5 minute read
HOW DOES AI FIT INTO YOUR BUSINESS?
/ By James Francis /
In January, Meta (formerly Facebook) revealed its AI Research SuperCluster. One might think: so what? You can log onto AWS or Azure and pick your AI service flavour. Why does it matter that Meta is doing the same?
Only, it's not quite the same and, in retrospect, we'll mark the SuperCluster as a significant milestone. For people in the commercial AI and high-performance computing (HPC) sectors, the SuperCluster is huge news, explains Adrian Wood, Business Operations Director at Data Sciences Corporation:
"I think two things stand out about the SuperCluster. Firstly, it's a massive, massive investment in AI infrastructure by a Fortune 500 company, not an academic institution or nation state. Second, Meta is not building this to sell services to users, as the hyperscale cloud providers do. The SuperCluster is an inward-facing strategic project. It's there to help drive Meta's business future."
The SuperCluster, dubbed by Meta as the RSC (AI Research SuperCluster), is a behemoth: 760 NVIDIA DGX A100 systems representing a total of 6,080 GPUs and over 200 petabytes of caching/storage capacity that can provide 16 terabytes a second of training data (with plans to scale to 1 exabyte a second). SuperCluster underpins Meta's metaverse plans. Real-time emotional speech translation in online meetings and highquality computer vision are just a few applications that will emerge from this AI training complex.
It's a strategic business investment - a sign that the future of digital enterprises will align closely with their AI investments. Of course, Meta's investment doesn't mean every business should start budgeting for a SuperCluster. But it suggests that AI be part of strategic conversations, says Werner Coetzee, Data Sciences Corporation's Business Development Executive:
"Until recently, companies could look at AI as something other than a central business enabler. That narrative is changing - the SuperCluster proves that AI infrastructure has become more commoditised and affordable. For example, Nvidia is one of the highest-valued chip manufacturers because it in part supplies chips for AI. We're entering the enterprise high-performance computing era. Whether businesses rent AI services from the cloud or start building their own mini clusters, they must think about how AI impacts their strategy and future business models."
Making sense of AI
Few technology topics can cause as much confusion and anxiety as AI does. Highly technical and broadly mythologised by Hollywood movies and sci-fi novels, AI sits outside the enterprise's comfort zone. Yet you don't need to wade into the AI undergrowth - a handful of concepts illustrate fundamental AI use cases and how they can fit into your plans.
Below is a primer on AI concepts that enterprises should know about:
Computer Vision
Computer vision is about 'seeing' objects. For example, many estates use AI via cameras to register car licence plates. Insurance apps can register a policy or claim, using computer vision to analyse photos of the insured items. Manufacturers deploy computer vision to spot defects on product lines. One of the current focus points is identifying complex visual elements, such as objects moving in 3D space and creating virtual duplicates of physical systems, often called Digital Twins.
Natural Language Processing (NLP) & Speech Recognition
Siri, OK Google and Alexa use speech recognition models trained by AI. If a service chatbot understands typed responses, that is natural language processing at work. You may have already encountered online presentations translated in real-time to another language. And if you use AI-powered transcriptions from audio files such as Otter.ai, you benefit from this branch of AI. Researchers are currently focused on analysing emotions in language that will translate real-time conversations with the nuances of human speech.
Augmented Reality & Virtual Reality
Augmented reality (AR) - which overlays digital information over what we see in front of us - is used for training, design and repairs. AR is also active in surgery and across the windscreens of high-end luxury cars. Mining, manufacturing and engineering firms have found numerous uses for AR. Likewise, virtual reality (VR) has become instrumental for designing cars, aircraft engines, and architecture. AI training helps create and manage the visual data that both technologies use, such as digital twins and virtual collaboration rooms.
Self-supervised learning & Transformer-based models
Self-supervised learning and more advanced transformed-based models enable AIs to adjust their training parameters. But they don't stray from their objectives - an AI improving computer vision won't go rogue. Despite the name, such systems are highly supervised. They just don't need human intervention to adjust their parameters, thus significantly speeding up AI model innovation. The ability of Google to translate web pages is a predecessor of these concepts.
Robotics & Embodied AI
When a robotic vacuum cleaner figures out its way around your home, that's an example of robotics and embodied AI. Embodied AI happens at the edge, detecting the physical world and making decisions based on that input. We're inclined to think of self-driving cars, but an embodied AI could be a search & rescue robot navigating an unknown space as it looks for survivo rs. Amazon Echo devices are a current example: they can dynamically lower or raise their volume in response to ambient noise. Robotics puts embodied AI in action around us, operating safely and effectively with the information detected by its sensors.
Multimodal AI
As the name suggests, multimodal AI incorporates two or more different AI concepts. Several above examples already illustrate the concept, such as robotics or virtual reality interactions.
What does any of the above mean for your business? That is a question only you and your peers can answer.
But Meta's big investment indicates that AI is inseparable from a maturing digital business world. Whether you use a few AI services or strategically adopt high-performance computing, start having conversations about AI and what it means for the future of your enterprise.