5

Sometimes Chess engines identify the best move as a move that’s mathematically the best but would be difficult for a novice/intermediate to pull off because it must be followed up by a specific, careful sequence of moves or else it’s not a good move.

Are there any Chess AIs that can identify the top engine move and also pick, say the second-best move, as being the “top practical move”?

1
  • 1
    A very interesting question, I think! :) Commented Apr 28 at 19:13

3 Answers 3

5

Maia Chess is a deep learning framework like AlphaZero/Leela. Its goal is to play the human move — not necessarily the best move. As a result, Maia has a more human-like style than previous engines, matching moves played by human players in online games over 50% of the time.

You can learn more about it here. If you'd like to play some games against Maia, lichess already has some featured bots for different rating levels. Here's the link.

1
  • Leela's WDL contempt might be closer to choosing practical lines.
    – qwr
    Commented Apr 28 at 18:57
1

Recent versions of Leela Zero have implemented a feature called Win/Draw/Loss (WDL) Contempt based on a rating difference between players.

The idea of WDL Contempt is that you tell Leela “Look I’m playing White against an opponent rated 400 ELO lower than me. Can you suggest some ideas in the opening to maximise my winning chances against such an opponent?” During analysis, Leela will look to sharpen the play to a degree appropriate to the ELO difference.

So high WDL will pick tricky sharp lines that have to be calculated accurately to navigate, and will give even experienced players trouble to calculate without preperation.

2
  • I could be misinterpreting, but isn't contempt just how Leela weights "100% chance of draw" vs "50% chance win, 50% chance loss"? I don't think it has anything to do with the human-ness of the position, just how willing it is to go into drawish endgames
    – Kaia
    Commented Apr 29 at 23:17
  • Contempt factors in opponent's strength. It chooses sharp lines that are harder for humans to navigate.
    – qwr
    Commented Apr 30 at 1:10
-1

If I understood the question correctly, it could be formulated this way: are there chess AIs that play moves that appear human? The answer is yes. Because a chess engine does not just “mathematically” find a move to play. If this were the case, all engines would play the same shot in the same position. We are far from it. The engine does explore the move tree, but it is so important that it is impossible to go through it all. To choose a move he must base himself on what we call evaluation. The evaluation begins with counting the weight of the game pieces. But it also associates various bonuses or penalties with specific configurations. In other words, it integrates positional knowledge. For example, a knight on one side, a blocked rook, a doubled, backward or isolated pawn will be perceived negatively by the evaluation. If the position rating is good, the motor moves to the next positions. Of course research and evaluation must be dynamically linked because evaluation is by nature static. If it turns out that an initially well-rated position results in a checkmate a little further away, the entire branch will be abandoned. On a good modern engine, such as Rodent IV for example, all of these processes give the illusion of a plan, a thought, a human game. Experience shows that these are not always the moves that humans would have played. Humans can play a fairly mediocre move and win because the opponent has not perceived the weakness of the position. The engine always assumes that the opponent's response will be optimal. But overall, the engine game will resemble a human game... if it is good! See : http://echecs-et-informatique.franceserv.com/glossaire-technique.html#signet_recherche

It's in French. Use Google Translate if necessary.

3
  • I'm talking about an "algorithmic" engine here. Maia is certainly the most human-looking AI. But it is a “neural” type of intelligence. Commented Apr 28 at 17:45
  • 1
    The premise of this answer is wrong, because traditional alpha-beta search and even neural evaluations are much different than how humans evaluate. Humans are much inferior at brute force calculations and have to rely on Shannon's chessprogramming.org/Type_B_Strategy
    – qwr
    Commented Apr 28 at 19:06
  • To say that "The premise of this answer is wrong, because traditional alpha-beta search and even neural evaluations are much different than how humans evaluate." is a tautology in my opinion. Because by very definition, an AI is non-human. We can't ask her to be human! Only to produce a game that appears to be. Commented Apr 29 at 8:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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