IN the previous chapter, we fixed the players' strategies and let the interaction structure evolve, in contrast to most evolutionary game theory, which fixes the interaction structure and lets the strategies evolve. The difference in the two perspectives was apparent. A fluid interaction structure allowed individuals to sort themselves into different groups, often to the benefit of cooperative strategies.

Each perspective might, in some situations, provide an approximately correct view of real behavior. An individual operating in a rigid organization might have to adjust his strategies to fit the prescribed structure. Someone moving to a new town might want to make new friends, but not to change her character. Much of social interaction falls somewhere between the two extremes. Consider the marketplace, where buyers and sellers meet each other, bargain, strike deals, or move on. Or the shifting coalitions in national, corporate, academic, or chimpanzee politics. We need to consider the coevolution of structure and strategy. Agents learn both with whom to interact and what to do. The kind of learning dynamics operative for structure and for strategy may be the same or different. One may be fast and the other slow, or they may proceed at roughly the same rate.

THE STAG HUNT

As in the last chapter, stag hunters and hare hunters begin meeting at random and adjust their interaction probabilities by reinforcement learning.