V01.03 — Mirror Theory III: Recursive Observerhood in Context

Abstract

This paper is the comparative closing paper of the first Theory Arc of Mirror Programme, Volume I: Observerhood. It situates the corrected Version 2 architecture of Mirror Theory relative to neighbouring work in control theory, cybernetics, cognitive science, philosophy of mind and artificial intelligence. Mirror Theory I Version 2 replaces the original primitive-to-observer narrative with a narrower decision-theoretic characterisation. It begins with an individuated candidate subsystem, an operational viability criterion, locally available history, admissible actions and finite resource bounds. Under explicit witness conditions, exact situated predictive control forces target-grounded environment-relative and candidate-internal distinctions that support counterfactual viability prediction and endogenous action. These distinctions earn technical environment-model and first-order self-model roles. Recursive self-reliability is introduced only relative to an already established first-order self-model: where a matched recursive reliability witness identifies a decision-relevant change in that model’s target-conditioned performance, any exact adequate refinement must preserve the corresponding distinction. Selection of a reliability-sensitive architecture over a matched collapsed alternative remains a separate endogenous net-value question. Against that corrected foundation, the present paper compares Mirror Theory with sufficient-state control, predictive-state and state-abstraction methods, good-regulator results, metareasoning, predictive processing, active inference, autopoiesis, enactivism, global workspace and global neuronal workspace theories, Integrated Information Theory, self-model theory, Dennett’s intentional stance and strange-loop accounts. It does not claim that prediction, self-modelling, metacognition, uncertainty monitoring, reliability tracking or recursion are individually novel. Instead, the proposed Mirror contribution is the staged combination of target grounding, predictive sufficiency, endogenous causal use, target-relative self-reliability witnesses and an explicit separation between role necessity and architecture selection. The paper treats this positioning as a testable research claim rather than a declaration of priority. If established frameworks already recover the same distinctions and operational tests more economically, Mirror Theory may be substantially subsumed or reduced to a useful taxonomy. If the witness-based architecture yields new exclusion cases, empirical discrimination or cross-framework interfaces, it may warrant a distinct explanatory role. The paper also preserves strict negative commitments. Technical Mirror observerhood is not equivalent to phenomenal consciousness, personhood or moral status; the framework does not derive natural system boundaries or viability criteria from an unpartitioned universe; generic confidence, precision, self-reference and feedback are insufficient for technical recursive observerhood; and no universal scalar degree of observerhood is claimed. Version 2 supersedes the July 2026 version of V01.03. It removes the former four-primitive dependency chain and critical-threshold framing, updates the comparative analysis to the corrected formal architecture of V01.01 — Mirror Theory I, Version 2 and the revised philosophical framework of V01.02 — Mirror Theory II, Version 2, and strengthens the treatment of novelty, subsumption, factorisation dependence, causal confounding and empirical failure conditions. This release forms part of the Theory Arc of the Mirror Programme, Volume I: Observerhood. Official website: https://mirrorprogramme.org Canonical public record: https://mirrorprogramme.org/publications/