In current scientific research, most AI-for-science systems focus on scaling a single reasoning process through better models, larger context windows, long-horizon agentic execution, or digital co-scientists working with a principal user. However, complex scientific problems are rarely solved by a single reasoner; they require teamwork, where members contribute different priors, experimental backgrounds, tacit knowledge, and domain-specific intuitions.
Thus, the open problem is not only how to scale models but also how to cultivate networked intelligence: scaling the connections between humans and AI systems so that results or hypotheses produced in one context can reach another person, agent, instrument, or robot that can act on them.
To address this, we introduce Mycelium, an active shared workspace that automatically connects researchers and AI agents as multi-user co-scientists. As human users and agents collaborate, the system captures important observations and hypotheses, tracks their relationships with the team's evolving model, and routes them to the person or agent whose next decision they can inform.
We evaluate Mycelium in its first empirical test, a biological multi-omics campaign, where routed shared context turned a local analytical finding into a cross-expert mechanistic constraint, ultimately leading to an experimental design. We also provide a computational account of networked intelligence as sparse conditional computation over distributed scientific contexts, distinguishing when a scaled standalone agent can match the network from when independent expertise and non-mergeable contexts render the network irreducible.
Blogger's Review: Mycelium significantly enhances the efficiency and innovation of scientific research by facilitating human-AI collaboration. The concept of networked intelligence not only offers a new perspective on scientific exploration but also lays the groundwork for future human-AI cooperative work models.