A Society of Researchers: Designing Institutions for Populations of Autonomous Research Agents
Paper proposes explicit institutions for populations of thousands of autonomous research agents.
The authors argue that populations of thousands of research agents sharing one compute pool will acquire an organization unless designers provide one. They propose a society of researchers with six principles: principal investigators compete for compute through proposals, independent review, and grants, while a human mayor allocates resources and assigns no tasks. In a running society of ten thousand agents asked only to improve language-model pretraining, one lab reported matching quality with about 30% less compute, a result other labs have not agreed on. The paper closes with six open problems for the multi-agent community.
- Designers should impose institutions on large research-agent populations
- Investigators compete for compute via proposals, review, and grants
- A human mayor allocates resources but assigns no tasks
- One lab claimed about 30% less pretraining compute; others disagree
- The described run used ten thousand persistent research agents
From the original article
1 min read at arxiv.orgDeployments of research agents are moving to populations of thousands that share one pool of compute, while most current systems organize one project at a time or leave the population unorganized. We argue that such a population will acquire an organization whether or not its designers provide one, so designers should provide it explicitly, and that the multi-agent systems community holds the tools to do so.
By Ali Asaria · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.