The future of chess could be right here. The Artificial Intelligence (AI) programme AlphaZero's recent, mind-boggling exploit -- destroying the highest-rated chess engine Stockfish in a 100-game match after 'learning' chess in only four hours -- has taken machine learning to new, inconceivable levels. If that sounds like hype from an overexcited journalist, here's what five-time world champion Viswanathan Anand has to say about the programme: "The chess world will get scrambled," he said. "The current rating list will be scrambled. I think it will happen gradually."
"It can definitely be a world-changing tool."
Stockfish is the go-to preparation tool for top players and being destroyed by the AlphaZero algorithm -- which has no domain knowledge barring the rules of the game -- is, to put it mildly, beyond unbelievable.
AlphaZero has been devised by the Google-owned company DeepMind. For someone like Anand (48), who belongs to the generation that was bred on the chessboard and later migrated to a technology-powered version of the game, the first reaction to the match result was pure astonishment. "To watch such a strong programme like Stockfish, against whom most top players would be happy to win even one game out of a hundred, being completely taken apart is certainly definitive," Anand told ESPN after finishing ninth at the London Chess Classic on Wednesday. "The important bit is the lack of human knowledge. AlphaZero did not base its conclusions on what we already thought and suspected, because that knowledge might be incomplete, inaccurate. So we've got a clean start and the implications are very interesting."
"I always wondered how it would be if a superior species landed on earth and showed us how they played chess... Now I know."Grandmaster Peter Heine on AlphaZero's win
AI in chess had its first historic moment exactly 20 years ago, when IBM Supercomputer Deep Blue beat reigning human world champion Garry Kasparov in a six-game match under standard time controls. Kasparov would later cry foul and raise doubts about the computer being a chess automation hoax and possibly being controlled by a human Grandmaster. Chess programmes have since leapfrogged way ahead of any human contest. They usually combine high-performance alpha-beta search and domain-specific adaptations with different positions, fed into it by its human creators.
But in a departure from how conventional chess engines function, AlphaZero used a vastly different and pioneering approach -- reinforcement learning. It can be likened to how humans learn, from experience and mistakes rather than labelled data. Mastering the game through self-play with no human intervention whatsoever, AlphaZero played 72 draws and went on to win 28 games to 0 against what's easily the highest-rated chess engine.
Also, unlike Stockfish, which searches 70 million chess positions per second, AlphaZero skims through just 80,000. While it's still an enormous amount when compared with humans, against other computer programmes it's a lot more discriminating, Anand suggested.
"It's like a robot being given 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."FIDE Master Mile Klein
The complete and unwavering dominance that the latest algorithm has shown even got some top players such as Sergey Karjakin and Wesley So enthusiastic about writing off big money. "I will pay very much to get access to this programme. Maybe $100,000 a day!" Karjakin told Chess.com. Anand, though, was typically more contained in his excitement, reasoning that beyond simply being able to ask questions and have them answered by the latest, super-intelligent programme, it's more a matter of what and how much he can reproduce on the board. "I would need to see AlphaZero play several matches with several programmes under different parameters before I can actually pull out my checkbook," he said.
And while he certainly foresaw a likely new world order in chess, Anand was clinical in spelling out the more immediate implications of AI's latest conquest, separating them from the long and more lasting impact.
"Among the short-term effects, I imagine it would be more opportunistic," he said. "If you're able to see some new games and see a new idea tried which hadn't even occurred to you before and if you are able to make sense of it before others, you might be able to use it here and there. But once this technology becomes available to all, you'll have a generation that will see the chessboard in a completely different way."