May 3, 2010Newsletter Date
Letters to the editor.
Special Issue
Page 10
“So you think you are the only one who has to follow laws ?� Page 8
Deep Blue Champion
TIME
Page 8
Throughout the Years: AI By Lea Balcerzak
Artificial Intelligence
Page 2
“Nanofly” Hand Made Robot
T
his whole robot was designed in a 3D computer program.
The main body is constructed from two 6” Plastruct hemispheres. Metal masters create the other part such as the feet, legs, arms, and thrusters. A small camera lens is used for the “eye”. This robot was sprayed with specially mixed metallic automotive lacquers and given a semi gloss coat. Buy the Nanofly today! Only $100,000,000,000. It will serve as a friend as well as help you. No other robot will suit your needs better! Call 214-574-5833 to order, and mention this ad to receive 10% discount!
Page 3
In This Issue...
2
Nanofly Robot: the robot for you
3
4
Table of Contents: explains where everything is
Main Article: talks about the 3 main AI approaches
8
10
8
11
Deep Blue Champion: the AI program that beat the world champion in chess
Three Rules of Robotics: the rules all robot have to follow
9
Turing Test: the machine test for intelligence
Letters to the Editor: Question from our readers
Artificial Intelligence Fun
Page 4
Throughout the Years: AI
T
he scientific understanding of the mechanism, underlying thought, intelligent behavior, embodiment in machines,” AI (artificial intelligence) is the ability for a computer system to process information in a manner similar to human thought or to exhibit human-like behavior. AI is a developing program which involves creating a computer program, which could be used in a robot so that it could operate by itself with an intelligence equal to a human one.
- Knowledge represented using numerical probabilities.
It is taken that the field of AI has its beginning with the publication of the British mathematician, Alan Turning’s paper “Computer Machinery and Intelligence.” (1950)
2. Functional AI
Main Approaches “Knowledge representation and reasoning are two core challenges in AI. They can be handled in various ways:” (Litman) Logical Approach - Knowledge represented by precise logical rules. Probabilistic Approach
Neural Network -Knowledge represented as a network of interconnected units that perform certain task by exchanging information. -Mimics the behavior of neurons (nerve cells) in the brain. The three most important types of AI are: 1. Symbolic AI
3. Relational AI
Page 5
Symbolic AI Success? “The earliest approach to AI is called symbolic or classical AI and is predicated on the hypothesis that every process in which a human being or a machine engages, can be expressed by a string of symbols that is modifiable according to a limited set of rules that can be logically defined.� (Harzfeld) Just as complex geometry is based on axiom and primitive objects such as lines and points, symbolists (scientists working with symbolic AI) predict that human thought can be broken down into basic rules and primitive objects. Symbolists believe that a simple idea can be directly expressed by a single symbol, while a more complex idea would be presented with more than one object, combined with certain rules. Step by step, rule by rule, scientists hope to construct a working AI program. Symbolic AI came to great success in areas where task could be described in simple rules. For example, a symbolic AI program, Deep Blue, beat the world champion in chess, (read more on page 8). Other successes of symbolic AI including: medical diagnosis,
mineral prospecting, chemical analysis, and mathematical theorem proving. Symbolic AI failed, however, not
in complex tasks such as passing a calculus exam, but in tasks a two-year old could perform, such as recognizing a face or understanding a simple story Critique Symbolic AI also came to be a figure of great critique. Why? This program achieved many things, and brought researchers so much further in technology, still some are not happy with it! Terry Winograd and Fernando Flores, in their 1986 book, Understanding Computers and Cognition provided a critique for symbolic AI, stating that human intelligence cannot be laid out by symbols
Page 6
or rules. Humans do not carry mental models around in their heads,” they state. Rather, these men suggest that humans use intuition, gained through multiple experiences in the real world. Can It Exist? Scientist now ponder if symbolic AI was such a great idea... “Should it exist, it is now clear to AI researchers that the set of primitive facts, necessary for representing human knowledge is exceedingly large.” (Harzfeld) What about intuition? It can’t be learned, but can a successful AI work without it? So can a good working, very intelligent, symbolic AI program, really exist?
Functional AI “The scientific understanding of the mechanism, underlying thought and intelligent behavior and their embodiment in machines,” is the politically correct definition of functional AI according to the Merriam-Webster Online Dictionary. It is also the basic foundation of functional AI. Many papers were written on AI. In 1980, John Searle, in his pa-
per, Minds, Brains, and Programs divided the field of functional AI into two categories: strong artificial intelligence and weak artificial intelligence. In his opinion, strong AI, works to develop a full human-like intelligence, while weak AI is used for the better understanding of human reasoning and to solve less complex problems. This type of AI has been the most helpful to scientists in performing experiments. For right now, researchers that this type of AI would not prove very helpful. Maybe...someday...in the future, functional AI will achieve success... Some Success So far functional artificial intelligence has achieved success such as: fuzzy logic, heuristic searching, machine learning via statistical methods, to practical problems. The Problem There are two problems with functional AI: 1. Determination whether a task should be considered an AI program or just simply a computer application. 2. Functional AI is rarely designed
Page 7
to communicate with other programs. Because of this it is usually unable to understand the techniques other programs use. How is that a problem? To have a fully working functional AI program, many different “parts” would have to combine. Presently, this is not possible.
