In December 2016 the White House released a report on Artificial Intelligence (AI) and its bullish effects on the economy, suggesting to earmark more investments in AI. The timing of the report is not surprising given that AI may be the single most disruptive technology the world may have seen since the Industrial Revolution. Although its seeds were sown back in 1956 at Dartmouth College, it’s only recently that AI has unleashed its potential to fundamentally alter how industries and humans work. Established companies and startups are now racing to use AI to build new products, services, and improve their operations.
The rise of ubiquitous computing, low-cost cloud services, near unlimited inexpensive storage, vast swathes of data, algorithms that learn and advances in Graphical Processing Units, all have put AI on an accelerated development path. Consider this, almost 40% of world’s population is online and uploads massive quantities of images, videos, and social media comments. All that creates labelled data that the machines ingest in what’s called Machine Learning (ML) that allows computers to find hidden insights (using statistical and predictive analysis) without being explicitly instructed where to look. Netflix recommendations, email spam filtering, trending topics and hashtags on social media, credit card fraud detection are some real world examples of ML. Throw in some intuitiveness and what you get is Deep Learning (DL), an advanced branch of AI with predictive capabilities that is inspired by the brain’s ability to learn (our new found love for neural networks!). DL algorithms can assess an object, digest the information and adapt to its variants. Let’s say you show a Deep Learning computer an image of a cat, and then modify a few percent pixels of the image, the computer will still recognize that there is a cat in the image (Schrödinger’s cat excluded!). This borderline science fiction technology with self-learning capabilities is fueling advancements in computer vision, speech recognition, cancer detection, self driving cars, user behaviour prediction, and weather forecasting to name a few fields.
2016 witnessed some notable AI accomplishments and will be remembered as the year when machine learning systems routinely drubbed humans in narrow domains of excellence. In March, the world’s best player of the complex board game Go, Mr. Lee Sedol was humbled by Google’s AlphaGo (a feat thought to be a decade away). In June, in a dogfight simulation, University of Cincinnati’s ALPHA AI trounced USAF top tactical expert Colonel Gene Lee with responses 250 times faster. In July, Google’s Deepmind reduced data center cooling bill by 40% (consider the implications when that knowledge is used to develop smarter energy infrastructure). With its ever increasing capabilities Deepmind can be deemed as an apotheosis of contemporary deep learning that continues to surprise us. In November, at the University of Oxford, Deepmind dug its teeth into lip reading BBC TV Shows that left the professionals in the dust.
For a tantalizing example of what higher intelligence might eventually look like we need not look any further than Google, yet again. In September 2016, its Translation Research division announced an enormous upgrade in the performance of ‘Google Translate’, using a system called Google Neural Machine Translation (GNMT) that achieved the feat of Transfer Learning. After a few hours of training its sophisticated algorithms had taught themselves a rudimentary new skill using indirect information. It’s like a machine that learns how to ride a motorcycle and then uses that knowledge to self-teach how to drive a car (though with elementary efficiency)! Elon Musk’s OPEN AI recently released an open source digital playground called Universe that virtually tests and trains AI using games and apps. It is designed to make use of reinforcement learning techniques to eventually start practicing Transfer Learning. A game might appear to be an artificial setting, but eventually the same techniques will be applied to solve the real-world problems.
AI spearheaded with Deep Learning will set the pace of technology innovation to the ‘Ludicrous Mode’, leaping closer to the utopian predictions. With each passing day machines collect more data and make smarter decisions for humans. We aren’t far away from putting deep learning into the reach of millions, potentially billions of consumers. At both ends of the spectrum the man-machine synergy is changing irrevocably, and along with it life as we know it. This is just the beginning!