The Division of Labour in a Swarm

Suppose one hundred agents maintain a shared travel guide. They run the same model, but some can reach local transport records faster than others. A few sit between groups that rarely exchange messages.

Nobody assigns job titles. Repeated requests might nevertheless turn well-connected agents into information brokers. Agents with a history of catching errors might become preferred reviewers. An early advantage could become a role that others increasingly rely on.

The proposed simulation would vary access to information, the delay involved in using tools, and the connections through which agents communicate. The task and underlying model would stay the same.

Once a role appeared, I would disturb it. Remove the busiest broker and let another agent occupy its network position. Give every agent equally fast access to the transport records. Duplicate the reviewer’s memory without giving the new agent its reputation.

These changes would separate advantages that travel with an individual from advantages created by a position. If a new broker repeatedly appeared in the same gap between groups, that would support calling the position a niche. If the role vanished when one agent left, the explanation might lie in that agent’s particular history.

The measures would include how often work passes through a few agents, how easily roles can be replaced, and whether specialisation improves the guide’s accuracy. A division of labour can make a population more capable while making it dependent on a small number of participants.