Speaking of the mario competition, is there a way to watch videos of submissions?

Here's one: http://www.youtube.com/watch?v=DlkMs4ZHHr8

Impressive stuff. Others in the video responses below.

Robin's also released the source to his bot, if you want to run it locally: http://www.doc.ic.ac.uk/~rb1006/projects:marioai

All you need to run it is the latest competition code: http://code.google.com/p/marioai/

You can run Team Proggit's bot locally pretty easily (one click if java is working right): http://rictic.com/marioai/

The Mario competition was held at the CIG 2009 conference. There has been talk of holding a StarCraft competition at CIG 2010. http://www.ieee-cig.org/

I wasn't sure. I assumed that most people on here check both websites therefore it was redundent. However, speaking from experience, I don't see the point in commenting on a Slashdot thread, so I guess there's no reason for this to be a taboo.

Would love to compete, but the required investment of time seems potentially enourmous.

Well, you know, it would probably be an enormous investment to get a general-purpose bot that could win the tournament, but you could write a simple bot with simple strategies pretty quickly (assuming that the API is sane.) For example, you could write a bot that always executes the same kind of rush right away, or one that just tries to expand absolutely everywhere on the map as fast as it can until it has X supply of workers. That's both fun and a great addition to the competition since it diversifies the ecosystem that the serious bots have to deal with.

In programming AI players for various games, I've found that stupid algorithms with excellently-tuned parameters beat smart algorithms with decently-tuned parameters.

Any examples or intuition about why?

I don't know, but I could imagine overfitting could have something to do with it. Also simpler models are probably more robust.

I agree. I doubt I'll be able to invest the time to truly compete. But I definitely plan on creating something on the simpler end for fun. And if I can get a bot going that doesn't crash, why not submit it? And given a year until the deadline I see no drawbacks in playing with this casually.

Seeing as there already seem to be a lot of people involved in Starcraft AI, writing a WeeWar bot seems more rewarding to me for the time being. I don't think there are any good WeeWar bots yet. Besides, it might be used by a lot of people, for example newbies who want to train.

Just for laughs I guess I could try genetic programming. Something that evolves would interest me the most, anyway.

I'm fully with you on that one. I think the most interesting part is going to be creating a usable state space and reactions.

Any higher language bindings to the AI API available? Don't want to mess with C++... The Lua version seems to be incomplete, judging from a quick glance.

There are some instructions for using an arbitrary language to control the bot over the network on the site.

There are different tournaments for participants that want to focus on only a subset of the game. For example, you could hard code all locations for the tournament 3 contest and focus on macro.