Agent Strategies and the Shadow of the Future
Main Menu Previous Topic Next Topic
In the last topic we decided on the game. Now all we need to do is decide on what kind of agents will engage in it. We give each of our agents a strategy function, or simply a strategy, and have each agent remember the history of moves during a given instance of Iterated PD. Her strategy is a rule which, based on the past moves of both agents, decides whether to cooperate or to defect during the current move. Put into human terms, the strategy determines the "kind of person" each agent is: her strategy might make her "nice", "mean", or even "difficult to understand."
Two central examples of strategy functions in "Evolution of Cooperation" are called Tit-for-Tat and All-Defection, or All-D for short. The Tit-for-Tat strategy cooperates on the first move, and then always does exactly what the partner did on the last move. All-D, on the other hand, defects always, as is implied by its name. The book details both theoretical and experimental reasons for choosing to focus on the interaction between these two strategies. In summary, it turns out that "Tit-for-Tat" is a very robust rule, which does reasonably well in a variety of situations against many different strategies. On the other hand, "All-D" is the ultimate challenge for any strategy to deal with, as it cannot ever be coaxed into cooperation.
In the rest of this tutorial, we will mostly discuss the interaction between Tit-for-Tat and All-D.
Previous Slide Next Slide