Human & AI Agents
Modeling, learning, and decision-making of social and AI agents — from boundedly-rational humans to large language models acting under strategic incentives.
Research
We design and govern complex, multi-agent sociotechnical systems.
Our research lives at the meeting point of three perspectives — agents, structure, and dynamics — and follows a single throughline: to design well, you have to understand and model first, and then govern the result.
Three research areas
Agents · Architecture · Dynamics.
Each area has its own methods, but the lab's projects almost always sit on at least two of them.
Network & Interaction Architecture
Structure, modularity, information flow, and interaction architecture in networked sociotechnical systems — and how those choices shape what the system can do.
System Behavior & Dynamics
Emergent structures and behavior, social norms, resilience, stability, coordination, and cooperation — the dynamics that ride on top of structure and agent design.
Four-stage approach
From understanding to governance.
A single project usually walks the same route — understand the system, build a model, design an intervention, then think hard about how to govern it.
-
Stage 1
Understand
Discover and analyze complex systems — what's actually happening, and why.
-
Stage 2
Model
Build analytical, simulation, and empirical models — including causal inference where it counts.
-
Stage 3
Design
Engineer architectures and interventions that work with the system's structure, not against it.
-
Stage 4
Govern
Align incentives and rules for better collective outcomes.
Methods
Tools we use.
Multi-agent reinforcement learning, graph representation learning, agent-based modeling, network formation games, NK-style fitness landscapes, and behavioral experiments — chosen by the question, not by fashion.
We also work hard on the explainable side: small, interpretable models, causal-inference workflows, and structured comparisons to keep big simulation pipelines honest.