Chess - 2070 CPUs vs 1 GM

jvarsoke writes "ChessBrain.net broke the world's record for 'largest number of distributed computers used to play a single game' by holding a chess match between Danish GM Peter Heine Nielsen and the equivalent of SETI@home (which similarly, has some people looking for a Mate). 2070 CPU's from 56 countries aided Black by running the chess program Beowulf, including a couple of University clusters. Their supernode ran Linux, and MySQL. The game was relayed by FICS. Results can be viewed here(1) and here(2)."

Comments

Re:Understanding vs. Processing

[vontrotsky]

We're getting closer and closer to the days when humans won't be able to compete with computer's at chess. Even so I don't think this is such a big deal. We haven't be able to compete with computers at arithmetic for half a century and this doesn't bother anyone.

Losing to computers in chess will be like losing to calculators in a addition match. People and computers aren't really in competition. They do very different things.

Results

[Stalyx]

"The game lasted several hours before resulting in a draw. Chess Grandmaster Peter Nielsen commented that he had set several traps for ChessBrain which computers normally fall for... but was surprised that ChessBrain refused them! "

So what does this tell us? Nothing really, however it would be interesting how the computer will perform in a 5 match series.

Although I still think the GM would win handily.

Re:What's the point?

[vidarh]

Any turn based board game are "really simple search-and-prune" systems. The problem is how to minimize the time taken by the search, and how to decide what a to prune, and how to decide which move to take.

Brute forcing a chess game tree based on basic alpha beta minimax for instance is no way to play well against an experienced human player - first of all you won't get many moves ahead, and a good player that know how the computer work can easily set up a trap that will make the board look good X moves ahead, to make the computer to do stupid moves they can't easily reverse later.

Second you face the problem of definining and weighting what a "good position" is. What is a good position depends on the strategy of the opponent.

Most modern chess programs will augment the basic search and prune with a lot of heuristics to guide the search and weighting of choices, exactly for that reason. They also often contain massive databases of games, sequences of moves etc., to hunt for known strategies that humans might try to recycle against it.

Chess isn't "simple". Chess is a game where it's easy to beat beginners, possible to beat intermediate players on modest hardware, and possible to face grand masters if you have lots of time and access to millions of dollars worth of hardware, and you can still expect to be surprised every now and again.

It makes it interesting, because you have a good foundation to research algorithm improvements on, and because a good algorithm will be more and more useful as hardware costs come down, but it certainly doesn't invalidate the need for better algorithms.

It's also interesting because better algorithms might help us appreciate how humans approach the problem, and as such benefit AI research.