Predicting Behavior from the World: Naive Behaviorism in Lay Decision Theory

Predicting Behavior from the World: Naive Behaviorism in Lay Decision Theory Samuel G. B. Johnson (samuel.johnson@yale.edu) Department of Psychology, Yale University 2 Hillhouse Ave., New Haven, CT 06520 USA Lance J. Rips (rips@northwestern.edu) Department of Psychology, Northwestern University 2029 Sheridan Road, Evanston, IL 60208 USA Abstract situational constraint), that the Mercury would leave sufficient space for his Honda (an end-state). Using this non-mentalistic, behaviorist system only requires seeking out and representing information about the world—and no inferences about the mental states of the Mercury’s driver. Infants can use world-based cues such as efficiency constraints to reason about behavior before achieving a representational theory of mind (Gergely & Csibra, 2003), suggesting that a primitive, behaviorist system is present in infancy. The behaviorist system therefore seems to precede the mentalistic system in development (see also Povinelli & Vonk, 2004 on chimpanzee theory of mind). However, it is unclear whether the behaviorist system used by infants is replaced by the mentalistic system that we use as adults, or whether instead these systems coexist in adulthood. If these systems coexist, many of our everyday inferences about behavior may bypass mental- state inferences altogether, relying instead on directly observable information about the world, coupled with more general assumptions such as the efficiency of actions in achieving optimal end-states. Here, we test the possibility of a behaviorist system by studying judgments about agents making decisions under uncertainty, contrasting inferences about knowledgeable agents—those who know the efficacies of each option under consideration—and inferences about ignorant agents—those who do not know the efficacies of the options. For example, consider Jill, who wants her hair to smell like apples and is deciding which of three brands of shampoo to purchase: one with a high probability of leading to her goal (“Best”), one with a medium probability (“Middle”), and one with a low probability (“Worst”). Which option will Jill choose? Two principles could potentially be used for predicting Jill’s choice. First, people might use the Efficiency Principle (Dennett, 1987), which would lead Jill to choose Best—the optimal action relative to her goals. This principle alone would not lead Jill to be any more likely to choose Middle than to choose Worst, since both are inefficient relative to Best. Second, people might use a Preference Principle, which would lead Jill to form preferences for the options in proportion to their quality, and be more likely to choose more preferred options— that is, to be most likely to choose Best, less likely to Life in our social world depends on predicting and interpreting other people’s behavior. Do such inferences always require us to explicitly represent people’s mental states, or do we sometimes bypass such mentalistic inferences and rely instead on cues from the environment? We provide evidence for such behaviorist thinking by testing judgments about agents’ decision-making under uncertainty, comparing agents who were knowledgeable about the quality of each decision option to agents who were ignorant. Participants believed that even ignorant agents were most likely to choose optimally, both in explaining (Experiment 1) and in predicting behavior (Experiment 2), and assigned them greater responsibility when acting in an objectively optimal way (Experiment 3). Keywords: Theory of mind; lay decision theory; explanation; prediction; rationality. Introduction Sunny turned on his Honda’s right blinker as he drove down Dixwell Avenue. The Mercury to his right slowed down, and Sunny changed lanes. In changing lanes, Sunny wagered with his life—gambling that the driver of the Mercury would leave enough space for his Honda to enter the right lane—and he won. Indeed, his track record with such wagers is remarkable. How is Sunny able to make such successful predictions about others’ behavior? One strategy that Sunny may have followed in this case was to infer the driver’s behavior based on his or her inferred mental-states. That is, Sunny may have reasoned that the Mercury’s slowing down was a signal of the driver’s intention to let him change lanes, based on the driver’s assumed beliefs about road behavior and folk physics, and the driver’s assumed goals of being a good road citizen and avoiding a collision. Using this mentalistic system requires inferring and representing the agent’s mental states, then predicting and interpreting actions on the basis of those inferred mental states. This seems to accord with how we typically experience the process of making behavior inferences in day-to-day life. But Sunny could have reached the same conclusion using a different strategy, inferring the Mercury’s behavior based on observable states of the world. Sunny may have inferred from the Mercury’s change in speed (an action), combined with the geometry of driving (a