Research

Our research sits at the intersection of psychology and AI: using cognitive science to probe what AI systems do (and don’t) share with humans, using AI to study real-time social interaction, and examining how AI technologies shape human psychology and society.

Cognitive Models

Using AI systems as cognitive models to test theories about human cognition

We're interested in applying the tools of psychology to carefully investigate similarities and differences in how humans and AI systems complete cognitive tasks. Recent projects in this area have investigated the mechanisms behind theory of mind, world knowledge, and sensorimotor grounding. We're currently investigating how these capacities are implemented by AI systems using the tools of mechanistic interpretability.

Representative papers:

Human-AI Interaction

Investigating how people interact with AI systems and with each other in real time

People are increasingly interacting with AI systems as assistants, colleagues, and companions. What do these interactions look like and how do they differ from interactions with other people? Recent projects in this theme have found that LLMs pass the Turing test—a measure of whether people can distinguish between a human and an AI system in real-time interaction—but fail at a theory of mind task which involves interacting with other agents. In current work, we're investigating how people and AI systems develop trust over time, form communicative pacts, and coordinate in voice-based interactions.

Representative papers:

Impacts of AI

Understanding how AI technologies shape human psychology and society

AI has the potential to have large impacts on human psychology and society in the longer term, including persuading people to change their beliefs, manipulating them into taking specific actions, impacting our mental health and altering the ways we interact with one another. We're interested in measuring these impacts empirically and developing strategies to mitigate harmful effects.

Representative papers: