Rock Paper Scissors Programming Competition

I participated in a Google hosted one a while back and it was great fun. There is a great push and pull between trying to be as random as possible while attempting to exploit non-randomness in your opponents.

I ended up going with a 'many demons' solution whereby I had many strategies voting on what the guess should be and I modulated the weight of their votes based on their success. I also used 'sicillian logic' to make meta-strategies that simply voted on a move that would win/lose to the guesses of my other strategies.

One thing I always wanted to try was to cheat. Flood the submissions with bots that would attempt a pseudo-random-looking handshake. If it failed the bots would act randomly. But if it succeeded they would perform a pre-determined pseudo-random pattern that my one true bot would exploit to boost its rating.

Reminds me of the clever strategy that won a prisoner's dilemma contest.

http://archive.wired.com/culture/lifestyle/news/2004/10/6531...

I wrote an entry about two years ago: http://www.rpscontest.com/entry/614005

The comment in the program is, of course, a joke. What I actually did was download every other bot on the site, pick out all the ones that made no calls to `random` and let a brute force search for a good constant run for a few days.

The trick is that random play wins about 50% of the time, and this hack, despite being totally deterministic, has play that is indistinguishable from random within the constraints of the contest unless specifically accounted for. For each other program that also has deterministic play, about half of all possible constants will win. Try enough constants and eventually you'll find one that wins against a moderate majority of the other entries.