Tournament players can "have a good tournament" because of the psychology involved, so their playing strength will vary from game to game. Do computers play at a constant strength all the time, or do they have the occasional bad tournament?
1 Answer
We're going to assume there is no hardware issue, as pointed by Tonny Ennis in his comment.
It's possible that a computer engine experiences a "bad" tournament, but it's extremely rare. When we say "bad", we don't mean collapsing like blundering pieces, it's more like not getting a good position suitable for the engine. Computer engine always play at a constant strength, dictated by the algorithm. They won't do anything other than what the algorithm tells it to do.
Let's give an example, it's well known Komodo is a better engine than Stockfish in a strategic position. If we randomly choose the openings in a match of four. We might see a closed position in three of those games. As expected, Komodo outplayed Stockfish in those three games. The final score would be 3-1 in favour of Komodo.
Does that mean Komodo is a much stronger engine than Stockfish? Probably not. Can we conclude Stockfish had a "bad day"? Maybe. But please remember the sample size is too small, what if we let them play more games? A typical computer chess tournament can have a match with over 1000 games. If we do it, we might see a very different result.
Conclusion: if we see a chess engine has a "bad day", usually it's because the sample size is too small or the testing condition is biased. The law of large number will erase unsystematic variation given enough sample size.
-
The question is whether the engine can have a bad day, given that the computers get to choose what opening they play.– limitsCommented Jun 19, 2015 at 22:55
-
The answer does include that. If computers can choose openings and they have a "bad day", most likely you don't have enough sample size. Commented Jun 20, 2015 at 6:22
-
I give you upvote, but the question is whether variation in playing strength will/can occur, not whether a large sample size will override variations.– limitsCommented Jun 20, 2015 at 18:43
-
-