CogLunch: Amrita Lamba "How credit assignment shapes learning in our uncertain social world"
Description
Speaker: Amrita Lamba
Affiliation: Simons Postdoctoral Fellow, Saxe Lab MIT
Title: How credit assignment shapes learning in our uncertain social world
Abstract: How do humans learn to adaptively function in a fundamentally uncertain social world? In this talk, I will present a neurocomputational study examining how humans attribute the outcomes of a social interaction to the appropriate cause—a process known as credit assignment. Using a trust game paradigm, I show that credit assignment substantially differs across individuals and varies across social and nonsocial contexts. People are more precise when crediting social agents compared to nonsocial objects (slot machines), a pattern that is mirrored by high-fidelity (i.e., distinct and consistent) neural representations in the prefrontal cortex. I will also present preliminary findings from an ongoing project in which learners coordinate with partners in a modified Nash Demand Game, the Asymmetric Social Exchange task, which requires learners to simultaneously assign value to their actions and a partner’s actions in parallel. Initial findings suggest this multi-agent credit assignment problem presents a significant obstacle for human learning given demands on memory capacity. Across this talk, I hope to demonstrate that several critical factors, namely how we implement credit assignment and how we respond to uncertainty, shape how we build mental models of others.
Location: 46-3310