Do you think a computer will ever be good enough to beat human players at Scrabble?
For some reason Matt L's latest travelogue (wherein he plays in a Scrabble tournament) reminded me of this, something I'd thought about over Christmas break.
Of course, I assume computers can very easily beat humans at Scrabble already, if only because a computer could very easily find the best possible single play for any board layout and rack. (Also a computer would be 100% accurate, if so programmed, with its decision whether to challenge a word. I assume it wouldn't even need ever to make illegal plays, since it could demolish a human on legal plays alone.)
This is assuming that the Greedy Algorithm is the best way to play Scrabble, though based on player comments in WordFreak (especially the experts who like a wide-open board), I expect that the point-maximizing play is the best play for a given rack/board situation 99 times out of 100.
The reason I ask this: Assuming such a Scrabble program exists, nobody seems to be concerned about it. Computers have long since been able to beat humans at checkers. And yet, the idea that a computer could consistently beat humans at chess makes a few people really paranoid and leads to all sorts of unwarranted apocalyptic jumping to conclusions about artificial intelligence.
Why is this? Is chess that exalted a game?
(On the flip side, I presume even the most probabilistically flawless computer would get destroyed by even intermediate level Hold Em poker players, and that this will continue to be true until, say, we can analyze an image or air sample with enough sophistication to detect "tells" among human players.)
Posted by Matt Bruce at January 10, 2005 05:08 PMA program called Maven exists that can beat any human opponent provided that Maven receives the correct tiles. While programmed by a human, Maven will always play what it calculates as a move that will keep its lead in place. Maven has the entire dictionary memorizes and even accounts for the endgame. Since the correct tiles are usually necessary to win at Scrabble, Maven, nor any other computer program will be able to beat an expert 100% of the time.
Posted by: matt levine at January 10, 2005 06:47 PMActually, there are already poker bots that can play limit Hold 'em better than most players, but I don't know of any that can play no limit. Poki is one such bot.
Posted by: David at January 10, 2005 08:58 PMKasparov came to Harvard one year-- slightly before your time? early nineties-- and gave a speech in which he made the following points:
1) A computer will never, ever beat the best humans at chess. Because chess is art, chess is creativity, chess incorporates the essence of humanity.
2) If a computer ever did beat the human world chess champion, we should all wrap up our affairs in a dignified fashion, take a lot of painkillers and vodka, and collectively throw ourselves off a bridge. Reason: see #1.
3) However staggering unlikely it may be that a computer would beat the world chess champion, it is even more unlikely that a woman would beat the world chess champion. Reason: see #1.
Posted by: Richard Mason at January 10, 2005 11:37 PMRe Richard: I think it was 1991 (I remember this as a longtime Chess Life subscriber). I got there fall '92.
Re Matt L.: For a game between experts, if you were told which player got the (Blank, Blank, S, S, S, S, Z, Q, J, X), but knew nothing of the players' relative strengths, with what probability could you predict the gamewinner from the tile distribution? (What more tiles would you want to know? I presume that if you knew *all* the tiles, you could figure out from quantity who'd gotten more bingoes.)
What I'm really asking: If the distribution of a particular subset of tiles predicts who "should" win, then in expert play how often will someone win a game despite the unfavorable tile distro?
Posted by: me at January 11, 2005 10:11 AM