Published June 19, 2026 | Version v1

The Recursive Observer Problem: Recursive Self-Certification and the Problem of Epistemic Independence

Authors/Creators

Description

The question of free will has traditionally been framed as a debate over whether human actions are causally determined. Recent work in neuroscience and philosophy—exemplified by Sapolsky's Determined (2023)—has marshaled extensive empirical evidence for biological determinism. This paper does not replicate that empirical case, nor does it seek to refute compatibilism or libertarianism. Instead, it proposes a conceptual framework through which questions of agency may be re-examined across biological and artificial substrates. The framework interprets both biological organisms and artificial intelligence systems as rule-governed information processes: carbon systems operate under constraints imposed by genetics, neurochemistry, and environmental history, while silicon systems operate under constraints imposed by code, architecture, and training data. The paper's central contribution is the formulation of the Recursive Observer Problem: a rule-governed system may be capable of accurately describing the rules that generate its observations, yet it lacks an internal procedure by which it can non-circularly validate that those observations are independent of the processes that generate them. The problem concerns validation rather than representation, and justification rather than truth. Throughout this paper, "independence" refers primarily to justificatory independence rather than causal independence. The Recursive Observer Problem does not show that freedom is impossible. It shows that any claim to have settled the question of freedom must confront the conditions under which that settlement is itself produced. The paper's arguments are advanced primarily on epistemic grounds rather than as claims of metaphysical impossibility. Whether agency survives such a framework remains an open question.

Files

ROP_Preprint_v1_2026-06-19.pdf

Files (174.3 kB)

Name Size Download all
md5:0b25432351b41a309c97a69a76d52c25
174.3 kB Preview Download