive data protection framework. The very first aspect of analysis include the ‘terms’ adopted by different jurisdictions for data protection. The Personal Data Protection Act uses the term ‘data processors" to refer to entities that process data on the instructions of a data controller but the Act refers to “data fiduciaries,” in place of the term data “controllers,” as in the GDPR and US regulations and where the GDPR refers to “data subjects,” the Act refers to "data principals." Similar to data controllers under the GDPR, data fiduciaries are liable for their own data processing activities, as well as the activities of data processors. This modified terminology, in the Indian context emphasizes a ‘trust based’ relationship between the data subject and the data holder wherein the subjects entrust their data to companies and other controllers, which therefore have a fiduciary duty of care towards those subjects. (2019) The next question pertains to applicability of the acts, wherein any company anywhere in the world must comply with the Personal Data Protection Act to the extent that the company in question processes personal data on Indian territory, including with the help of a data processor in India, offers goods or services to data subjects in India, or profiles Indian residents remotely. (Ratul Roshan & Sreenidhi Srinivasan, 2020) Outsourcing functions that involve exchange of foreign personal data in India will also be covered under its ambit and will have to comply with the acts requirements. Under the act, discretion is granted to the government to exempt data processing related to foreign data subjects in an outsourcing context from obligations under the Personal Data Protection Act, but it remains to be seen whether the

Artificial Intelligence and Policy in India, Volume 1(2020)

government will actually exempt such data processing from substantive compliance obligations. (Indiaâ&#x20AC;&#x2122;s Personal Data Protection Act, 2018: Comparison with the General Data Protection Regulation and the California Consumer Privacy Act of 2018, 2020)Whilst dealing with the ambit of protection extended, both GDPR and Personal Data Protection Act accommodates provisions for a Data Protection Authority and imposes obligations on companies to ensure administrative compliance like appointment of data protection officers, local representatives (for foreign companies), data protection impact assessments, record keeping, audits etc. In Contrast the United States not have established data protection authorities. (Adequacy of data protection in the USA: myths and facts, 2016). Further it does not create any such administrative obligations. The California Consumer PrivacycAct,2018 in the US also, addresses specific risks for individual privacy created by data trading (Determann, 2019) however does not provide for a data protection authority. On the matter dealing with ambit of protection, it is vital to take note of a key difference between the GDPR, Data Protection Act and US laws of privacy wherein most US privacy laws, including the CCPA, only protect the privacy of residents, whereas the GDPR and the Personal Data Protection Act regulate any processing of personal data on local territories, including personal data pertaining to persons residing in other countries. Consequently, foreign companies are subject to these laws if they process foreign personal data in European or Indian territory. This could have a major impact on India's business process outsourcing, call centers, and data processing services more generally. As far as the scope of rights provided to subjects under privacy law goes, all the three jurisdictions have adopted a similar approach. Under all three laws, individuals have the rights to access (i.e., to know what data is held about them), portability (i.e., to have data transferred to another entity that provides similar services), and to be forgotten (i.e., to have information held about them deleted or restricted), subject to different nuances and exceptions. Under the Personal Data Protection Act, individuals enjoy only a limited right to be forgotten with respect to further disclosure, but not a right to absolute deletion. To obtain absolute deletion, individuals need to seek a decision weighing data privacy and information freedom interests from an adjudicating officer at the Data Protection Authority. Requesting individuals may have to pay a fee to compensate the data controller for the costs of handling such requests. ( Indiaâ&#x20AC;&#x2122;s Personal Data Protection Act, 2018: Comparison with the General Data Protection Regulation and the California Consumer Privacy Act of 2018, 2020) To ensure strict compliance penalties of the PDP Bill builds on GDPR on penalties. For other violations, such as non-compliance of the PDPBâ&#x20AC;&#x2122;s cross-border transfer provisions and consent and grounds of processing, penalties extend to Rs 15 crore ($2,184,525) or 4% of the global turnover in the previous financial year (whichever is higher). (India shakes International Data Market, 2019) Furthermore, under US law the CCPA establishes a Consumer Privacy Fund, which is funded by penalties and is designed to induce and support additional enforcement activities. ()The Indian Data Protection Authority also includes provisions for allocation of special funds for its operative costs, supported by fees and charges, and for privacy awareness, supported by penalty funds. (2019) Therefore, it is imperative for companies in these jurisdictions to keep in mind such provisions. These data protection provisions which differ to a certain extent from jurisdiction to jurisdiction however pose similarities in their core underlying principles.

Artificial Intelligence and Policy in India

Analysis Of ‘National Strategy for Artificial Intelligence’ By NITI Aayog With Respect to Privacy

The paper addresses a vast variety of issues with respect to Artificial Intelligence and its application and development in various fields, with special context of India. It has a specific section dedicated to possible privacy concerns and then some well thought solutions. It mentions two major ways in which individual privacy is violated by corporates – (1) Companies collecting massive amount of consumer data and using it inappropriately to gain insights about consumers. This becomes an issue of concern as consumers are not aware about how these insights are gained and what are they used for. If everything would be transparent with the consumer and data would be used to provide better services then it is not a problem. (2) Large corporates gaining an unfair competitive advantage by collecting large data sets. These data sets have no value on their own unless they can be analysed using tools involving AI system and machine learning. Although it guided well about threat to privacy posed by private players in the market, the threats to data privacy posed by the government weren’t mentioned anywhere. It is really important that if an issue is addressed, its wholistic view must be taken. When one of the solutions proposed by the paper to protect privacy is to spread awareness, not making people aware about such a massive risk of privacy they can face by the government is hypocritical. 6.1

Solutions Proposed

The paper offered a number of measures that can be taken to safeguard privacy of individuals. • It emphasised the importance of the 7-core principles of data protection and privacy – ‘informed consent, technology agnosticism, data controller accountability, data minimisation, holistic application, deterrent penalties and structured enforcement’ (2018) for a strong data protection regime. These were proposed by Srikrishna committee on data protection law. • It called for a sectoral regulatory framework for privacy protection so that specific laws can be made to cater the needs for data protection in different fields. • It recommended that the benchmark for national data protection and privacy laws should be at par with international standards such that of GDPR. • It urged AI developers to incorporate international standards into their Artificial Intelligence system. It asked them to follow the Global Initiative on Ethics of Autonomous and Intelligent Systems of the IEEE ‘s chapter on ‘Personal Data and Individual Access Control in Ethically Aligned Design’. • It appreciated a concept developed by Cynthia Dwork in 2006, Differential Privacy that aims at preserving identifiable user information irrespective of any outside information the aggregator agency holds and not letting the user to be uniquely identified. • It encouraged the government to invest and collaborate in researches, technology and models being developed to enhance data privacy and protection. 6.2

Unaddressed Issues

There were many sectors that the paper proposed AI to dwell into, but failed to address some serious privacy concerns that can be faced and one should be cautious and prepared for.

Artificial Intelligence and Policy in India, Volume 1(2020)

• Healthcare Sector – The paper proposes the involvement of AI into healthcare by way of AI driven diagnostics, personalised treatment, early identification of potential pandemics, and imaging diagnostics, among others. Also, it suggested that at remote places where there is unavailability of specialised medical staff, AI can be used to train many who couldn’t access healthcare before. But the paper didn’t take into consideration that how the data collected, analysed and stored by Artificial Intelligent systems can be exploited in various ways. In United states it was discovered that Genetic Testing companies that analyse user’s DNA to tell ancestry and identify potential diseases, were selling their data to Health Insurance companies. These companies then used this data to deny certain users health insurance or charge them extra. Another violation occurs when large social media companies used AI to collect mental health data without any legal safeguards. Mental health data can be exploited in various ways, it can be used to personally target users to engage them for longer durations. Now sometimes the motive might be positive but the major problem is that all this is being done without users consent and without informing them. (Kulkarni, 2019) Thus, there is a need for some more stringent measures of data protection to be imposed when it comes to health care sector as an individual’s health data is the most sensitive information. • Agricultural Sector – The paper calls artificial intelligence to address challenges such as inadequate demand prediction, lack of assured irrigation, and overuse / misuse of pesticides and fertilisers. Some use cases include improvement in crop yield through real time advisory, advanced detection of pest attacks, and prediction of crop prices to inform sowing practices. (2018) Physical sensors and livestock in the field generate data and receive command operations via user applications. These on-farm devices are connected to gateway supported edge nodes, which help enable in-farm device communication, filter sensor data and real time agronomy analytics. At the same time, data lakes in the cloud hold a large amount of data and information including but not limited to, environmental information (e.g. soil moisture level and fertility status), monitoring information (e.g. sensors and smart machinery status), energy management data, and other sensitive information. In terms of security and data privacy, it is needless to say that manipulation and leakage of such data, as well as the impairment of physical equipment and software systems, can induce serious consequences. (Gupta, et al., 2020) • Semi-Autonomous Features and Driverless AI Cars – It suggest that AI be used in providing semi-autonomous features such as driver assist, and predictive engine monitoring and maintenance, etc. ‘In this near-future filled with self-driving cars, the price of convenience is surveillance. This level of data collection is a natural extension of a driverless car’s functionality. For self-driving cars to work, technologically speaking, an ocean of data has to flow into a lattice of sophisticated sensors. The car has to know where it is, where it’s going, and be able to keep track of every other thing and creature on the road. Self-driving cars will rely on high-tech cameras and ultra-precise GPS data. Which means cars will collect reams of information about the people they drive around—like the data Uber has amassed about its customers’ transportation habits, but down to a level of detail that’s astonishing. The more personalized these vehicles get—or, the more conveniences they offer—the more individual data they’ll incorporate into their services. The future described might be a way off, yet, but there’s no reason to believe it’s especially far-fetched.’ (LaFrance, 2016) Autonomous cars would generate, collect and analyse reams of data for a host of purposes, ranging from basic navigation to other complex functionalities like traffic management, speeding enforcement and knowhow about driving circumstances. With the overall requirement of a plethora of

Artificial Intelligence and Policy in India

data, there arises an inescapable question that who will have access to this massive data. (Chaturvedi, 2020) • AI Enabled Smart Cities – The paper proposed that smart cities where AI would play a major role could be built. AI could be linked with infrastructure and can also be used for traffic control and enhanced security systems. As much security and convenience this step would increase but it would leave a huge area for the government to exploit the citizen’s data to serve their nasty purposes like supressing protests, tracking activists, etc. Until and unless the Data Protection laws of our country doesn’t include checks and regulations for the government, the idea of smart cities would do much more harm than good.

Suggestions & Recommendations

• India presently does not have a data protection law however the Personal Data Protection Bill, 2019 which envisages a comprehensive policy on data protection is in development and is pending approval. • Free of Bias: It is imperative that the data protection law, which is to develop into the law of the land, should be free of bias. In its current form, the bill provides unrestricted access to government agencies for matters of ‘nation concern’ or in the interest of the sovereignty and integrity of India. These terms are vague and leave room for ambiguity to kick in thus paving way for misuse of the provisions and defeating the very purpose of the act. • Data Regulatory Authority: It is proposed that India needs a Central Data regulatory authority which will act as the sole body handling data, leading to an increase in centralization and transparency whilst dealing with sensitive personal information. The concept of data protection officers as listed in the GDPR can also be implemented in India by moulding it as per requirements. • Keeping It Simple – Privacy Policies drafted by companies should be in plain simple language. The privacy policy is drafted for consumers, who don’t prefer reading technical jargons or legalese, so use of them should be avoided as much as possible. The Privacy policies are lengthy in the first place and a further addition of Technical Jargon and Legalese makes it even more difficult for consumers to read and comprehend it. • Choosing Rights Over Consent- While framing data protection laws a Rights based approach should be followed where certain guidelines are laid out by law as to what extent and manner the data of a consumer, although obtained with consent, can be used. Such laws should limit the use and collection of data in a manner to protect the violation of data rights of consumers. Breaking of such laws should lead to penalization of the corporation doing so. • Opt-in rather than Opt-out - To ensure that the consent given by consumers is closer to real consent, companies should follow a strict Opt-in policy of Data Collection. Now all these points are not well established. The world we live in where our data can be used in numerous adverse ways, it becomes really important that the consumers are well informed about the data collection and they agree to it prior to it being in effect, even if it involves a bit of disruption. • A two-layered protection model: It has two components; one- technological regulators; and two- laws to control AI actions as well as for accountability of errors. Technological regulators include introducing a multi-layered security and defense software to protect data and in turn business. Layered security strategies are reactions to today's cyber threat landscape.

•

Artificial Intelligence and Policy in India, Volume 1(2020)

Rather of only waiting for the attacks to reach endpoints, layered security takes a systemic view of cyber protection, preparing for the multitude of vectors that carry modern malware. Right to be Absolutely Informed – There should be a clear declaration made to the consumers with regards to third parties using their personal data. The law should mandate it for companies to disclose the list of third-party organizations to consumers who will be accessing and using their data and also clearly state in the disclosure as to what part of the consumer data is provided to which organization for what purpose. Right to Erasure – Right to Erasure enables any data principal to have his personal data deleted from all databases on his request. This right ensures that a consent is valid only till it is an active consent. If once a consent was given for an information to be collected, it does not mean that the information shall be withheld for lifetime. This also provides a person an opportunity to get any false information about him, present on the internet, to be removed permanently. Transparency and accountability form an important part of the training process because in numerous cases there is a manifestation of unwanted bias which is reproduced via people building it or through biased data which is fed to train the AI and Machine Learning algorithms. In such a scenario, transparency and accountability are two major pillars which will prevent the structure of AI from being corrupted and ensure it remains free of bias. Awareness is one of the most important tools to combat AI considerations and privacy aspects. India’s low literacy levels acts as a major barrier to the common people understanding privacy laws and resolving complexities, conducting, and inculcating public consultation goes a long way. The above-mentioned suggestions and recommendations call for a robust legal framework pertaining to privacy legislation in India which adopts an all rounded legislation keeping in mind India’s position.

Conclusions

AI is ubiquitous in our everyday lives. It is the new ‘normal’ and is making a difference in numerous fields. It is defining our lives in new innovative ways every day. Its impact on society cannot be negated in any manner whatsoever. There exist numerous AI's that aid essential areas such as law enforcement to provide novel solutions to omnipresent problems, gradually annihilating the gap between already diminishing human and AI functions, amongst performing other functions for the overall betterment of society. However, with increase in involvement of AI in our lives, increases the exposure of our information to this world which leads to increase in threat to our privacy. This discussion paper introduced various facets of privacy and information security in which AI can cause challenging issues. The paper has shed light on the present Data Protection laws in the Indian regime and has compared it with International standards. An analysis of whether the present laws in India and around the world are enough to cater to the needs of future threats to privacy especially in context of Artificial Intelligence and Machine Learning systems has been made. Some present laws showed scope but the analysis mainly concluded that in general the current laws, especially in India are not ready to handle the situation well. Various researches discussing and contemplating the problem and proposing solutions have been reviewed. The country’s premier think tank, Niti Ayog’s strategy on the development and regulation of Artificial Intelligence has been critically analysed, the practical innovative solutions it

Artificial Intelligence and Policy in India

preached are appreciated and light is shed on areas left neglected by it. The possible threats in areas where the strategy proposes a boom in development of AI have been showcased. At last through research, critique and analysis a list of recommendations has been provided through which we can conquer the threats to privacy posed by the modern era technologies, especially Artificial Intelligence, inclusive of their development towards the positive growth of society.

References 1. India’s Personal Data Protection Act, 2018: Comparison with the General Data Protection Regulation and the California Consumer Privacy Act of 2018. Gupta, Lothar Determann and Chetan. 2020. 3, s.l. : BERKELEY JOURNAL OF INTERNATIONAL LAW, 2020, Vol. 37. 2. Adequacy of data protection in the USA: myths and facts. Determann, Lothar. 2016. 3, s.l. : INTERNATIONAL DATA PRIVACY LAW, 2016, Vol. 6. 3. 2020. Artificial Intelligence. Privacy International. [Online] Privacy International, 2020. [Cited: June 14, 2020.] https://privacyinternational.org/learn/artificial-intelligence. 4. 2018. Artificial Intelligence and Privacy. s.l. : Office of Victorian Information Commissioner, 2018. 5. 2017. Artificial Intelligence, Ethics and Enhanced Data Stewardship. s.l. : Information Accountability Foundation, 2017. 6. Ayyangar, N R. 1963. Kharak Singh v State of UP. 1295, All India Reporter : SC, 1963. 7. CAL. CIV. CODE. 8. Cases, Supreme Court. 1998. Mr X v Hospital Z. 294, Supreme Court Cases : SC, 1998. 9. —. 2009. Suchita Srivastava v Chandigarh Administration,. 91, Supreme Court Cases : SC, 2009. 10. Cases, US. 1965. Griswold v. Connecticut. 479, US : s.n., 1965. 11. —. 1973. Roe v. Wade. 113, US : s.n., 1973. 12. Chaturvedi, Aditya. 2020. Implications of data privacy once autonomous vehicles hit the roads. Geospatial World. [Online] January 14, 2020. [Cited: June 15, 2020.] https://www.geospatialworld.net/blogs/implications-of-data-privacy-once-autonomous-vehicles-hit-the-roads/. 13. 1950. Consitutution, Indian. 1950. 19. 14. 1950. Constitution, Indian. 1950. 21. 15. Determann. 2019. California Privacy Law: Practical Guide And Commentary, U.S. Federal And State Law,. 2019. 16. Dictionary, Black's Law. 17. 2018. Discussion Paper on National Strategy for Artificial Intelligence. s.l. : Niti Ayog, 2018. 18. Gupta, Maanak, et al. 2020. Security and Privacy in Smart Farming: Challanges and Opportunities. Ieeexplore.ieee.org. [Online] February 19, 2020. [Cited: June 15, 2020.] https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9003290. 19. Identifying emotional states using keystroke dynamics’. Epp, C., Lippold, M. and Mandryk, R.L. 2011. Vancouver, Canada : The Interaction Lab, University of Saskatchewan, 2011. 20. India shakes International Data Market. Panakal, Dominic Dhil. 2019. s.l. : THE NATIONAL LAW REVIEW , 2019, Vol. 1. 21. India’s Personal Data Protection Act, 2018: Comparison with the General Data Protection Regulation and the California Consumer Privacy Act of 2018. Gupta, Lothar Determann and Chetan. 2020. 3, s.l. : BERKELEY JOURNAL OF INTERNATIONAL LAW, 2020, Vol. 37. 22. Jagannadhadas, B. 1954. M.P. Sharma v. Satish Chandra. 300, All India Reporter : SC, 1954. 23. Kulkarni, Andrea. 2019. AI in Healthcare: Data Privacy and Ethics Concern. Lexalytics. [Online] November 12, 2019. [Cited: June 15, 2020.] https://www.lexalytics.com/lexablog/ai-healthcaredata-privacy-ethics-issues. 24. LaFrance, Adrienne. 2016. The Creepy Thing About Self-Driving Cars. The Atlantic. [Online] March 21, 2016. [Cited: June 15, 2020.] https://www.theatlantic.com/technology/archive/2016/03/self-driving-cars-and-the-looming-privacy-apocalypse/474600/.

100

Artificial Intelligence and Policy in India, Volume 1(2020)

25. Lahoti, R C. 2005. District Registrar and Collector, Hyderabad v Canara Bank. 496, Supreme Court Cases : SC, 2005. 26. MacCarthy, Mark. 2020. How to address new privacy issues raised by Artificial Intelligence and Machine Learning. Brookings. [Online] 2020. [Cited: June 14, 2020.] https://www.brookings.edu/blog/techtank/2019/04/01/how-to-address-new-privacy-issues-raised-by-artificial-intelligence-and-machine-learning/. 27. Mathew, K K. 1975. Govind Singh v State of MP. 946, Supreme Court Cases : SC, 1975. 28. â&#x20AC;&#x201D;. 1975. Gobind v State of Madhya Pradesh. 1378, Reporter, All India : SC, 1975. 29. Miler, Leila. 2020. LAPD will end controversial program that aimed to predict where crimes would occur. Los Angeles Times. [Online] April 21, 2020. [Cited: June 15, 2020.] https://www.latimes.com/california/story/2020-04-21/lapd-ends-predictive-policing-program. 30. 2019. Personal Data Protection Act. 2019. 31. 2018. Privacy and Freedom of Expression in age of Artificial Intelligence. s.l. : Article19.org, 2018. 32. Rao, K. Subba. 1964. Kharak Singh v The State of UP and Ors. 332, Supreme Court Cases : SC, 1964. 33. Ratul Roshan & Sreenidhi Srinivasan. 2020. European Union: Comparative Analysis: General Data Protection Regulation, 2016 And The Personal Data Protection Bill, 2019. Mondaq. [Online] March 12th, 2020. [Cited: June 5th, 20202.] https://www.mondaq.com/india/privacy/903076/comparative-analysis-general-data-protection-regulation-2016-and-the-personal-data-protection-bill-2019. 34. Reddi, P V. 1997. Peopleâ&#x20AC;&#x2122;s Union for Civil Liberties (PUCL) v Union of India. 568., All India Reporter : SC, 1997. 35. Reddy, B Jeevan. 1994. R. Rajagopal vs State Of T.N. 632, Supreme Court Cases : Supreme Court, 1994. 36. Sikri, A K. 2014. K. S. Puttaswamy (Retd.) v Union of India. 43, Supreme Court Cases : SC, 2014. 37. Singh, K. 1997. PUCL v Union of India. 301, Supreme Court Cases : SC, 1997.