Relational AI “ A third approach is to consider intelligence as acquired., held, and demonstrated only through relationships with other intelligent beings.” (Harzfeld) Alan Turing was one of the most involved and outspoken researchers of relational AI. He suggested that programmers should stop trying to create a program with the intelligence of an adult human being, and instead, construct a mind that would simulate one that of a child. Such a mind then, if given proper education, would develop into an adult mind. This idea came to be called functional AI. Rodney Brooks of the Massachusetts Institute of Technology expanded on Turing’s idea and created a few robots including Cog and Kismet. Their programming is distributed among the various physical parts. Each joint
has a small processor that controls only the movement of that joint. The processors are then connected to a larger processor which controls the whole. So far these robots have been trained to perform tasks as those of an infant: eye-hand coordination, handling simple objects, and face recognition. Through social interaction with a team of researchers. The Turing Test In 1997 in his paper Computing Machinery and Intelligence, Turing suggested a test which has now been generously accepted as the machine test for intelligence ( more on page 11 ) . Neural Network A branch of relational AI. This area of AI shows much promise, it works by imitating the neurons found in the brain. “Even though, this field is also a bit limited, since the human brain has billions of brain cells, and scientists are yet to fully understand how they work. It is though that the neural network has shown the most promise in areas such as speech or image recognition, and learning.
Page 8
Deep Blue Champion
I
n 1997, a chess-playing computer AI program, Deep Blue, won a match against the Chess World Champion. This symbolic AI program was able to beat the champion because “the game of chess takes place in a world where the only objects are thirty-two pieces moving on a sixty -four square board according to a lim-
ited number of rules.” (Herzfeld) The limited options gave the program the advantage of looking ahead, seeing all (and I mean ALL) the possible moves, and choosing the one which would provide it with the largest benefit. This accomplishment gave researchers hope that symbolic AI, could prove helpful.
The Three Laws of Robotics
S
o you think you are the only one who has to follow laws? Not true! In Isaac Asimov’s science fiction books, he presented three main “laws” that all robots had to follow. These laws were first presented in his book “I, Robot: Runaround”, published in 1942.
“The Three Laws of Robotics: 1. A robot may not injure a human being, or through inaction, allow a human being to come to harm. 2. A robot must obey any orders given to it by human beings, except where such orders would conflict with the First Law. 3.
A robot must protect its own existence as long as such protection does not conflict with the First or second Law.”
The Laws, so popular and sensible, are now being used by modern robot constructors, as the foundation of the robot’s artificial intelligence.
Page 9
Turing Test
I
n 1997, in his paper Computing Machinery and Intelligence, Turing suggested a test which has now been generously accepted as the machine test for intelligence. (from page 6) In this test, a human interrogator is connected through one terminal each, a human and a machine with an installed AI. The interrogator ask both ends the same questions. If he fails to distinguish as often as he succeeds in determination who the machine and who the human is, the machine/ computer is considered to have a well working artificial intelligence. “The Turing Test is not based on the completion of tasks or the solution of problem by
the machine, but on the machine’s ability to relate to a human being in a conversation.� (Herzfeld) When Turing first presented the idea of the test, he predicted that by the year 2000, computers could fool at least 30% of the interrogators. So far no computer has came even close to passing! After acknowledging that no program could pass his test, Turing preformed more research and predicted (again!) that it would take about 300 more, human years to construct a program which could score the Turing Test with a perfect score!!!
Page 1010
Letters to the Editor Dear Editor,
Dear Editor,
Do AI programs have names?
What is AI, anyway?
Paulette T.
Camden A.
Dear Paulette,
Dear Camden,
Many AI programs have been created, so it would be very hard for researchers to remember all their names. Only the most successful AI programs have been given a name.
AI is the ability of a computer or machine to act, but mostly think like a human.
Editor
Dear Editor,
Editor
Is a computer or video game AI?
Dear Editor, What are AI programs used for? Clint K. Dear Clint,
Martin S. Dear Martin, No, a computer or video game is not artificial intelligence. It is simply a computer/TV application or program.
Researchers hope that in the future AI programs will make our daily life easier. Editor So far, AI is grouped into the following categories, according to what the program Dear Editor, does: Do AI programs have attitude? knowledge representation and reasoning Lilliane F. speech and natural language processing Dear Lilliane, planning and problem solving So far, no. AI programs are very machine learning loyal to their creators. Lets hope that in the future, artificial intelligence programs don’t computer vision develop one. robotics
Editor
Editor
Page 11
Artificial Intelligence Fun!