My attempt at a partial-knowledge AI: Dark Chess 960

[Dave]

I thought it would be cool to try to write an AI that only has partial knowledge of its environment. A lot of effort has gone in to making chess playing computers that are great at chess and some chess variants, but little effort has gone into variants of chess that have incomplete information.

So, I decided it'd be cool to write an AI for Dark Chess. To spice it up a little, I combined it with chess 960. The result, a half-decent AI that plays without cheating at all, and can do okay (at least it can beat me sometimes when I play fairly casually).

Here's a page I put up for it: http://www.silvertreerpg.org/dark/. (Includes source code and Windows binaries).

Any comments/feedback on it are welcome.

David

[mog]

The page has the following meta-tag, which makes reading it a bit challenging:
<meta http-equiv="refresh" content="5; URL=http://">

[Dave]

The page has the following meta-tag, which makes reading it a bit challenging:
<meta http-equiv="refresh" content="5; URL=http://">

Ahh yes, sorry; fixed.

[baruk]

I just beat it. The AI gave me a head start of 2 knights and a queen, before letting me queen a pawn. Think it needs to focus more on freeing pawn moves rather than just dancing its knights around the board. A game replay feature would be nice as well.

[Lorbi]

There is things it could know but ignores.
It moves its knight on a square where my knight was able to capture it.
well this is bad luck and may happen but it should see than that my knight captures it's knight ....
well i did so and next turn it moves the queen on a square where it should know i can capture it with aforesaid knight

i hope i made clear the point : )
besides that i like it : )

[Elvish_Pillager]

It's spent 5 minutes so far and from the look of the progress bar it's going to take another 10 or 15 minutes on this turn.

EDIT: update: That turn only took about another 4 minutes, but it took another four minutes for the first bit of red to even appear in the progress bar on the next. I reckon it would take over two hours to finish; instead, I declare victory and kill the process, to go do something else

(We're maybe a couple dozen moves into the game... not sure how many ; there ought to be a way to view the game history. Does it try to calculate every possible current state given what it knows?)

[Dave]

Does it try to calculate every possible current state given what it knows?)

Yes, exactly. This is what makes it take so long in some positions: the set of possible states might be huge.