AI: Beyond Alpha Go !

“Game Over! Godlike AI Officially beats the #1 Go Player”

Perhaps the most seminal event in the recent past years which catapulted AI to its recent fame and glory has been the win of AlphaGo over the then reigning world champion Lee Sedol. It is the most oft quoted event in seminars & talks (even by renowned experts of the likes of Andrew McAfee and Jeremy Howard) of AI making a resurrection from its dark winter and proving undeniably that it is here to stay this time. Alpha Go’s win was trumpeted to be a true turning point in the recent history of AI as it demonstrated that given enough data, compute & algorithmic intelligence, AI could conquer the toughest of challenges albeit still in a narrow & specific domain.

We may still be far away from Artificial General Intelligence (varied forecasts including the Singularity forecast put this closer to the Year 2030) but Alpha Go has moved on in its quest to scale ever greater heights in its narrow intelligence domain. In its current avatar christened as AlphaGo Zero, it achieved a super human status in just 70 hours and is apparently now the best ever Go Player in the entire world surpassing any human. What’s amazing is the fact that it did this entirely through self-play and with no prior historical data & human intervention. This feat does raise huge expectations from AI to be a force multiplier which if coupled with human ingenuity can provide us domain agnostic & scalable solutions to address some of the most pressing challenges facing industry & society at large. 

The transition to super human levels of maturity for an AI Gaming application may definitely be symbolic from an optics perspective but it doesn’t quite even begin to address the wider set of technological, societal and economic implications that take AI beyond the realm of complex gaming into the messy real world.

AI has increasingly captured the attention of the elite and masses alike. AI has its band of die-hard supporters & evangelists but also has a lot of cautionary voices (some very notable ones like Elon Musk and Stephen Hawking). AI has been the cynosure of all eyes at prestigious events like the recently concluded World Economic Forum where it has captured the audience’s attention. AI has also become an imperative board room topic across industries. The crescendo around AI indeed has been gaining a ground swell and now extends far beyond mere gaming supremacy.

So what should we expect from AI in the short to mid-terms. Here’s a look at how AI will evolve in the near future:

1. The AI blackbox will begin to unravel: One of the key challenges towards the active adoption of AI has been the inherent opaqueness of the AI algorithms. There has been pressure from multiple quarters that these algorithms perpetrate bias and inequity and further propagate an unfair world. Author Cathy O’Neil’s book ‘Weapons of Math Destruction’ is a scathing account of how these algorithms have gone awry and continue to perpetuate a unhindered vicious & reinforcing loop (e.g. consigning a once convict to a repeat offender or scavenging off the poor). Other challenges have been regulatory aspects which require these algorithms to provide justifications before any of their recommendations can be applied large scale in a live environment. The AI Blackbox needs to unravel if AI has to achieve mainstream status. Concepts like LIME (Local Interpretable Model Agnostic Explanations) which address the same have been around for a while but are increasingly gaining prominence. There is also an active thrust on ‘Explainable AI’ which should help promote wider adoption and deployment of AI. ‘Distill’ a new online Open Science journal aimed at further promoting human understanding of machine learning was launched by a collaboration between Google Brain, Open AI, Deep Mind and others. In the near future, Application of AI in mission-critical or regulatory environments will squarely place the onus on creators of AI models to ensure that they are explainable and hence completely transparent and unbiased. The contrarian viewpoint does state that unraveling the blackbox may amount to dumbing down AI to mere human understanding levels thus reducing its utility but the outcry to reduce bias, inequity and other ill effects of AI algorithms keeps getting louder and supercedes the contrarian camp.
2.  AI will go unsupervised: Most of the AI advancements till date have focused on supervised learning approaches which are premised on the existence of tons of labeled data to train the AI algorithms. This has limited the progress in data constrained domains where it is difficult to find labelled data. Liberating AI from the confines of supervised learning approaches is the need of the hour to help catapult AI adoption multifold and into unchartered territories. Training techniques like deep reinforcement learning (a technique attributable for the success of AlphaGo Zero) and transfer learning will help widen the net of use cases to prioritize areas of high impact but data constrained environments. GAN (Generative Adversarial Networks) which are another type of unsupervised learning technique which helps tide over limitations on existence of labeled data are also gaining ground for applications like cyber security. Hybrid learning models are also on the rise to support things like uncertainty modeling. New kinds of multi-modal models which can perform multi-task learning (e.g. across audio, video and text) in the same model are also beginning to emerge. Capsule Networks now on the anvil purport to do away with the limitations of Neural Networks and need less training data and are better able to deal with ambiguity specifically in the computer vision domain. In short, AI is being unfettered from the bounds of supervised learning & existing data constraints. AI will increasingly leveraging unsupervised learning techniques to help propel the AI revolution multifold over the coming years.
3.  AI will move beyond the geeks: Over the past couple of years, various technology vendors have chased the utopian vision of making AI accessible to the masses on a large scale basis. We are not quite there yet but fast approaching that reality. Citizen Data Scientist will no longer be an esoteric concept but is translating into reality soon. Google’s AutoML (Automated Machine Learning) is a key initiative in the democratization of AI enabling business users to develop machine learning models without much of a programming background. The added benefit of such a toolkit being that it also helps the data scientist community too. There is an increasing repository of notable tools promoting the democratization of advanced machine learning tools coupled with Cloud based consumption models to bring AI to the masses.
4.  AI will continue taking next generation architectural leaps: GPU architectures have been on the ascendancy since the past year and will predominantly be the hand that rocks the AI cradle for the near future. Newer architectures are also beginning to emerge. NVIDIA the market leader in the GPU space has launched the next generation GPU architecture called VOLTA which provides enhanced optimization for deep learning applications. Constantly evolving architectures will ensure reduced latency & enhanced throughput leading to the real time AI nirvana that we have been constantly pining for.
5.  AI will finally go mainstream: For past decades, AI had suffered credibility issues inhibiting its wide spread adoption. The current sentiment towards AI keeping in line with the broader economic outlook (barring the intermittent stock market gyrations) is definitely upbeat. With leading governments (US, China, India and others) now providing a full backing through supportive regulatory regime & allocation of significant investments, AI now has the necessary fillip to grow at a rapid clip. China in particular is making stupendous investments and is making rapid strides to gain supremacy in AI leadership. The industry conversation has gravitated towards tapping AI as a source of competitive advantage. All recent reputable surveys point towards increased adoption or increased willingness towards adoption of AI. The leading technology companies are also placing heavy bets on AI adopting an ‘AI first’ approach which augurs well too. AI is addressing a diverse array of application use cases across industry segments and the leaders of tomorrow will be the ones that can truly harness the power of AI in a creative manner. Investments in AI focused ventures has already shot to a 5+ Billion Dollar level in less than 5 years and is expected to grow at a rapid clip. As end consumers too, we stand enamored with the charms of AI enabled gadgets, immune to the wider manipulative implications it holds for our privacy and the various other ills it may subject us to. In short, we all (Governments, Industry, Investor Community and End users) are providing the needed impetus to advance the AI agenda.

AI is now proving its mettle far beyond the gaming domain into the real world. AI is getting real and every facet of life is beginning to feel its impact. AI will create winners and losers across industries. AI will fundamentally alter the basis for competition & differentiation in industry much like specialization, lean & automation have been in the past. But AI shouldn’t just stop there! AI should not only transform and disrupt industries. It should become the basis for creation of entirely new industries of the future which transform society and that’s where the true potential of