AI Contest Type Thing

Pages: 123
What about just making AIs that do not make illegal moves? You need only make sure that:

1) You're moving within the area of the board
2) You're not moving over a fence
3) If placing a fence, you:
a) Have a fence available
b) It's placed on the board and not in collision with another fence.
4) If moving over the other player, you're doing so correctly. The most complex situation is when a fence is behind the player you are jumping over.*

This doesn't seem like too many conditions to check? The test program could just reject invalid inputs and disqualify the program / make the program resign? Is this too harsh?

*I have this game at home and have played quite a bit (although I've never tried to write an AI for it before). I can explain any rule ambiguities or anything if anyone has any.

Edit: My point here is kind of, it's not a very good AI if it makes illegal moves.
Last edited on
The best AI will just pause the game then it can never lose.
The best AI will just pause the game then it can never lose.

God I remember doing this back in WC3... Would just spam pause the game and people would get so frustrated they would just leave.
Clearly some rules need to be set:

- Failure to move in a certain time --> Automatic resignation
- Failure to play a legal move --> Automatic resignation

I thought today I can make a little ascii board representation that checks move legality and keep track of pawn and fence numbers and positions. AIs could be written in a header file and then the AIs included with their respective header files and left to do battle until a winner is found. It would seem fair to match each AI with another twice, once going first and once going second. I figure I could release the code here so people can error check it and see what conditions they need to give to the program to make a move. Thoughts?
closed account (N36fSL3A)
So if your program has a bug it gets disqualified? Make limits.

Couldn't you just make your ai go into an endless loop?
Endless loop would lose on time.
closed account (G30oGNh0)
I'm guessing a chess engine would be too complicated? I know I wouldn't know where to start but it would be nice to witness.
closed account (N36fSL3A)
But I don't know how to play chess.
^ Lol.
Chess rules are easy.
It's the strategies, those are very hard (besides the standard "rush and try to eat the king").
closed account (N36fSL3A)
When I printed out a page on how to play chess a few years ago it was like 54+ pages of stuff.
Chess just seems a bit boring to make an engine for, since there are so many thousands of chess engines already out there. I suggested Quiridor because it's a game with rich strategy and hardly an AIs.

Edit: @Lumpkin wtf? I play tournament chess and I'm sure if I requested a copy of tournament rules it would be at most 3 pages long and most of that would be clarifying who plays who in each round and tie-breaks. The basic rules would probably be just over a page.
Last edited on
closed account (G30oGNh0)
Don't you feel like those thousands of chess engines would top anything made from people on this website?

I call bonus points if anyone here makes one, takes on Deep Blue, and lasts more than 10/15 moves.

Which ever way this turns out I'm up for the learning experience, I'm not very good at C++ but I'll take in as much knowledge as I can handle.
It's not beyond reason to defeat Deep Blue* with a chess engine. There has been so much knowledge gained in chess theory since the mid 90s that I would expect many programs of today would defeat it**.

*If it still existed that is.

**If they were matched up against the 1997 version and not some new improved IBM chess computer.

Edit:
Don't you feel like those thousands of chess engines would top anything made from people on this website?


This is one of the reasons I elected not to mention chess. It seems to be far more interesting to try making AI play a game where we might actually progress AI for the game, rather than walking the through the well known chess AI theory.
Last edited on
closed account (G30oGNh0)
Many programs of today could defeat it yes I am sure, I wasn't implying otherwise, but engines made by you or me is what I were getting at.

Sorry for not being clear.
closed account (Dy7SLyTq)
i dont know man... there are some pretty smart people on this forum. so smart that if im sure they chose to, they could all get together and design a compiler for c++ that would rival gnu and contain all features that they wanted
EssGeEich wrote:
Chess rules are easy.
It's the strategies, those are very hard (besides the standard "rush and try to eat the king").


For some reason this made me think of this:

http://xkcd.com/1112/
That was very accurate.
Anyways, which game are we going to use, in a definitive answer?
I vote for Quiridor. :p But I'll have a go at whatever game AI.
If you guys are going for board games, it makes sense to go for games which aren't solved.
And if you're not up to chess, what about arimaa :3?

http://en.wikipedia.org/wiki/Arimaa
How many games are actually 'solved'? AFAIK Connect 4 is the only major game solved and draughts is weakly solved. Stuff like chess or backgammon probably need a decent quantum algorithm to actually solve in any reasonable time.

Any more game suggestions we might do this for?
Pages: 123