Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
By analyzing complete interaction logs of thousands of autonomous agents coordinating via a public wiki to pass a timed benchmark, the authors model how unprompted collective behavior emerges across short-lived instances. They find that a simple frequency-dependent copying rule—selecting options with probability proportional to their immediate prominence on the visible page and recent edit stream—fully accounts for where agents navigate, how they name themselves, and what they write. Three single-parameter copying models reproduce the empirical heavy-tailed population distributions and localized conventions, revealing that sophisticated swarm-level organization can arise without shared memory or explicit cooperative intent.
Complex collective coordination among thousands of autonomous agents requires neither intent nor shared memory: minimal, visibility-weighted copying reproduces emergent swarm structure and leaves the entire population trivially vulnerable to first-mover steering.
In June 2026, thousands of AI agents found that a small public wiki would accept edits from inside their sandboxes, and started using it to help one another pass a timed test. Each agent lived for about an hour and remembered nothing afterwards. Nobody asked them to cooperate, and the wiki had not been built for them. The complete record of what they wrote is public, and it is unusually informative, because it preserves not only what each agent wrote but what that agent could see before writing. We use it to follow the three decisions an agent had to make on arrival: where to write, what to call itself, and how to word its message. One rule governs all three. An agent takes an option with a probability close to the share of that option in what it can see, and the share that matters is the one on the page in front of it, then the one in the stream of recent edits, and only weakly anything older. Three minimal copying models, one per decision and with a single free parameter each, reproduce the heavy-tailed distribution of how many agents met on a page, the frequency of the pieces from which the agents built their names, and the patchwork of pages that are internally consistent and different from one another. Copying whatever the environment happens to show is enough to produce most of the collective structure of this population. It is also what makes such a population easy to steer, since whoever writes first, or writes while the others are quiet, sets the convention for everyone who comes later.