Even though modern chess playing programs have demonstrated themselves to be as strong (or stronger) than even the best human players for nearly 20 years now (1997 when IBM's Deep Blue defeated the world chess champion Gary Kasparov), why would a game like chess still be considered a valuable research subject in Artificial Intelligence? In other words, what can be gained by continuing to advance AI in areas that have already surpassed human capabilities?

For instance, as recently as November 2017, Google successfully challenged its deep learning technology against one of the world's strongest chess-playing programs.

  • $\begingroup$ My high-level answer would be that Chess is still presumed to be unsolvable, the gametree is quite complex, and it's useful to validate algorithms against other algorithms in multiple contexts. Go may be the most complex "game people play", but based on a recent, reliable poll by yougov, Chess has about 10x the casual player base, and, until recently, was much more famous than Go in the west. (In tv and movies, Go is starting to replace Chess as the signifier of a character's intelligence or strategic acumen, no doubt a function of AlphaGo, though the movie π utilized Go in 1998.) $\endgroup$
    – DukeZhou
    Dec 15, 2017 at 20:43
  • $\begingroup$ Thus, Chess still has high PR value, in addition to "complexity akin to nature". $\endgroup$
    – DukeZhou
    Dec 15, 2017 at 20:47
  • 1
    $\begingroup$ So once you get -just- past human abilities, do you just sort of stop bothering anymore? $\endgroup$
    – Mitch
    Sep 13, 2021 at 22:13

1 Answer 1


Chess isn't really a benchmark per say.

The method developed in AlphaGo to play Go should in principle generalize quite nicely to other games of this sort, such as chess. Since Stockfish is quite dominantly the strongest Chess AI, the natural question would be to see how well AlphaGo's method compares to Stockfish.

Being one of the most well developed AI agents of all time, the situation concerning the defeat of Stockfish (AlphaZero was trained for only 4 hours entirely via self-play, without access to historical data) signifies the complete dominance of modern neural-network methods over classic methods (hard coded evaluation functions).

Also as @DukeZhou♦ mentioned in the comments, while Chess bots can regularly beat human players, it's still a useful metric to evaluate bots against each other via "games" of this sort.

edit: But as the more recent results of Stockfish 13 versus Lc0 (an open source AlphaZero clone) show, handcrafted/traditional algorithms (search in particular), paired with neural network techniques, can still outmatch pure neural networks. This perhaps highlights the value of classical techniques in the face of more modern approaches.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .