Poster C108, Sunday, March 25, 1:00-3:00 pm, Exhibit Hall C
Integrating incomplete information with imperfect advice
Natalia Vélez1, Sajjad Torabian1, Hyowon Gweon1; 1Stanford University
Social learning—learning from others—can often help us make better-informed decisions. The present study combines behavioral, computational, and fMRI methods to examine how human learners combine their own knowledge with the knowledge hidden in other people’s minds. Participants (N = 20) played a simple card game where they could choose to “stay” with a card of known value or “switch” to a card of hidden value. Participants received advice from an “advisor” whose information access varied in three within-subjects conditions: the advisor could see no cards, both cards, or only the card hidden to the participant. Behaviorally, we find that participants strategically used both the advice and their own knowledge (i.e., the known card) based on the advisor’s access to information. Participants’ choice behavior is well described by a Bayesian model of Theory of Mind that uses that advisor’s advice and information access to infer the advisor’s unobservable beliefs (i.e., the value of the hidden card). Consistent with prior work and model predictions, activity in brain regions that support both Theory of Mind (e.g., right temporoparietal junction) and reward-guided choice (e.g., dorsal anterior cingulate cortex, striatum, and frontopolar cortex) track the inferred value of the hidden card, even after accounting for the value of the visible card, the advisor’s access to information, and the difficulty of the decision. Our work provides novel insights into the neural and computational mechanisms that support learning from social information: human learners put “two heads together” by using mental-state inferences to guide value-based choice.
Topic Area: THINKING: Decision making