What we learn from chess, car crashes, and disease

It was December 2017. We turned a new page in technology evolution. Google’s AlphaZero program beat Stockfish 8 program at the traditional game of chess. But what is the big deal? Well for starters, Stockfish 8 was the world’s computer chess champion for 2016. It had access to decades of human chess experience. What is more surprising is that AlphaZero’s creators never taught it any chess. AlphaZero used machine learning (specifically reinforcement learning) to play matches against itself. And later, out of 100 games they played, AlphaZero won 28 and drew 72. It lost none of the games.

Okay. Our supposed-underdog must have sweat years in practice to bleed less in war right? So how long did it take to reach this level of mastery? 4 hours ; the collective time one would take to finish watching Kabhi Khushi Kabhie Gham and hardly have time to rewatch their favourite scenes.

If you are not pulling your hair at this point, it’s probably because you don’t have any. We created a machine that with four hours of self-practice beat centuries of collective wisdom at a game we have crowned as the epitome of human intelligence. That is the power of machine learning (ML) .

From the given context, we can infer that ML is about kick-ass decision making. More precisely, ML uses computer algorithms to analyze data and make intelligent decisions based on what is learned. At a ‘simpler’ level, this would be how streaming sites recommend music — comparing what one listener likes to others with similar tastes.

Deep learning (DL) is the big brother and does the heavy lifting. It does the same job (decision-making and prediction) but uses more complex relations on high dimensional, unstructured data (images, audio, video). Currently we make use of artificial neural networks to achieve this, and function similarly to the neurons in our brains.

The famous IBM Watson uses DL. It is used to predict when elevators will break down and proactively fix them; help banks deploying virtual agents respond faster to customer inquiries, etc. Watson can also tell an insurance company how exactly a car has been damaged after a crash by comparing it to the same model undamaged car. It can identify damages with very high specificity. How is ML and DL related to Data Science?

Data science (DS) covers a spectrum of activities from data collection and processing to describing and modeling. It overlaps with ML and DL in the aspect of modeling data, specifically algorithmic modeling. Algorithmic modeling focuses on prediction based on underlying features. For example, we could diagnose disease by studying the various symptoms reported by a patient. How it stands out from traditional statistical data modeling, lies in the wide range of data used, the high complexity of relations between different features, etc. This shows the convergence of the three- ML, DL, and DS.

Since its introduction in late 2017, AlphaZero went on to beat world-champions in chess, shogi(Japanese chess) and Go. The highlight of its success was that many of its winning moves and strategies were unconventional to human eyes. This untapped creative genius that the convergence of Ml, DL and DS offers is a source of many innovations for the future.