Research network / collective intelligence
Designing new forms of collective intelligence.
Equilibria Network is a research network studying how groups of people and AI systems come to act together, and building the tools to design better ways of doing it. The work runs from theory to simulation to practice.
Cooperation is a property of the network.
To understand how a system works, you have to understand how it cooperates at larger scales. Whether people and AI systems act together depends less on any one of them than on the network around them.
Read the full thesis →- 01
Loops spread, good and bad.
Once a few actors start poisoning the well, it spreads through the network. Cooperative clusters exist too, and they spread the same way.
- 02
Start the positive loops early.
Computational social science has shown that it matters to start the positive loops early. As AI agents are integrated into society at larger and larger scales, we want positive environments to be what comes out of it.
- 03
One language across disciplines.
Existing theories each cover one corner. We are building a cross-disciplinary way of seeing these systems, where democratic decision-making, social choice theory, economics and mechanism design can be modelled in the same language and from the same perspective.
Five phases, from foundations to self-sustaining networks.
The programme moves from understanding coordination failures, through building and testing new mechanisms in simulation, to validating them in real domains. Each phase is named for a thinker whose work it leans on. Detail thins out further from the present, on purpose.
Phase 1 · HumeFoundationUnderstand Multi-Scale Coordination
Phase 2 · WienerConstructionBuild new coordination mechanisms from cybernetic principles
Phase 3 · BeerGenerationTest mechanisms through simulation in RL and LLM environments
Phase 4 · OstromValidationTest in multiple domains to build confidence
Phase 5 · NashEquilibriumSelf-Sustaining Networks
A simulation library where institutions are objects you can compose.
The Collective Intelligence Library treats a market, a network and a democracy as the same kind of thing: mechanisms acting on one population, scheduled over time. Compose a world, run it, measure it. It is where every research area gets tested.
Inside the library →
Four research areas, each shown as work in progress.
When does a group of agents count as one agent? How fast can a collective change, and what sets the limit? How much variety does an institution need to keep up with what it governs? And the library where all of it gets tested. Each area has its own page with the open questions and the pieces so far.
Feb 2026Spectral Signatures of Gradual DisempowermentAI disempowerment operates across markets, networks, and governance at once, but analytical tools do not cross those boundaries.
Feb 2026The Atoms of Knowledge Aren't UniversalWhy DeSci should stop searching for universal verification and start building compositional translations.
Sep 2025System Level Safety EvaluationsAdversarial evaluations and safety measures at the level of whole systems.
Aug 2025A Phylogeny of AgentsHow simple elements give rise to complex wholes with emergent properties.- Jul 2026The Viable System Model & Multi-Scale AgencyStafford Beer and the quest for understanding general agents.
- Jul 2026Models of Society Are Built on Models of AgentsPart 1 of a series on process alignment, a different way of thinking about agents at every scale.
- Mar 2026A Taxonomy of AgentsWhat kinds of agents exist, and a request for feedback on the classification.
- Mar 2026A Compositional Philosophy of Science for Agent FoundationsA philosophy-of-science approach to agent foundations research.
- Feb 2026Systemic Risks and Where to Find ThemTodd has a presentation in London on Thursday and three academics, some of them dead, will not stop arguing about root fungi.
A small team, with advisors across disciplines.
A research network, run as a Swedish non-profit: a core team of two and four advisors from neighbouring fields.
About the network →
Get in touch
We are open to collaboration.
Want to use system dynamics or agent-based models to make better policy proposals? Interested in the foundations of the science we are doing? Write to us. We are happy to talk about either, and about most things in between.