Intentionality detection and "mindreading": why does game form matter?

By around the age of 4 years, children "can work out what people might know, think or believe" based on what they say or do. This is called "mindreading," which builds upon the human ability to infer the intentions of others. Game theory makes a strong assumption about what individual A can expect about B's intentions and vice versa, viz. that each is a self-interested opponent of the other and will reliably analyze games by using such basic principles as dominance and backward induction, and behave as if the normal form of an extensive form game is equivalent to the latter. But the extensive form allows intentions to be detected from actual sequential play and is therefore not necessarily equivalent psychologically to the normal form. We discuss Baron-Cohen's theory of the mindreading system [Baron-Cohen, S. (1995) Mindblindness: An Essay on Autism and Theory of Mind (MIT Press, Cambridge, MA)] to motivate the comparison of behavior in an extensive form game with its corresponding normal form. As in the work of Rapoport [Rapoport, A. (1997) Int. J. Game Theory 26, 113-136] and Schotter et al. [Schotter, A., Wiegelt, K. & Wilson, C. (1994) Games Econ. Behav. 6, 445-468], we find consistent differences in behavior between the normal and extensive forms. In particular, we observe attempts to cooperate, and in some treatments we observe the achievement of cooperation, occurring more frequently in the extensive form. Cooperation in this context requires reciprocity, which is more difficult to achieve by means of intentionality detection in the normal as opposed to the extensive form games we study.