Loomis wrote:
artfizz wrote:
Chess learning computers are quite interesting - provided that they have started out from not even knowing how the pieces move - in the same way that people have to. Maybe you like to torture new players, but I always teach them how the pieces move and the object of the game. Surely computers (even "learning" ones) deserve the same courtesy. In fact, when new players show interest in doing so, I'm happy to give them a few strategy tips such as ... I could see arguments for and against letting computers have the same benefit of our learned wisdom. ...
There are several different approaches to computer learning. The Neural Net approach is probably the most similar one to the way a baby learns - by trial and error. Rather than burden the computer with any preconceived ideas about 'good' and 'bad' strategies, you could just cut it loose to move any piece anywhere. If it makes an illegal move, it gets penalised. If it makes a legal move, the game proceeds. It will doubtless lose a few games initially, but after a while, it ought to figure out for itself by pattern analysis which moves are legal. After a few million games, it should have worked out which combination of moves help it win games, and which sequences of moves make it lose games.
If you try to help the computer by instructing it, for example, to keep knights away from the edge of the board, that may prevent it from discovering some totally new, alien strategies: "it's OK to have a knight on f7 provided you also have a bishop on e4 and a pawn on g5, and your opponent has ...", say.
I dont think that will happen after all it is people who gave computers instrutions.