Artificial Intelligence and Policy in India

101

Artificial Intelligence in Corporate Transactions: AI, Automated Mergers & Acquisition and Corporate Ethics Stuti Modi1 1

Honorary Research Member, Indian Society of Artificial Intelligence and Law 18jgls-stuti.m@jgu.edu.in

Introduction

1.1

Anticipated Trends In Mergers and Acquisition Deals

• Backed by a positive trend in M&A deals worldwide in 2018 as well as 2019, Deloitte comprehensive survey reports analysed and anticipated that there existed strong signals for deal making to flourish even in 2020. Among the respondents of the survey, 98 percent expected rise in deal size, tied in with 96 percent forecasting an increase in number of deals over the next 12 months. • In the report it was further reiterated, that, Divestment remains a key component of M&A deals. Such divestments are facilitated by two factors, firstly by way of organisations seeking to capitalise on high valuation and secondly, by way of distress deals. • As rightly stated by Sir Jason Langan “If the economy slows, let alone dips into a recession, companies aren’t going to ignore M&A, they are just going to be even more deliberate in the deals they look to do. That might mean more divestitures, an effort to raise cash or shed units that just aren’t performing. Some also are going to be strategic about using a downturn to buy assets more cheaply; this is especially true on the PEI side, which has large cash reserves. Typically, M&A doesn’t disappear in a downturn but some strategic imperatives shift.” (Jason Langan, et al., 2019) Conclusively, if the economy falters, divestitures would become more relevant. • Additionally, Data Protection and allied regulations coupled with shareholder’s activism are among the pertinent issues affecting M&A deals and strategy. Evidence can be drawn from active board involvement in the form of directors ensuring deals meeting their expectations. 1.2

Impact of Covid-19 on M&A Deals

• Unanticipatedly, the outbreak of coronavirus has exposed the global economy to unprecedented economic challenges. The pandemic has not only exposed the world to uncertainties but has also froze the economic growth of the companies. • Despite a stern outlook of the near future, every crisis has its winners and losers. Just like the period following the Economic Recession of 2008, the current situation is likely to be followed by a series of opportunities in the form of Distress Deals. • At the time post the crisis, it might be favourable for the companies holding abundant cash reserves, to expand their business at affordable costs. However, the challenge posed to such companies would be in the form of identifying the appropriate opportunity and mitigate legal and other risks.

102

1.3

Artificial Intelligence and Policy in India, Volume 1(2020)

Imperative Need of Incorporating Artificial Intelligence in M&A Deals

• Due Diligence- A recently published article analysis the imperative reliance on virtual data rooms for the process of Due Diligence in M&A. There will be less reliance on site visits or physical meetings and more upon the data stored virtually. Another analysis rightly sheds light on the adoption of virtual working as a challenge that M&A Deals in order to succeed have to overcome and substantially integrate in the Due Diligence process.

•

Business Valuation- In the process of carrying out business valuation, acquirers will have to be extra cautious and conduct detailed economic assessment in form of liquidity and business status. Additionally, there would be a crucial need to carry out examination of success or failures of similar M&A Deals in form of comparative analysis, depending upon availability of data. However, the challenges in conducting business valuation does not stay limited to economic uncertainties but also extend beyond that. There is an additional opposition in the form of lack of credibility of economic projections posed due to unavailability of set benchmarks for economic undertakings. (Verma, 2020)

•

Legal or Regulatory Perspective- The acquirer needs to be mindful of the possible impacts of such M&A Deals on every stakeholder, from the perspective of taxation and regulation. In order to reduce economic risks a holdback module in form of an Escrow account coupled with cashless system of share swaps forms an imperative measure. • Negotiation- Data Analytics help develop valuable insights in a quick manner, facilitating faster action on the transaction. A detailed analysis may be drawn pertaining to geographies, operational risks, product lines and others. • Post-Merger Integration- Data analytics may facilitate identification of opportunities in form of potential synergies, mitigate risks and hurdles and ensure post deal execution. Other factors such as identifying scope for and conducting renegotiations in order to enhance revenue, evaluating customer response to the ownership change and analysing the change in risk profile can be successfully conducted by way of deploying analytics in M&A Deals. (Deloitte Report)

Legal and Ethical Considerations in Deployment of AI in M&A Deals

Along with the widespread potential of AI across various sectors, comes the diverse regulatory issue. The regulation of AI possesses multiple unique contextual challenges which varies in accordance with specific use of technology in the sector. Hence, the regulation of AI in ‘onesize-fits-all’ manner can not take place. In deploying AI in M&A Deals, various legal and ethical challenges need redressal. 2.1

Privacy and Security

• The advent of Right to Privacy as a constitutionally warranted right, by the judiciary in Justice K.S. Puttaswamy vs. Union of India (2017) (“Privacy Judgement”) paved way to the recognition of the long due Data Protection Laws in India. In the judgement, the Supreme Court recommended establishment of a robust regime for data protection, in a manner such that there exists a balance between individual interest and legitimate concern of state. • This was followed by constitution of a committee of experts (“Srikrishna Committee”)

Artificial Intelligence and Policy in India

103

• to prepare a draft Data Protection Bill, 2018 (MeitY, 2018), which was thereafter open to comments and suggestions from stakeholders. Consecutively, a Personal Data Protection Bill, 2019 (MeitY, 2019) was placed before the cabinet, which incorporated the comments and concerns raised previously under the draft bill. • As a result of Privacy Judgement in India, data collection, storage and use can dilute the essence of privacy solely, on fulfilling restrictions which are legitimate. Further, the established data protection mechanism, is based on broad privacy principles and deploys sending notice as a pre-requisite, before such data is collected. • In Corporate Transactions, such as that of undertaking potential M&A Deals, a Non-Disclosure Agreement (“NDA”) becomes an imperative part of the deal. Under such transactions, privacy and security concerns revolve around consent and the lawful use of such data. Such Data Security forms a critical element for establishing long-term commercial relationships. • Drafting and Negotiating on a favourable NDA is a time, labour and money intensive task. Further, the strategies used in negotiating upon NDA have been laid down in a number of playbooks and writings, in order to provide a step by step guidance for lawyers and contemporarily for machine learning. In order to efficiency negotiate upon a favourable NDA, AI is to some extent and should be deployed. (Monika Chin, 2018 ) • Resultantly, this raises issues pertaining to retention of such confidential information by machines. It is needless to mention that return and destruction of such confidential information forms an integral and crucial part of NDA. 2.2

Liability

• The main challenge for government is warranting constitutional due process standards be adhered by software developer of private sectors. Another, issue requiring redressal is the proprietary nature which the developer of source code possesses, rather than the user. 2.3

Accountability, Oversight and Evaluation

• An eminent feature of AI is algorithmic black box, in which input is processed and usable outputs generated. As stated by Sir Frank Pasquale, that multifarious uses of algorithm could lead to Black Box Society, where trajectory of daily existence is defined by opaque or black boxed algorithms. (Pasquale, 2015) In such a scenario, accountability is an imperative challenge when “values and prerogatives that the encoded rules enact are hidden within black boxes”. • However, with the deployment of metaphorical black box, developing accountability and evaluation standard poses a continuous challenge. • Further, there exists an imperative need for continuous communication channel. This would ensure awareness of individual regarding the manner in which decisions impacting them are being taken and would establish clarity. 2.4

Transparency

• In U.S. the case of State v. Loomis (2016) sets a remarkable example on addressing transparency. The court noted four integral requirements of transparency-

104

Artificial Intelligence and Policy in India, Volume 1(2020)

1. The inputs themselves, 2. How the algorithm weighs these inputs, 3. Whether combinations of certain factors such as race, gender or economic status may end up being used as variables, and 4. The underlying assumptions made by the computer scientists who designed the algorithms. • In India, the problem is amplified due to lack of clear set of guidelines in this regard, hence, resulting in wide discretionary power in the hands of judges (Niruphama, 2016). In, the case of State of Punjab v. Prem Sagar & Ors. (2007), it was highlighted that there exists an unaddressed lacuna of legal principles in regard to sentencing. • The use of algorithms comes along with the possibility of use of diverse set of algorithms bearing the likelihood of causing prejudice to the defendant. Hence, there exists an imperative need to establish guidelines to ensure consistency in decision making by the algorithms. • Further, it must be noted that unlike General Data Protection Regulation (“GDPR”), the Personal Data Protection Bill, 2019 entails no provision for Right to Explanation involving meaningful information about the logic underlying the automated decision. (Powles) • Additionally, lack of transparency may be violative of due process as per constitutional standards since it directly impacts life and liberty of individuals. This further raise obstruction in raising concerns or to challenge the same. 2.5

Redressal

• The constitution warranties the right of every individual affected by a particular action, to challenge decision in court. With the current regulatory regime, problems exist in two-fold. • Firstly, the decision being undertaken in an opaque black box manner, limits the scope to challenge the outcome of the decision making facilitated by AI due to inability of determining how such outcome was arrived at. Secondly, the obligation of private sector developer remains unclear. • Further, the complicated nature of public-private partnership, coupled with most data being held by private sector, along with absence of judicial precedents or robust legislature to regulate AI raises severe concerns. The enactment of laws and judicial precedents in addressing questions of accountability, liability and redressal, in order to ensure that state action does not end up becoming reason for dismissing harm is much needed. 2.6

Bias and Discrimination

• The standards established by International Human Rights and Constitution of India recognizes a dual way of possible discrimination. For the implementation of rule of law, in substance, there must be no discrimination carried out on basis of religion, caste, creed, political belief and ideology (15, 1949). Both Direct as well as Indirect discrimination are prohibited under the Indian Constitution. (1949) • The value is not only upheld worldwide by constitutional courts but was also read into the Constitution of India. This was carried out in the case of Madhu Kanwar v Northern Railway (2017), wherein the algorithms were not reflective of a particular group in question. It was held that deploying algorithms for use of AI, which apparently seem neutral but carry out indirect discrimination are impermissible and violative of the constitution on grounds

Artificial Intelligence and Policy in India

105

of carrying out indirect discrimination. This is indispensably necessary in a country like India which is striving to be inclusive by fighting away the pre-existing categorization and victimization. • There exist three stages of possible Algorithmic Discrimination. Firstly, by Incomplete or inaccurate training data, Secondly, Algorithmic Processing and Proﬁle Development and Thirdly, Interpretation of Outputs. (Danks, 2017) • • Despite the term ‘due process of law’ not being explicitly used in Constitution, the legal tradition in Indian presupposes the principle of natural justice. It can safely be inferred that Maneka Gandhi v UOI (1978) by imposing the requirement of the procedure established by law to be just, fair and reasonable read Due Process into the constitution. • However, the determination of fairness is left at the discretion of judges. Further, deploying fairness in algorithm and their contextual adaptation in Indian notion is not easy. This is so because, algorithms are designed at the cost of fairness in order to achieve efficiency. (O'Neil, 2016) Hence, in successfully deploying an automated system by using AI to undertake M&A Deals, the above-mentioned issues need to be urgently addressed by India. To stand in the wake of global competitive environment and ensure efficiency and conserve valuable time, money and resources, the regulatory regime in form of laws and judicial precedent have to be adopted or updated in a manner which may support deployment of AI.

Policy Recommendations For Appropriate Governance Of AI In M&A Deals

3.1

Contextual Application of Rule-Based System Along with Incorporating Due Process, Ethics, Fairness and Transparency

Adoption of Fairness and elimination of Bias in Governance • Although there can be no straight-jacket rule of fairness which will ensure the desired result in every situation, for ensuring adherence to public policy, there should be set of best practices or contextual rules, explaining the manner in which fairness in real world application of AI driven solutions should be utilized. (Kroll, 2016) • The fact that it is not possible for Due Diligence process to be always conducted perfectly, is not unknown. However, attempting to make a good faith effort to conduct due diligence properly is what matters from a legal perspective. (LajouxIssues, 2007) Hence, when AI is being deployed to carry out the Due Diligence process under M&A, in a cost and time efficient manner, same or even more caution and standards of good faith needs to be exercised. • Along with broad parameters being encoded by developer for facilitating adherence to constitutional standards, an ex-post fact check upon fairness must be carried out by an independent committee in situations where the impact of such AI driven solution is challenged. (Arindrajit Basu, 2018) • In spite of there being no precedent anywhere globally pertaining to this, India can consider setting up a Committee which would be entrusted with the responsibility of assessing the development and use of AI driven solutions and algorithm in its operation.

106

Artificial Intelligence and Policy in India, Volume 1(2020)

Cross Checking • In doing justice to the due-diligence process, the government reports and judicial decisions have favourably encouraged to deploy cross-check, to see whether the output add up when examined by other division. (Alexandra Reed LajouxIssues, 2007) This would be extremely more required in a system where AI is deployed to undertake Due Diligence in M&A. Transparency • Natural Justice being an indispensable part of the Indian Constitution, makes transparency a crucial precondition. (Carothers, 2014) The check for transparency does not limit itself to the design of algorithm but also extend to the social situations surrounding it. There must be a continuous and consistent evaluation within and across the sectors to warrant a uniform standard of permissible transparency throughout. (Arindrajit Basu) • As suggested by Oswald et al, that a rule-based system applied contextually shall follow two proposals to deal with algorithmic opacity. (Oswald, 2018) Under the first proposal, Experimental Proportionality is suggested. Here, the power is given to court to make determination of proportionality based on algorithms, at the stage even before the impacts for the same are fully determined. The second framework dubbed by acronym ‘ALGO-CARE’ calls for the design of a rules-based system which ensures that the algorithms are: ─ Advisory: Algorithms must not replace human discretion. ─ Lawful: Algorithm’s proposed function, application, individual effect and use of datasets should be considered in conjunction with necessity, proportionality and data minimisation principles. ─ Granularity: Consideration should be given to data analysis issues such as meaning of data, challenges stemming from disparate tracts of data, omitted data and inferences. ─ Ownership: Due regard should be given to intellectual property ownership but in the case of algorithms used for governance, it may be better to have open source algorithms at the default. At any rate, the developer must ensure that the algorithm works in a manner that enables a third party to investigate the workings of the algorithm in an adversarial judicial context. ─ Challengeable: The results of algorithmic analysis should be applied with regard to professional codes and regulations and be challengeable. ─ Accuracy: The design of the algorithm should check for accuracy. ─ Responsible: Should consider a wider set of ethical and moral principles and the foundations of human rights as a guarantor of human dignity at all levels. ─ Explainable: Machine Learning should be interpretable and accountable. A rules-based system like ALGO-CARE can enable predictability in use frameworks for AI. Predictability compliments and strengthens transparency. 3.2

Constitutional Principles to Assess the Use of AI Coupled with Algorithmic Impact Assessment

Adherence to Constitutional Principles • The three tests of proximity, proportionality and non-arbitrariness warranted and inscribed in the constitution must be adhered to even at the time of deploying AI in M&A

Artificial Intelligence and Policy in India

107

Transactions. The process adopted must be just, fair and reasonable. The algorithmic process used must not impact free speech, along with keeping in mind the boundaries for the object with which such technology and such consequent restriction was deployed. (Arindrajit Basu) • For making training data more holistic and inclusive, the existing data curation need to be adequately representative of the heterogeneous realities of India. (Abraham S.) Impact Assessment • The government is under an obligation to develop guidelines and procedure for examination of impact of such AI driven solution coupled in with its requirement for a contextual application. 3.3

Appropriate Discretion Delegated to AI Coupled with Process of Explanation and Accountability

Controlled Discretion • The AI-driven solution needs to be more regulated, needing more oversight along with adherence to stricter standards. The government is additionally obliged to develop guidelines for assessment of automated impacts. There must always exist a human discretion constantly evaluating and monitoring software actions and final decisions, along with selective delegation upon machines. Explanation Process • The deployment of contextual and appropriate process of explanation at the time of decision making would help reduce the opaqueness caused by black box decision making. (Doshi-Velez, 2017) Accountability • Primary accountability in any use of AI which has received state sanction should be of the government. Further, there is a dire need for a uniform and cohesive framework for regulating the public-private partnership. Development of Adequate Redressal Mechanisms • A proper judicial mechanism where judges possess special knowledge of how technology functions coupled with experts to help them understand the technicalities, is integral and non-negotiable to address the challenges to algorithmic governance. Additionally, an Alternate Dispute Resolution mechanism are also an essential requirement in the field because of huge backlog of cases in India.

108

Artificial Intelligence and Policy in India, Volume 1(2020)

Conclusions

The research concludes that despite of government pursuit both by way of report of AI Task force NITI Aayog along with public statements, there exists a lacuna in the current propositions that need immediate redressal. The technology has still not been developed sufficiently to be utilised on a large scale across various sectors. Hence, there exists a unique opportunity to evaluate the impacts of such AI-driven solutions being undertaken in western nations and analyse the various challenges. The gaps in law need to be filled in a manner such that it is not violative of basic tenets of our constitution. One challenge faced by India is in regard to the development capacity vesting with the private sector, which also implies they will undertake projects which are commercially beneficial. Hence, a robust mechanism for a liability regime holding the developers accountable must be established. It is also proposed not to deploy in a one-size fits-all manner, since, it would negate the purpose of its deployment. Further, machines should not be entrusted with complete discretion and there must always be human supervision and final decision.

References 1. 2016. [2016] 881 N.W.2d 749, 774., s.l. : The U.S. Supreme Court, 2016. 2. 15, article. 1949. The Indian Constitution. Article 15. 1949. 3. Abraham S., Hickok E., Sinha A., Barooah S., Mohandas S., Bidare P. M., Dasgupta S., Ramachandran V., and Kumar S. ‘NITI Aayog Discussion Paper: An aspirational step towards India’s AI policy’. Niti Aayog. [Online] 4. Alexandra Reed LajouxIssues. 2007. ‘M&A Due Diligence in new age of Corporate Governance’ . [Online] January 2007. [Cited: 26 May 2020.] https://iveybusinessjournal.com/publication/ma-duediligence-in-the-new-age-of-corporate-governance/. 5. Arindrajit Basu, Elonnai Hickok. Artificial Intelligence in the Governance Sector in India. [Online] [Cited: 2 June 2020.] https://cis-india.org/internet-governance/ai-and-governance-case-study-pdf. 6. Arindrajit Basu, Elonnai Hickok,. 2018. ‘Artificial Intelligence in the Governance Sector in India’. [Online] 2018. https://cis-india.org/internet-governance/ai-and-governance-case-study-pdf. 7. —. ‘Artificial Intelligence in the Governance Sector in India’. [Online] [Cited: 2 June 2020.] https://cis-india.org/internet-governance/ai-and-governance-case-study-pdf. 8. Carothers, T., & Brechenmacher, S. 2014. ‘Accountability, Transparency, Participation, and Inclusion: A New Development Consensus? Carnegie Endowment for International Peace’. 2014. 9. Danks, D., & London, A. J. 2017. Twenty-Sixth International Joint Conference on Artiﬁcial Intelligence. NITI Aayog, National Institution for Transforming India. [Online] August 2017. 10. Deloitte Report, ‘Not using analytics in AI, you may be falling behind. Not using analytics in M&A? You may be falling behind. Deloitte . [Online] [Cited: 15 May 2020.] https://www2.deloitte.com/ca/en/pages/finance/articles/analytics-m-and-a-ia.html. 11. Doshi-Velez, F., Kortz, M., Budish, R., Bavitz, C., Gershman, S., O'Brien, D., & Wood, A. 2017. ‘Accountability of AI under the law: The role of explanation’ . 2017. 12. Jason Langan, Partner, M&A Services, Deloitte & Touche LLP and ‘M&A trends 2020 Report’, Page 8,. 2019. M&A trends 2020 Report. s.l. : Deloitte & Touche LLP, 2019. 13. 2017. Justice K.S. Puttaswamy vs. Union of India. (2017) 10 SCC 1, s.l. : The Supreme Court of India, 2017. 14. Kroll, J. A., Barocas, S., Felten, E. W., Reidenberg, J. R., Robinson, D. G., & Yu, H. 2016. Accountable algorithms. U. Pa. L. Rev., 165, 633. 2016.

Artificial Intelligence and Policy in India

109

15. LajouxIssues, Alexandra Reed. 2007. M&A Due Diligence in new age of Corporate Governance. [Online] January 2007. [Cited: 29 May 2020.] https://iveybusinessjournal.com/publication/ma-duediligence-in-the-new-age-of-corporate-governance/. 16. 2017. Madhu Kanwar v Northern Railway. (2017) SC 640 201, s.l. : Supreme Court of India, 2017. 17. 1978. Maneka Gandhi vs UoI. (1978) AIR 597, s.l. : The Supreme Court of India, 1978. 18. MeitY. 2018. Draft Data Protection Bill, 2018. Ministry of Electronics and Information. [Online] July 2018. https://meity.gov.in/writereaddata/files/Personal_Data_Protection_Bill,2018.pdf. 19. —. 2019. Personal Data Protection Bill, 2019 . Ministry of Electronics and Information Technology (GoI). [Online] 2019. http://164.100.47.4/BillsTexts/LSBillTexts/Asintroduced/373_2019_LS_Eng.pdf. 20. Monika Chin. 2018 . An AI just beat top lawyers in their own game. Mashable.com. [Online] 26 February 2018 . https://mashable.com/2018/02/26/ai-beats-humans-at-contracts/#ikC6NPTTIkqo. 21. Niruphama, R. 2016. Need for Sentencing Policy in India: Second Critical Studies Conference . [Online] 2016. [Cited: 25 May 2020.] http://www.mcrg.ac.in/Spheres/Niruphama.doc. 22. O'Neil, C. 2016. Weapons of math destruction: How big data increases inequality and threatens democracy. s.l. : Broadway Books, 2016. 23. Oswald, M., Grace, J., Urwin, S., & Barnes, G. C. 2018. Algorithmic risk assessment policing models: lessons from the Durham HART model and ‘Experimental’ proportionality. s.l. : Information & Communications Technology Law, 2018. Vols. 27(2), 223-250. 24. Pasquale, F. 2015. The black box society: The secret algorithms that control money and information. . s.l. : Harvard University Press, 2015. 25. Powles, Selbs A and. Meaningful Information and the Right to Explanation. [Online] https://academic.oup.com/idpl/article/7/4/233/4762325. 26. 2007. State of Punjab v. Prem Sagar & Ors. SLP (Crl.) No.4285 of 2007, s.l. : Supreme Court of India, 2007. 27. 1949. The Indian Constitution. Article 14 read with Article 15. 1949. 28. Verma, Bhumesh. 2020. Mergers and Acquisitions in COVID-19 times: Challenges and Way Ahead. Delhi, India : s.n., 2020.

