These results are just delightful! Twenty years ago people started running a tournaments to find the best algorithm for playing a repeated prisoner dilemma game of random length. Pretty quickly the winning strategy (nice is better than mean) emerged and nothing much has changed since them.
This year one of the entrants found a new approach!
Instead of entering one algorithm they entered 60. In the metaphor of they game they didn’t send one criminal to be repeatedly arrested; they sent a gang. Members of this mafia were very loyal to each other and they methodically ganged up on the other players and the police. When matched with an opponent in the game these gang members would try to puzzle out if they were playing against one of their own or not. In the real world you identify others of your kind with various signalling devices: secret handshakes, gang tags, etc.
In this game they can only signal to each other via the game. Using patterns in their game play mafia members recognize each other. At that point they can game the police in ways that to raises the gang’s overall score.
It’s just wonderful. This is the classic model of all game theory! And even in this tiny little dishpan model collaborative groups form and once they form they out compete the players that fail to collaborate. As Dave Weinberger once pointed out, we are a species that will form communities even if it means tapping out the alphabet on the wall of our cell.