World chess champion Garry Kasparov during the first six-game match against IBM supercomputer Deep Blue, Pennsylvania Convention Center, Philadelphia, February 10-17th, 1996
First people thought that chess engines could never beat humans, and then Deep Blue showed them that this is wrong. This changed the way grandmasters thought about computers, and soon enough chess engines became a part of every top-level player’s toolbox. This eventually affected the way tournaments are held as well, because the rules had to be changed so that games have to be completed during one day and they cannot be adjourned, because computer analysis during the break would affect the game too much.
But that’s not all. Engines such as Stockfish, which can readily defeat even the strongest human players got so revered that they changed the way grandmasters evaluated certain moves and positions. Basically, it was thought that if the engine calculates the move to be bad, then it is bad, end of story.
And then came AlphaZero and changed everything again. Orignally developed as an engine to play Go (and it beat the ruling world go champion in 2017), AlphaZero’s approach to the game is different from traditional chess engines. In fact, it was never programmed any way to calculate the positions, it was indeed taught nothing but instead taught itself chess by playing millions upon millions of games, and after four hours of studying it beat StockFish 8, held to be the world’s strongest traditional chess engine (and a descendant of the style of engine that Deep Blue was). You can check out (some of) the games (they played hundreds of games) between the AIs on Youtube analyzed and explained for example here.
What’s so fascinating about AlphaZero is that as a self-learning system, it doesn’t play like an engine does. Old fashioned engines rely on brute force calculation and decision trees. Whereas AlphaZero makes moves that look silly, it sacrifices material, it plays for space and position. And it runs circles around traditional engines.
This changed the way the grandmasters approach the game again. Moves that were held ‘objectively dumb’ (such as pushing pawns ahead aggressively in mid-game to control space, even if it means potentially sacrificing the pawn) because old school chess engines hated them, are in fact played by AlpaZero in winning games against old-school engines, which proved that they are indeed viable, and are now played in top-level play by leading grandmasters such as the reigning world champion Magnus Carlsen.
It took computers decades to eventually learn the art of chess from humans, but it’s now gotten to the point that we’re learning from them. The difference is, whereas a decade or so back, everyone was ‘trying to play like Stockfish’, now everyone is ‘trying to play like AlphaZero’.
