Using AI-based social simulation to understand the emergence, stability, and transformation of norms, institutions, and social order.
Research coordinates
The guiding questionHow can large-scale cooperative social orders emerge from heterogeneous individuals, and under what conditions can institutions make such cooperation stable across scales?Explore the programme ↓
01 / Research programme
The programme connects a methodological question about what simulators know to a theoretical question about how social order becomes possible.
Ask what a model-as-social-scientist understands: not whether an LLM agent resembles a person, but whether a simulator captures distributions, relationships among variables, their importance, and their change over time.
Move from reproducing population-level regularities to specifying how heterogeneous beliefs, expectations, social influence, and interaction generate—and are reshaped by—macro social order.
Explain how norms and legitimacy make cooperation durable, how institutional orders become self-sustaining, and when they lose authority or transform—from local communities to global governance.
What is being simulated?
The current project does not hinge on whether an LLM agent “acts like a real person.” It treats simulation as learning a population distribution—and the relationships, differences, and temporal changes that structure it.
That shift makes the simulator answerable as a model-as-social-scientist: can it predict, recover relations, recognize what matters, track change, and eventually support a genuinely generative explanation?
02 / Paper sequence
Each paper earns the next one. Benchmark performance is a foundation for explanation, but it is not itself an explanation.
Now · Foundation
Define what it means for an AI social simulator to understand society using survey datasets. Evaluate predictive support, conditional relationships, variable importance, cross-year change, and the boundary between fidelity and explanation.
Next · Capability
Build a simulator—potentially diffusion-based and trained for the benchmark—that models population distributions and social relationships unusually well. This is still a methodological contribution: a stronger model-as-social-scientist.
Then · Explanation
Use simulation to answer a consequential social-science question: how legitimacy emerges, how institutions stabilize cooperation, or how cooperation can persist across scales. Make mechanisms explicit and validate them against multiple real-world patterns.
03 / Reading map
Read for arguments and mechanisms, not coverage. For every text, ask: What are the actors? What do they know? How do they interact? What stabilizes the macro outcome? What evidence would distinguish this mechanism?
What must a simulator learn beyond a population distribution in order to reproduce the process that generated it?
Thomas C. Schelling
The clearest entry into how modest individual preferences can generate surprising collective patterns.
Mark Granovetter
A compact mechanism for understanding cascades, participation, and discontinuous macro outcomes.
James S. Coleman
The classic macro–micro–macro bridge: situations shape actors, whose actions aggregate into social outcomes.
Peter Hedström & Petri Ylikoski
A rigorous vocabulary for separating correlations and predictions from mechanism-based explanation.
Michael W. Macy & Robert Willer
A bridge from variable-centered explanation toward heterogeneous interacting actors.
Damon Centola
Shows why network structure and complex contagion matter for the diffusion of social behavior.
How can cooperation survive when actors, interests, institutions, and enforcement are distributed across scales?
Elinor Ostrom
The foundational account of how communities build rules, monitoring, sanctions, and durable self-governance.
Robert Axelrod
A minimal starting point for thinking about repeated interaction and the emergence of cooperation.
Robert O. Keohane
Why international institutions can sustain cooperation even without a dominant power.
Scott Barrett
A sharp treatment of scale, incentives, participation, and enforcement in global collective action.
Elinor Ostrom
Connects local institutional diversity to global problems through overlapping centers of governance.
Zhao Tingyang
A philosophical provocation: what would it mean to make the world, rather than the state, the unit of political order?
When does legitimacy become a self-sustaining source of institutional stability rather than mere strategic compliance?
Cristina Bicchieri
Separates empirical expectations, normative expectations, and conditional preferences—essential for modeling norms.
Tom R. Tyler
A core empirical account of procedural justice, legitimacy, and voluntary compliance.
Mark C. Suchman
A useful map of pragmatic, moral, and cognitive legitimacy and how organizations gain or lose each form.
David Beetham
Treats legitimacy as rule-conformity, justifiability, and expressed consent—not simply a belief or approval score.
Ian Hurd
Connects legitimacy directly to international order and asks why states follow rules without centralized enforcement.
Avner Greif
Institutions as self-reinforcing systems of rules, beliefs, norms, and organizations—highly compatible with generative modeling.
If a simulator predicts well, what exactly has it learned—and what additional evidence would justify calling its output an explanation?
Joshua M. Epstein
A short manifesto separating modeling from prediction and naming explanation, discovery, and theory clarification as independent aims.
Joshua M. Epstein
The canonical statement of generative explanation: grow macro regularities from explicit local rules and interactions.
Nigel Gilbert & Klaus G. Troitzsch
A practical map of simulation methods, model construction, verification, and interpretation.
Paul Windrum, Giorgio Fagiolo & Alessio Moneta
Frames validation as a relationship among theory, the model, and the target system—not merely output matching.
Volker Grimm et al.
A strategy for confronting mechanisms with multiple empirical patterns across scales.
Bruce Edmonds et al.
Why prediction, explanation, description, theory exposition, and policy advice require different standards of success.