úterý 28. října 2008

Partial Chess

Some minutes before, I have meditated about artificial intelligence and the game theory, so to say. I have had an idea about chessplay.

One step was, it would be great to watch a competition.
Not between a human player and a classical chess playing program (which uses huge memory of already played games and the computer's computation speed to overcome the human player); but between such a classical program on one side and a program using some kind of connectionist or emergentist method like a neural network.

Beside the amusement of watching "who wins", and "how the intelligent one learns", this method would be great to train some kind of more human-like opponent. It's always great for chess disciples to have an appropriate opponent. The classical programs are predetermined by the way how their algorithms are designed and they just "think" in completely different ways so when a human player wants to be trained well to play against other human players, he or she needs human opponents to practice. It's still the best way.

Some people however — when a computer is closer than a human to play with — may prefer to play with it, e. g. when they cannot fall asleep, when they travel by plane or when they just don't feel "sociable" at that moment… or when they don't have access to the Internet for some reason at that moment and so cannot play chess on line.

So for some chess apprentices it would be great to have a computer opponent that would be still closer to the human way of thinking. They could even teach it (or him or her, depending of the sight of view) how to circumvene their own tricks that they are settled to use so that they would be pushed to change their strategy as their "program" opponent grows in experince with themselves ☺

Learning chessplay is practically a never ending process for a human. For the computer, using standard deductive and enumerative approach, it's all solved already, though theoretically. It's only a matter of computation now. But an inductive, intelligent program stands in the same situation as a human being: the process nearly never ends for it. Let's experiment with that. It's going to be a lot of fun ☺ It's also heavily probable that somebody has (or many have) already tried that.

The second step is to narrow and simplify the chess down. It's mere a matter of the game theory now. Again, I think someone must have been already engaged by that idea. Maybe I discover America again and invent the wheel again. But therefore I write it here, to query if someone knows such simplified chess to be a formalized type of game or a teaching approach.

There's already a response I know of: For the purpose to study chess, the specialists build up model situations that emerge at the end of a game. The specialists study the possibilities and combinations of moves that the players can perform to win. Such models already constitute some form of such "simplified chess" I'm in serach of.

Just imagine that the chess game doesn't start with all the figures on the field but only with two knights, each player having one.
Let's try also to imagine a chess field not with that classical 8×8 measure but only a 4×4 square. Wouldn't that be just a childsplay to learn the outcoming rules? Yes, and that's how probably children could be better taught to play chess.

The level of knowledge and experience of the opponent could be a satisfying simplification… but there's another reason: a basic study of the elements of chessplay and chess strategy.

What would be this all good for? Just to study the game of chess and to learn chess basics more instructively and smoothly. This would help especially me to learn chess better ☺, me being a chaotic evergreenhorn ☺ And … to return to the first step … such a type of chess game would be suitable for the first steps with the intelligent machine-learning chess programs to go. If I had the time, I wrote such a program. But I don't have it at the moment and have other plans — so whoever is interested herein, go and code it and become inspired ☺