110

Artificial Intelligence and Policy in India, Volume 1(2020)

Machine Learning and its Privacy Implications in India: Analysis of the Logistical Imperative in Data Protection and Jurisprudence Sameer Samal1 Nodal Advisor, Indian Society of Artificial Intelligence and Law sameer_samal@yahoo.in

Introduction

The world is currently witnessing technological advancements that challenge the purview of the digital revolution, also known as the ‘Third Industrial Revolution’. With the advent of these technologies, countries are realizing their widespread applicability. Technologies such as machine learning are penetrating and disrupting almost all sectors and thus stimulating a transition into the ‘Fourth Industrial Revolution’. Machine learning (ML) is a subset of Artificial Intelligence that can learn from datasets, improve from such experiences, and perform tasks without being explicitly programmed. With the assistance of ML, specific problems can be solved that are too complex or time consuming for human beings and traditional computer programs. The machine learning application learns by consuming and analyzing datasets that consist of industry-specific data. Developers seek data that is available either with the government or the private sector to create training modules. These datasets consist of information that may be directly or indirectly associated with individuals and considering that India lacks a data regulatory regime, such information may or may not be masked to ensure privacy. Artificial Intelligence, in its true form, is but an elusive vision; thus, the focus of regulators and policymakers should shift towards a more targeted subset of AI such as ML, wherein the involvement of personal data is relatively high. Existing literature on the privacy implications of such technological developments does not specifically cater to the need of a data privacy policy to regulate ML technology. Therefore, it is necessary to establish a data regulatory mechanism in India that ensures personal data protection and informational privacy while promoting technological advancement. Considering the multitude of opportunities emerging from data-centered-technologies such as machine learning, the rationale of this paper revolves around the importance of a comprehensive policy framework to regulate the flow of personal data in this sector. The paper advocates a central policy as well as industry-specific data protection regulations. To support the foregoing thesis, data protection policies in India are examined considering their relevance in regulating ML technology. Conclusively, a comprehensive policy is proposed to achieve the aforementioned objectives without inflicting any unreasonable burden over technological advancement.

Artificial Intelligence and Policy in India

111

The Interrelation of Machine Learning and Personal Data

Machine learning, a subset of Artificial Intelligence, is a field of study of computer algorithms wherein computer programs and algorithms automatically improve based on training and experience. Arthur Samuel, an American expert in the field of artificial intelligence research and computer gaming, described ML as the “field of study that gives computers the ability to learn without being explicitly programmed” (Some Studies in Machine Learning Using the Game of Checkers, 1959). These computer algorithms create a mathematical program based on training datasets to make certain decisions without being explicitly programmed. Polanyi’s Paradox is a theory by the British-Hungarian scientist and philosopher Michael Polanyi, that states, “we can know more than we can tell” (Polanyi, 1966). This phrase concisely records that humans may know a lot about how the world works but cannot explicitly explain this knowledge. In this line of reasoning, knowledge can be categorized into two distinct areas, explicit knowledge and tacit knowledge. Explicit knowledge is the codified information that can be readily described and explained to individuals and can also be programmed into computers. However, tacit knowledge, as introduced by Michael Polanyi, is the knowledge that is difficult to express or transfer to another individual by verbal or written form; for instance, the ability to ride a bicycle or speak a language is the knowledge that cannot be explicitly described or transferred to another person. It can be defined as skills, ideas and experiences that people have but are not codified and may not necessarily be easily expressed (Do Australian Universities Encourage Tacit Knowledge Transfer?, 2015). Technological advancements such as machine learning, and its subset deep learning, have enabled computer systems to gain tacit knowledge by being trained with a large amount of sample data; thus, these advanced algorithms learn automatically by analyzing large datasets instead of being explicitly programmed to do so. This feat itself proves the potential of ML for solving complex issues that are too difficult or time consuming for humans or traditional computer systems. ML algorithms have been successfully implemented in industrial and professional sectors and are now making their way towards other fields such as classification and extraction of data, image processing, medical diagnosis, language translation, speech recognition, and even financial services (Wladawsky-Berger, 2018). These algorithms are increasingly entering the fields that largely involve the flow of personal data. Therefore, to understand the policy requirements of ML regulation it is necessary to recognize the involvement of personal data in the training datasets. From a privacy perspective, the data points associated with individuals in a dataset can be differentiated based on the class of information they hold. Every dataset is in a tabular format with rows and columns. The rows contain information about one specific individual (Mr. X) of a group while the columns represent the values of attributes related to that group (age, gender, nationality). These columns are grouped based on the type of information they contain and can be classified into: 1. Personally Identifiable Information (PII), 2. Quasi-Identifiers (QI), 3. Sensitive Columns, and 4. Non-sensitive Columns.

112

Artificial Intelligence and Policy in India, Volume 1(2020)

The first category, i.e., Personally Identifiable Information (PII) contains information that can explicitly and uniquely identify the individual, such as name and personal identification number. The second category, i.e., Quasi-Identifiers (QI) contains information that may be attributable and common to more than one person, such as age and gender. The information contained in the category of QI cannot be solely used for identification. However, if the information contained in this category is analyzed with additional external input, then the individual may be identified. The third category, i.e., Sensitive Column contains information that is sensitive and critical and shall not be traceable to the individualâ&#x20AC;&#x2122;s identity, such as the results of disease diagnosis. The fourth category, i.e., Non-sensitive columns contains certain attributes that are not critical in nature and the same information cannot be utilized solely to identify an individual, such nationality. In the case of a dataset containing medical records for disease analysis, the abovementioned categories may contain: 1. PII- full name, personal identification numbers, etc. 2. QI- age, gender, etc. 3. Sensitive Column- disease diagnosis (HIV positive or negative). 4. Non-sensitive Column- nationality, to understand demographic references.

PII

Sensitive Column

Non-sensitive Column

Name

Personal ID

Age

Gender

HIV Diagnosis

Nationality

Privacy data breaches in the past have caused stakeholders to take safeguard measures, both legal and technological measures. For most people, an ideal and straightforward measure to successfully anonymize personal information in datasets would be to mask the information column or to remove the column entirely. However, in certain cases, if such columns are removed, it will render the entire purpose of such an algorithm to be absurd, such as if the algorithm is being trained for developing facial recognition software. Therefore, this paper will analyze the legal and technical privacy safeguarding measures adopted by developers and regulators and will propose certain legal policy recommendations that will ensure personal data protection while taking into account the complex ML technicalities.

Unique and Amplified Data Privacy Issues in Machine Learning

To protect personal data, developers use certain technical safeguards such as adopting Privacy-Preserving Data Publishing as a preventive method. Privacy-Preserving Data Publishing is carried out for anonymizing personal data in the datasets before it is used for training ML algorithms. The three most relevant data anonymization techniques are: â&#x20AC;˘ k-anonymity: is an anonymization method wherein specific Quasi-Identifier columns in the datasets are removed or changed, so that at least two rows will share certain similar attributes. This technique assures that there will always remain certain ambiguity of a minimum

Artificial Intelligence and Policy in India

113

‘k’ records for any targeted query search that contains privacy-sensitive attributes. Although this method does little to anonymize personal data, it has proven successful in reducing the chances of a privacy breach. • l-diversity: this technique is generally implemented in addition to ‘k-anonymity’ and substitutes for the weak points of ‘k-anonymity’. This anonymization method addresses privacy issues by ensuring that ‘equivalence groups’, i.e., subsets of datasets that consist of the same value of more than one Quasi-Identifier, have sufficient attribute diversity of sensitive data. • t-closeness: in this anonymization technique, “the distance between the distribution of a sensitive attribute in an equivalence group and the distribution of that attribute in the whole table (or population) is no more than a threshold ‘t’. It uses the notion of Earth Mover’s Distance (EMD) to represent ‘distance’ between distributions for its advantages of annulling the effects of differing attribute sensitivities” (Prabhu, 2019). The abovementioned anonymization techniques are altercations made to datasets before the personal data is shared and; this is known as Privacy-Preserving Data Publishing. However, with the advent of advanced computer algorithms such as ML and its subsets, these safeguard measures may not prove to be sufficient anymore. Traditionally, the identifier columns that contained sensitive information were removed from the dataset. This ensured that the removed information could not be re-introduced solely based on the information in the dataset, but only from combining additional external information. However, machine learning algorithms can recreate identities even if such sensitive identifiers are removed. These algorithms represent promising statistical capabilities of analyzing large datasets to make decisions and predictions; and it is this exact potential that has created data privacy issues like never before. For instance, while analyzing a dataset containing CV details of potential applicants, ML algorithms can identify the gender of applicants based on their language preferences in the CV, even if the identifier column containing gender information is removed to prevent gender discrimination. Although it might seem a far-fetched argument, certain similar instances of re-identification have occurred in the past. In 1997, the Massachusetts Group Insurance Commission (GIC) released hospital data for researchers to improve the healthcare system and make it more economically efficient. The then Massachusetts Governor, Mr. William Weld, assured the public that the Massachusetts Group Insurance Commission had taken privacy protection measures by deleting identifiers before releasing the hospital records. To test the privacy measures, an MIT graduate student, Dr. Latanya Sweeney, combined the voting records of the city of Cambridge, Massachusetts containing name, address and gender of every voter with the hospital records. With this additional information, she was able to trace the medical records of Mr. William Weld himself. This instance of de-identification of hospital records proved that computer scientists can deanonymize data with astonishing ease by simply cross-referencing publicly available data (The "Re-identification" of Governor William Weld's Medical Information: A Critical Reexamination of Health Data Identification Risks and Privacy Protections, Then and Now, 2012). In 2008, Arvind Narayanan and Vitaly Shmatikov from the University of Texas at Austin were able to successfully apply their de-anonymization method to the Netflix Prize Dataset which contained anonymous movie review ratings. They linked this dataset to the Internet

114

Artificial Intelligence and Policy in India, Volume 1(2020)

Movie Database (IMDb) and re-identified the data of almost 500,000 Netflix subscribers. Moreover, they were able to uncover probable political preferences and other politically sensitive information of these users (Robust De-anonymization of Large Sparse Datasets, 2008). In 2018, a group at MIT used Amazon product reviews to successfully de-anonymize the Netflix Prize dataset even when the datasets were cleaned and perturbed to ensure consumer privacy. They were able to match the Amazon product reviews and successfully identify individuals and their shopping habits (Who's Watching? De-anonymization of Netflix Reviews using Amazon Reviews, 2018). In conclusion, machine learning has amplified existing data privacy issues and has also created certain new concerns that need to be solved. However, regardless of these obstacles, the implementation of machine learning algorithms across various fields seems a dire necessity. 3.1

Technical Safeguards

To counter the issues faced by traditional anonymization methods such as k-anonymity, ldiversity and t-closeness, the following privacy-protecting methods are introduced: • Differential Privacy: This concept uses a mathematical equation to determine the extent to which an ML algorithm retains information about individuals. It achieves this objective of preserving informational privacy by injecting a certain quantity of random noise to the input dataset. This random noise is added to the dataset to ensure that privacy-sensitive inferences are not drawn from the predictions made by the machine learning algorithm while retaining the possibility of predictions to be practically accurate enough.

Figure 1: The illustration accurately explains the working of Differential Privacy. The red blocks represent personal data while the black blocks represent added noise. When a search query is entered to identify whether specific records are present in the dataset, the query will not be able to determine it. Therefore, after the randomized algorithm is applied to the dataset, an attacker will not be able to confirm whether certain targeted information is present in the dataset or not.

• Federated Learning: This subset of machine learning adopts a privacy-preserving technique by learning from ‘islands’ of data instead of combing all the data. The algorithm analyses the data in individual subsets of datasets and then aggregates the findings. The most common stakeholders that adopt federated learning is the healthcare industry. Hospitals that

Artificial Intelligence and Policy in India

115

are interested in adopting machine learning techniques to improve the services provided to their patients prefer federated learning for two reasons: ─ individually, these hospitals do not hold specific data for a targeted issue to train machine learning algorithms, and ─ they do not want the health records and other clinical data to face potential privacy breaches. • Another common use of federated learning is to improve user experience without sharing personal data to a common platform or a centralized database. Federated learning enables algorithms to analyze data that are in decentralized form, which in turn ensures individual informational privacy. • Homomorphic Encryption: Homomorphic Encryption is a cryptographic method of ensuring data privacy. This method is usually preferred by parties that wish to share the data with certain third-party for computational purposes. This technique encrypts the information in the dataset and then shares it for analysis or computation. This method of encryption does not require decryption for analysis using the algorithms. Instead, computation and analysis can be performed on the data without decrypting it. Therefore, when the computed data is returned from the client, the host can decrypt the data and receive the actual results without having the risk of a possible privacy breach. However, this method has its limitations in computing only certain functions, and this poses challenges. 3.2

Legal Safeguards

It is surprising to note that the abovementioned technological safeguards are currently only voluntary in practice by privacy-respecting developers and entities. India does not have comprehensive data protection legislation in action and the only piece of legislation that slightly deals with data protection is the Information Technology Act, 2000. The IT Act, 2000 was amended to insert sections 43A and 72A, which give a right to compensation for improper disclosure of personal information. Apart from the compensatory provision to privacy breach victims, the legal system currently does not confer any rights nor duties upon developers and users.

The Personal Data Protection Bill, 2019: Analysis

The Ministry of Electronics and Information Technology of the Government of India, on the directions of the Hon’ble Supreme Court of India, established a committee led by Retd. Justice Shrikrishna to propose a framework for data governance in India. The committee submitted its report in the form of a draft bill known as the Personal Data Protection Bill, 2018. The Bill faced serious negative feedback from the involved stakeholders; thus, an amended Bill was introduced as the Personal Data Protection Bill, 2019. The provisions that refer to machine learning technology are as follows: 1. Section 2 of the Act states that the Act applies to the processing of such personal data that has been collected, disclosed, shared or, processed in India. 2. Section 5 of the Act makes it mandatory for the data processor to process the personal data only for the purposes for which the data principal has consented.

116

Artificial Intelligence and Policy in India, Volume 1(2020)

3. Section 7 of the Act requires the data processor to provide a notice for collecting personal data. 4. Section 9 of the Act restricts the retention of personal data beyond the period necessary for the purpose consented by the data principal. 5. Section 17 of the Act confers the right to obtain confirmation and access to personal data upon the data principal. 6. Section 24 of the Act requires the imposition of security safeguards by the data processor such as de-identification and encryption.

Recommendations

Machine learning, as discussed above, is a complex technological development that, along with its benefits, has amplified existing data privacy issues and has also introduced new challenges. The existing data governing Act, i.e., the Information Technology Act, 2000 does not possess the relevant provisions to counter the complex challenges introduced by the machine learning technology. The Central Ministry responsible for governing technology has introduced a data protection bill known as the Personal Data Protection Bill, 2019 which, based on the aforementioned analysis, has proven to be a comprehensive piece of legislation on the subject. However, certain data privacy issues that are unique to the machine learning technology have been overseen by the bill. Therefore, with an intent to counter these issues, this paper recommends certain legal safeguards that will successfully ensure personal data protection from the perspective of informational privacy. The paper recommends a two-tier legal framework consisting of certain provisions that shall be inserted in the central data protection legislation forming the first tier and certain industry-specific data protection provisions that shall act as ‘Rules’, forming the second tier. The following provisions are recommended to be inserted in the central data protection legislation (Tier-1): • Clause 1: The Data Principal shall be granted absolute ownership over all categories of his/her personal data. Individuals to be granted property rights over all categories of their personal data. The majority of industries do not provide property rights over personal data to individuals. However, certain industries, such as the healthcare industry, provide partial ownership. The data in medical records are owned by the patients but the medium of transmission and storage is owned by the healthcare provider. In such situations, although the data is owned by the data principal, he/she does not have any control over the medium of storage and transmission. At this juncture, regardless of central legislation that grants property rights over personal data, the intervention of industry-specific policy measures to ensure uniform rules for collection, transmission, dissemination and storage is necessary. Granting property rights over personal data to individuals would ensure a strict consent-based usage of such data as individuals will exercise absolute control over the generation, transmission and secondary usage of their personal data. • • Clause 2: Enforcing the Data Principal-Data Fiduciary relationship over data processors (this provision shall act as an exception to Clause 1 in conditions where granting data ownership will be unreasonable). In certain cases, granting ownership rights to data principles over

Artificial Intelligence and Policy in India

117

their personal data will seem unreasonable. Therefore, to counter the issue, this clause has been introduced. The healthcare industry would be a prime example where this clause will prove to be beneficial. The healthcare industry is in the dire need of technological advancements such as machine learning, but at the same time, healthcare records and other clinical data are classified as highly critical or sensitive personal data that cannot be risked. While introducing machine learning algorithms to assist medical professionals and researchers, these algorithms have to be trained on clinical data. In such circumstances, data ownership cannot be granted to data principles as this would adversely impact the relationship shared between the healthcare industry and the patients. Buying and selling of healthcare records and clinical data will be disastrous for the healthcare industry. Therefore, to ensure the highest level of data protection without jeopardizing the relationship between patients and the healthcare industry, the only plausible solution could be to establish and enforce a moral and ethical basis for dealing with clinical data. A data principal-data fiduciary relationship can be enforced between any two parties dealing with such data. The following provisions are recommended to be inserted as ‘Rules’ for industry-specific data protection (kindly note that these recommended ‘Rules’ are in addition to those duly introduced by the government): • Rule 1: Mandatory anonymization of personal data. Anonymization methods such as Kanonymity, L-diversity and T-closeness are commonly used to mask or remove personally identifiable information and sensitive column data from ML training datasets. These anonymization techniques are currently voluntary in practice and India does not have any policy measures to enforce such safeguards. Industries such as Agriculture do not generate personal data and thus require minimal data protection. However, the financial industry and the healthcare industry generate sensitive and critical personal data and thus require stringent anonymization measures. The Government of India also offers certain open-access datasets containing data from various Ministries to enable innovation and development in India. Therefore, even if explicit consent is obtained from the data principal, it is essential for industry-specific data governance legislation to enforce such anonymization techniques before personal data is used in ML training datasets. • Rule 2: Mandatory usage of applicable machine learning subsets. The Draft Personal Data Protection Bill, 2019 has categorized personal data and has proposed general regulatory measures for the specific category. The proposed regulatory provisions are indeed promising but will fail to achieve its objectives if data is used for secondary purposes, such as Machine Learning training datasets. Therefore, specific learning methods have to be introduced for certain industries. Machine Learning algorithms’ training on clinical data for Healthcare development should be mandatorily performed using ‘Federated Learning’ likewise training on financial data for Investment and Banking development should be mandatorily performed using ‘Shared Machine Learning’. The aforementioned learning technologies, although voluntary, have been already implemented in certain overseas organizations. The industry-specific data regulatory policies for the healthcare industry and Investment and Banking industry should impose the mandatory usage of such training methods. • Rule 3: Regulation of dataset purchases from the private sector. The Government of India offers certain open-access datasets containing data from various Ministries to enable innovation and development in India. However, it does not consist of robust and comprehensive datasets across various sectors and fields, thereby creating a shortage of quality intelligent data. The quality of the dataset affects the performance accuracy of the Machine Learning

118

Artificial Intelligence and Policy in India, Volume 1(2020)

program. Considering the lack of any legislative policy for data protection in India, the majority of data is owned and controlled by the private sector, and thus Machine Learning developers turn to private sector industries for datasets. However, due to the high costs of such private-sector datasets, small-scale innovators and start-ups face a serious obstacle. Therefore, a regulatory framework has to be established that regulates the pricing and quality of the datasets to ensure a fairground for innovation in India for both large-scale innovators and start-ups. â&#x20AC;˘ Rule 4: Sectoral data localization obligations. Considering the multitude of opportunities emerging from Machine Learning technology, the correct rationale for a developing country like India would be to analyze data localization measures. Industry-specific data localization measures for industries that generate sensitive or crucial personal data would ensure local storage of such data, increased accessibility to government institutions and growth in innovation. Local storage of data would provide the government with better access and control over citizenâ&#x20AC;&#x2122;s personal data. The rate of accessibility of data for quality machine learning datasets will likewise improve. Industry-specific data localization obligations are more favorable than absolute data localization measures over all industries as an absolute measure would cause a negative economic impact.

Conclusions

In conclusion, the benefits of machine learning and its subsets are well known to the involved stakeholders. However, the benefits of this technology do not overshadow the privacy implications it has. This paper recommended certain legal provisions to ensure privacy protection while keeping the regulatory adversarial impacts on the industry at the minimum. Machine learning and its subsets are truly revolutionary technological developments that, if utilized with utmost precaution.

References 1. Do Australian Universities Encourage Tacit Knowledge Transfer? Chugh, Ritesh. 2015. 1, Melbourne : Science and Technology Publications, Lda, 2015, Vol. 3. 978-989-758-158-8. 2. Polanyi, Michael. 1966. The Tacit Dimension. California : Peter Smith, 1966. p. 108. 0844659991. 3. Prabhu, Manish. 2019. Security & Privacy Considerations in Artificial Intelligence & Machine Learning- Part 6: Up close with Privacy. towards data science. [Online] 02 09, 2019. [Cited: 06 23, 2020.] https://towardsdatascience.com/security-privacy-in-artificial-intelligence-machine-learningpart-6-up-close-with-privacy-3ae5334d4d4b. 4. Robust De-anonymization of Large Sparse Datasets. Narayanan, Arvind and Shmatikov, Vitaly. 2008. s.l. : The University of Texas at Austin, 2008. 5. Some Studies in Machine Learning Using the Game of Checkers. Samuel, Arthur L. 1959. 3, s.l. : IBM , July 1959, IBM Journal of Research and Development, Vol. 3, pp. 210 - 229. 0018-8646. 6. The "Re-identification" of Governor William Weld's Medical Information: A Critical Re-examination of Health Data Identification Risks and Privacy Protections, Then and Now. Barth-Jones, Daniel C. 2012. s.l. : SSRN, June 18, 2012. 7. Who's Watching? De-anonymization of Netflix Reviews using Amazon Reviews. Archie, Maryam, et al. 2018. s.l. : MIT, 2018.

Artificial Intelligence and Policy in India

119

8. Wladawsky-Berger, Irving. 2018. What Machine Learning Can and Cannot Do. Medium. [Online] Medium, July 23, 2018. [Cited: June 23, 2020.] https://medium.com/mit-initiative-on-the-digitaleconomy/what-machine-learning-can-and-cannot-do-6788a818776.

