1
$\begingroup$

I've been researching the following topic. Or rather, I would like to but I can't find anything because I'm not sure what to look for.

I am interested weather there are some concepts or models that explain how humans (or cognitive systems in general) trade off remembering what to do in a particular situation versus thinking about the board state and devising a new solution. Consider the following example:

A person plays chess. On their turn they must decide what to do. They could act according to one of the following extreme strategies:

  1. Remembering all possible board configurations and how the game ended (we'll call that Memory)
  2. Use the known set of rules to simulate the game in their mind and choose the optimal path (Optimization)

I'd say the best choice is a trade off between these two. But how? When to choose one of the strategies?

Is there a research field that is working on these kind of questions? What would I look for?

$\endgroup$
2
$\begingroup$

The The Oxford Companion to Chess has entries on only 700 named openings, and lists another 1327 opening variations in the index, and I wouldn't be surprised if someone out there had them all memorized. For an algorithm, however, storing that number of openings is trivial, and Chess algorithms traditionally made use of high-quality "game books" which are move-by-move records of chess games.

The problem with intractable models, like chess is that they are unsolvable, so brute-force is not an option, and any decision on a move, whether "memory" or "optimality" based, at least in the context of current advances, are probabilistic. (The exception is endgames where the the problem becomes tractable, and perfect play can be achieved.)

Essentially, the memory of a board state is only a small piece of the puzzle because the unsolvability of the game means that particular board state from a recorded game, where that state led to a successful outcome, will not guarantee a similar result in a new game, because new, more optimal strategies can always emerge.

Secondly, because of the extreme variation engendered by non-trivial games, it is extremely unlikely any current game would mirror the previous "memorized" game.

It is my understanding that "memory" as you define has been utilized in Chess engines, accessing a database of previous games to evaluate a position and derive a good move, but the intractability of Chess in general makes this method sub-optimal, and it has been definitively supplanted by newer methods (machine learning, neural networks).

In the famous Giraffe Chess paper, Matthew Lai states:

The job of the evaluation functions is to assign scores to positions statically (without looking ahead)

But the method is highly nuanced and probabilistic. Nevertheless, I think you'll be interested because the problem presented in the paper is rooted in the human vs. AI play method issue:

...the way computers play chess is very different from how humans play. While both humans and computers search ahead to predict how the game will go on, humans are much more selective in which branches of the game tree to explore. Computers, on the other hand, rely on brute force to explore as many continuations as possible, even ones that will be immediately thrown out by any skilled human... How can a human searching 3-5 positions per second be as strong as a computer searching 200 million positions per second? And is it possible to build even stronger chess computers than what we have today, by making them more computationally efficient? Those are the questions this project investigates.

There is no simple answer to the trade-off question due to the complexity of the problem. But this is an area where tremendous advances are being made.

I'd strongly recommend reading Giraffe Chess (September, 2015) as a primer. It produced very exciting results, synopsized in the MIT Tech Review article "Deep Learning Machine Teaches Itself Chess in 72 Hours, Plays at International Master Level". (Giraffe's creator currently works as a Research Engineer in Machine Learning at DeepMind.)

$\endgroup$
2
$\begingroup$

If you can remember everything and there's no randomisation in your outcome like chess, there is absolutely no reason not to do that. Anybody who can remember all the possible board configurations in chess, by definition plays perfect chess. A perfect player would never lose.

Unfortunately, most practical problems can't be solved by brute-force, and that includes chess. All chess players learn general chess principles. Nobody could memorize all possible opening variations.

If you can somehow memorize every possibility, why not? But can you?

$\endgroup$
  • $\begingroup$ Of course you can not do that. But you can, let's say, memorize 10 prototypic configurations. And then starting from them you can go on and optimize it. The question is now: how much should you memorize optimally. I'm just looking for a field that's working on these kind of questions. $\endgroup$ – hh32 Oct 25 '17 at 12:27

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.