-8

I know that modern programs are highly use historical plays databases to have openings and ends lists. Also chess algorithms often use AI to learn how to better calculate the move. Anyway I am not still sure if they also use plays database to learn the algorithm.

I know that modern chess programs are way better than the best players but I am wondering what would happen if we used only e.g. minimum 50 years old games to learn computers how play (so learn the algorithm only on the games since 1966). The quality of the human play improved since 1966 so IMO the questions makes sense.

3
  • 3
    This might be already an answer: chess.stackexchange.com/questions/13048/… Sep 30, 2016 at 8:55
  • It may be helpful to know that (apart from some relatively new and still experimental work with genetic algorithms): 1) engines use an opening book put together by a few experts, who use existing theory and add some new ideas, and 2) engines evaluate moves based on what they accomplish in the position. They use weighting factors that are the result of trial-and-error experimentation, usually against other engines. To use only games that are at least 50 years old would involve starting from scratch. And it would be pointless to test against other engines; defeat would be certain. Why bother?
    – jaxter
    Oct 1, 2016 at 23:55
  • I also note that you spotted a nearly-identical question on the same day you posted this one. Ideally, it's best if you do your research here and elsewhere first, and then post the question if you can't find relevant information.
    – jaxter
    Oct 1, 2016 at 23:58

1 Answer 1

6

This question is based on a number of misconceptions.

One, engines are not self-learning software, they are not "trained" on a set of games. They just go through a huge number of possible variations and evaluate the end positions based on fixed factors set by the human programmers.

Second, engines can be configured to use opening databases, but they hardly need them to beat humans -- they play above world class level without them as well, nowadays. The main reason to include them is to introduce some variety (they'll pick a random well known opening move) because the programs are deterministic and otherwise they would play the same move in the same position every time.

Endgame tablebases are sometimes turned off because the time it takes to look up a position in them may not be worth it and it doesn't help the engine differentiate between a bunch of moves that all lead to a draw against best play; but they are perfect, of course.

Anyway the way it usually goes is that nowadays engines win whatever you do to them... for instance this january top GM Nakamura played Komodo in various odds games (twice starting a pawn ahead, once an exchange ahead, and once four moves ahead) and lost 1.5-2.5 ( https://www.chess.com/news/komodo-beats-nakamura-in-final-battle-1331 ). Those advantages are huge.

Not the answer you're looking for? Browse other questions tagged or ask your own question.