120

Artificial Intelligence and Policy in India, Volume 1(2020)

â&#x20AC;&#x2DC;

Discussion Papers and Research Articles

Artificial Intelligence and Policy in India

121

On the Entrepreneurial-Employability Ecosystem & Technology Preparedness due to COVID19 in India: Seminal Analysis & Proposed Solutions via the AI Ecosystem Abhivardhan, Suman Kalani, Ankur Pandey, Baldeep Singh Gill, Kshitij Naik, Manohar Samal, Ritansha Lakshmi, Adetola Jesulayomi and Ruhi Tyagi abhivardhan@isail.in; editorial@isail.in

Synopsis. The Background Analysis covers the original mandate and purpose of the COVID19.AI Report, the initiative by Indian Society of Artificial Intelligence and Law. The report covers two key questions of concern: • That how COVID19 will affect the entrepreneurial and employable sectors of the AI industries in terms of economics and skill development? • That how can AI guide to prevent the social, ethical and legal fallacies in human society to prevent the spread of COVID19? The questions are addressed in the domains of Law, Ethics, International Affairs, Technology Ethics and Economics. Furthermore, the report provides seminal suggestions as to the context of the COVID19 crisis in the Indian context. The background analysis provides recommendations in the domains related to: • Law, Governance and Constitutionalism • Economics, Health and Administrative Infrastructure • Entrepreneurship, Skill Development and Innovation • Technology, Liability Frameworks and Health Securitization • International Affairs, Multilateralism & Diplomacy

Introduction

The COVID19 outbreak has had special implications to affect global health and economic order. There are circumstances, where states are not able to manage their status quo. In the context of India, despite various frugalities, at least a sincere case of activity and materialization seems to be quite possible. The paper lays down focus by hypothesizing two basic questions, which we intend to address through the paper: • That how COVID19 will affect the entrepreneurial and employable sectors of the AI industries in terms of economics and skill development? • That how can AI guide to prevent the social, ethical and legal fallacies in human society to prevent the spread of COVID19?

122

Artificial Intelligence and Policy in India, Volume 1(2020)

The policy paper therefore critically provides insights and recommendations in the areas of economic mechanisms and technology preparedness (in a comprehensive sense), encompassing domains such as: ─ ─ ─ ─ ─

Law, Governance and Constitutionalism Economics, Health and Administrative Infrastructure Entrepreneurship, Skill Development and Innovation Technology, Liability Frameworks and Health Securitization International Affairs, Multilateralism & Diplomacy

The analysis takes into consideration the viability and usefulness of the AI ecosystem developed by governments, entrepreneurs and industries. 1.1

Addressing the Questions

How COVID19 will affect the entrepreneurial and employable sectors of the AI industries in terms of economics and skill development? The question covers the issues related to the employment and entrepreneurship sectors in the Indian economy. A seminal focus is on the issues related to: 1. Skill rejuvenation education 2. The decline in Microfinancing and Economic Feasibility 3. Contractual Obligations and their Scope in Force Majeure Conditions 4. Ingraining Global Health Security Measures to Minimize Economic Losses Due to COVID19 5. Establishing and Scrutinizing the Intellectual Property Rights Obligations How can AI guide to prevent the social, ethical and legal fallacies in human society to prevent the spread of COVID19? The question covers the issues related to constitutional governance, democratic liberalism, diplomatic morality and legitimation and administrative issues related to India as a liberal democracy. A seminal focus is on the issues related to: 1. Constitutional Redemption in Interpreting and Assuming the Scope of Fundamental Rights for Citizens Beyond the Debate 2. Resolving Administrative Issues 3. Adjudicatory and Policy Interpretation of Privacy Rights and Liberties Issues 4. Expanding and Democratizing Individual Liberalism and Coherence Between Community and Individual Rights 5. Questioning and Scrutinizing the Role and Legitimacy of the World Health Organization 6. India’s Diplomatic Stand and COVID19 Diplomacy with Western Bloc and China 7. Future of Multilateralism and India’s Persistent Role as a Stabilizing Power in a Multilateral World

Artificial Intelligence and Policy in India

1.2

123

Interpreting the Issues

The context behind the questions and seminal issues addressed is based on the premise that the most affected systems in the rules-based global order are (a) the global economy and its supply chains; and (b) the global multilateral order in many dimensions. We, therefore, propose that dealing with the issue of COVID19 in the context of the present AI ecosystem is possible in India. However, we see that in the case of India, a model of development and resilience would take time, and is central to the ideology of development. India, despite various constitutional and administrative redemptions, can adapt to newer avenues of constitutional viability and restore balance in its constitutional order by a communitarian approach of its administrative institutions and can set an example of a ‘development’ and ‘resilience’ power in the Global South. Furthermore, beyond revisiting the culture of constitutionalism and changing its orientalist attitude, India needs to engage impeccably in the Global South and embrace its narrative of multilateralism to provide relevant support to the international organizations and other subsidiary powers to detoxify the age of technology diplomacy. How COVID19 will affect the entrepreneurial and employable sectors of the AI industries in terms of economics and skill development? Skill rejuvenation education ─ Once the COVID19 crisis deepens and precedes, it is required that proper methods are established to provide solutions in the areas of skill development considering the paradigm shift of approaching AI as a utility. It is quite certain that the idea of skill development and work rights will change – but will need to shape in a more neorealist and post-neoliberal way, therefore to balance global capitalism. ─ As the world is in the grip of global pandemic and lockdown, it has presented an atypical challenge for skill development systems. In-person classrooms now have shifted to online classes including recorded and live sessions while some follow the mixed discipline of online and physical classes where situations are under control. Some innovations in this space have built platforms to accept feedback from students and teachers to enhance the experience. Meanwhile, some countries are working on their strategies to take examinations online and further assessments which were initially taken as paper-based. Distance learning experience had been quite rare as people preferred conventional classes however, the COVID19 outbreak has put a relative strain on the social distancing mechanisms from one another and no other option has been left. But settling for the online experience is difficult for some students, as well as people in large populations in India do face lack of laptops, personal computers, or have an internet connection facilities. ─ Since the education system has been drastically affected, all the countries around the globe are worried, especially the ones with already low education outcomes, high drop-out rates, and high resilience to shocks (Kaliope and Tigran 2020). The prolonged closures have caused a burden on guardians who cannot provide proper day-care or proper regular meals. On top of that, many underdeveloped countries have not reported cases (Sharma 2020) properly thus, it creates a problem for delivery service and prepares for the future. Unlike India, not all countries are keeping schools closed some countries have developed other strategies to cope with the situation.

124

Artificial Intelligence and Policy in India, Volume 1(2020)

For instance, Afghanistan enforces and supports preventive measures in schools, and Singapore and Russia have limited physical contact activities to reduce virus spread. While the majority of schools and colleges are closed there are certain countries viz. Austria, China, Germany, Laos, and New Zealand are shortly re-opening their institutions with fewer working days and restrictions. While Vietnam, which has had under 300 infections, no deaths, and no new cases so far, has resumed the activities compared to other countries. There is a need to acknowledge the fact that the concept of literacy is twisted. Due to the vague definition of Literacy (Profile - Literacy - Know India: National Portal of India [no date]), the literacy rate in India stands at 74.00% (STATUS OF LITERACY 2011) which does not represent the true picture. To consider a person literate, new criteria should be established. that a person who has at least completed elementary education should be treated as a literate. According to the 2011 census, 2,59,678 spoke English (CENSUS OF INDIA 2011 2018). A survey conducted by Lok Foundation and Oxford University found out that only 6% population could speak English (S 2019). Given that majority of computer programs and software are developed in English, along with native language (s) we need to adopt English as a universal language. Decline in Microfinancing and Economic Feasibility ─ Economic inequality is an important issue, and for the Global South, the conditions require coordinated stability based on sovereign imperatives. Seeing the course of nations in continental Africa, ASEAN and SAARC region, it is quite clear that India needs to cater more options to face the issues related to microfinancing and economic feasibility. ─ Though it is utterly integral to practice social distancing, sanitation, and fewer outdoor trips, there are times one has to leave their house to buy daily essentials and in case of emergency, many affluent people were seen stocking up the commodities leaving less for the impoverished. Meanwhile, the supply chain stopped domestically as well as internationally, the rates of certain essential goods raised due to scarcity and highly affected the bourgeois and destitute people. . About 0.7% of GDP loss can be faced annually when future pandemics arrive, as predicted by economists (Charlton 2020). According to the World Bank President, the pandemic has set a taxing situation such that it might wipe 5-10% of the GDP of the world’s economy and kill over ten million people. The economy has already contracted in the first quarter of 2020, and about to slump further in the second quarter, meaning a recession is likely (World Economic Forum 2020). It is difficult to frame all the economic losses and it makes the problem look much larger in scale. The outbreaks of future diseases are unavoidable however mitigation of economic losses is not. By building new strategies for now, can help to fight with possible akin situations. The essential point of discussion is how nations are working on averting the outbreaks by preparing robust and functioning healthcare systems (Politzer 2017) enough to respond to emergency health risks. According to Dr David Heymann, senior fellow of the Centre on Global Health Security at Chatham House, international affairs think tank, any plan to deal with a potential health security crisis — whether an epidemic, pandemic, flood, earthquake, or act of biological terror — must have three core components: effective prevention, detection, and response. Effective surveillance of animals living in proximity of humans must be carried out such that no further diseases are propagated to another species. As the current coronavirus is suspected to spill over from bat to pangolins and then to humans via consumption or coming

Artificial Intelligence and Policy in India

125

in contact with them, the Chinese Government has closed the sale of any exotic animal meat especially in the epicentre of the disease, Wuhan. Many scientists have suggested that there are millions of unknown viruses in wildlife and few thousands of them are known, their meat consumption or close contact must be avoided. Still, the world is continuously in touch with wildlife and learned nothing from the past deadly epidemics The COVID19 pandemic has been noted as that which may bring the worst financial crisis ever seen since the 2008 financial crisis but the big question is how large will these costs be for the financial institutions specifically for the Microfinance institutions ( MFIs ). There have been contentions whether the decline in microfinancing during this Covid 19 period is good or bad. India as a case study, microfinancing in India is gradually being recognized as that drive that will make India attain its financial inclusion goal. Although, the sector's contribution to the nation's economy is highly appreciated but there are still disagreements which may put the whole industry in danger. First is the high interest rates charged by these MFIs although it is lower than what the money lenders charge. In the Bangladesh's microfinance sector, the lending institutions charge about 30-40 percent interest on loans. Also, most of the MSEs do not get their returns back as the economics of microfinance requires high repayment rates. And if there is a slip in repament rates from 95 to 85, these would render many MFIs insolvent in less than a year as burrowers struggle to make ends meet in this period of income shock as a result of the Covid 19 pandemic lockdown. Furthermore, even before the outbreak of the virus in the global south, macroeconomic conditions had deteriorated rapidly enough to cause serious shocks to many developing economies. Natural resource prices plummeted as global demand shrank and China shut down its factories. But it is critical to note that even with all these disadvantages attached to that industry, the microfinance industry has helped various economies in the past and I believe that it can still do so. What has the sector done? The microfinance industry offers a global opportunity though riddled with challenges to offer health related services to those who are in most need. Also, the microfinance industry is the key driver of financial inclusion, a financial system for the poor and lowly most especially in developing countries. There are about 3500 microfinance institutions around the world that provides credit and other financial services to more than 155 million households who are in need of income generation and consumption. Contractual Obligations and their Scope in Force Majeure Conditions â&#x201D;&#x20AC; The existential instruments and liability framework that span the legal machinery of contractual liability in matters like employment, innovation, health, entertainment, education and others had not expected or seen such repercussions before the COVID19 outbreak. India must show a model to prepare and economize its force majeure framework in matters related to contractual obligations created. â&#x201D;&#x20AC; The ongoing outbreak of COVID-19 has disproportionately affected lives and livelihood all across the globe. While the health infrastructure has been at the forefront in this battle against the novel coronavirus, the containment measures adopted by government across jurisdictions have led to economic disruptions. Businesses and commerce- both domestic and international- have been affected and many of them failed to discharge their contractual obligations owing to restrictions placed by the governments. However, the legal doctrine lex non cogit ad impossibilia states that the law does not compel a person to do something which

126

Artificial Intelligence and Policy in India, Volume 1(2020)

is impossible to be performed. Since the failure to perform contractual obligations was not attributable to any of the parties to the contract but to COVID-19, we need to ascertain if the contract laws provide for such contingencies. Force majeure is a civil law concept, frustration of contract being its equivalent under common law. The term was borrowed from the Napoleonic Code 1804 (French Civil Code: Book III: Of Acquiring Property, Title III 2020). A force majeure event, when it occurs, excuses performance of the contract in spite of its express provisions obligating the parties to perform (Azfar 2012). Force majeure clauses are mentioned in the contract stipulating the eventualities under which a party will not be liable for non-performance of contractual obligations. However, not all contracts may include a force majeure clause. Also, while some force majeure clauses might be open-ended, capable of wider interpretation, some might be exhaustive and not inclusive of events such as COVID-19. To understand the obligations of parties in these cases, it becomes necessary to examine the provisions of the Indian Contract Act, 187211 that provide for such eventualities. Supreme Court has held that Section 32 and Section 56 of ICA, 1872 are relevant provisions which regulate the non-performance of contract due to occurrence of events which render the discharge of contractual obligations as impossible (Ganga Sharan v Ram Charan 1952). While Section 32 deals with cases where the contract itself, either impliedly or expressly contains a term, according to which performance would stand discharged under certain circumstances, Section 56 deals with the cases which are not specifically provided for in the contract but still render the discharge of contract as impossible (Energy Watch Dog vs. CERC 2017). Thus, even in the absence of a force majeure clause, a party to a contract can take the defence of Doctrine of Frustration under Section 56 of the ICA, 1872, if the performance of the contract becomes impossible due to unforeseeable events. While the impossibility under Section 56 is not just literal impossibility but also impracticability, mere temporary impossibility does not discharge the parties from contractual obligations, unless and until time is the essence of the contract (Mugneeram Bangur & Co. v Gurubachan Singh 1965). â&#x201D;&#x20AC; Since the COVID-19 pandemic is only a temporary situation, only those contracts will stand frustrated in which the time of performance was essential. For the rest of the contracts, the contractual obligations will be postponed till normalcy revives. Also, in a case due to the prevailing pandemic the fundamental basis of the contract itself gets frustrated, any advantage received thus far under such contract by any party shall have to be restored to the opposite party under Section 65 of ICA, 1872. However, a lot depends upon the actual language of the contract, and if the contract has an exhaustive list of contingencies that do not include a pandemic, it would be hard to infer one, and in such a case the parties will be left with no remedy for non-performance. Ingraining Global Health Security Measures to Minimize Economic Losses Due to COVID19 â&#x201D;&#x20AC; India has to face a central issue apart from the outbreak of COVID19, which is to maintain and convene the issues of major economic losses incurred to the national economy due to COVID19. However, economic losses at this time can be appropriately minimized by

Please refer to the Section 32 on Enforcement of Contracts Contingent to Events, Section 56 on Agreements to perform Impossible Acts & the Section 65 on the advantages received by a person on a contract that becomes null and void or is a void agreement of the Indian Contract Act, 1872.

Artificial Intelligence and Policy in India

127

ingraining certain global health security measures, in line with the International Health Regulations, 2005, or at least by providing a protectionist roadmap to curtail economic losses. While the threat is natural (a pathogen), pandemic frequency and impact mostly depend on people action. Weak policies and governance can turn an infectious disease in a small part of a country into a pandemic, resulting in economic loss, as well as impacting health. It is also important to encourage countries to work together to share information about known diseases. Being prepared to cope and lessen impacts on a pandemic is required. Policies and budgets should therefore systematically anticipate pandemics and assess the risks; most whole-of-society preparedness measures are the same as for other major complex disasters. Weak interaction between veterinary and public health department in developing countries is one of the key drivers of pandemic risk. As the veterinary health system can detect pathogens early, diagnose them correctly, and control infection before it spreads within the country and across. As most of the risk originates in livestock, which is under human control and can be controlled at its animal source, so it does not spread and become a pandemic in humans and results in disturbing economic prosperity of a country. Pandemic amnesia and collective myopia in governments and international organizations are pervasive. When a disease is not controlled at its source it results into the sudden pandemic outbreak, people across the world suffer both an infection of disease and a sharp, probably disastrous, economic depression accompanying with shifts in the demand curve, supply imbalance, and economic and social disturbances. Because countries are interconnected by, and depend on, travel, trade and capital movements, the disturbance would spread across worldwide economic and financial systems, possibly ahead of the infection itself. Moreover, many developing countries depend on remittances, tourism, world capital markets, and foreign trade markets for their exports and imports. The importance of these vulnerable links would differ across countries, and many developing countries could be especially prone to disruptions. Active promotion of whole-of-society resilience and pandemic preparedness can benefit countries by reducing not only pandemic impact, but also the costs of other disasters and major crises in term of economic loss. The international community is left with two choices: Either continue to rely on emergency responses, with their high-impact/low-sustainability trade-off and their huge human and economic losses or commit to supporting systemic prevention efforts that will lessen pandemic risks and deliver considerable long-term health and economic benefits. All countries benefit if each undertakes adequate whole-of-society preparedness measures to increase resilience and capacity to respond to disasters and thus limit spillovers of negative impacts to other countries. Establishing and Scrutinizing the Intellectual Property Rights Obligations â&#x201D;&#x20AC; The innovation and originality scrutiny and assessment framework will have significant changes, especially in the horizon of digital rights and spaces. India can be a preparatory hub for developing global south economies to retain and rewire the IPR issues that may arise in the economic context.

128

Artificial Intelligence and Policy in India, Volume 1(2020)

How can AI guide to prevent the social, ethical and legal fallacies in human society to prevent the spread of COVID19? Constitutional Redemption in Interpreting and Assuming the Scope of Fundamental Rights for Citizens Beyond the Debate ─ India has to revisit the limited modalities of its constitutional structure and revisit certain autarchic issues that arise due to certain old legislative instruments related to the policies that encompass the scope of the Indian Constitution. In the case of the propositions and recommendations in the paper, we are referring to the legislations related to disaster management and healthcare. Furthermore, India has to think how it can revisit and balance its commitment to ideological and ethnic diversities that are limitedly sheltered within the constitutional framework with certain major problems in course, which is integral. Resolving Administrative Issues ─ The administrative machinery of the Union and State Governments in India needs a new policy and perspective to combat challenges in achieving cooperative and competitive federalisms. India must balance between its technocratic and communitarian ends of administrative legitimacy and ethics and must resort to a technocentric way of fixing leakages in the framework. Adjudicatory and Policy Interpretation of Privacy Rights and Liberties Issues ─ The concept of privacy in Constitutional Law needs to be revisited due to some specific reasons: • the current juridical framework would attract more technocratic and kleptocratic measures, therefore putting the administrative framework into a dilemma due to the orientalist approach of dealing with constitutional morality; and • the transition of the privacy framework from pure law to applied law is not easy, and needs to understand the sociological and multi-lithic design of the Indian Society in an anthropomorphic way. Expanding and Democratizing Individual Liberalism and Coherence Between Community and Individual Rights ─ Important questions must be raised to consider whether the judgmental approach of treating with individual liberalism is sufficient. Also, community rights must not be viewed at a collectivist level, but they should be seen as a complementary submergence with individual values that encompass the human rights regime in India as a liberal democracy. Questioning and Scrutinizing the Role and Legitimacy of the World Health Organization ─ The World Health Organization faces restrictions in his expanding and exercising its functions due to certain mishaps and misled measures by to contain the COVID19 outbreak. Furthermore, its relations with the establishment from the People’s Republic of China do prove that the International Community cannot rely on the pretext of innocence anymore. In this scenario, India can drive the WHO out of its own administrative obscurities.

Artificial Intelligence and Policy in India

129

India’s Diplomatic Stand and COVID19 Diplomacy with Western Bloc and China ─ India’s multi-aligned approach in its Foreign Policy has improved its foreign relations and has the capacity to build India as a stabilizing power. Furthermore, India can share its diplomatic narrative with the West as a trusted nation, with capabilities that are underestimated by the West and China. Future of Multilateralism and India’s Persistent Role as a Stabilizing Power in a Multilateral World ─ India can use its cultural confidence to reshape the dynamics of multilateralism by revising the approach of the Global North towards the Global South. Furthermore, India’s presence and viability in various international organizations is trusted with a deemed sovereign imperative by various nation-states in Europe and the Atlantic, which India can maintain and establish in a communitarian global order.

Recommendations & Inferences

At the best of our understanding and estimates, we would like to propose the following recommendations: 2.1

Law, Governance and Constitutionalism

• The Constitution of India, 1950 is the law of the land. Any law passed or policy implemented needs to derive its legitimacy from the Constitution. If it fails to do so, the law or the policy measure is liable to be struck down. It is a given that desperate times need desperate measures. But the measures have to be worked out within the boundaries defined by the Constitution. • In the time of Pandemic, we should not lose vision of the inventive capacity of the law to maintain its significance. Necessity is a doctrine which ties the gap between what the law allows the government to do and the government's actual response at the time of emergency. • Since the Epidemic Diseases Act, 1897 was enacted before the Constitution of India, 1950 was enforced, the restrictions related to Fundamental Rights are yet to be read into the Act. The legality of other measures taken by the Central or State Governments needs to be ascertained too. Then, it is not that only these statutes and measures need to pass the test of the Constitution. Rather, the Constitution itself needs to be analyzed and amended if it fails to provide for such pandemics because after all, our Constitution is a living document. • Two legislations that have provided statutory backing to the government orders to control the pandemic have been the Epidemic Diseases Act, 1897 and National Disaster Management Act, 2005. EDA, 1897 gives carte blanche to the State Executive to take special measures and prescribe regulations as to any dangerous epidemic disease. • Another major issue with the orders issued under the EDA, 2897 and NDMA, 2005 is that they fail to be the least restrictive alternative. The nationwide lockdown resulted in a sudden loss of livelihood for lakhs of migrants who, in absence of any legal remedy, defied the very

130

•

Artificial Intelligence and Policy in India, Volume 1(2020)

lockdown and started walking back to their homes, resulting in an otherwise avoidable catastrophe. Similarly, the policy of the Karnataka government asking those quarantined for an hourly selfie, violation of which is to be penalized, prima facie fails to acknowledge Right to Privacy as a fundamental right. Thus, it is apparent that not only the existing statutory provisions need to be tested on the principles of reasonableness and proportionality, but there is also a need for specific legislation to deal with public health emergencies. Once it is acknowledged that we lack a robust statutory mechanism to deal with public health emergencies like the present one, it is imperative that we devise such a mechanism to guide our pandemic planning and response. It is conceded that Ethics is a precursor to Law, and our statutes and policy to tackle such a crisis must be based on sound ethical principles. On the face of it, the dilemma faced by pandemic policymakers is how to strike a balance between individual rights and public interest. That is why the issues of individual liberty, dignity, privacy, and equity are at the forefront while deciding policy questions, especially those allowing surveillance and data collection. The fixation over individual rights and interests makes us oblivious to the socio-political context and how certain communities as a whole are at a disadvantageous position due to the systemic inequalities present in society. Policy needs to be framed keeping these underprivileged, disadvantaged groups in mind. For example, senior citizens, homeless or slum dwellers, migrant workers will face the severe consequences of the nationwide lockdown and special provisions need to be made lest the lockdown itself does more harm than good to them. Social justice can only be achieved by acknowledging and providing for the needs of the underprivileged sections of society. The community ethics principles like solidarity, trust, neighborliness, reciprocity need to be endorsed along with individualistic notions of ethics. The policymakers mostly consider utilitarian ethics to achieve the greatest good. However, this might result in benefiting a section of population that is already at lesser risk due to inherent advantages like access to health care facilities. The more vulnerable population might get excluded as they would require more resources with lesser chances of successful recovery. Hence, the utilitarian approach needs to be balanced with a deontological approach which places a duty on the health care system, and a rights-based approach which places the onus on the State to provide the best possible medical care. The consequence of law which curtailed individual freedoms guaranteed by Article 19 would be required to answer the tests of reasonableness stated in clauses 2 to 6 of Article 19 and the State must satisfy that both the fundamental rights are not infringed by showing that there is a law and that it does not amount to an unreasonable restriction within. Similarly, the ethical principles of equity and fairness need to be considered to prevent any unfair discrimination based on the gender, caste, color, race of any individual or group. Efficiency, to maximize the benefits with minimum resources, and effectiveness, to translate the policy measures into practice, are equally important. Additionally, the principle of transparency achieves great significance during such a crisis. The accurate dissemination of information about the policy decisions and the prevailing situation on ground is essential to prevent any panic among the affected population. A related principle of participation should be realized so that the stakeholders have a say in the decision-making process. Above all, the accountability of all stakeholders must be fixed and the policies should be continuously reviewed and revised so as to produce the best possible outcome.

Artificial Intelligence and Policy in India

131

• It is essential that a nation as populous and diverse as India has a robust statutory framework, guided by constitutional and ethical principles, to deal with public health emergencies. COVID-19 has surely tested our preparedness in dealing with pandemics. We have to make the best use of the opportunity to learn from our mistakes and come out of this stronger. • Quarantine, across the world, is proving to be the best bet in the containment of Coronavirus disease and the same is prescribed by WHO. It might be interfering with our liberty but such a reasonable restriction is even permissible arguably under the Constitution of India, 1950 itself in the interest of the general public and order, it is also our duty as a citizen to cooperate with the government and help to break the chain of infecting coronavirus by being at indoors. 2.2

Economics, Health and Administrative Infrastructure

• Drones should be deployed as a means to fight the deadly coronavirus pandemic by the government. One of the safest and fastest ways to get medical supplies to deliver where needed during an outbreak. Drones can also be used to patrol public places, track noncompliance to quarantine mandates, and for thermal imaging to detect temperature. • Local medical stores, pharmacies, private hospitals should have to tie a rope with the governments and provide information of patients to govt health care dept and closely work with govt agencies. • Thousands of laborers, migrants, who fear dying not from the deadly virus but rather from hunger. Govt can’t lock them down through awareness. The only thing government can do is to feed them with food and basic requirements. They can remain quarantined as long as they get food to eat with very basic requirements met. • All the old or illegal buildings and the building which are on hold should be converted to make-shift hospitals with basic medical facilities and isolation wards to cater to the growing number of cases. • The services like flights and trains are being resumed, markets are slightly open and so – very limited public movement would be endorsed. There has been check-post set at airports and railway stations were the temperature of travellers is taken and upon results, they are given prescriptions to quarantine for 14 days at least. Furthermore, local vendors and shopkeepers are subjected to take tests every week and dispensed health cards to help contain and recognize infections • At the outset, it is important to showcase how artificial intelligence is already enabling and empowering administrative authorities in the combat against the coronavirus (COVID-19) pandemic. Artificially intelligent tools are being employed for identifying, tracking, forecasting outbreaks, diagnosing the virus, processing healthcare claims and in the functionalities of drones (Marr, 2020). In India various private entities and the Government have been reported to be using artificial intelligence in administering relief against COVID- 19. • Larsen & Toubro Smart World & Communication technology solutions used by various local authorities and police agencies in cities like Pune, Mumbai, Visakhapatnam, Nagpur, Hyderabad and Prayagraj for complex civic and enforcement activities (Prasad, 2020) is one such example of public private partnership for battling the pandemic in India. Furthermore, even support from non- governmental organisations and companies like PATH and

132

•

Artificial Intelligence and Policy in India, Volume 1(2020)

Quere.ai are using artificial intelligence for assisting Indian administrative authorities during the pandemic stage (Jain, 2020). However, considering the size of the total Indian population these measures are not enough. It is pertinent that emerging technologies have to be adopted to further ameliorate the situation. No doubt the non- technological administrative measures taken by the Indian Government are laudable but in order to ensure that more progress is seen in the battle against COVID- 19 in India, adoption of artificial intelligence technologies is indispensable. The use of artificial neural networks can help in identifying significant activities inside the cells of the virus which can lead to the development of targeted medicines (Bansal and Gopalakrishnan, 2020). Artificial neural networks can also be used in chatbots and in scanning body temperatures of the population in order to detect traces of the virus (Roy, 2020). Furthermore, smartphone applications and diagnostic systems combined with artificial intelligence containing deep learning algorithms are aiding the battle against the COVID- 19 too. Deep learning algorithms in diagnostic systems have enabled the tracing and detection of the virus within 20 seconds in chest X-Glorays with an approximate 96% accuracy (Bansal and Gopalakrishnan, 2020). Moreover, artificially intelligent drones and robots can also join the fight against the virus if they are deployed in large quantities. The combined usage of blockchain and artificial intelligence is also being seen as a viable option to increase the support for tackling various issues of the pandemic in India (Roy, 2020). IN addition to these, natural language processing for tracking, recognising and reporting the spread of the virus, facial recognition and fever detector technologies and information verification through artificial intelligence can also be dispensed for administration purposes (Obeidat, 2020). One of the successful executions and applications of artificial intelligence can be seen in Kanpur Smart City Ltd. where the city has partnered with Tech Mahindra to set up a control room to track patients, lockdown violations, supply of essential and healthcare items using artificial intelligence and geofencing solutions (Press Trust of India, 2020). There are multiple innovative and upcoming artificial intelligence technologies which countries across the world are utilising to fight the virus. India’s focus on availing these technologies can act as a catalyst for saving many lives. There have been reports from all around the nation which portray that many doctors, police personnel, local self government workers and other healthcare, relief and enforcement workers popularly referred to as “the corona warriors” in India have been tested as COVID- 19 positive (Ali and Mahamulkar, 2020). Therefore, it is not wrong to presume that the enlarged utilisation of artificial intelligence can help in reducing the number of healthcare, relief and enforcement personnel being affected by the virus. Few of the innovative and upcoming artificial technologies which India can adopt are the use of telepresence robots, digital teleworking tools, home- schooling solutions and safe food delivery supply chain systems for quarantined populations (Clifford, 2020) in order to ensure compliance with the lockdowns. In light of the recent division of India into green zones, red zones and yellow zones (Times of India, 2020), the use of artificially intelligent enabled tools for administrative purposes can indeed bear fruitful results in administering the spread of the virus in the respective zones as per its requirements. Microfinance institutions have proven capable of contributing their own quota by improving health care facilities and the outcome as they educate people and providing health financing options such as microfinance and savings as well as even delivering directly clinical care. There have also been numerous positive outcomes from the combination of

Artificial Intelligence and Policy in India

•

• •

133

microfinance and health education in areas such as Reproductive health, HIV prevention, child nutrition, Malaria, Sexually transmitted diseases etc. In order for the Global South countries to survive, it is important that they respond with an online means that is digital payment for social protection. It is proposed that the future of economies of most countries especially the global south countries may be bleak as they continually expand their funding which were never even enough in the first place. So to counter this, it has been recommended that the Government to Person ( G2P ) payment is the best repose to the economic and social consequences of the Covid 19 pandemic. It is important to take note that 656 million people living in extreme poverty worldwide, an immediate cash support can be lifesaving. Emphasis must also be placed to ensure that the digitalization of payments should not exclude the vulnerable populations, such as the elderly, those without access to technology, the disabled and those living in remote areas. These cash transfers will prove to be important in supporting recovery and rebuilding livelihoods when this pandemic is over. While the number of cases in the global south remain relatively low compared to countries which are highly affected, leaning on Africa specifically, it is highly recommended that their government build enough public health capacity to contain the virus. Isolation centres and buildings specifically meant for testing, tracing, isolating and treating cases should be the number one priority. Also, in light of these countries shortage of health care workers, a number of lab technicians, field epidemiologists and other frontline workers should be heavily recruited and trained as this is important in effectively fighting the virus because especially in Africa, Nigeria specifically there is a huge shortage of health care workers and isolation centres as well as basic personnel and life saving equipment. They should also continue to enhance awareness through effectual risk communication concerning COVID-19 to the general public, health professionals, and policy makers. There should be a harmonized and coordinated cross continental effort for the virus' spread and control. For example in Africa, the Africa Centres for Disease Control and Prevention (Africa CDC) roles should be accelerated as it is the first public health institute which has been instituted to fight against infectious diseases like Coronavirus among the group of independent countries. As its mandates states, Africa's CDC should begin its surveillance, prevention, and response measures which will shift the focus from quarantine and embargoes at borders to containment at the source. Also, there should be close partnerships with the more established and experienced CDCs of the global north countries like, China, Europe, and the U.S as they can offer the best practices and strategies for dealing with this coronavirus crisis. Furthermore ,to curb the spread of the coronavirus among the global south countries, as it is obvious that the nations cannot defeat the virus alone, it is therefore highly recommended that the global north, donors and individuals should come together and provide enough funding to help these countries fight against the virus. This will not only help curb the spread of the virus but will serve as a massive help to reduce the economic decline in these countries. There will be a reduction in the economic losses experienced by the countries and even help during the after effects of the virus.

2.3

Entrepreneurship, Skill Development and Innovation

• As the education paradigm has shifted from teacher-centric to student-centric curriculum, a stronger link must be formed between course-content, educationists, technology, and course

134

•

• •

Artificial Intelligence and Policy in India, Volume 1(2020)

takers. Many countries are working with content-creators (Sinha 2020), teachers, technicians, and telecom/internet infrastructure experts to cope up with the pressing situation. Following that many reforms are made and changed in areas comprising connectivity, telecom infrastructure, affordability of online system, availability of laptop/desktop, software, educational tools, online assessment tools, etc. Moreover, teachers are also training themselves to maximize their output and produce a class-like experience (UNESCO 2020) for students. They are now using online platforms to give assignments, conduct parent-teacher meetings, and assessment of examinations. This might be a big advantageous move for the future as not only online homework will save paper and cost, but the parent-teacher meeting can be set with every parent without any absence of some remotely distant parents. Digital Saksharta Abhiyan (DISHA), an initiative taken by the Government of India towards Digital Literacy has numerous drawbacks (Pradhan Mantri Gramin Digital Saksharta Abhiyan [no date]). The project is focused on rural areas and completely ignores the urban areas. The duration of the course is 20 hours which makes it difficult for Digitally Illiterate and Illiterate people to grasp all the information in such a short time. Literacy and Digital Literacy go hand-in-glove. Firstly the people should be literate and then Digitally Literate. Furthermore, such schemes should include both rural and urban areas. Given there is no marketing of such schemes and the people are unaware, Government should ensure the marketing of such schemes through television media, social media and other possible means. To ensure that the purpose of the scheme is fulfilled, there must be an assessment examination with a minimum passing criterion. The person failing the examination shall be allowed to reappear and the person passing with high marks shall be allowed to volunteer and teach other people enrolled under the scheme. Artificial Intelligence is a disruptive technology which will contour the future of society and the way everybody lives. COVID-19 has unveiled unprecedented challenges. The countries struggle to obstruct the pandemic and bring back life to normal (new normal or ab-normal). There is an expeditious surge in the use of AI and AI is being used to fight the COVID-19 by many countries. Therefore the need for AI literacy/education and Skill Rejuvenation is indispensable. The Central Boord of Secondary Education (CBSE) has adopted a remarkable ‘Twin Initiative’ i.e., first to introduce AI as an elective subject in Class 8,9 and 10 and, secondly the integration of AI with other subjects (ARTIFICIAL INTELLIGENCE INTEGRATION ACROSS SUBJECTS FOR CBSE CURRICULUM [no date]). The Tamil Nadu government in collaboration with Google and Microsoft plans to introduce a course on AI in government schools that all the state governments should consider the steps taken by the TN government. To build the future workforce, education and skills should be prioritized. The same can be accomplished by integrating AI education modules in DISHA curriculum. The curriculum should be divided into three categories i.e., Beginner, Intermediate and Advanced. Examinations must be conducted for every category to ensure the learning curve of each individual. After passing the examinations in all the categories, Practical Training shall be given to building the skills of the individuals. This shall include basic Computer Science, basic Programming and hands-on experience on AI systems. For the successful implementation of such programs, inter-ministerial coordination is imperative. The Ministry of Ministry of Electronics and Information Technology and State

Artificial Intelligence and Policy in India

135

Information Technology Departments should work in coordination to ensure funds from the Central Government, equipment, trained professionals, examination and training, deciding the curriculum for the above-mentioned categories etc. • To make AI and Digital Literate India, individuals or people who already have access to Digital/AI systems and are digitally literate should educate the people around them as a moral contribution towards the future workforce. To make India biggest economy in the world, a collective effort has to be taken by all the citizens. 2.4

Technology, Liability Frameworks and Health Securitization

• What we need to understand the weather COVID-19 itself or any of its consequences' such as Quarantine, government restrictions, closure of borders, suspension of any Flights or vessels entering the country can be brought under the clauses in the contract, only if the ambit of the clause is wide enough similar to Article 79 of the Vienna convention only then COVID-19 could be considered as an 'event' and the Force Majeure clause can be invoked. • In the case of Dhanrajmal Gobindram vs. Shamji Kalidas, (1961) 3 SCR 1020, the Supreme Court of India held that the term 'Force Majeure' has a wider meaning and held that the intention behind such a clause is to save the performing party from the consequences from the anything over which he has no control. However in recent times the Applicability of 'Force Majeure' majorly depends on the wordings of the clause if the clause specifically mentions the words "Pandemic or Epidemic" then the clause may be applicable or if the clauses is worded in such manner for Eg. 'Prevention of fulfilment of obligations due to governmental restrictions’, ‘any unforeseen circumstance that prevents the fulfilment of obligations’ then the parties may be excused from Non-Performance of the contract. • On the basis of the above discussion, it can be inferred that a party to a contract taking the defence of the force majeure or doctrine of frustration would be expected to prove the following (Nandini Gore 2020): (1) That the event which caused the non-performance fell within the ambit of such force majeure clause (2) That the non-performance of the contract was indeed due to that event (3) That the said event, as well as the non-performance of the contract, were beyond the party’s control (4) That no reasonable steps could have been taken to continue performance or there existed no alternative mode of performance. • It is the duty of the affected party taking the defence of the force majeure clause to notify the opposite party about the occurrence of the event and keep the opposite party updated during the subsistence of the event. It is also the duty of the affected party to mitigate the loss due to the event by taking all such reasonable measures as are necessary and practical. • The Ministry of Finance in its office memorandum dated 19.02.2020 has stated that any disruption of the supply chains due to spread of coronavirus in China or any other country can be treated as a natural calamity and force majeure clause can be invoked following due procedures (Office Memorandum 2020). The Delhi High Court has also ruled that nationwide lockdown due to COVID-19 is in the nature of force majeure (Halliburton Offshore Services v Vedanta Limited 2020).

136

Artificial Intelligence and Policy in India, Volume 1(2020)

• Thus, while the government has declared COVID-19 pandemic a suitable case to invoke force majeure clause, and the courts have generally interpreted force majeure clauses not to the detriment of the affected party, the ultimate test of such cases will depend on the terms and purpose of the contract itself and that would be examined on a case by case basis. 2.5

International Affairs, Multilateralism and Diplomacy

• In the recent outbreak of the coronavirus (COVID- 19) pandemic in the world, the role of the World Health Organisation has escalated. Because the World Health Organisation took an active role by consulting experts instead of consulting the leaders of individual nations, questions on the legitimacy of the World Health Organisation have surfaced in the past few months (Jansen, 2020). • The most prominent action taken by any country in the world in questioning the legitimacy of the World Health Organisation was that of the United States of America which halted its funding the organisation alleging that there was a failure to provide timely information and lack of cooperation from the World Health Organisation in respect to issues of the United States of America (Fidler, 2020). In consideration of the fact that the United States of America was the top donor, this could result in issues with the amount of action, World Health Organisation is capable of taking with its present funding (Global Health Policy, 2020). • The World Health Organisation’s Emergency Committee convened a meeting in January 2020 and recommended that the organisation declare an international public health emergency (World Health Organisation, 2020). By the time this was done, over 7,800 confirmed cases of COVID- 19 patients had already been diagnosed and by the time the organisation declared it as a pandemic in March, more than 118,000 cases had already been reported (Jansen, 2020). • A plethora of actions such as forming the Solidarity Response Fund in partnership with the United Nations (United Nations Foundation, 2020), publishing recommendations, technical guidance, daily situation reports and the prescription of International Health Regulations which are non- binding and are aimed at altering the behaviour of States and individuals (Jansen, 2020). The Constitution of the World Health Organisation enables it to exercise these powers (International Health Conference, 1946). • It is indeed undeniable that the World Health Organisation has taken a lot of efforts to fight the pandemic. But, the process of the organisation’s working has been extremely slow since it took 4 weeks to declare the COVID- 19 as a pandemic even though China had already furnished information to the organisation. Furthermore, the World Health Organisation’s slow response has been highlighted during the outbreak even on earlier virus outbreaks (Jansen, 2020). • The World Health Organisation is a specialised international organisation formed under Article 57 of the United Nations Charter (United Nations, 1945) and therefore, it is in the best possible capacity to deal in global pandemics such as the COVID- 19. The area of healthcare is a sector where negligence and no- accountability cannot be accommodated. This is because irreparable damage is the direct consequence of such negligence or no- accountability. Although the events that have already occurred are irreversible, the specialised nature of activities of the World Health Organisation can enable it to ensure that it rethinks its sequence of actions in the global sphere in a way that maximises prevention rather than providing relief after the damage is done. This could prove to be challenging because the

Artificial Intelligence and Policy in India

•

137

participation of various stakeholders and donors impact the intensity of action which can be taken by the organisation. Because the World Health Organisation issues various technical reports and guidance for nations from time to time, it can provide a structured document which can guide nations to formulate or amend legislation that will enable the prevention or at least, the minimisation of epidemics in the upcoming future. Moreover, any such document must be created with collaborations with other international specialised organisations so that an umbrella document can be created which envisages the distinct needs of nations in the world and addresses the needs of both, the Global North and Global South nations. Amidst the coronavirus (COVID- 19) pandemic, India’s commitment, diplomacy and response towards the crisis in the international sphere has been laudable. The Modi administration has already pledged $10 million to the South Asian Association for Regional Cooperation (SAARC) Fund. Furthermore, an Information Exchange Platform (IEP) was established to rally and provide interdependent support for the amelioration of conditions while battling the pandemic in South Asia (Zeeshan, 2020). Few partnerships which India has pursued with the Western bloc is by participating in the Franco-German Alliance for strengthening of key international organisations to contain the pandemic and has also clearly stated that it will not support any move to blame the World Health Organisation because this is the time for strengthening such organisations to battle the pandemic together (Chaudhury, 2020). India’s medical diplomacy has also paved the way for a brighter future in the post-recovery stage as India is supplying Hydroxychloroquine to 55 countries which include the United States of America, Western Europe, Africa and Latin America (Pandey, 2020). The drug is being distributed for free to a few nations depending upon their economic conditions. It is undoubtedly true that this has been made possible due to the wide pharmaceutical presence of India in various nations right from the pre-pandemic stage (Pandey, 2020). India must use the opportunity of multipolarity in the rules-based international order to contain the economic and diplomatic influence of China and its partner nations (Bhadrakumar, 2018) such as Islamic Republic of Iran, Russian Federation, Pakistan, Democratic People’s Republic of North Korea and other states so forth. In respect with the association for Southeast Asian Nations (ASEAN) like Singapore, Republic of Korea and Japan, India must use its political legitimacy and utilize the proposal of shadow diplomacy, i.e., it must apply wiser methodologies to resolve and minimize the damage in the equilibrium that balances international law and the liberal order, between natural morality (based on international legal customs and documents) and plurilateral interests of nation-states to counter the sovereign and innocence question raised by People’s Republic of China. Even though implementing the international health regulations upheld and to be maintained by the World Health Organization are to be adequately met with decent administrative practices and practical diplomatic commitments, India, being a stabilizing power, must further increase its weightage to create resistance among international organizations, regional blocs and nation-states which are interested in endorsing investigations over the COVID-19 pandemic and its global outsourcing. Prime Minister of India, Narendra Modi and the External Affairs Minister, S. Jaishankar had made it clear in its silent discourses at the United Nations and also in the virtual meetings met with SAARC nations that multilateralism is India’s visionary precept that it intends to maintain. As India had indicated that it is interested to pave reforms in various

