Author: STEPHEN A. BOAK
Date: 18:10:00 12/23/98
Go up one level in this thread
When computers play other computers in sets of specific openings, the results must be treated with care. Some thoughts about the subject: 1. If the reason a computer doesn't handle a certain opening very well is that the computer doesn't understand that opening, this seems to imply that the computer doesn't know the relatively long term *plans* and planning *ideas* that lead to relatively better results in such openings. Of course, along with such *ideas* are immediate tactical motifs. This difficulty is inherent in programs (virtually all of them I think) that don't understand the making of plans but merely calculate lines and evaluate positions to a relatively limited depth. It seems that a well planned opening will steer the course of a game to later positions (perhaps late opening, middlegame, or even far off end game positions) that are playable (equal), if not with an advantage (outplayed the opponent); however, the desktop computer, limited by processor power and available time, can not see deeply enough to properly evaluate the *final* course of the game, since it only sees positions that are within its horizon and cannot plan well for subsequent positions that may arise. For example, in a Gruenfeld Exchange Var, can a computer understand the power of Black's 2 to 1 or 2 to 0 or 1 to 0 pawn majority on the queen side *in a potential future endgame* that lies beyond the horizon of current tactical skirmishes in the center or on the king side? Would the computer continually (as Black) try to maintain that long term advantage until it could pay dividends, or would it (as White) seek to nullify that advantage perhaps by ignoring it through vigorous center and kingside play 2. Two computers, similarly lacking in understanding, that play each other i a match over set openings may result in one of them being the victor by a large margin. Is that result because the opening, or the middlegame, or the endgame was played better than the opponent? Or was it some combination of all three? Often the mistake in a middlegame or endgame bears no relationship to how well the opening was handled, thus game results are not totally indicative of how well the opening was played. 3. If two computers, equally lacking in understanding, generate lopsided results after a match, does this mean the computer with the better results understands that opening well? Or would both computers perhaps have played the opening poorly (by human understanding) but one was superior in the resulting middlegame or endgame play? These considerations require that strong human players evaluate the strength of the opening positions achieved as a result of how well the computers *understand/play* the opening, and discount the effects of later midgame play or endgame play. The computers may show new tactical motifs, not yet discovered by the strong human players, and pave the way to a better understanding of an opening. Finding theoretical opening novelties might then cause the human evaluator to give the computer high marks for *understanding* such an opening. The opposing computer would receive high marks if it countered such *improvements* with good methods to neutralize or prove the variations unsound.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.