I have yet to find a chess AI that plays like a human opponent at an amateur level. Most AIs seem to play a tactically perfect game but throwing in a really bad move every now and then. Has anyone undertaken (and had any success with) the project of writing a chess computer that plays like a human?

To make the question precise, is there a chess-bot that would pass a turing test?

  • 1
    You are really going to need to define what plays like a human means. Yes, computers don't play like humans do, not in the least, but that is probably by design. If you could articulate what you mean by, "play like a human," it might be possible to make a computer act like one. For example, in the Turing test, computers try to trick judges from believing that they are chatting with a human instead of a chat bot. The best chat bots fool judges about 30% of the time (humorously enough, some humans are mistaken for computers). – user1873 Mar 7 '14 at 5:42
  • 1
    A Turing test would be fine--are there computers out there that cannot be distinguished from a human opponent? – kuzzooroo Mar 7 '14 at 12:41
  • 2
    @user1873, To be fair to the OP, Turing's initial proposition of what eventually became the "Turing Test" began with "It is not difficult to devise a paper machine which will play a not very bad game of chess..." – Brian S Mar 7 '14 at 14:50
  • 3
    @user1873 Turing tests are absolutely not limited to NLP. The suggestion to use a Turing test here is a good one. As with all Turing tests, a time limit must be imposed in order to determine pass/fail. I suggest "three games" as the time limit. If an unknowing chess player can be fooled for the duration of three online games against an AI, then the AI is human enough for me. – Rainbolt Mar 7 '14 at 15:45
  • 3
    @kuzzooroo It shame this question was closed. I was really looking long time for this type of engine and definitely the free Pro Deo (UCI enging) top-5000.nl/prodeo187x.htm It plays very close to human. You can also setup to it to play like novice (without the pattern you described), club player it has also a few interesting personalities. If you are stronger player you can try to play in a full strength with ANTI GM mode. Even I am not very strong player I played a lot interesting rapid games where I was able to gain +- advantage. I prefer v1.87 over v1.88 due the antigm mode. – Pawel Dubiel May 16 '14 at 23:44

The Chess Master series is arguably capable of doing this. It creates personalities that are defined by their personal valuation of the various pieces. Starting with the standard, rough approximations that are typically used (Queen = 8 points, Rook = 5 points, Bishop = 3 points, etc...), each AI personality has these numbers adjusted, often subtly (down to small fractions, such as Queen = 8.08 points, or Rook = 4.73 points). These different weights lead to different (and often non-optimal play) where moves are not-necessarily "bad" per-se, but rather simply valued in a way that a master would not value them.

  • 3
    I always found chessmaster to be notorious for unnatural ai. At lower levels it would make dumb moves, but check many moves ahead to avoid a checkmate. This would mean every game would have to be a full peace clear before a win. – Andrey Mar 7 '14 at 14:17
  • I remember seeing "... sees mate in 11" and then have it blow the line and a few turns later "... sees mate in 14". – corsiKa Mar 7 '14 at 18:57

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