AFAIK the strategy was slightly different: You start collaborative, and then simply repeat the other person's last action. This means that, if they screw you, you keep screwing them until they turn collaborative, at which point you collaborate in the next iteration.