Quote: Original post by seanw
Goto your University library and read through some of the latest AI journals. Pay particular attention to the parts in articles which say what further work could be done and what could have been done better. There will be many obscure and abstract article topics, but you should be able to find a few concrete ones. That's how a lot of new research starts, by understanding the current progress in a particular field and trying to improve on it. I think that would make a very good impression on your professors as opposed to picking a random project or basing it on non-peer reviewed games material you find on the internet.

Hey that's a really good idea, in fact, I seem to remember now my teacher suggesting something similar. I'll probably check that out, but I think I'm gonna try the racing AI first, and if the programming part looks like too much to fit into my schedule this semester I should still be able to do quite a bit of research on it. Racing AI may not be the most important thing in research today, but I could also include the AI reaserch they are doing for driving real cars, which has also always interested me.

Quote: Original post by Rixter
Hey that's a really good idea, in fact, I seem to remember now my teacher suggesting something similar. I'll probably check that out, but I think I'm gonna try the racing AI first, and if the programming part looks like too much to fit into my schedule this semester I should still be able to do quite a bit of research on it. Racing AI may not be the most important thing in research today, but I could also include the AI reaserch they are doing for driving real cars, which has also always interested me.

It's fine to do that, I mean that you should research what the current state of racing car AI is. For example, find articles in established AI journals that concern this. References are also a great source of data. If you can find one very relevent paper, go and grab all the papers that one references and you should have plenty to read. For your research and software to be purposeful, you need to be able to place it within established literature. You shouldn't just write a racing car AI for the sake of it, it should have some goal or message to achieve. For example, to try and recreate the research of someone else to verify that it works in siutations they did not try it in, to redo someone's research you didn't think wasn't done very well, to extend someone elses research, to use someones research in a novel way (e.g. like using real car driving AI in a computer game might come under this) etc. Most people are just going to do random things for no real purpose, so you'll get serious brownie points if you take an academic appoarch to this.

We have been doing a crossword-solver for our semestral project.

The goal was to fill 77 randomly generated pattern (sized 5x5 to 77x77) with words from a (fixed) dictionary (~300000 words) either in horizontal or vertical direction, so that overlapping letters match. All this in 1 hour.

The results were horrible: we could solve only the 12 smallest/simplest patterns. But maybe because it was our first ai project...

In one course we had to write an anti-chess program. This was a rewarding system to write as the whole class submitted their programs for competition.

The basic rules of anti-chess are to remove all of *your* pieces first. If there is a move whereby you can take your opponent’s piece you must.

This project is similar to what's needed in chess but the games are typically much faster (meaning your look ahead is closer to the end of the game than real chess). It also provides some different challenges than normal chess.

I would highly recommend not doing Go. Real people don't even play that game well. It would be very difficult to make anything that even appeared to be playing the same game as its opponent. You would get very little reward out of that project.

If you choose a board game, choose something simple and straightforward. That means don't choose chess or go. They both have subtle rules that are easy to mess up. Chess has to deal with awkward moves, or "exceptions". Go has many different rule sets, and creating a program that would demonstrate even a beginner's level of intelligence is not an easy task.

Personally I'd choose amazons. It uses chess queens and is a game of territory, so it has similarities to both chess and go. The rules are simple and straightforward. You could focus on making the program play better instead of trying to write code to handle exceptions and complicated rules. Hex is another game I would consider. Dots and boxes is another I find interesting. I don't know the status of them. Obviously you should check to see the research potential before choosing one (i.e. not much potential for new advances in researching 8x8 checkers A.I. at this point in time, even though it isn't technically "solved").

If you can, make a chatbot!

I've been working on mine(Polly) for 3 days strait (and you can download it via a link the tread in the ai forum), and it can currently do:
question responce with simple learning
keyword search
A semantic net! (thanks to wiki)
A conjugator (thanks to benryves), and
Tagging (scriptlets inside the responces, so something like {addraw(right(inp, 3))} would put in everything but the first three characters of the input.

Its actually quite easy to make... (only 665 lines, including command codes). Lots of googleable stuff. And if you find a bot that you like, you can download it, and disect it to find useful tricks to use on your bot! (very useful for eliza, alice and co)

From,
Nice coder