Description:

 Common knowledge and curiosity-driven conversation

Humans are estimated to spend several hours every day in conversation, most of it curiosity-driven. We seem to have some appetite for gaining conversational common ground, but traditional models of common ground make it hard to see why this is so. Classical ‘iterative’ models of common knowledge cap the epistemic power of the conversational dyad at the level of the weaker performer, or worse: it has been argued that even ideal reasoners cannot have any substantive common knowledge, if the classical iterative model is correct. Inspired by models of bimodal sensory integration, and by research on eye contact, I propose a new model of dyadic common knowledge in which the epistemic power of the dyad outstrips the individual power of either participant alone. By connecting social learning with reinforcement learning, this model also explains why we spend so much time in curiosity-driven conversation.