Neural signatures of emotion intent and inference align during social consensus

Humans effortlessly transform dynamic social signals into inferences about other people’s internal states. Here we investigate the neural basis of this process by collecting fMRI data from 100 participants as they rate the emotional intensity of people (targets) describing significant life events. Targets provide self-ratings on the same scale. We then train and validate two unique multivariate models of observer brain activity. The first predicts the target’s self-ratings (i.e., intent), and the second predicts observer inferences. Correspondence between the intent and inference models’ predictions on novel test data increases when observers are more empathically accurate. However, even when observers make inaccurate inferences, the target’s intent can still be predicted from observer brain activity. These findings suggest that an observer’s brain contains latent representations of other people’s socioemotional intensity, and that fMRI models of intent and inference can be combined to predict empathic accuracy. People try to infer how others feel, with varying success. Here, the authors show that others’ feelings and our inferences can both be predicted from our brain activity, and that we are more accurate when these brain patterns align.