To study this, we recently developed two computational models:
🔵 A bottom-up GNN that makes social inferences by relationally structuring visual input
đź”´ A generative inverse-planning model that infers agents' goals, beliefs, and relationships by simulating what rational agents would do
In everyday life, we can effortlessly tell if people are interacting - and whether they're being friendly or fighting. We know “social perception” and “mentalizing” regions are involved (like pSTS and TPJ respectively), but what computations are they actually performing?
This challenges a strict division-of-labor story for the social brain. Both "social perception" and "mentalizing" regions carry out a combination of relational bottom-up and higher-order inferential computations!
Heading to #vss2026? Check out the presentations from our lab @vssmtg.bsky.social