history
Rock, Paper, Scissors
by Eric Mattes
Fri 18 Apr 2003 18:46:31
There's a whole world of RPS out there that I never knew about! Although it's a pretty trivial game for most people, computer scientists (being who they are) have been exploring the strategy behind this seemingly random game. However, you've got to admit that when you play, you do have a strategy of some sort. It is this fact that the computer RPS opponents will attempt to exploit. Check out this page for the current champion:<br><br>
<a target="_blank" href="http://ofb.net/~egnor/iocaine.html">http://ofb.net/~egnor/iocaine.html</a>
<br><br>also check this one out for some real-time action that you can play on the web:<br><br>
<a target="_blank" href="http://www.csua.berkeley.edu/~emin/writings/lz_rps/">http://www.csua.berkeley.edu/~emin/writings/lz_rps/</a><br><br>
-E<br><br>
comments
anders pearson - Fri 18 Apr 2003 18:47:31
RoShamBo is fun to think about from a game theory perspective precisely because it is so simple. similar to the RoShamBo tournaments were iterated prisoner's dilemma programming tournaments. i remember reading in my game theory textbook about the early prisoner's dilemma tournaments. one of the first programs that just cleaned up was called TIT-FOR-TAT. from <i>Game Theory and Strategy</i>, by Philip D. Straffin: <blockquote> Axelrod invited professional game theorists to write computer programs to play iterated Prisoner's Dilemma, and then conducted a tournament in which each program played all the other programs. Fourteen programs were entered, some of considerable complexity. The winner was a four line program submitted by Anatol Rapoport, called TIT FOR TAT. Its instructions were: <ol><li>Start by choosing C [cooperate].</li> <li>Thereafter, in each round, choose whatever your opponent chose in the previous round.</li> </ol> In other words, "Do unto your opponent what your opponent has just done unto you." <i>[...]</i> A more general conclusion comes from Axelrod's analysis of programs which did well. They tended to share four properties with TIT FOR TAT. To do well, a strategy should be <ul><li>Nice. It should start by cooperating, and never be the first to defect.</li> <li>Retaliatory. It should reliably punish defection by its opponent.</li> <li>Forgiving. Having punished defectionn, it should be willing to try cooperating again.</li> <li>Clear. Its pattern of play should be consistent and easy to predict. </li> </ul> It is tempting to speculate about the extent to which these four properties might characterize successful social behavior in competitive situations. </blockquote> another great part is that apparently at the next year's tournament, everyone entered programs whose strategies were specifically designed to defeat TIT FOR TAT. TIT FOR TAT still managed to win that tournament. optimizing their strategies against TIT FOR TAT weakened them so much against each other and other programs that no one program was able to get a clear lead and TIT FOR TAT came out with the most overall points.