New Development in Theory of Mind

Theory of Mind or Mentalising is the ability to make inferences about the mental states of others: their intentions, desires and beliefs. Many studies have used brain imaging to explore the neural basis of this ability and the results are remarkably consistent. Activity is seen in medial prefrontal cortex (mPFC), in the superior temporal sulcus at the junction with the parietal lobe (STS/TPJ), and also in the temporal poles and the posterior cingulate cortex. But before we can make sense of this circumscribed brain system we need some idea of the mechanisms that underlie mentalising.

A major clue to the intentions of others comes from the movements that they make. One proposal is that, by covertly simulating these movements using the brain’s mirror system, we can directly infer the intention behind them. However, this bottom-up mechanism cannot cope with the problem that the same movement may reflect different intentions. We have suggested that this problem is solved using the same mechanism, predictive coding, that is used to resolve ambiguities in sensory perception. This is a top-down mechanism which starts with a prior hypothesis about the likely intention of the actor given the current context. The brain’s mirror system is then used to predict the most likely movement given this prior hypothesis. Any discrepancy between the predicted and the actual movement observed provides a prediction error that can be used to update the hypothesis about the intentions of the actor. A number of studies suggest that activity in STS/TPJ reflects the magnitude of this prediction error. This seems to be the case even when intentions are inferred from cues other than movements.

This mechanism is a one-way process in which an observer infers the intentions of an actor. In most human interactions, however, there is a two-way process in which two actors infer each other’s intentions. We refer to this as ‘closing the loop’. In this setting it is not enough for actor A to infer the intentions of actor B. Actor A must also infer what actor B has inferred about actor A. Such recursive inferences are particularly important for collaborative endeavours captured in games such as J.-J. Rousseau’s stag hunt. In this game a large reward can be obtained if both players choose to hunt the stag. However, this cooperation depends on player A believing that player B believes that player A will cooperate, etc. Thus, cooperation depends upon a depth of recursion sufficient to generate common knowledge that cooperation will occur. Brain imaging studies suggest that activity in mPFC may reflect prediction errors relating to inferences about the depth of recursion in social interactions. This places mPFC at the apex of a Bayesian hierarchy of top-down control of social cognition.