AI Research Project

[gametaku]

As part of my course on AI I need to do a research project, and well I'm not looking to get published (though the professor did mention it) I am looking to use it as learning experience for a project I would want to try and get published. However I can't seem to come up with any somewhat unique ideas, specifically ones that deal with the topics covered in class

• Search
• Logic
• Uncertainty
• Learning

I took an AI course as undergraduate, and worked on a program that plays the game Go* so I'm looking to do something more advanced** then that. I've though of making an AI that plays Super Mario Brothers but that's been done it seems.

*The game board was smaller then normal as to reduce the number of states, and was played against other students AI's.
**The game it self is fairly simple, and all of the programs used a fairly simple strategy to determine the next move.

[oops]

Are you limited to specific problems in the CS domain, or could you work with people in another field? I know a couple people at my university that are working on AI applications for biotechnology. I'm not precisley sure what their project does, but I know it's used to model protiens, and the AI helps fill in gaps in the amino acids.

[gametaku]

Are you limited to specific problems in the CS domain, or could you work with people in another field? I know a couple people at my university that are working on AI applications for biotechnology. I'm not precisley sure what their project does, but I know it's used to model protiens, and the AI helps fill in gaps in the amino acids.

No I don't see any reason to limit my self to the CS domain,
however I can't work work with someone in another field.

[Meem1029]

This is probably going to be an impossible challenge, but you could try applying AI to problems using Foldit. It is a protein folding program that crowdsources the job by making it a game. It might be possible to write an AI for, but it might also be insanely difficult.

[somebody already took it]

You could make a bot that plays hateris (not to be confused with hatris and wins (or rather doesn't lose). Perhaps, there is a maximum number of points that is is possible to get before you are destined to loose, you could do an exhaustive search to find out what that number is. And if there is a point cap, what if you change the size of the arena or blocks available? When does it become possible for the AI to go on playing indefinitely?

[gametaku]

This is probably going to be an impossible challenge, but you could try applying AI to problems using Foldit. It is a protein folding program that crowdsources the job by making it a game. It might be possible to write an AI for, but it might also be insanely difficult.

Wasn't Foldit created because of how poorly computers do protein folding compared to humans?

You could make a bot that plays hateris (not to be confused with hatris and wins (or rather doesn't lose). Perhaps, there is a maximum number of points that is is possible to get before you are destined to loose, you could do an exhaustive search to find out what that number is. And if there is a point cap, what if you change the size of the arena or blocks available? When does it become possible for the AI to go on playing indefinitely?

I get that Hatris is a hard version of Tetris that always gives a bad piece but unless the AI determines the piece the game is going to give I'm not sure how an AI for it would be different then any other Tetris game. Though that does make me think of doing some sort of competitive Tetris game that players/bots picks the opponents next piece.

[WarDaft]

I get that Hatris is a hard version of Tetris that always gives a bad piece but unless the AI determines the piece the game is going to give I'm not sure how an AI for it would be different then any other Tetris game. Though that does make me think of doing some sort of competitive Tetris game that players/bots picks the opponents next piece.

Hateris is deterministic. It doesn't just give a bad piece, it gives the worst piece (IE, the one whose highest point must be at least as high as any other piece no matter where it is placed) That means you can actually control the piece you're going to get next. People have gotten scores of up to 30 on it. A typical tetris AI, not taking into account the fact that where it puts the piece is going to determine future pieces, is definitely not optimal.

[gametaku]

Hateris is deterministic. It doesn't just give a bad piece, it gives the worst piece (IE, the one whose highest point must be at least as high as any other piece no matter where it is placed) That means you can actually control the piece you're going to get next. People have gotten scores of up to 30 on it. A typical tetris AI, not taking into account the fact that where it puts the piece is going to determine future pieces, is definitely not optimal.

I know that Hateris is deterministic that's why I was was wondering if this concept should be used by the AI playing Hatris or if it should just deal with current piece. I do like the idea however it does seemed to be based almost entirely on the search part of AI, and does not have include uncertainty or learning.

[WarDaft]

Depending on the size of the search space, modeling it as uncertain rather than enormous can actually help (such as in the monty-carlo Go AIs,) and if you just don't tell it everything it needs to know about the game, then it can learn too.

[gametaku]

Depending on the size of the search space, modeling it as uncertain rather than enormous can actually help (such as in the monty-carlo Go AIs,) and if you just don't tell it everything it needs to know about the game, then it can learn too.

So I've been thinking about the idea, and think I'm going to divide the project into two parts.

Part 1 will will be deterministic search based, which will be unconstrained by any game time. This will be useful in providing a comparison for part 2.

Part 2 will be constrained by game time, and will use modeling parts as uncertain as well as incorporating learning into the AI.

Tentative part 3: Seeing how well the AI goes against Bastet.