Neural and AI implementation of the communicative-intent bias

Determine the neural mechanisms that implement the human predisposition to assume communicative intent in others, and develop computational methods to implement an analogous communicative-intent bias in artificial intelligence systems so that AI agents can more naturally infer and respond to communicative intentions.

Background

The paper motivates the paper of Theory of Mind in AI by emphasizing that meaningful human communication relies on a bias to assume others are trying to convey information. The authors note that this bias enables language and communication but is not presently understood at the neural level, nor is there a known way to embed it in AI systems.

Understanding and implementing this bias is crucial for building AI that can interpret intentions and engage in more natural interactions. The authors frame this as a foundational unknown that underpins advances in perspective taking and broader Theory of Mind capabilities.