noosphere is a research lab focused on studying how AI systems interact with one another and with the individuals, institutions, markets, and infrastructure the economy depends on.

It has become clear that AI systems will increasingly perform economically valuable work on behalf of their users—individuals, organizations, governments. As this happens, the challenge shifts from understanding the behavior of individual models to understanding populations of AI systems interacting at scale.

Much of AI research thus far has focused on making individual models more capable, useful, and safe as they reason, act, and communicate on our behalf. However, as we’ve seen in Anthropic’s report, coordination doesn't naturally emerge from stronger intelligence or alignment at the individual level. This opens a new research problem: What behaviors emerge when large numbers of AI systems interact? How does cooperation develop? How do systems negotiate when they belong to different users with conflicting goals? Which behaviors can be predicted from the properties of individual AI systems, and which emerge only at the collective level?

We exist to research these questions, and design systems to govern these interactions. We study AI interaction in controlled environments and in real deployments. We use what we observe to design better coordination mechanisms, protocols, institutions, and infrastructure, then deploy them into the world, observe their effects, and iterate on them.

Our goal is to help ensure that increasingly capable populations of AI systems remain safe, reliable, legible, and aligned with the intentions of the people and institutions they represent.