The Things You Do: Internal Models of Others' Expected Behaviour Guide Action Observation

Abstract

Predictions allow humans to manage uncertainties within social interactions. Here, we investigate how explicit and implicit person models-how different people behave in different situations-shape these predictions. In a novel action identification task, participants judged whether actors interacted with or withdrew from objects. In two experiments, we manipulated, unbeknownst to participants, the two actors action likelihoods across situations, such that one actor typically interacted with one object and withdrew from the other, while the other actor showed the opposite behaviour. In Experiment 2, participants additionally received explicit information about the two individuals that either matched or mismatched their actual behaviours. The data revealed direct but dissociable effects of both kinds of person information on action identification. Implicit action likelihoods affected response times, speeding up the identification of typical relative to atypical actions, irrespective of the explicit knowledge about the individual's behaviour. Explicit person knowledge, in contrast, affected error rates, causing participants to respond according to expectations instead of observed behaviour, even when they were aware that the explicit information might not be valid. Together, the data show that internal models of others' behaviour are routinely re-activated during action observation. They provide first evidence of a person-specific social anticipation system, which predicts forthcoming actions from both explicit information and an individuals' prior behaviour in a situation. These data link action observation to recent models of predictive coding in the non-social domain where similar dissociations between implicit effects on stimulus identification and explicit behavioural wagers have been reported.

Fig 1. Trial sequence.

Each trial started with a fixation cross (400 ms.) and a brief blank screen. Each action started with an image showing one of the two individuals (John, left; Claire, right) in one of the two situations (at a computer, top; near a soccer ball, bottom). They then either interacted with the object or turned away from it, with one individual typically interacting with one object and turning away from the other, and vice versa for the other individual.

Fig 2. Experiments 1a and 1b RT and liking results.

Top panel: average response times in Experiment 1a (left panels) and 1b (right panels). In each panel, the left bars show identification of actions towards objects (typing on a computer or kicking the soccer ball) and the right bar shows withdrawals from these objects. The black bars reflect actions expected of this individual in the given situation, and white bars show the action expected of the other individual. Error bars show the standard error of the mean. Middle and bottom panels: correlation between prediction effects in the RTs for actions towards objects and the corresponding differences in perceived object liking and interaction frequency, for individuals who either identified (unfilled diamonds) or did not identify (filled diamonds) the behavioural pattern.

Fig 3. RTs and error rates for Experiment 2.

The black bars represent trials which followed the hypothesis and the white bars represent trials, which are the opposite of the hypothesis. The left side shows the response times and the right side shows error rates. The top row indicates actions towards objects and the bottom row indicates withdrawals. Error bars show the standard error of the mean.