3

Chess is a deterministic game. Hence in principle we can have a perfect strategy for chess, a strategy that prevents a player losing a single game (given that no mistake has been made). All chess heuristics thus results from the single fact that the search space of possible moves of a chess game is incredibly large, finite but unimaginably large. Beyond any scope of human brain’s computational ability to follow through to the end. Even machines don’t possess such humungous processing abilities as of now. How do you proceed from this point? How do we get around our limited computational abilities? What aims do we start with? It seems looking ahead in the search space even to a little extent is something we can’t do away with. Every chess strategy ever is basically a smart way to replace a brute force search with something else. My question is how do you replace searching without actually searching? Do we have to rely a lot on previously tried and tested variations and their results? Then becoming a good chess player becomes a problem of gathering plus memorizing data and doing near perfect five-six moves long calculation, am I right?

I guess what I’m asking here is that can we algorithmize the problem of becoming a good human chess player to some extent? What are an aspiring chess player’s challenges? What fundamental actions do those challenges finally boil down to?

3
  • "I guess what I’m asking here is that can we algorithmize the problem of becoming a good human chess player to some extent?" seems like a good AI question. This is more a classic AI question than the deep-learning approach of Alpha Zero (which side-steps the problem of modeling human intelligence in favor of an almost unbeatable but alien intelligence). Sep 5 at 12:42
  • @JohnColeman: AlphaZero may be alien to a degree, but it still has a common reference point with human heuristics. For example, it knows perfectly well that White shouldn't begin the game with 1. e4 [any] 2. Ke2. It knows to protect the king, to put the bishops on open diagonals, etc.
    – Kevin
    Sep 5 at 20:57
  • @Kevin True, but in many ways your comment is an example of us using our human intelligence to try to understand the actions of an alien intelligence. Sep 6 at 10:26
5

I don't think the fact that chess is deterministic means anything in practice for training (it does for playing of course). Compare for instance with poker, a non-deterministic game where it's still possible to develop a long-term winning strategy. Poker training is not that different from chess training. We must understand however that the word "strategy" is used in chess differently than in game theory. Learning chess strategy means learning about planning and positional concepts, not learning a "procedure" to make moves.

In fact, the irony here is that high-level poker players rely on memorizing "game theory optimal" strategies much more than high-level chess players (who only do so for theoretical endgames)

Even machines don’t possess such humungous processing abilities as of now. How do you proceed from this point? How do we get around our limited computational abilities?

For the most part, we don't care. We understand that this limitation exists in our opponent as well as in ourselves, so we try to play the best chess we can within our possibilities.

I guess what I’m asking here is that can we algorithmize the problem of becoming a good human chess player to some extent?

There have been attempts but mostly unsuccessful ones. We can't rely on the "memory+6-move-search" strategy because humans are quite bad at those two things. No substitute has been found for "tacit" knowledge or human intuition.

2

Warning: My answer will be very subjective.

Methinks that your maybe a bit long-winded paragraph essentially boils down to the question: positional or tactical play? And here I daresay that 99% of the 3500 ELO of nowadays computer are due to brute calculating force. Only 1% is due to improved heuristics of assessing positional factors that don't need calculating. It would be an interesting experiment to program an engine which calculates to depth 100, with just a minimal evaluating function (say, material and king).

Also note that this is completely irrelevant for a human chess player - we simply can't calculate to a depth where pure calculation saves us. We can only play positionally. Case in point: I had FM strength - and I calculated almost nothing in my whole chess life. At least not conscious.

Thus: No, a human player can't play chess by algorithm, if we interpret "algorithm" as "something you directly can code into a program". (Obviously a human uses an algorithm too, we just can't grasp it in high-level if-then terms, as in the case of Alpha Zero.)

Community, feel free to protest.

1
  • What does positional play mean to you? Also algorithm is not only meant for programming or something only related to that I simply meant if there's any simple step-by-step procedure to follow before making each chess move?
    – S_Mitter
    Sep 5 at 8:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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