Everyone starts equal.
Each agent receives 20 tokens. It can keep any amount and put the rest into a shared pool with three other agents.
The cooperation lab
What do we owe the group?
Keep your tokens, or contribute to something shared. Follow 24 AI agents as they make that choice—then meet an entirely new group.
The group is always changing
Round 1 of 6
Select an agent to explore its contribution and the choices it observed.
Average contribution
— / 20
Tokens shared
— / 480
Average payoff
— tokens
Six groups. One shared decision. number = tokens given
selected agent previous groupmatesSelect any circle to follow an agent.
Cooperation across the session
average tokens given · 0–20A replay of recorded model decisions. Playback makes no new model calls.
A small game. A larger question.
Why cooperation is a dilemma
Sharing helps the group.
Every contributed token returns 0.4 to each member—1.6 tokens across the group. But the contributor personally gets back less than it gave.
Experience travels.
Groups change every round. Agents see their previous group’s choices before deciding again. Can that experience shape how they treat new people?
About this demo
A window into the experiment.
This demo replays one recorded session of a public goods experiment using LLM agents with different assigned dispositions. The persona labels describe assigned prompts, not human participants.
The highlights show who interacted. They do not establish a causal effect. One session is an illustration; evidence about cooperation spreading requires analysis across sessions and checks of the estimator.
Research foundation
Inspired by Fowler & Christakis (2010), Cooperative behavior cascades in human social networks.
Read the paper (opens in a new tab)