Neural regions that underlie reinforcement learning are also active for social expectancy violations

Abstract

Prediction error, the difference between an expected and an actual outcome, serves as a learning signal that interacts with reward and punishment value to direct future behavior during reinforcement learning. We hypothesized that similar learning and valuation signals may underlie social expectancy violations. Here, we explore the neural correlates of social expectancy violation signals along the universal person-perception dimensions trait warmth and competence. In this context, social learning may result from expectancy violations that occur when a target is inconsistent with an a priori schema. Expectancy violation may activate neural regions normally implicated in prediction error and valuation during appetitive and aversive conditioning. Using fMRI, we first gave perceivers high warmth or competence behavioral information that led to dispositional or situational attributions for the behavior. Participants then saw pictures of people responsible for the behavior; they represented social groups either inconsistent (rated low on either warmth or competence) or consistent (rated high on either warmth or competence) with the behavior information. Warmth and competence expectancy violations activate striatal regions that represent evaluative and prediction error signals. Social cognition regions underlie consistent expectations. These findings suggest that regions underlying reinforcement learning may work in concert with social cognition regions in warmth and competence social expectancy. This study illustrates the neural overlap between neuroeconomics and social neuroscience.

Keywords:

• Trait attribution
• Social expectancy violation
• Trait warmth and competence
• Striatum
• Reinforcement learning

Acknowledgements

We thank the Center for Brain Mind and Behavior at Princeton University for funding this research and technical support. We also thank Bruce Barcelow, Jian Li, and Samuel McClure for feedback on earlier versions of the manuscript, and Mina Cikara and Lulu Kuang for assistance in collecting the imaging data.

Notes

¹Assessments of warmth and competence satisfy the components of the Rescorla-Wagner model: V _new=V _old+η(R – V _old). Here, R is a scalar quantity that is an assessment of goodness or badness, consistent with warmth assessments that specify the valence, and competence assessments that specify magnitude or value as a function of warmth.

²These are correlates of warmth and competence. Though sociability is a separate dimension of warmth (Leach, Ellemers, & Barreto, 2007), morality underlies the same dimension (Fiske et al., 2007). Therefore, we used behavioral sentences rated high on morality.

³We reverse the conventional order, presenting the behavior first and then the social target, because of the nature of our independent variable, ANOVA-styled sentences leading to dispositional attributions. Previous work suggests that people make dispositional attributions using ANOVA-styled sentences 9-14 s after the sentences are presented (see Harris et al., 2005). Therefore, we could employ a block design in order to capture the attributions across the entire 20-s period, regardless of when they occurred. However, given the variance in making the attribution, it would be very difficult to estimate precisely when the violation occurred if the order were reversed. By the time the social target is presented, the participant has made an attribution. There is a cleaner, more precise response to a picture as an isolated event in a stream of sentences.

⁴We use just some of our data to define the ROIs, to allow unbiased comparisons within our ROIs. Therefore, we expect the biased simple effect to be significant within the larger ANOVA model. However, no other simple effect, main effect, or interaction is biased by this ROI selection strategy.

⁵The nature of our contrasts makes it difficult to determine whether we are reporting positive or negative prediction errors. For instance, given our paradigm, one might predict that negative prediction errors should result from warmth expectancy violations. However, the low warmth social groups include both high and low competence groups. Therefore, it is possible that the high competence groups could lead to a positive warmth prediction error. It is difficult to make either case as our contrasts do not allow for independent exploration of positive and negative prediction errors.

⁶There is a highly contentious debate as to the exact role of dopamine in reward, specifically when present in striatal regions, considering these regions receive afferent inputs from other regions beside the substantia nigra of the basal ganglia. We raise this possibility as a potential subsequent study, not a definitive statement about the role of dopamine in social learning.