Julian Kiverstein :"The role of sensorimotor contingencies in mutual prediction"

Thursday, 15 September, 2016 - 10:00 to 11:30
When two or more people engage in interaction in collaborating on a joint task or in having a conversation their actions must coordinate. My talk will start from the hypothesis that coordination requires mutual prediction – each interaction partner must be able to accurately and precisely anticipate what the other is going to do, and respond accordingly. We can do this if we can each accurately and precisely predict the effects (i.e. contingencies) of the other’s actions. It is these regular sensory effects of actions that I will refer to as “sensorimotor contingencies”. In coordinated social interactions like a tango dance, my brain must be able to anticipate the sensory consequences of your actions, and your brain must be able to do the same for my actions, if we are to avoid tripping up over each other.
Do I need to build up a complex model of your goals and intentions in order to predict the effects of your actions? According to active inference models in motor processing (associated with Karl Friston’s work) the brain first predicts top-down, the sensory inputs it desires on the basis of the agent’s goals and intentions. It then produces the motor behaviours that are most likely to reduce the mismatch between its predictions and actual inputs. This reading of active inference would suggest that in order for my brain to predict the effects of your actions (and vice versa) it will need to model the goals and intentions behind your actions.
I will make a different proposal based on an interpretation of active inference as a process that reduces disattunement in the animal-environment system as a whole. The key difference this makes is that we must understand sensorimotor contingencies in a normative context in which the contingencies of actions are evaluated according to how the contribute to reduce disattunement or improve the grip the agent has on its environment. In the case of social interaction, active inference occurs in the context of perception-action coupling between people in which stable and regular patterns of interactions arise over time that entrain the actions of individuals. These patterns of interactions are often found in the cultural environment (such as the practice of dancing tango) and form the basis for mutual prediction. The upshot of my talk will therefore be that to socialise sensorimotor contingencies means reconceiving them in the context of social and cultural environment.