Deep Blue vs. Kasparov 10th Anniversary

qeorqe writes "For the tenth anniversary of Deep Blue's victory over the world chess champion Garry Kasparov, Wired has an interview with Deep Blue developer Murray Cambell. The discuss the power of the now-aging supercomputer (equivalent to just one Cell processor), and the nonexistent future of PC vs. Human chess contests. 'It's almost the end of the story for chess in the sense that matches between chess machines and grand masters are becoming less interesting because it's so difficult for the human grand masters to compete successfully. They're even taking relatively dramatic steps like giving handicaps to computers, making them play the game with a pawn less or playing the game with less time. We're past the stage where there's a debate about who's better -- machines or grand masters -- and we're just looking for interesting ways to make the competition fairer.'"

the problem is chess

[St. Arbirix]

and we're just looking for interesting ways to make the competition fairer.

Make them play go.

Re:It was inevitable

[j0nb0y]

Humans are really bad at being random, so a computer could stomp a human at Rock/Paper/Scissors, provided that there were enough rounds.

Re:Actually no

[QuantumG]

A lot of intelligent pruning techniques can be applied in chess.

And that's what is worth studying (from an AI perspective).

Re:Scientific American in 1992

[daveewart]

Almost all chess programs now have an "opening library" of opening move strategies, so it's not that far to extend that library to 10-15-20-50-100 moves...

Actually it *is* a big deal to extend it: remember also that opening libraries are not necessarily *perfect* moves, they're just commonly-played and probably-OK moves. The opening book represents an incredibly small subset of all legal moves in the opening.

If one assumes approximately 50 legal moves in any position during the opening (close enough for me to make the following point), an opening book to cover all possible first 10 move pairs is going to need to contain 50^20 entries. This is *huge*. But, to include *50* move pairs, one needs 50^50 entries, which is 50^30 times *larger* than 'huge'. It *is* a big deal to try to store. I suspect you're gonna start approaching "atoms in the known universe" analogies pretty quickly...