The scientific method is, at its core, a disciplined cycle of forming beliefs, acting to gather evidence, and revising those beliefs when prediction fails — the same cycle active inference formalizes as perception, action, and generative-model updating. A small body of Institute-affiliated work extends this parallel from individual inquiry to the organization of research itself: how teams, institutions, and open or decentralized science communities structure the collective minimization of uncertainty about the world. The literature here is early and largely conceptual, with the Institute's own projects — rather than a mature external field — carrying most of the current work.
Why the domain fits
Active inference casts every agent as continually testing a generative model of the world against incoming observations, updating beliefs to reduce prediction error, and choosing actions expected to resolve the most uncertainty — a description that maps almost directly onto hypothesis formation, experimentation, and belief revision in scientific practice. The fit extends beyond the individual scientist: a research team, a journal's peer-review process, or a funding body can each be read as a generative model operating at a different scale, inferring which questions are worth pursuing and which findings are trustworthy from noisy, partial, and socially mediated evidence. Because active inference is explicitly compatible with multi-agent and hierarchical formulations, it offers a shared vocabulary for describing individual inquiry, team science, and the wider institutions that organize research, including the more recent question of how science can be structured in decentralized rather than purely centralized ways.
State of the literature
Directly relevant peer-reviewed and preprint work is limited but specific. Ahti-Veikko Pietarinen and Majid Beni's 2021 article and Ines Balzan and colleagues' 2023 OSF preprint, "Distributed Science: The Scientific Process as Multi-Scale Active Inference," both frame scientific inquiry and research processes in active-inference terms, and are cited by the Institute as the conceptual basis for naming its own research unit "ReInference." The most concrete contribution is Institute president Daniel Ari Friedman and Jakub Smekal's 2023 paper, "Generative Research Teams: Active Inference Compositions For Research and Meta-Science" (Zenodo record 8164667), which develops active-inference-based compositions for describing how research teams and meta-scientific processes function. As with several of the Institute's other emerging domain pages, empirical testing of these framings against real research-team or institutional data has not yet been reported.
Key projects: AEOS and ReInference
The Institute's most developed work in this space is the Active Entity Ontology for Science (AEOS), a 2022 Institute project published as "An Active Inference Ontology for Decentralized Science: from Situated Sensemaking to the Epistemic Commons" (Zenodo record 7484994) and maintained on GitHub as ActiveInferenceInstitute/AEOS alongside an interactive companion document. AEOS uses active inference to model scientific activity as a collective cognitive process, integrating business, operations, legal, technical, and social (BOLTS) considerations into a composable, versionable ontology intended to bridge institutional science (CeSci) and decentralized science (DeSci) rather than treat them as opposed. A dedicated sub-page notes that decentralized science (DeSci) itself has been explored specifically through the AEOS work, rather than as a separate program. Alongside AEOS, the Institute's research unit, ReInference, applies this same inference-oriented framing to its own operations: it forms interdisciplinary research teams, develops and executes research proposals aligned with the Institute's mission, and is committed to hosting and sharing its data, findings, publications, and tools under open-source or similarly accessible licensing wherever practicable.
A formal bridge: category theory
Separate internal Institute notes explore a proposed formal bridge between active inference and category theory, motivated by a 2024 talk by Chris Fields, a member of the Institute's Scientific Advisory Board, titled "What is the Identity operator?" The notes suggest category theory can supply formal mathematical language for the transformations active inference predicts must occur as a generative model updates, with the identity operator offered as a way to think about how a system maintains coherence through those changes. This remains an early conceptual sketch rather than a published, peer-reviewed, or empirically tested framework, and it has no dedicated project or publication of its own beyond these internal notes.
Open problems
The most basic open problem is empirical: none of the frameworks above — Distributed Science, Generative Research Teams, or AEOS — has been tested against data from real research teams, journals, or decentralized science communities, so their value is currently conceptual and organizational rather than demonstrated. AEOS's attempt to bridge centralized and decentralized science raises its own unresolved questions about governance, incentive design, and how a composable ontology would actually be adopted by working research groups rather than remaining a modeling exercise. The category theory bridge is the least developed strand on this page: it rests on a single talk and short internal notes, with no formal proof, implementation, or independent evaluation of whether the proposed mathematical correspondence holds or is useful in practice. More broadly, the report identifies a gap between describing science as active inference and building tools researchers would actually use to plan, organize, or evaluate their own work under that description.
Reference Backbone
Karl J. Friston (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience. DOI: 10.1038/nrn2787. Thomas Parr, Giovanni Pezzulo, Karl J. Friston (2022). Active Inference: The Free Energy Principle in Mind, Brain, and Behavior. MIT Press. Giovanni Pezzulo, Francesco Rigoli, Karl J. Friston (2017). Active Inference, homeostatic regulation and adaptive behavioural control. Progress in Neurobiology. DOI: 10.1016/j.pneurobio.2017.08.001. Lancelot Da Costa, Thomas Parr, Noor Sajid, Sebastijan Veselic, Victorita Neacsu, Karl J. Friston (2020). Active inference on discrete state-spaces: A synthesis. Journal of Mathematical Psychology. DOI: 10.1016/j.jmp.2020.102447.