Comments on: Checkers Solved?

By: Johan_C

Johan_C — Fri, 27 Jul 2007 07:33:28 +0000

Walt said:
“What do you mean, Johan? Is there a problem with it that we need to fix?”

well, I did (unvoluntarily) manage to submit a post without typing the two words required. Frankly, I hadn’t even noticed that reCAPTCHA gizmo. Not able to replicate this particular feat, so possibly some kind of fluke.

By: Neil

Neil — Thu, 26 Jul 2007 20:10:47 +0000

It’s explained properly in the actual paper. They (claim to) have ‘weakly solved’ checkers, meaning that from the starting position, their program can never lose. They have not ’strongly solved’ checkers, meaning that they can’t guarantee that their program will play optimally if you give it a position to start from. In other words, if you play for a bit before letting Chinook take over, and make some moves that Chinook never would, you can get to a state they have not analysed. But as long as Chinook plays from the start, it doesn’t matter what moves the opponent makes - Chinook will obtain the optimal outcome (that, at least, is what is claimed when they say they’ve weakly solved checkers).

By: Walt

Walt — Thu, 26 Jul 2007 19:01:18 +0000

What do you mean, Johan? Is there a problem with it that we need to fix?

By: TL

TL — Wed, 25 Jul 2007 19:35:12 +0000

“So if Chinook plays against Chinook both can’t loose?”

Yes, they said a perfect game by both sides would lead to a draw. The claim’s an overreach though as the article shows.

By: Johan_C

Johan_C — Wed, 25 Jul 2007 10:25:37 +0000

BTW your reCAPTCHA doesn’t work either

By: Johan_C

Johan_C — Wed, 25 Jul 2007 10:24:18 +0000

“…have shown that Chinook cannot ever lose at checkers …”

So if Chinook plays against Chinook both can’t loose ?

By: Isabel

Isabel — Tue, 24 Jul 2007 14:48:33 +0000

(My previous comment is broken — I didn’t close my “a” tag — but just click on the now-too-long link to see what I’m talking about.

By: Isabel

Isabel — Tue, 24 Jul 2007 14:47:45 +0000

Actually, looking at the applet at , which displays the game tree, certain positions are called “unresolved”. These appear to be only positions in which white has the move, though. So my guess is that this program can play perfectly if playing black, but not if playing white.

By: Isabel

Isabel — Tue, 24 Jul 2007 14:45:41 +0000

The impression I got was that they only consider positions that actually occur in games. It doesn’t seem immediately obvious to me that that cuts down on the number of possible positions by a lot, but I guess it does.

As for whether go can be done by brute force: I don’t see why not, given enough computing power. But I have no idea whether enough computing power exists. And there are clearly games which are complicated enough that the number of possible game positions is greater than the number of particles in the universe (just take go on a large enough board).

By: icecube

icecube — Tue, 24 Jul 2007 13:56:24 +0000

I think go has proven rather resistant to AI players (given that computers can’t yet beat masters, to my knowledge)