Rubin, S., Roli, A., Stano, P., & Damiano, L. (2026). Genealogies of organizational closure: from logical paradox to mind in life. Frontiers in Artificial Intelligence 9. DOI 10.3389/frai.2026.1806435
Key takeaways
Read in full (NOTE-727).
- “Closure” names different things. Ashby closed information, Maturana and Varela operations, Rosen efficient causes, Montévil and Mossio constraints, Kauffman catalysts. Table 3 classifies the notions by what is closed, the causal regime, materiality, computability and where semantics sits.
- Closure is not feedback. First-order cybernetics equated circular causality with feedback loops; the BCL and Rosen each broke that equation. For Maturana and Varela, regulation belonged to the observer. Catalytic closure is not closure of efficient causes.
- The thesis. Closure is “the circular entailment of efficient causes that fabricate and regenerate one another while maintaining their own boundary conditions”, and it cannot be exhaustively characterized as a state-determined syntactic procedure. The authors say this is not a claim that life cannot be simulated in some respects.
- The non-computability argument is conditional. A halting-problem argument against a computable model of Rosen’s closure (Mossio, Longo & Stewart 2009, not LIT-570) holds “if and only if computability is defined by halting algorithms”, and the authors call it provisional.
- The AI verdict. Agentic loops, memory and RLHF add circularity and adaptive regulation, but “Their feedback loops do not regenerate the constraints that make those loops possible”. They are “engineered approximations of closure”, drawn from the definition, not from analysis of any system.
Why this is here
A11 offered it as an adversary: computational systems may simulate cognition without the endogenous organization biological autonomy requires. That is what it says. THEORY-221 records the thesis.
It supports CLAIM-288 and goes further. Policy updates and revised objectives are still, on this account, recurring information flows, not production of efficient causes, so the answer that counter allows would not satisfy these authors. Whether CLAIM-284 needs closure in this material sense is a choice the record has not made. On CLAIM-116 it agrees that regulation is not closure, but it follows Maturana in equating life with mind, a claim about cognition, not a will. It offers no intervention for QUESTION-086; note 35 says closure verdicts change with the level of description, which is CASE-076’s complication.
The filing’s summary, written before reading, was too narrow. The roots are in seven traditions, not mathematical logic alone, which the paper treats as a precursor (“not deductive”), and the summary left out the paper’s second claim, the irreducibility thesis and the AI verdict.