Winner of 2011 Google AI Challenge Describes Released Code

I had fun playing around with the challenge and spent most of the time poking around with map exploration as implementing min-max once again isn't too much fun. I ended up having the most fun playing around with the idea of an anti-object . The simplicity of anti-objects combined with the low memory and low cpu overhead was very nice. Wrote it up in a little blog for others:

http://benjamin-meyer.blogspot.com/2011/11/using-collaborati...

Any ideas for a type of challenge that wouldn't be reduced to mostly min-max for the ranking, but could explore other concepts?

Now this is what I go to HN for! The most interesting part for me was that he did not focus on some huge strategy. Just individuals and small groups with clear goals and constant revaluation. Keeping in mind that the only thing kept between rounds were missions and those were constantly reevaluated as well. I think this is a vivid, albeit simplified, example of how successful organizations of anything should operate.

EDIT: grammar

I see a few people saying that they didn't have much time (or that they didn't want to spend too much time) for this.

Is that really actually true, or is it some kind of 'modesty'[1]?

When people talk about IQ there's usually a list of people saying something like "IQ is meaningless, it's a flawed concept (but my IQ is 138)". Are there any people saying "I worked really hard at this, I spent a lot of time and effort. I ranked 984th."?

Because I'd love to see the results if those smart people got together and spent some serious time and effort on it.

Lots of popular science projects suffer from a heavy churn rate, which leads to people being introduced to the subject, producing an introductory program, doing a bit of reading, but then moving onto something else. A few people stay, and work on intermediate and advanced level projects, but they suffer from lack of interest and expertise. See, for example, all the millions of artificial life / evolution softwares derived from Martin Gardners Bugs. (See also the death of Fractint, which was excellent software but did not transition to Linux or Windows or modern display programming.)

[1] Not the right word but I'm not sure what fits.

> Otherwise if the ant was just spawned sitting on my hill, the target border tile is determined using some complex calculation considering the ratio of own/enemy ants near that border position, the possible distance to enemy hills, how many ants are already on a mission to a near border tile and how long they do need to reach their desitnation... Oh, I'm kidding, it was random. Completely random. Like in target = area.border.get(turnRandom.nextInt(area.border.size()));

Nice pun targeted at over-thinking of things. If you try to take many things into account without any concrete knowledge of how they influence outcome you'll get random result. So you may just start with random and see if it's good enough.

Unfortunately the link appears to be temporarily dead. Luckily before you read about his bot you should watch it play anyways!

Here are two replays of the same map. The first is played by average players and the second is played by xathis (and other top bots).

Avg: http://aichallenge.org/visualizer.php?game=335973&user=8... Top: http://aichallenge.org/visualizer.php?game=329760&user=4...

The first lines seemed familiar:

http://quotenil.com/Planet-Wars-Post-Mortem.html

No content inside (but ads)