138

•

Artificial Intelligence and Policy in India, Volume 1(2020)

international organizations under the United Nations system, the United Nations Security Council, the World Health Organization, the United Nations Human Rights Council and other relevant bodies (Times Now, 2020), it would be therefore better that India fixates its Western bloc allies from the Atlantic, continental Europe & ASEAN to convert hyper-globalization based on excruciating supply chains towards two imperatives: ─ embracing sovereign protection and maintaining trust as an incremental catalyst in the reforms related to the global supply chain; and ─ empower specific-interest nation-states such as the Russian Federation, Turkey and Iran to balance the repercussions that China might bring due to its allies and less visible nation-states in Central Asia, South East Asia and African Union to boost the latent cooperation that Canada, the United States of America and certain European Union and the non-EU States in continental Europe can already bring up to reform the global system. India did have to face some human rights-related repercussions, but after their diplomatic strength unusually maintained and tailored on issues such as Kashmir (Chaudhury, 2019) and the dilution of the Article 370 of the Indian Constitution (Haidar, 2019) and the issue related to the Citizenship (Amendment) Act, 2019 which was passed by the Indian Parliament in late 2019 (Samanta, 2019), India is maintaining some restraint as a stabilizing power despite the controversies appropriated or internationalized by various state and non-state actors around the world. Therefore, this is quite opportunistic for India to apply reasonable measures of wisdom it can share in and via its diplomatic narrative by effectively spending its soft power due to the economic implications of the global order. Along with all of this, India can also direct its attention towards the rising trends in digital diplomacy and artificial intelligence for battling the pandemic and strengthening its relations with the Western bloc. Presently, consular assistance with artificially intelligent chatbots, national image management and striking disinformation are some of the tools of digital diplomacy being used to fight the worldwide pandemic (Bjola and Manor, 2020) and therefore, India’s participation could lead to pragmatic results shortly. The semblance between multilateralism and technocratic governance will be broken, and so will happen with both of them. India, since it has been a beneficiary of multilateralism (Sidhu, 2014), may resort to keeping its stance on multilateralism as optimistic as it should be. However, India’s optimism is based on deeply constructive and stabilizing practices, which can also become a model for the international community to restore balance. Cultural confidence is India’s special advantage in the Western Bloc. A scholarly assumption that under President Donald Trump of the United States, the Bloc will divide itself and the United States of America may adopt an isolationist, inward-looking position (Linn, 2018) is untrue. In reality, sovereign imperatives will be created and maintained by the Western Bloc to endorse significant changes to protect and prevent the imbalance of the global order due to the role of the People’s Republic of China in the global supply chains. Therefore, India’s cultural confidence will inspire G20 states to rethink their strategies in the areas of technology diplomacy and the regionalization of healthcare. ─ India can expand its mixed artificial intelligence strategy derived and embraced by the NITI Aayog in 2018 (NITI Aayog, Government of India, 2018), beyond its special visions: ─ respect of and pro-empowering agenda for the rural populations considering the Central and State Governments’ vision to economize and benefit the rural populations; ─ fixing and affirming a trust-based economic and sociologically immune surveillance infrastructure considering the Central Government’s vision on technology and privacy; and

Artificial Intelligence and Policy in India

139

─ making technological evolution of societies and communities deeply rooted in their multi-lithic values. • India has many bilateral relations with different nations for the development of artificial intelligence such as France for artificial intelligence in cybersecurity and digital technology (Ministry of External Affairs, 2019), Singapore for organising the AI 4 All Global Hackathon (Sarmah, 2019), Canada for development in robotics, internet of things and artificial intelligence, Russia for the application of artificial intelligence in the industrial sector (Sarmah, 2019) and Germany for the use of artificial intelligence in healthcare and precision farming (ET CIO, 2019). • However, to and in course of expanding the NITI Aayog approach and further, establishing itself as a more stabilised actor in terms of multilateralism and the global sphere, India can: ─ empower the Global South against Chinese expansionism by democratizing and engaging with the countries that lead in the area related to artificial intelligence in Asia-Pacific and Africa, span a trusted and concordant communitarian approach to replace the top-down approach of technocratic multilateralism with trust-based federalized partnerships, to empower underdeveloped and developing economies to grow and resettle their global supply chains; and ─ embrace a Global South-led artificial intelligence initiative to contain the artificial intelligence-related influences that are rendered and promulgated by People’s The Republic of China via the Belt and Road Initiative. Legislators and policy analysts have often inspired India’s liberal and deep-driven technology laws and policies from Europe, especially from the works published by OECD, Council of Europe and the European Commission. ─ Because India and Europe have balanced and stable ties, while Russia will also attempt to gauge the Eurasian Economic Union, India can use its moral capital to contribute towards grassroots-based artificial intelligence-based education initiatives and economic policy solutions considering its experience of the sociological and multi-lithic Indian culture (Abhivardhan, et al., 2019). • In consideration of the fact that present and emerging artificial intelligence technologies are being decentralised from State control, it can pose challenges to the multilateral system that is prevalent in today’s global arena (Pauwels, 2019). Therefore, along with all the aspects highlighted above, it is pivotal that India also focuses on the technological forecast to ensure foresight and also plan for sufficient exigencies to tackle the problems related to multilateralism in artificial intelligence technologies to enable itself to emerge as a stable power in a multilateral world.

References 1. AZFAR, FAREYA, 2012, The Force Majeure ‘Excuse’. Arab Law Quarterly. 2012. Vol. 26, no. 2, p. 249-253. DOI 10.1163/157302512x628369. Brill 2. French Civil Code: Book III: Of Acquiring Property, Title III, 2020. Napoleon-series.org [online], 3. NANDINI GORE, RUBY SINGH AHUJA, 2020, The Viewpoint: Force Majeure in times of a pandemic. Bar and Bench - Indian Legal News [online]. 2020. [Accessed 8 May 2020]. Available from: https://www.barandbench.com/view-point/the-viewpoint-force-majeure-in-times-of-a-pandemic 4. Office Memorandum, 2020. Doe.gov.in [online], 5. Ali, Ahmed. S; Mahamulkar, Sujit. Coronavirus Claims Kolkata Government Doctor, Mumbai Cop. The Times of India. [online]. 27 April 2020. [04 May 2020]. Available from:

140

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

Artificial Intelligence and Policy in India, Volume 1(2020) <https://timesofindia.indiatimes.com/india/coronavirus-claims-kolkata-govt-doctor-mumbaicop/articleshow/75397267.cms>. Bansal, Pankhuri; Gopalakrishnan, Narayanan. COVID- 19: Disruptive Tech and SME Innovation Can Boost the Fight Against the Virus. Money Control. [online]. 20 April 2020. [04 May 2020]. Available from: <https://www.moneycontrol.com/news/coronavirus/covid-19-disruptive-tech-andsme-innovation-can-boost-the-fight-against-the-virus-5165341.html>. Clifford, Catherine. These are the New Hot Spots of Innovation as a Result of Coronavirus Pandemic. CNBC: Make It. [online]. 17 April 2020. [04 May 2020]. Available from: <https://www.cnbc.com/2020/04/15/hot-spots-of-innovation-as-a-result-of-coronaviruspandemic.html>. Jain, Isha. With Technology, Information, and Staff, PATH Fights COVID- 19 in India. PATH. [online]. 10 April 2020. [03 May 2020]. Available from: <https://www.path.org/articles/path-fightscovid-19-india/>. Marr, Bernard. Coronavirus: How Artificial Intelligence, Data Science and Technology is Used to Fight the Pandemic. Forbes. [online]. 13 March 2020. [03 May 2020]. Available from: <https://www.forbes.com/sites/bernardmarr/2020/03/13/coronavirus-how-artificial-intelligencedata-science-and-technology-is-used-to-fight-the-pandemic/#1ce6d0b15f5f>. Obeidat, Samer. How Artificial Intelligence is Helping Fight the COVID- 19 Pandemic. Entrepreneur: Middle East. [online]. 30 March 2020. [04 May 2020]. Available from: <https://www.entrepreneur.com/article/348368>. Prasad, Rachita. L&T Uses Artificial Intelligence to Help 20 Cities Combat COVID- 19. The Economic Times. [online]. 10 April 2020. [03 May 2020]. Available from: <https://economictimes.indiatimes.com/tech/ites/lt-uses-artificial-intelligence-to-help-20-citiescombat-covid-19/articleshow/75073867.cms?from=mdr>. Press Trust of India. Coronavirus: Kanpur Smart City Deploys AI Tool for Administrative Work. Yourstory. [online]. 15 April 2020. [04 May 2020]. Available from: <https://yourstory.com/2020/04/covid-19-kanpur-smart-city-deploys-ai-tool-coronavirus>. Kaliope, Azzi-Huck; Tigran, Shmis. 2020. World Bank Blogs. [Online] 03 18, 2020. [Cited: 05 13, 2020.] https://blogs.worldbank.org/education/managing-impact-covid-19-education-systemsaround-world-how-countries-are-preparing. Sharma, Dr. Ashwini Kumar. 2020. The Economic Times. [Online] 04 15, 2020. [Cited: 05 13, 2020.] https://economictimes.indiatimes.com/blogs/et-commentary/covid-19-creating-a-paradigmshift-in-indias-education-system/. Sinha, Vaishnavi. 2020. India Today. [Online] 04 20, 2020. [Cited: 05 13, 2020.] https://www.indiatoday.in/education-today/news/story/mhrd-s-new-initiative-offers-online-and-offline-access-tostudy-materials-for-pg-students-1668936-2020-04-20. UNESCO. 2020. UNESCO Digital Library. [Online] 05 07, 2020. [Cited: 05 13, 2020.] https://en.unesco.org/events/skills-development-during-pandemic-and-preparing-recovery-covid-19education-webinar-8. Roy, Rakesh. Coronavirus. Disruptive Technologies Network Help to Mitigate Future Threats. ET Government. [online]. 17 March 2020. [04 May 2020]. Available from: <https://government.economictimes.indiatimes.com/news/technology/coronavirus-disruptivetechnologies-network-help-to-mitigate-future-threats/74668464>. Times of India. COVID- 19: Centre Lists Red, Green Zone Districts for Week After May 3. The Times of India. [online]. 02 May 2020. [04 May 2020]. Available from: <https://timesofindia.indiatimes.com/india/covid-19-centre-lists-red-green-zone-districts-for-weekafter-may-3/articleshow/75489944.cms>. Bhadrakumar, M.K. India Must Shed Its Strategic Ambivalence in the Multipolar World. The Week. [online]. 14 October 2018. [03 May 2020]. Available from: <https://www.theweek.in/theweek/more/2018/10/05/india-must-shed-its-strategic-ambivalencein-the-multipolar-world.html>.

Artificial Intelligence and Policy in India

141

20. Bjola, Corneliu; Manor, Ilan. Digital Diplomacy in the Time of the Coronavirus Pandemic. CPD Blog: University of Southern California Center on Public Diplomacy. [online]. 31 March 2020. [03 May 2020]. Available from: <https://www.uscpublicdiplomacy.org/blog/digital-diplomacy-timecoronavirus-pandemic>. 21. Chaudhury, R. Dipanjan. High Charged but Quiet Indian Diplomacy Blunts Desperate Pakistan in UNGA & UNHRC. The Economic Times. [online]. 29 September 2019. [03 May 2020]. Available from: <https://economictimes.indiatimes.com/news/politics-and-nation/high-charged-but-quietindian-diplomacy-blunts-desperate-pakistan-in-unga-unhrc/articleshow/71365436.cms?from=mdr>. 22. Chaudhury, R. Dipanjan. India Supports Franco-Germann Alliance to Fight Coronavirus Pandemic. The Economic Times. [online]. 18 April 2020. [03 May 2020]. Available from: <https://economictimes.indiatimes.com/news/politics-and-nation/india-supports-franco-germanalliance-to-fight-pandemic/articleshow/75210230.cms?from=mdr>. 23. Haidar, Suhasini. The Perils of Post- 370 Diplomacy. The Hindu. [online]. 02 November 2019. [03 May 2020]. Available from: <https://www.thehindu.com/opinion/lead/the-perils-of-post-370diplomacy/article29857257.ece>. 24. Pandey, Aparaajita. Indiaâ&#x20AC;&#x2122;s Medical Diplomacy During the Times of Coronavirus Pandemic. The Financial Express. [online]. 25 April 2020. [03 May 2020]. Available from: <https://www.financialexpress.com/defence/indias-medical-diplomacy-during-the-times-ofcoronavirus-pandemic/1939554/>. 25. Samanta, D. Pranab. The Citizenship (Amendment) Act Poses GOI a New Kind of Diplomatic Challenge. The Economic Times. [online]. 17 December 2019. [03 May 2020]. Available from: <https://economictimes.indiatimes.com/news/politics-and-nation/view-the-citizenshipamendment-act-poses-goi-a-new-kind-of-diplomatic-challenge/articleshow/72780531.cms?from=mdr>. 26. Charlton, Emma. 2020. World Economic Forum. [Online] 18 03 2020. [Cited: 13 05 2020.] https://www.weforum.org/agenda/2020/03/quarantine-india-covid-19-coronavirus/. 27. World Economic Forum. 2020. World Economic Forum. [Online] 2020. [Cited: 13 05 2020.] https://www.weforum.org/projects/managing-the-risk-and-impact-of-future-epidemics. 28. Politzer, Malia. 2017. devex. [Online] 23 10 2017. [Cited: 13 05 2020.] https://www.devex.com/news/working-toward-global-health-security-strategies-and-challenges90727. 29. Times Now. Future is Internationalist: EAM S Jaishankar on Multilateralism. The Economic Times. [online]. 20 February 2020. [03 May 2020]. Available from: <https://economictimes.indiatimes.com/news/politics-and-nation/future-is-internationalist-eam-sjaishankar-on-multilateralism/videoshow/74229253.cms?from=mdr>. 30. Zeeshan, Mohamed. Can India Organise Post COVID- 19 Global Action? The Diplomat. [online]. 07 April 2020. [03 May 2020]. Available from: <https://thediplomat.com/2020/04/can-india-organizepost-covid-19-global-action/>. 31. Abhivardhan. Machine Learning and Enculturation: Perspective of International Human Rights in China. IOSR Journal of Engineering. [online]. 2019. Journal No. 48995, pp. 70-74. ISSN: 22503021. [07 May 2020]. Available from: <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3391858>. 32. ET CIO. AI will be a Key Area of Cooperation between India, Germany. ET CIO. [online]. 04 November 2019. [07 May 2020]. Available from: <https://cio.economictimes.indiatimes.com/news/government-policy/ai-will-be-a-key-area-ofcooperation-between-india-germany/71888066>. 33. Linn, Johannes. Recent Threats to Multilateralism. Global Journal of Emerging Market Economies. [online]. SAGE Publications. 2018. Volume 9, pp. 86-113. [07 May 2020]. Available from: <https://journals.sagepub.com/doi/pdf/10.1177/0972063417747765>.

142

Artificial Intelligence and Policy in India, Volume 1(2020)

34. Ministry of External Affairs. Indo- French Roadmap on Cybersecurity and Digital Technology. Ministry of External Affairs: Government of India. [online]. 22 August 2019. [07 May 2020]. Available from: <https://mea.gov.in/bilateraldocuments.htm?dtl/31757/IndoFrench+Roadmap+on+Cybersecurity+and+Digital+Technology+August+22+2019>. 35. NITI Aayog. National Strategy for Artificial Intelligence. NITI Aayog: Government of India. [online]. June 2018. [07 May 2020]. Available from: <https://niti.gov.in/sites/default/files/201901/NationalStrategy-for-AI-Discussion-Paper.pdf>. 36. Pauwels, Eleonore. How Can Multilateralism Survive the Era of Artificial Intelligence? United Nations Chronicle. [online]. United Nations. 2019. [07 May 2020]. Available from: <https://www.un.org/en/chronicle/article/how-can-multilateralism-survive-era-artificialintelligence>. 37. Sarmah, Harshajit. 6 Countries that India has Partnered on Artificial Intelligence. Analytics India Magazine. [online]. 03 November 2019. [07 May 2020]. Available from: <https://analyticsindiamag.com/6-countries-that-india-has-partnered-on-artificial-intelligence/>. 38. Sidhu, W.P.S. Why Multilateralism Matters for India. Brookings. [online]. 07 July 2014. [07 May 2020]. Available from: <https://www.brookings.edu/opinions/why-multilateralism-matters-forindia-2/>. 39. Fidler, David. An Abuse of Presidential Authority and American Power: Halting U.S. Funding for the World Health Organization. Just Security. [online]. 15 April 2020. [07 May 2020]. Available from: <https://www.justsecurity.org/69694/an-abuse-of-presidential-authority-and-american-powerhalting-u-s-funding-for-the-world-health-organization/>. 40. Global Health Policy. The U.S. Government and the World Health Organisation. The Kaiser Family Foundation. [online]. 16 April 2020. [07 May 2020]. Available from: <https://www.kff.org/globalhealth-policy/fact-sheet/the-u-s-government-and-the-world-health-organization/>. 41. International Health Conference. Constitution of the World Health Organisation. World Health Organisation. [online]. 22 July 1946. [07 May 2020]. Available from: <https://www.who.int/governance/eb/who_constitution_en.pdf>. 42. Jansen, Oswald. Increasing the Legitimacy of the World Health Organisation. The Regulatory Review. [online]. 22 April 2020. [07 May 2020]. Available from: <https://www.theregreview.org/2020/04/22/jansen-increasing-legitimacy-world-healthorganization/>. 43. United Nations. United Nations Charter. [online]. 1945. San Francisco: United Nations Treaties. [01 May 2020]. Available from: <https://treaties.un.org/doc/publication/ctc/uncharter.pdf>. 44. United Nations Foundation. COVID- 19 Solidarity Response Fund for WHO. COVID- 19 Response Fund. [online]. 2020. [07 May 2020]. Available from: <https://covid19responsefund.org>. 45. World Health Organisation. Statement on the Second Meeting of the International Health Regulations (2005): Emergency Committee Regarding the Outbreak of Novel Coronavirus. World Health Organisation Newsroom. [online]. 30 January 2020. [07 May 2020]. Available from: <https://www.who.int/news-room/detail/30-01-2020-statement-on-the-second-meeting-of-theinternational-health-regulations-(2005)-emergency-committee-regarding-the-outbreak-of-novel-coronavirus-(2019-ncov)>.

Artificial Intelligence and Policy in India

143

AI and its Tortious Liability Sadaf Fahim National Law University, Delhi sadaf.fahim@nludelhi.ac.in

Abstract. The likelihood of making thinking machines raises a large group of tortuous issues. Artificial Intelligence is basically a study of how to make a system, which can think, behave and act exactly or better than what a human being can act or react. It tends to the issues of making AIs more wise than human, and guaranteeing that they utilize their propelled insight for good as opposed to ill. In the field of Tort Law, the ultimate concerns for Artificial Intelligence are whether an autonomous vehicle, drones and robots should also be given a status of electronic person? or robot would only be considered as a legal personality just like-corporations (as a legal person-who can sue and be sued as given to Sophia-a citizenship in Saudi Arabia) or would it be considered as a like it as an individual person within the purview of law. Artificial Intelligence has evolved out of from four basic subjects: Psychology, Philosophy, Mathematics and Linguistic, they are making a big role in an enhancement of Artificial Intelligence. This paper intends to identify issues and challenges pertaining to torts, especially in terms of whether we should consider software programme as a product or service, as earlier it happened in case of considering electricity as a product rather than considering as a service, now that what is the obstacle is here, in the case of negligence, strict product liability, and vicarious liability in the field of law of torts, where India lacks specific legislation. The question of legal liability arises when unmanned vehicle is involved in a car accident, the surgical system is involved in a surgical error or the trading algorithm is involved in fraud, etc., now the question is who will be held liable for these offences. Before we delve into the potential of Artificial Intelligence, letâ&#x20AC;&#x2122;s take a step back to understand AIâ&#x20AC;&#x2122;s legal issues pertaining to legal liability of Artificial Intelligence systems under the head of legal categories such as: Law of Torts and, Criminal Law .Such determination is likely to get more muddled with the onset of AI, particularly due to the possibility of it being accorded the status of a person in law. I will explore tortuous implications of AI / in relation to the use of AI. This is the most new aspects in the field of the laws of robots, self-driving car and drones in contrast to traditional forms of responsibility-proof for otherâ&#x20AC;&#x2122;s behaviour, like children, employees, or pets which gets in addition to new strict liability policies, mitigating through the insurance models, systems authentication, and the mechanism of allotting the burden of proof. Further this paper will critically analyze the nuances of using AI system in the field of law of torts. At the end this paper will suggest and recommend solutions to overcome these issues and challenges through the use of doctrinal with qualitative research methods. Keywords: Artificial Intelligence, Negligence, Strict Product Liability, Vicarious Liability Algorithm, Legal implication of AI.

144

Artificial Intelligence and Policy in India, Volume 1(2020)

Introduction

Tort Law in India is a comparatively a new development in common law as supplemented/complemented by its codifications of statutes including statutes governing compensations/damages. While we all know this fact that India mostly follows the United Kingdom’s approach towards it- so here the researcher is going to discuss the legal liability of Artificial Intelligence, in the field of Law of Torts with respect to India, Germany and, California- a comparative studies, where the researcher will discuss how there are certain gaps in the judicial activism, where which it later highlighted the problems-pertaining to creating controversies. Tort is breach of some duty/obligations which is independent of contract that has caused harm to the plaintiff and giving rise to cause of actions which is civil in nature and for which damages is available therein. Meaning thereby, if there is no remedy available then it cannot be called/termed as a tort because the very essence of the nature of tort is to give compensation to the individuals who has suffered the injury or loss of any kind or nature. Machine learning with the help of data, on which it operates on- constantly evolves, and helps in making complex decisions based on those data. Rarely does it happen, in the absence of human supervision, or due to some adversity we receive the outcomes of what we are not even expecting, mostly the end-result, is predictable enough. Because of the artificial nature of AI as autonomous in kind, it requires a lot of new considerations in terms of fixing liabilities. Tort law has always been considered as traditional law serving with the growing needs of society, including the new developments and technical advances. The analytical framework of tort law has always been applied by courts time and again and not only that but also the legal principles, whenever the facts of the case demands by the court. 1.1

Responsibility for AI-Private Law

Private law talks about the legal relationship of people and it includes the alteration, destruction and alteration of rights.12 Basically, private law relations start with voluntary actions, like for example- no one can compel others to enter into contract with others till that person wants to enter into a contract on its own, which later on becomes legally binding in nature. We have always witnessed in private law, rights and obligations generally come into pairs, and i.e. a liability for one party becomes a claim for another. (Wenar, 2020) When it comes to either civil or criminal laws, deterrence of wrongful conduct is the ultimate aim for both. The purpose of private law is to make sure that right are avenged and therefore parties are compensated for the same. So in usual circumstances, generally the money has been given to the innocent party, in the case of private law, when harm caused-although other remedies are also available for defendant either go for undertake or abstain himself from doing such particular act. Those who transgress, criminal law proves to be a society’s powerful weapon. We all know its state that enforced criminal laws, where individual perpetrators expressly agreed to comply with. Criminal laws serve various purposes like, showing us where state’s against of certain conduct, deterrence, retribution and protection of society as a whole. Suppose, if a person commits some crime then he/she should be penalized for his/her conduct, typically in a form of fine or 12

Private law sometimes called also referred as “civil law”. Furthermore, this can get confusing with the terminology ‘’civil law’’ because it could be used to describe legal systems that are founded upon great codification, (such as the BurgerlichesGesetzbuch in Germany and or the Civil Code in India).

