HN2new | past | comments | ask | show | jobs | submitlogin

It is! See the International RoShamBo Programming Competition: https://webdocs.cs.ualberta.ca/~darse/rsbpc.html


Hmm, the little I know about game theory, wouldn't that competition actually be about "who writes the best (most uniform) random number generator?"


It could be, if they were all trying to be random.

It could also be about "trying to exploit others' RNG" or more aptly "trying to exploit the tendencies of other competitors."

If I entered as a human participant I would get crushed.


If I recall correctly, an effectively designed strategy performs at least as well as a random number generator when playing against a random player, and can do better against a non-random player.

So if someone used the metastrategy of entering a shill competitor that always chooses rock, in a field where everyone else is perfectly random, then a true competitor that always chooses paper would win the competition. Without the shill, the winner would be random.

Therefore, you have to assume there is at least one strategy in the competition that can be consistently beaten, and that perfectly random players cannot be beaten by strategy and are therefore ignorable as part of the overall field. So the winning metastrategy is to design a non-random, yet unpredictable strategy.

As I recall, "Iocaine Powder", which created an internal model of the behavior of the other players and then played to beat the model's prediction, was an early dominator. There have also been players that won by "cheating", such as by examining the stack of the other player's program. Either way, the best way to win is to read your opponent's mind.

So after Iocaine Powder wins, the next year's competitors could put a copy of it into their programs, so that they can detect whether Iocaine Powder 1.0 would be able to play to beat them and then change their play to something else. Meanwhile, Iocaine Powder 2.0 also has a model of Iocaine Powder 1.0, and whenever it detects that someone is playing to beat IP1, it plays to beat that instead.

After several years, everybody knows that everybody knows that everybody knows that everybody knows that everybody knows that you were going to play rock, so I was going to play paper, so you were going to play scissors instead, so I was going to play rock instead, so that you had to play paper, so that I had to play scissors, so you end up playing rock. It's way more complex than "international rock-paper-scissors competition" makes it sound. If a human competitor entered, the bots should be able to predict what that person would order for lunch for the next week. It's crazy.


One particularly pithy implementation of that idea:

http://www.rpscontest.com/entry/614005




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: