On the drawing board
Emergent Societies
Agent-based models you can specify, replicate and defend.
How does a pattern no one intended arise from many local interactions? Classic agent-based models answer with simple rules. Language-model agents promise richer behaviour, and also new ways to fool ourselves. Emergent Societies puts both in one workbench, with the discipline that makes a simulation evidence rather than an animation.11Epstein, J. M. & Axtell, R. (1996). Growing Artificial Societies. MIT Press. Axelrod, R. (1997). The dissemination of culture. Journal of Conflict Resolution, 41(2).
What it does
- Classic models as baselines
- Cultural dissemination (Axelrod), bounded-confidence opinion dynamics (Deffuant) and Sugarscape economies as reference implementations.
- Language-model agents, next
- Agents driven by language models, with your own model or provider, run under the same protocol so they can be compared against the classic baseline. On the roadmap.
- Experiments, not demos
- ODD specifications, seeds, parameter sweeps and Monte Carlo replications, with measures of emergence exported for analysis.
The method behind it
A plausible animation is a starting point, not a validation.
The project grows out of a research line on artificial societies: under which conditions an interaction mechanism produces a collective pattern, and what evidence lets us explain or predict it. It is closer in spirit to the academic tradition of NetLogo and Mesa than to large-scale generative simulations, and it is meant to make LLM-based simulation reproducible.
Who it is for
- Computational social scientists
- Researchers exploring generative agents with methodological care
- Teachers of complexity and simulation
Where it stands
An idea with working prototypes of the classic models. Join the list to follow its development.
Get notified about Emergent Societies
We will write when there is something you can try. Your interest also tells us what to build first.