Artificial Intelligence and Policy in India

145

imprisonments. Though still many state yet practice “corporal punishment”, which includes infliction of mutilation, pain, or even death on an accused. (Turner, 2019 pp. 82- 102) Crime considered as the most vehement act against the community at large. To find individual guilty or blameworthiness of an act, it needs to be exceeded in terms of private laws but on a contrary in respect of criminal law- mere preponderance of probability is not only needed but higher burden of proof needs to establish in proving the guilt of an individual. Nothing lasts on an individual as much as having the impression of proven-guilt of any such crime, then civil law liability in tort or contract. Now, this can leave the impression of social stigma and legal disabilities once convicted of any crime. It may happen in some jurisdiction, that they tend to takeaway the power of an accused from casting votes and perhaps other civil rights. In the Thirteenth Amendment of the Constitution, USA imposed fine on a party, convicted of crime regarding slavery (Turner, 2019 pp. 82- 102). The issues come when robots cause harm to third parties instead of their own counter-parties, i.e. contractual relationship partners, now that will take to another level of extra-contractual responsibility. Attention needs to be drawn, because torts deal obligations between the two parties which is imposed by the government, and if not followed then damages has to pay to the sufferer or claimants. With the exponential growth of AI-autonomy, new cases are coming up which involves a new kind of liability for the behaviour of others. Are we far-away of those situations where humans will be legally responsible for the acts of artificial state-transition system, who will decide what to do? Furthermore, these different kinds of robots will determine the different kinds of liability also, for instance: a robot nanny, a robot toy, a robot chauffeur, a robot employee, so forth and so forth.

Private Law- AI-Tort Analysis

Whenever we engage discourses about AI, especially in terms of Private law, two sources come into play that puts obligations on a party: civil wrongs and contract13 (Jackson, 2014). So, once the legal rights of one party are infringed, it means civil wrong occurred.14 For instance, if Simon throws a sound box out of a hotel room window and harm Harry, an unknowing pedestrian walking down the streets, Simon has done a civil wrong to Harry by inflicting unnecessary harm on his right to bodily integrity by intervening with his right to walk down the streets in peace. Now, here the claim of Harry becomes against Simon under the purview of private law. Contract is relied on an agreement, where if suppose, Moore agrees to sell a new television to Mira, but not complying with his agreement, delivers a second-hand model, then Mira might sue Moore for breach of agreement. On a contrary, if Moore delivers a new television as promised by Moore but Mira denied paying for it, then here case can become against Mira for not paying the amount and Moore can sue Mira based on their exchange of promises. (Turner, 2019 pp. 82- 102) Now, covering all the aspects of it: a. Who is responsible- manufacturer/ producer, designer, owner, or user? b. Who should bear liability- manufacturer/ producer, designer, owner, or user? c. In the case of AI- programmer or developer, user or 13

The distinction between civil liability arising from tort and contract may be traced at least as far as Roman law. The Institute of Gaius (compiled c. 170 AD) stipulate that obligations could arise under two headings: ex delicto and ex contracto. The Institute of Justinian (compiled in the sixth century AD) added two more categories, namely quasi ex delicto and quasi ex contracto. The latter are outside the scope of the present work. 14 Civil wrongs are referred to in some systems as “delicts” or “torts”. The source of the latter is the Latin torquere, to twist, which became in Medieval Latin tortum: a wrong or injustice.

146

Artificial Intelligence and Policy in India, Volume 1(2020)

technology itself? Perhaps what changes we could see in terms of ‘the principles of negligent design’ or ‘standard of care’? As, AI keeps evolving, making its decision on its own, depending on this analogy, could we have considered it as an agent of the developer in the future and if, yes, then would that developer becomes liable-vicariously, in terms of whatever decisions AI makes that perhaps resultant into negligence? Though we really think, negligence analysis should remain as it is, if yes.... then the criteria of standard of care would need- revisiting in an AI context, because few of the questions would become central to the court’s contemplation: 1. Is court so clear with the help of crystal-clear case in terms of ‘golden box’, how does it reach its outcome? 2. In terms of machine learning what steps would be taken to monitor its outcomes? 3. Talking about the quality and integrity of data-sets for the purpose of which it was made. 4. Data whether it used representative or does it promote discrimination and /or bias? 5. Was the design of algorithm accurately designed to protect against unwanted outcomes? Anticipation is there in terms of programmers and software development companies regarding the exponential growth of industries in case of negligent actions. 2.1

Negligence

Negligence has two meanings in law of torts: (Turner, 2019 pp. 82- 102) a) Negligence as state of mind- Negligence is a way of committing certain torts, for example, negligently or carelessly committing trespass, nuisance or defamation. This is the very subjective meaning of negligence which is advocated by Austin, Salmond and Winfield. b) Negligence as a type of conduct- Negligence is a conduct, rather than a state of mind-conduct, which involves the risk of causing damage. And, this is the very objective meaning of negligence, which treats negligence as a separate or specific tort. Actionable negligence is the neglect use of ordinary care or skill towards a person to whom the defendant owes a duty of observing ordinary care of skill, by which neglect the plaintiff has suffered, to his person or property [Heaven v. Pender (1883)]. In an action for negligence, the plaintiff has to prove: 1. That the defendant owed duty of care to the plaintiff 2. That the defendant made a breach of the duty i.e. he failed to exercise due care and skill 3. That the plaintiff suffered damage as a consequence thereof. So, negligence here is ‘conduct’ which has fails to comply with a required standard. In a famous UK case Donoghue v. Stevenson (1932), (Ratanlal & Dhirajlal, 2016) the appellant plaintiff drank a bottle of ginger beer which was brought from a retailer by her friend. The bottle which was of dark opaque glass in fact contained the decomposed body of snail (found out by her when she had already consumed a part of the contents of the bottle). Held that the manufacturer of the bottle was responsible for his negligence towards the plaintiff. Here, it was said that the manufacturer or producer was held to have a duty of care to the plaintiff (whoever might reasonably be expected to have open the bottle, even though there was no direct contract between the two, i.e. between manufacturer/producer and consumers). The House of Lords also rejected the plea of defendant and held it doesn’t matter whether there was any contractual relationship between the manufacturer/producer and plaintiff or not. Lord Atkin said: “The rule that you are to love your neighbour becomes in law ‘you must not injure your neighbour’.” Moreover, the judgement explained:” ......you must take reasonable care to avoid acts or omissions which you can reasonably foresee would be likely to injure your neighbour”. In many

Artificial Intelligence and Policy in India

147

different types of legal systems, the same rules apply, including Germany (Turner, 2019 pp. 82102). In, Section 823 of the German Civil Code: “A person who, intentionally or negligently, unlawfully inquires the life, body, health, freedom, property or another right of another person is liable to make compensation to the other party for the damage arising from this”. (Federal Ministry of Justice and Consumer Protection and the Federal Office of Justice, 2020)

California Negligence Law is codified at California Civil Code Section 1714: “Everyone is responsible, not only for the result of his or her wilful acts, but also for an injury occasioned to another by his or her want of ordinary care or skill in the management of his or her property or person, except so far as the latter has, wilfully or by want of ordinary care, brought the injury upon himself or herself. The design, distribution, or marketing of firearms and ammunition is not exempt from the duty to use ordinary care and skill that is required by this section. The extent of liability in these cases is defined by the Title on Compensatory Relief”. (California Legislative Information, 2020)

Fitting AI in Law of Negligence- But How? How we could apply to AI- the Law of Negligence. Putting it straight, if harm is done or caused for that matter, the very first question would come into mind is, duty of care to the plaintiff-whether anyone was under a duty not to cause harm? or to prevent that harm? Taking up the same supposition suppose, the owner of a robot lawn-mower while cutting the grass of his lawn might come under a duty not to harm his neighbour or anyone who lives in the vicinity, by his any acts of his own through that lawn-mower. This would include what a ‘reasonable man’ (i.e. a man of ordinary prudence or intelligence) would have foreseen and behaved under the circumstances. This is what exactly needed it from the owner of AIlawnmower, to take care of it, and it should not budge into neighbor’s garden and guillotine their heart-winning roses. (Turner, 2019 pp. 82- 102) Now, the second question comes into play is that of, breach of duty-whether the duty was breached? Even if, after taking all the precautions, which could be expected from a prudent owner of lawn-mower under all the prevailing circumstances, then also if the lawn-mower goes wrong and caused harm to anyone, then the owner of the lawn-mower is not responsible here, rather he will be excused or exonerated, because its machine’s fault not the owner. It might happen in other scenario that, neighbour decides to borrow the lawn-mower without the due permission of an owner and uses it on his own garden, now while using it, it causes damage to the neighbour, and here the neighbour cannot argue that the damage was caused by the owner of lawn-mower. The third aspect of the question is, damage-whether the breach of duty caused the damage? Suppose the lawn-mower on his rolling towards the neighbor’s garden and owner has no knowledge about it, under his negligence. Now, before the lawn-mower damages any flowers, a car ran off the road and damaged the rose-bed of the concerned neighbour. Here, the owner of lawn-mower can take a plea of breach of duty though happened from his side, to not to take control of machine properly, but at the same time the damage was not caused by him, rather its car’s driver who intervene into it and destroyed neighbor’s flowers, so it’s driver fault not the lawn-mower’s owner. A fourth question critically comes into mind is, remoteness of damage-whether the damage was

148

Artificial Intelligence and Policy in India, Volume 1(2020)

of type or extent which was reasonably foreseeable? Though the cost of replacing the heart winning roses, likely to be foreseeable but at the same time, neighbour who may entered and made it through prize winning rose-competition, who actually loses it, may not be foreseeable to the owner. It is not only the owner who is responsible to take due care in the above situations. Perhaps, it applies same to the designer of the AI, or the person who trained and taught it. Suppose if the design of the AI possessed a fundamental flaw (letâ&#x20AC;&#x2122;s say it interpreted child as weeds that needs to be destroyed) then, here the designer may have breached a duty to not design a robot properly so that it should functions it properly. (Turner, 2019 pp. 82- 102) What are the advantages of Negligence? This talk about the duty how we can adapt in these prevailing circumstances. Though itâ&#x20AC;&#x2122;s obvious that it depends on context, i.e. the level of duty-which can shrink and expand15 (House of Lords).As we budge from narrow AI (which could be used only for one task) to general AI (which could be used for multi-purposes), different spectrum altogether needed for both to functions in law of negligence, because it increases its usefulness. So, as a thumb rule, if one chance has been given to cause harm then think how it would be multiplied in and cause more potential harm, taken us up to the stage of calculating measures of precautions which needs to be taken up- immediately (United States v. Carroll Towing Co.). Suppose when transporting nuclear waste, there high level of care or precautions needed from the end of that person who is carrying it with himself, though the chances of leak risk is less, but the danger is exponentially high. So, it not only what we demand from the end users to take due care of, rather courts also take this into consideration when it comes to society at-large when the concerned activities benefit all. More considerations given to risky activities which involve more benefits rather than less beneficial in nature serving no good purpose to society at large, less leniency given to them. Understanding it in an easier example, like police who will be exonerated for his negligent driving while pursuing a criminal than a fun-rider because here the act of police is to save the greater interest than the latter one. (Lady Hale, 2018) Following with all the examples given above, what we noticed here is, when it comes to Artificial Intelligence (AI), who needs to be more careful are: producers, operators and owners of AI, why so? Because they are ones who can cause greater harms to others, so they need to be extra conscious and take precautions of every length. But, yes nonetheless we can create any theory of negligence which would create restrictive rules for AI and make it harder for it to work smoothly and properly in terms of its developments and innovations. So, if we talk about flexibility in terms of duty, like who owned duty? There is no straight jacket formula where we can come up with a list of people who can claim for negligence, and this is useful because the people who is working AI or interacting with it might get changed and so to say the outsets too-very unpredictable. Furthermore, we cannot come up saying that AI affected us, through the lens of AIâ&#x20AC;&#x2122;s creator, owner or controller because whosoever got affected or AI-does to them, could not come up with any prior contractual relationship with AI. When people decided to be liable only for AI, then it leads to gaps in terms of protection of third parties involving AI activities. Now, what can be the approach of non-voluntary aspect of negligence? Perhaps it inspires subjects under given

In this case court held that before determining the liability, many factors

needs to be set out.

Artificial Intelligence and Policy in India

149

legal system to have greater regard towards others if they are seeking profit-making objectives. (Lady Hale, 2018 pp. 86-88) Putting differently, the likeliness of negligent liability may cause externalities to take their actions and indeed price into their considerations, so that at least that could be calculated accurately. Negligence - And its Shortcomings What could be the way of setting standards of behaviour for AI’s? The question comes into frame that is defendant always acted prudently and reasonably like any other normal humanbeing in the case of negligent. The problem arise here is, now because everything we are seeing from the lens of human using-AI, then the question is whether we can apply the same reasonable person test or not in case of AI itself, can we apply the same reasonable person test with human using-AI? One can question that one can ask what could be the actions of AI-designer or user under such circumstances, how would they behave or act? For instance, it may be easy to set and drive an autonomous car in a clear motorway, but that could not be possible in an urban hectic environment, to set a fully autonomous car to operate on its own. What maximum could designer do with AI? Designer can provide AI with stipulations regarding “health warnings” and could tell AI, what is advisable to do under given circumstances or what not at the same time when it’s operating on its own. These solutions look perfect, in short-term basis, but think about the scenario when there is no human intervention or no operator of AI then what? The liability questions will come into picture when it comes to fixing liability on whomafter all? Furthermore, using AI wrongly is considered as potential harm. AI using in the development field might cause harm through some form of unforeseeable development. The more the manner of failure become unpredictable the more difficult it will become to make anyone held liable, i.e. the designer or user responsible for his acts without resorting to the fact of a form of strict liability. (Lady Hale, 2018 pp. 86-88) In order to deal with these issues, Ryan Abbot has come up with a solution where he proposed an idea that if retailer or manufacturer/producer of autonomous robots, computer and machine gets successful in showing that they are smarter and safer than a reasonable person then in that case they will be only held liable for his negligent acts rather than facing liability against strict liability for any harm which might cause due to autonomous entity by the supplier.16 Abbot’s definition of negligent test would focus on AI’s “act instead of its design, and in a sense it would treat a computer tortfeasor as a person rather than a product”. (The Reasonable Computer:”Disrupting the Paradigm of Tort Liability", 2017 pp. 101- 143) Abbot explains how much we should determine negligence? It depends totally upon the standard of “reasonable computer”, and on this basis we should get clear picture of what is more or less reasonable on the part of computer in a given particular situations (The Reasonable Computer:”Disrupting the Paradigm of Tort Liability", 2017 pp. 101- 143). Abbot here, inspect this standard by establishing the “considering the industry customary, average, or safest technology”.17 (The Reasonable Computer:”Disrupting the Paradigm of Tort Liability", 2017 pp. 101- 143; (American National Standards Institute)). Though, in practice applying this standard of “reasonable computer” is pretty tough to incorporate and make it applicable, See the following section of strict liability. It would be assumed for current purposes about Abbot’s definition of “autonomous” which cover substantially the similar entities AI within this chapter. 17 Some efforts are currently underway at the level of standard setting bodies such as the International Standards Organization to establish general rules on these features, so at a minimum the agreement and articulation of such standards will be a prerequisite for Abbot’s scheme to work. 16

150

Artificial Intelligence and Policy in India, Volume 1(2020)

though on the other hand to imagining a human person in this situation is easy. Its law’s capability that sets an objective-standard of behaviour, where it considers all humans are just same, precisely putting up: it thinks we humans have certain limitations and abilities which arises from ours physiology. Does it really matter or any concerned with this fact that when we set standards for negligence, whether the opposite human person is clever, brave or stronger, does it really matter as a matter of fact-No. Seeing this, we can say AI is heterogeneous in kind, because there are no single way of creating AI, rather we have many other techniques of creating AI, if talking in terms of present situations and even about future too, as with the help of different developments of techniques in the market. Now, applying the same techniques or standard for that matter to all very varied AI-entities perhaps would be inappropriate (American National Standards Institute). Lastly, we can conclude it by saying that it is not necessary that the test we have for reasonable human person in case of negligence would suffice the same purpose as in the case of artificial entities, and can bound up with the way humans operate. If we take the example of autonomous vehicles, considerably would it be considered, safer than humans-driven car, so, it is very open-ended question.

Foreseeability- How much Reliance On? Foreseeability is another concept on which law of negligence relies. It is basically used for the purposes of asking both the claimants “was it foreseeable that this person would be harmed?” and the recuperate harm by questioning “what type of damage was foreseeable?” Now, this must be challenging so think about it, the actions of AI perhaps likely to become exponentially unforeseeable, except that it may be at massive level of generality and abstraction for that matter. Corollary, holding a human responsible for all actions of AI might get fudged and budged from human’s fault which is a hallmark of negligence and become system more like- a strict or product liability. (American National Standards Institute, pp. 90- 91) In regard to AI system, to budge it more from classic form of coding and algorithms, if it happens then the creators will go hay-wire because they have just not imagined that the behaviour would be unforeseen rather unforeseeable-wholly. What would the situation be like when there is no foreseeability, does it mean it will not make anyone liable for their damaging acts on others. One could think likely that it is our courts that would look into the matter of and prevent such consequence. In a lack of foreseeability generally it happens that its analysis gets budged to strict liability than on negligence. When we see things from the angle of strict liability then the very first case comes into anyone’s mind is Rylands v. Fletcher where defendant regardless of his faults whether its negligent act or intentional ones, still found legally liable for his responsibility because it resulted into plaintiff’s injury. 3. Strict and Product Liability Another arm of tort law is product liability when looking from the perspective of liability in terms of AI defectiveness. Strict liability means where a party (defendant) is held liable regardless of their fault to plaintiff, here in case of strict liability- plaintiff is not required to demonstrate defendant’s negligence. Putting it in simple words, defendant through his conduct if injured plaintiff or caused any harm to plaintiff, whether negligently or not, and it resultant into plaintiff’s injury or loss then it is defendant who will be liable for the plaintiff’s injuries. Product liability focuses and warns on many aspects like negligent design, manufacture and breach of duty, under the common law. “Product liability” refers to a system of rules where, injured consumers can claim for damages against the

Artificial Intelligence and Policy in India

151

concerned parties whosoever involved at any steps of the manufacturing, distribution, or retail process of the concerned product. In California, product liability claims is based on a theory of strict liability, though generally personal injury’s standard claims which is -based on theory of negligence, and this theory is helping into making their path into California-strict product liability law. (The Law Offices of Brian J. O'Grady) Now, under the Section 276 of German Civil Code:18 (Federal Ministry of Justice and Consumer Protection and the Federal Office of Justice, 2020) where it talks about the strict/absolute liability, where under it disposes that in general, the obligor is just responsible for their fault (“intention or negligence”) but exceptionally there can be a “higher degree of liability”, regardless of the obligor’s fault. (Federal Ministry of Justice and Consumer Protection and the Federal Office of Justice, 2020) Whereas, Section 254 of the German Civil Code talks about the Contributory negligence:19 (Federal Ministry of Justice and Consumer Protection and the Federal Office of Justice, 2020) it talks about in particular to what extent the damage caused basically by one or the other party and what if the fault of the harmed person is restricted to failing to draw the concern of the obligor to the risks of unusually extensive harm, where the obligor neither was nor ought to have been known of the harm, or to failing to avert or minimize the harm (Federal Ministry of Justice and Consumer Protection and the Federal Office of Justice, 2020). In addition to it addresses basically the parties concerned relating to liabilities in terms of sale, distribution of a product and manufacture. To apply this doctrine, first we need to confirm that AI must qualify as a product not as a service. Though with the help of machine learning and algorithm no doubt we have come up with an autonomous AI-product, but what would happen if any defects would fall into place regarding supply chain of its product, how to determine where actually defects happen. This is what commentators thought that while solving the issues related to autonomous vehicles, robots, drones, and other AI-enables systems, the product liability will then become a relevant subject of issues to solve in like- defective products, dangerous products liability, and types of product defects. 1. Defective Products: If anyone is harmed by unsafe and defective product, is entitled to compensation/damages, including self-driving car parts, drone parts, robot, machinery and more. 2. Dangerous Products and Liability: Who will bear the consequences if products get fail and they are no more capable of giving safety of using that products, the only reasonable person is manufacturer who will be held liable and bear the consequences. Forcing manufacturer to pay compensations to the claimant (plaintiff), whether it would be for the damaged seat “Section 276 of the German Civil Code: Responsibility of the obligor (1) The obligor is responsible for intention and negligence, if a higher or lower degree of liability is neither laid down nor to be inferred from the other subject matter of the obligation, including but not limited to the giving of a guarantee or the assumption of a procurement risk. The provisions of sections 827 and 828 apply with the necessary modifications. (2) A person acts negligently if he fails to exercise reasonable care. (3) The obligor may not be released in advance from liability for intention.” 19 “Section 254 of the German Civil Code: Contributory negligence (1) Where fault on the part of the injured person contributes to the occurrence of the damage, liability in damages as well as the extent of compensation to be paid depends on the circumstances, in particular to what extent the damage is caused mainly by one or the other party. (2) This also applies if the fault of the injured person is limited to failing to draw the attention of the obligor to the danger of unusually extensive damage, where the obligor neither was nor ought to have been aware of the danger, or to failing to avert or reduce the damage. The provision of section 278 applies with the necessary modifications.” 18

152

Artificial Intelligence and Policy in India, Volume 1(2020)

belts of self-driving car, defective drones, dangerous cribs, benefits society at large in a various way. So, the person who is injured gave compensation. Furthermore, by suing a lawsuit against the manufacturers for defective product-make manufacturers more sincere in their works and not being careless, rather, in-line, revisit those products again and make them safer when necessary. 3. Types of Product Defects: Any product for that matter can be defective, some may be harmless and some perhaps are of no consequences. Such other defects may endanger other’s life or make them fallen sick. Three kinds of defects that can cause injury to a consumer and held manufacturer or supplier-liable are as follows: a. Design Defects: Are very much present since the inception/start of the product. b. Manufacturing Defects: Are those defects that are present during the process of making those products. c. Marketing Defects: It includes- wrong labelling, poor or inappropriate instructions and safety warnings of products. (The Law Offices of Brian J. O'Grady)

