I expect computer chess programmers to find these rules controversial or onerous, or to cry "not fair". But recent results have demonstrated that any traditional "fair" match between a human and a computer to be pretty much a certain victory for the latter. So we need a new computer chess challenge, one which might seem as insurmountable now as the Fredkin prize seemed when it was proposed decades ago, but eventually won by Deep Blue against Kasparov.
The CAPITALIZED WORDS are "free variables" whose values I chose arbitrarily. The match organizer may adjust the values to something more convenient or appropriate.
1. The match shall consist of an unlimited number of games in a SEVEN-day period.
2. The human may not receive any computer assistance during the entire duration of the match, including between games.
3. The human needs to win only one game to win the match.
4. The human may abandon a game at any point with no penalty.
5. The computer must remain unmodified for the duration of the match; that is, the computer must not receive any human assistance during the match.
6. The computer must play deterministically; that is, it must not deviate from any move played in a previous game of the match. In order to keep things moving along, the computer should replay previously played moves immediately.
7. The computer is limited to THIRTY SECONDS per move. There is no accumulation of time from previous moves. (Rationale: A computer may virtually get more time by using more processors, so THIRTY SECONDS should be more enough. A constant time limit per move also avoids tricks like the human trying to induce the computer into time trouble, which is not an interesting way to demonstrate human superiority at chess.)
8. The computer may not ponder. (Rationale: Every move, the human should should face an opponent of constant strength. This avoids the human trying to make the computer play weaker by steering to positions it has not pondered.)
9. The human faces no time control other than the length of the match.
10. The human may propose any position as the starting position of a game, with white to move. The computer shall have ONE MINUTE to decide which color it wishes to play, and if choosing white, to give its first move.
(Rationale: computers are often called upon to analyze positions they did not play into. These days, computers are thought to play certain positions worse than humans, which makes muddy the question of who is ultimately the best chess player. Human-computer games in the past have typically seen the human making suboptimal moves in order to steer the computer to a position it will play poorly. We allow the human to "fast-forward" directly to such positions. If the computer can outplay the human from any position of the human's choosing, then it will conclusively lay to rest who is the better player.
No doubt the arbitrary starting position will result in the human proposing many unusual "chess problem" positions or positions with deep home preparation. The idea is for any problem that has a winning strategy simple enough for a human to memorize, the computer should be able to rediscover it within ONE MINUTE.)
11. If the computer believes a starting position to be equal, it must defer the choice of side to the human.
12. If a game ends in a draw, the human may return to any previous position of the game and switch sides deviating from what the computer played, thus attempting to win from a position that the computer could only manage a draw. (Rationale: because the human must win a game to win the match, the computer might seek to always play for a draw, making suboptimal moves. This rule penalizes such a strategy.)
13. Embargo. This rule is awkward, and a different rule may be substituted for the same effect if one can be found:
The human should be discouraged from having the computer play against itself. Because the human may switch colors and deviate after a draw as explained above, and because the computer may not deviate from any previous game, it may be possible to play a series of games following a draw to achieve the effect of the computer playing against itself. Maybe this can be achieved by the same position achieved two different ways through transposition. If a game reaches a position with the computer to move that the computer has previously never faced, but the same position has been reached in a previous game with the human to move, the computer's move is embargoed (not divulged) for ONE HOUR. The computation time remains THIRTY SECONDS.
14. If the computer discovers it is in a losing position (forced mate), it must resign the game (and consequently the match). (Rationale: Lessen the importance of the human memorizing long endgame sequences. This rule also stresses the computer's mate-detection code which, if incorrect, will cause it to resign non-lost positions.)
15. The human is permitted a separate analysis board.
16. The human player is actually a TWO-human player team who are free to discuss with each other. (Rationale: Teams of people are often much more capable than a single person alone, and the resulting creation, for example, a chess win over a computer, is clearly still a product of human ingenuity.)
No comments :
Post a Comment