A computer chess contest, though involving humans.
1st step: create a computer chess program tuned to play at a certain Elo rating level. Test that the tuning is correct by playing against humans on an internet chess server. During a test, the program is not allowed to use feedback of its wins and losses to change its level of play. This step could be quite difficult as the humans may learn how to beat the computer, especially if it is programmed to make certain kinds of mistakes to keep its rating lower than full strength.
2nd step: Among programs all at the same rating, seek to minimize computational power to play at that rating. A practical way to measure this is to maximize the number of simultaneous games the program (or multiple instances of the program) can play on a standardized computer while maintaining the specified rating.
No comments :
Post a Comment