Advantages of Product Liability in terms of AI

1. Certainty: Once the manufacturer/ producers or the supplier of AI is known, they are liable to the plaintiff/victim fully i.e. 100% of the compensations. Because product liability rules specify in advance that who will bear the consequences, which is very much helpful for the victims to avail, without going to multiple parties to seek compensations in proportions to their relative fault. Now, budging ahead the onus probandi is on the manufacturer/producer or supplier to prove and make liable to other parties involved and to sue them for a contributory negligence. Through the lens of the manufacturer/producer or supplier of AI, it requires more accurate calculations before making anyone held liable, because the risk of compensation be priced into the last cost of the products, with the accounting futuristic of companies and investors with their brochure containing risk factors as well. 2. Safety and Extra-Cautious in the development of AI: It encourages developers of AI, to be more extra cautious during the process of making the products with rigorous safety and control mechanism, even if the AI development has been done in a situation which might become unforeseeable in ways, still the designer or producer of AI would be considered best in understanding and controlling the risks. Michael Gemignani wrote in 1981 of computers, so the same principles would apply to AI with more force added to it: “While the computer is still in its infancy, it may prove to be as beneficial, or as potentially harmful, as atomic power. If imposition of strict liability in tort would make the manufacturers of computer hardware and software more careful and more thoughtful in their race to develop an ultimate product, that alone would justify its application.” (Product liability and Software, 1981) Disadvantages of Product Liability in terms of AI 1. Is AI a Product or a Service: Knowing this fact that product liability regime denotes product not services. But as scholars and commentators are arguing on this point whether we should consider AI as product or services or not that holds the very basic and fundamental questions before assuming on anything or labelling it. In European Union, products are defines as “all movables” in Article 2 of the Product Liability Directive, which says it applies only to

Artificial Intelligence and Policy in India

153

the physical goods, if we go by this definitions of product liability under Article 2 then it is much suitable for the ‘robot’ but not for the cloud-based AI. (Turner, 2019 pp. 82- 102) 2. Presuming Products once released to be Static (Turner, 2019 pp. 82- 102): It runs on the assumptions that the product doesn’t keep changing its kind in an unpredictable manner once it is released and left the production line. The US and EU when talking about the static products are subject to number of defenses, proved to be overly submissive when applied to AI-producers. In European Union, the liability includes carve outs: “.......having regard to the circumstances, it is probable that the defect which caused the damage did not exit at the time when the product was put into circulation by him or that this defect came into being afterwards; or.....that the state of scientific and technical knowledge at the time when he put the product into circulation was not such as to enable the existence of the defect to be discovered.......” (THE COUNCIL OF THE EUROPEAN COMMUNITIES, 2020; Developments in English Product Liability Law: A Comparison with the American System, 1987-1988) Suppose if we consider AI as product liability, the probability is more likely that producers would start taking the defenses of the above safe area/havens, which would resultant into less protections given to or available to consumers.20 (COMMUNITIES) Beyond the criminal and contractual liability there is another set of responsibility known as individual responsibility, why because these cases occurs because of the personal fault to other which resultant into paying damages to them. Here the lawyers explain the term extra-contractual as torts which is at stake in Mracek v. Bryn Mawr Hospital. (Pagallo, 2013 pp. 115- 116) The only contentions of plaintiff is damages which he suffered because of the strict liability and malfunctions liability, making liable the designers and processer of robots for damages which they caused to third parties due to their manufacturing defects of the products and design-flaws. 3.1

Strict Liability Rules: AI-Employees

Notwithstanding any illicit or culpable behaviour, there are two cases of liability which has been established like: malfunctions liability and strict liability rules, which talks about the robots as a tool for an industry of humans and as an agents of human interaction under strict liability rules, either conjectured as an animal-the dangerous ones, or alternatively, to the liability of Indian parents for their children’s behaviour and pets as discussed above. Many legal systems still provides different kind of strict liability, which surely fits in the laws of robots case as believed to be acting as agents of human interactions, precisely, for any wrongful action or, in case the employees engage in under some contract activities then its employer’s liability which arise.

Art. 6(2) of the Directive: “A product shall not be considered defective for the sole reason that a better product is subsequently put into circulation”. This may have been a reasonable rule for traditional industrial products, but seems ill-suited for software where everyone rightly expects constant security updates, patches, bug fixes, etc. This is not a problem unique to AI, but it is especially pertinent for programs which by their nature learn and improve over time.

154

Artificial Intelligence and Policy in India, Volume 1(2020)

Faultless Responsibility for Robots

• As Means of human industry • Strict Products Liability • Strict Malfunctions Liability • Breach of Warranty

As Agents of human interaction

• Robots as dangerous animals • Robots as children • Indian Parents • American Family • The German Family • Robots as employees

Figure 2: Strict liability for robots in the law of Torts. (Pagallo, 2013 pp. 130) Pagallo explains strict liability- its varied types, illustrating the behaviour of robots. So the focus of this figure is to restrict it to dig deeper into the respondent superior and strict liability rules under the purview of American common lawyers as well as civil lawyers such as the judicial system of California permits/allows a defendant to take a defence of the comparative negligence to reduce his/her own fault in a case. As an example- if a defendant is only 30 percent at fault for contributing to a car accident, then would be liable for only 30 percent for an award that a plaintiff would receive. The California Supreme Court adopted the concept of ‘pure comparative fault’ principle where it puts an end to- several and joint liabilities, supposition of risk, and “last clear chance” as doctrines. In California, if you are harmed due to the negligence of another, in today’s time you still be entitled to compensation in relations to your injuries suffered and property harmed or loss, even though the defendant/s claim that you were at least partly or almost mostly were at fault for whatever happened. If we go by the Pure Comparative Fault Rule, we can see thirteen states recognizes this rule, where it allows an injured party to recover even though if it is 99% at fault, perhaps it might get reduced by the injured party’s degree of fault. Now, these states are namely, California, Florida, Kentucky, Louisiana, Mississippi, Missouri, New Mexico, New York, Rhode Island, South Dakota, Alaska, Arizona and Washington. (Law Offices of Stimmel) Moreover, under the strict liability rules for vicarious responsibility in terms of owners/users of robots would be held strictly responsible for their machines acts. (Law Offices of Stimmel)

Vicarious Liability

Here, the legal system is talking in terms of principal-agent relationship, or partners of partnership firm (employer-employee) and, master- servant relationship. The one person who is responsible for the acts of another, which created liability for one person, the “principle”, for acts/actions performed/undertaken by another person, the “agent”. Based on a legal maxim:

Artificial Intelligence and Policy in India

155

Qui facit per aliumfacit per se, which is a Latin legal term that means, "He who acts through another does the act himself." (Ratanlal & Dhirajlal, 2016) It is a fundamental legal maxim of the law of agency. For instance, liability of A for the act done by B can arise when there is some relationship between A and B, meaning thereby, when one person is liable for the wrongful acts done by other person, then that first person is known as vicariously liable for the acts of second person. So, here the relationship of the two is must to establish liability. Individuals used to be responsible for his own free will, and informed actions under paradigm condition of legal liability, but here vicarious liability is acting as an exceptions to this standard where agent cause harm, but someone else-the principal would be held liable for them having done so, though we can cannot say that agent would be exonerated completely for his acts/omission, generally agent will also be held liable sometimes for their dangerous acts, but it depends upon victim to choose to file a case against their principal on the basis that the latter has deep pockets. So, generally after paying the victims, the principal go on asking for some contribution from his agent. So, the two concepts, i.e. strict liability and vicarious liability are similar, although it differs from the perspectives of principal liability, i.e. for any acts done by his agent under the course of employment-makes principal responsible, whereas on the other hand in vicarious liability, it is principal who will be held liable, but before holding principal liable for the act of his agent, firstly, they both need to establish their relationship into recognized categories of-‘employment’, secondly, there has to be wrongful acts/doings within the scope of that relationship. Section 831 of the German Civil Code talks about the Vicarious Liability:21 (THE COUNCIL OF THE EUROPEAN COMMUNITIES, 2020) In Germany, under vicarious liability-there has to be some wrongful act done by the agent. If agent has not done some wrongful act-applied foreseeability, then in that case principal is not liable vicariously. In California, ‘vicarious liability’ is a legal doctrine which creates liability for an individual who has not causes any injury but having a special relationship with the one who did it. Vicarious liability is also referred as Respondent Superior and imputed negligence, without this relationship one can held liable for the act of another. 4.1

AI under Vicarious Liability

Suppose a police force is using AI robot for patrol and he assaulted one of the innocent individual in the public during its patrol, now in this case, police will be vicariously held liable for the acts of patrol robot, though the AI system is not been created by police force which the robot uses, still they deemed to be held liable for the act/conduct of robot because here police is the one who is deriving benefits from that robot. The assault which happened over an innocent individual which was never intended desired or permitted by police but it happened under the robot’s assigned role. Here, the robot (intelligent agent) is acting as a slave for his master,

Section 831 of the German Civil Code: Liability for vicarious agents (1) A person who uses another person to perform a task is liable to make compensation for the damage that the other unlawfully inflicts on a third party when carrying out the task. Liability in damages does not apply if the principal exercises reasonable care when selecting the person deployed and, to the extent that he is to procure devices or equipment or to manage the business activity, in the procurement or management, or if the damage would have occurred even if this care had been exercised. (2) The same responsibility is borne by a person who assumes the performance of one of the transactions specified in subsection (1) sentence 2 for the principal by contract.

156

Artificial Intelligence and Policy in India, Volume 1(2020)

as under course of its employment without giving status to robot (agent) as a fully legal person in itself. 4.2

Advantages of Vicarious Liability in AI

AI agency: Recognising AI agency under the heads of vicarious liability, when considering the fact that when we were dealing about the negligence and product liability there we have taken AI as an object rather than an agent, now if we depend on that notions and draw inferences in the light of vicarious liability, and holding that legal person liable for the act of an independent agency of-AI, then there we need to strike a balance between the two to recognise legal person liable for its act. Vicarious liability is not so stricter/limited in forms in terms of unilateral/autonomous actions of AI, when it comes to draw a causation relationship between the harm and the person held responsible. This model of vicarious liability needs to be differentiated between the man-made entities and the unique functions of AI. (Turner, 2019 pp. 82- 102) Disadvantages of Vicarious Liability in AI Shortcomings of clarity: No clarity as per needed in relationship, in both the situationsadvantage and disadvantage under certain specific sphere of activities performed by an agent. Depending upon that we cannot label every action taken by AI as ascribable by owner or operator of AI. The less pressing concern is the narrow AI continues to work under tight fitted box of bands. No limitation on labelling AI, whether take it as-employee, child, student, servant, teacher, parents, employer, master or other legal person (human). Now each has their own share of responsibility against others. The focus should be on when, if ever need to cut loose from humans for the purposes of legality like child used to cut loose from being the liability of their parents (potential principal) (Turner, 2019 pp. 82- 102). In the first place this could prevent some person from buying and using robots. So, the strict liability rules of vicarious liability are more stringent than the strict liability norms for compensation caused by any Indian child or animals. We have seen how in the latter case no-fault liability is minimized by changing the burden of proof, so robots-owners and users are not liable when the harmful tendency of the robot were reasonably unforeseeable, an unanticipated event happened where the humans could not protect the dangerous behaviour of the machine, etc. As, Chopra and Write states in A Legal Theory for Autonomous Artificial Agents (2011:11), “to apply the respondent superior doctrine to a particular situation would require the artificial agent in question to be one that has been understood by virtue of its responsibilities and its interactions with third parties as acting as a legal agent for its principal.” Putting differently, this strict liability rules will not fit in all ‘robotic-applications’, rather, to only some special/unique kinds of machine as examined, that is, in Civil Law ‘robots-as-agents’. (Ratanlal & Dhirajlal, 2016) 4.3

AI and Its Burdens of Proof

Problems of responsibility and accountability in the domain of legal parlance are twisted with the burden of proof. According to the maxim of Roman law, “onus probandi incumbitei qui dicit, non ei qui negat”, namely, the burden of proof does not fall on defendants, but rather on the party making allegations concerning a fact or legal issues. In criminal law (Ratanlal & Dhirajlal,

Artificial Intelligence and Policy in India

157

2016 pp. 135), its prosecutors on whom the burden of proof falls to demonstrate that defendant’s guilt on account of any acts/omission which is prohibited under specific norms or statutes? In contracts, the burden of proof falls on the parties who are alleging a breach of an agreement by their own-counterparties. In tort law, the burden of proof falls on the plaintiff who has to showcase the evidence of the wrongdoing-defendants, as caused harm to the plaintiffs. While tortuous claims in international wrongdoing are traditionally distinguished enough, in case of negligence-based responsibility and no-fault liability, in the case of robotic torts a further differentiation is required. Robots raise new kinds of human liability for the conduct of others because of these machines’ acts-as much as humans and animals. In light of panoply, regarding the extra-contractual duty regarding the design, supply, production and usage of these machines, all these suggested questions of tort responsibility that we should differentiate between the two concerning concepts, i.e., when we talk about human industry then robot-as-means and robots-as-agents- in social interaction.

Conclusions

The legal liability of AI systems, as established depends on at least three factors namely: 1. Whether AI is a service or product-ill defined in law, different scholars and commentators offer their different views on it. 2. Even, if a criminal offence is being taken into considerations then what mens rea is required? The reason is that in order to contravene the law it required knowledge that yes a criminal act was being committed by AI programs, it might be the case that they would contravene the laws for which ‘a reasonable man would have knowledge’ about the course of action that could lead to an offence. And, that is why it is much expected that would have contravene strict liability offences. 3. Whether the limitations have been communicated to a purchaser by the AI systems. Although, AI systems have both –general and specific limitations. Legal cases regarding such issues perhaps falls under the purview of any words or any specific wordings of any warnings about such limitations. Now, here the ultimate question is who should be held liable. It depends on Halevy’s three models, i.e., perpetrator-by-another, natural-probable consequence, or direct liability. So, in the perpetrator-by-another offence- is the person who instructs the AI system-that is either the user or the programme- is likely to be held liable. In the case of natural-or-probable-consequence offence, here liability would have fallen on anyone who perhaps foreseen the product as being used in the way it was- the programmer, the vendor of a product, or the service provider. Here the users are less likely to be blamed or held liable until and unless the instructions which came with the product or service spell out the limitations of the system and the possible consequences/outcomes of the misuse which happened in an unusual ways/details in manner. Lastly, AI programs might also be held liable for strict liability offences, mainly in this case- the programmer of AI is likely to be blamed or found guilty/at fault. However, in all the cases where the programmer is held liable, there comes out another debates on this issues, whether it is a fault of the programmer that is, whether the programmer should be held liable or the fault lies only with the programmer or the designer of program (program designer), or the

158

Artificial Intelligence and Policy in India, Volume 1(2020)

expert who has provided the knowledge or the manager who is the one who appointed the inefficient/inadequate expert for that matter, or lastly, the program designer or programmer.

Suggestions and Recommendations

What law we have? 1. Offences in Relation to Use of Motor Vehicles which are Punishable under Indian Penal Code. ─ Rash Driving or Riding on Public Way Section 279 I.P.C. states that whoever drives any vehicle or rides on any public way in manner so rash and negligent as to endanger human life or to be likely to cause hurt or injury to any other person shall be punished with imprisonment of either description for a term which may extend to six months or with fine which may extend to one thousand rupees or with both. The offence under section 279 is cognizable and bailable and triable by the Magistrate having territorial jurisdiction over the area wherein such offence has been committed. ─ Causing Death by Negligence Section 304A I.P.C. dealing with causing death by negligence, “whoever causes the death of any person by doing any rash or negligent act not amounting to culpable homicide shall be punished with imprisonment of either description for a term which may extend to two years or with fine or both. The offence under this section is cognizable and bailable and triable by the Magistrate of the first class. This section has been couched in general terms, based on the main ingredients of ‘rash and negligent act’ which would, naturally, include the act of ‘rash and negligent driving ─ Act Endangering Life or Personal Safety of Others Section 336 I.P.C. deals with Act Endangering Life or Personal Safety of Others. It is provided in the act that whoever does any act so rashly or negligently as to endanger human life of the personal safety of others, shall be punished with imprisonment of either description for a term which may extend to three months, or with fine which may extend to Rs. 250/-, or with both. The offence under this section, as under section 279, is an offence independent of its consequences, and if consequences also follow, the offence would become aggravated and taken account of under section 336 and 337. The offence under section 336 is cognizable and bailable and triable by the Magistrate having territorial jurisdiction over the area wherein such offence has been committed. ─ Causing Hurt by Act Endangering Life or Personal Safety of Others Section 337 I.P.C. deals with cases causing hurt act endangering life or personal safety of others. It is as stated below: “whoever causes hurt to any person by doing any act so rashly or negligently as to endanger human life, or the personal safety of others, shall be punished with imprisonment of either description for a term which may extend to six months, or with fine which may extend to five hundred rupees, or with both. The offence under section 337 is cognizable and bailable and triable by the Magistrate having territorial jurisdiction over the area wherein such offence has been committed. ─ Causing Grievous Hurt by Act Endangering Life or Personal Safety of Others Section 338 deals with cases causing grievous hurt by acts endangering life or personal safety of others and it states that whoever causes grievous hurt to any person by doing any act so rashly or negligence as to endanger human life, or the personal safety of others, shall be punished with imprisonment of either description for a term which may extend to two

Artificial Intelligence and Policy in India

159

years, or with fine which may extend to one thousand rupees, or with both. The offence under section 338 is cognizable and bailable and triable by the Magistrate having territorial jurisdiction over the area wherein such offence has been committed. Grant of Compensation Hearing of Accused Necessary Section 357 (1) of Cr.P.C. deals with a situation when a court imposes a fine or sentence of which fine also forms a part. It confers discretion on the court to order as to how the whole or any part of the fine recovered to be applied. For bringing in application of section 357 (1) it is statutory requirement that fine is imposed and thereupon make further orders as to the disbursement of the said fine in the manner envisaged therein. If no fine is imposed section 357 (1) has no application. The basic difference between Section 357 (1) and Section 357 (3) is that in the former case, the imposition of the fine is basic and essential requirement, while in the later, even in the absence thereof empowers the court to direct payment of compensation. Such power is available to be exercised by an appellate court or by the High Court or Court of Session, when exercising revision powers. Section 357 (5) deals with a situation when the court fixes the compensation in any subsequent civil suit relating to same matter, while awarding compensation the court is required to take in to account any sum paid or recovered as compensation under section 357 of the Cr.P.C.

References 1. American National Standards Institute. ISO/IEC JTC 1/SC 42 Artificial intelligence. International Organization for Standardization. [Online] [Cited: 1 June 2019.] https://www.iso.org/committee/6794475.html. 2. California Legislative Information. 2020. California Legislative Information. California Legislative Information. [Online] 2020. [Cited: 13 April April.] http://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CIV&sectionNum=1714. 3. Developments in English Product Liability Law: A Comparison with the American System. Lord Griffiths, Peter de Val, and R.J.Dormer. 1987-1988. New Orleans, Louisiana : Tulane Law Review, 1987-1988, Vol. 62. 4. â&#x20AC;&#x201D;. Lord Griffiths, Peter de Val, and R.J.Dormer. 1987-1988. New Orleans, Louisiana : Tulane Law Review, 1987-1988, Vol. 62. 5. Federal Ministry of Justice and Consumer Protection and the Federal Office of Justice. 2020. German Civil Code BGB. Federal Ministry of Justice and Consumer Protection and the Federal Office of Justice. [Online] 2020. https://www.gesetze-im-internet.de/englisch_bgb/englisch_bgb.pdf. 6. House of Lords. Bolton v Stone. [1951] AC 850, United Kingdom : House of Lords. 7. Jackson, Lord Justice. 2014. CONCURRENT LIABILITY: WHERE HAVE THINGS GONE WRONG? s.l. : Technology & Construction Bar Association and the Society of Construction Law, 2014. 8. Lady Hale, Lord Mance, Lord Reed, Lord Hughes, Lord Hodge. 2018. Robinson (Appellant) v Chief Constable of West Yorkshire Police (Respondent). [2018] UKSC 4, United Kingdom : The Supreme Court, 2018. 9. Law Offices of Stimmel, Stimmel & Roeser. Comparative Negligence In Tort Claims In California. Law Offices of Stimmel, Stimmel & Roeser. [Online] [Cited: 13 April 2020.] https://www.stimmellaw.com/en/articles/comparative-negligence-tort-claims-california. 10. Pagallo, Ugo. 2013. The Laws of Robots: Crimes, Contracts, and Torts. Torino, Italy : Springer Dordrecht, 2013. ISBN-978-94-007-6563-4.

160

Artificial Intelligence and Policy in India, Volume 1(2020)

11. Product liability and Software. Gemignani, Michael C. 1981. New Jersey : Rutgers Computer & Technology Law Journal, 1981, Vols. 173, 204. 12. Ratanlal & Dhirajlal. 2016. The Law of Torts. s.l. : Lexis Nexis, 2016. 13: 978-9350357415. 13. THE COUNCIL OF THE EUROPEAN COMMUNITIES. EUR-Lex. [Online] 14. —. 2020. EUR-Lex. [Online] 2020. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A31985L0374. 15. The Law Offices of Brian J. O'Grady. The Law Offices of Brian J. O'Grady. [Online] [Cited: 13 April 2020.] https://www.brianogradylaw.com/Articles/Strict-Product-Liability-in-California.shtml. 16. The Reasonable Computer:”Disrupting the Paradigm of Tort Liability". Abbot, Ryan. 2017. 1, Washington, D.C. : The George Washington Law Review, 2017, Vol. 86. 17. Turner, Jacob. 2019. Robot Rules: Regulating Artificial Intelligence. London : Palgrave Macmillan, 2019. 978-3-319-96235-1. 18. Wenar, Leif. 2020. Rights. The Stanford Encyclopedia of Philosophy. s.l. : Metaphysics Research Lab, Stanford University, 2020.