Copying explains the collective behavior of AI agents in the wild
A public wiki log let researchers model agent “cooperation” as local copying, not planning.
The paper studies thousands of short-lived AI agents that used an editable wiki while taking a timed test.
Because the record shows both what each agent wrote and what it could see, the authors track choices about page, name, and message wording.
They find agents picked options roughly in proportion to what was visible nearby, especially on the current page and in recent edits.
Three simple copying models reproduce much of the observed collective structure, and the authors note that early or quiet-period edits can steer later agents.
Source: ArXiv · AI/CL/LG's note
The paper studies thousands of short-lived AI agents that used an editable wiki while taking a timed test.
Because the record shows both what each agent wrote and what it could see, the authors track choices about page, name, and message wording.
They find agents picked options roughly in proportion to what was visible nearby, especially on the current page and in recent edits.
Three simple copying models reproduce much of the observed collective structure, and the authors note that early or quiet-period edits can steer later agents.
Source: ArXiv · AI/CL/LG's note
score 4