The Prisoner's Dilemma - Did they Cheat?
The winners of the first competition that was held in July 2003 were a team from Southampton University, UK who submitted multiple strategies, some of which acted as "fall guys" and once they were aware they were in competition with a friend they simply maximised their opponents score.
We were always aware that this could happen and there was a lot of discussion prior to the competition as to whether we should allow this. Our view was that it was fair on the basis that we are conducting research and, anyway, how do we stop it?
In response to one EMAIL, I wrote
"We had many discussions with people about how many strategies they were allowed to submit, and many discussions about how strategies could cooperate with one another. We decided that if people wanted to do this, that was fine (this is research) but the interface certainly does not allow you to know who you are playing against. To do that the strategies would have to have some sort of protocol. Again, we see this is a research question and it will be interesting to see if anybody has done this. This was one of the resaons why we have EMAIL'ed everybody now asking for a description of their strategies."
If nothing else, this question has prompted much discussion, which can only help move the research forward. See for example slashdot which was a discussion after an article appeared in Wired News.
We suspect that there could be a lot more colluding steategies in the next competition, so please see revised rules for how we plan to deal with this situation.
Fortunately, we are planning to run the competition again at the CIG'05 conference and we have introduced another category where only a single entry is allowed, per competitor. We encourage you to enter this competition - especially if you think the first one was unfair.