Representing action in the human brain

Optimal decision-making entails that we maximise rewards and minimised punishment. Two classes of mechanisms enable such optimization; hardwired, or Pavlovian, policies directly tie affectively important outcomes, together with predictions of these outcomes to valence-dependent behavioural responses. Secondly, a more flexible, goal directed controller learns choices on the basis of their contingent consequence. These controllers tended to favour the same choices rending action fast and efficient. However their underlying workings are often revealed by striking sub-optimality.

In this talk I will address how we learn to perform optimal actions, how actions are encoded in the aforementioned systems and how sub-optimalities interfere with the efficacy of action. I will sketch how computational models can successfully tease apart the distinct contributions provided by these controllers. Lastly I will suggest that neuromodulatory influences exert selective effects upon these controllers.