14 Dec 2017 14:00 IST

Chess wiz ‘alien’ AlphaZero is an AI game-changer

Not real aliens, of course!

Been binge-watching fantasy serials on Netflix?

Yes, but I’m not spaced out. Just last week, our very real world got a whole lot more fantastic than anything a scriptwriter could have conjured up.

Lemme guess: because of chess-playing aliens?

Not real aliens, of course. But last week, an Artificial Intelligence program, AlphaZero, devised by Google-owned company DeepMind, gave an intimation of the future of AI and machine learning in a way that had chess Grandmasters — and AI geeks — in rapture, as if they had sighted an alien life-form.

Hyperbole overmuch?

I’ll have you know that Grandmasters are seldom given to exaggeration. But after hearing of AlphaZero’s mind-boggling exploits, Grandmaster Peter Heine Nielsen told BBC: “ I always wondered how it would be if a superior species landed on earth and showed us how they played chess. Now I know.”

What exactly did AlphaZero do?

Now we’re talking. AlphaZero’s programmers did not teach it to play chess: all they did was acquaint it with the game rules. Yet, by repeatedly playing against itself, it sharpened its machine learning skills to the extent that within 24 hours, it defeated Stockfish, the current highest-rated chess program, by 28 games to 0 in a 100-game match (the 72 other games were drawn).

So a machine learnt to play a game? That’s it?

You’re missing perspective here. As FIDE Master Mile Klein noted on chess.com, it’s comparable to “a robot being given access to thousands of metal bits and parts, but no knowledge of a combustion engine, then it experiments numerous times with every combination possible until it builds a Ferrari.” All this in a day! As former world champion Garry Kasparov, who lost to IBM supercomputer Deep Blue in 1997, noted, AlphaZero’s achievement has history-shaping potential. “The ability of a machine to replicate and surpass centuries of human knowledge… is a world-changing tool.”

Hmm, yes, I’m beginning to get excited.

There’s more. As DeepMind’s programmers demonstrated in a research paper — which is yet to be peer-reviewed — AlphaZero’s achievement is not about brute computing power. The paper, posted on the Cornell University website, notes that Stockfish, which lost to AlphaZero, actually had the more powerful program, which could run 900 times faster than its opponent’s. Stockfish could calculate about 70 million positions per second, against AlphaZero 80,000 positions per second.

And yet it won?

It did – by focussing selectively on the most promising variations by instinctively “weeding out” potentially less fruitful explorations. To chess analysts, this is how humans, with limited processing power, play the game, and in that sense, AlphaZero adopted a more “human-like” approach rather than just rely on processing power.

What does all this mean away from the chess board?

Oh, a lot. Earlier this year, the AlphaGo Zero program, developed by DeepMind geeks, beat the world’s top-ranked player of the Chinese board game Go (or weiqi). Having demonstrated its machine learning capabilities, AlphaGo Zero is now working with scientists on drug discovery by figuring out how proteins fold.

As chess writer Albert Silver noted, the ability of open-ended AI systems to learn from the least amount of information and take it to such elevated levels holds the promise that their analysis of problems of disease and famine could lead to genuine solutions.

The bottomline?

The chess wiz ‘aliens’ could well lead us to a brave new world.

(The article first appeared in The Hindu BusinessLine.)