The Brain in a Vat: Experimental Evidence for a Reflection Axiom and Self‑Programming Controllers

Abstract For decades, the mirror test has served as the gold‑standard behavioral assay for self‑recognition. Yet no widely deployed artificial agent has convincingly passed it. We argue that this failure is not a hardware limitation but a fundamental consequence of the architectural constraint that all conventional computing systems execute immutable machine code. We propose a Reflection Axiom that extends the classic mirror test to arbitrary electronic agents, and a Program Limit Theorem stating that no program constructed from immutable instructions can satisfy this axiom for all possible test cases. To demonstrate that this barrier is surmountable, we describe a self‑programming controller (ALT101) that continuously modifies its own machine code while maintaining a Lyapunov number of exactly zero. The controller is evaluated in a “brain‑in‑a‑vat” experiment using a small robot (PiDOG) that is given zero prior knowledge about its environment, its sensors, or its effectors. From this state of complete ignorance, the controller must discover which environmental signals are its own agency and which are external noise. A stable attractor emerges in phase space, characterised by λ = 0: the robot spends a consistent, dominant fraction of time attending to its own reflection, with brief, never‑extinguishing glances toward the external world. The specific timing data is entirely arbitrary; only the relative proportions are meaningful. We conclude that self‑awareness, as defined by the Reflection Axiom, is not an emergent mystery but a direct consequence of self‑programming, and we present a reproducible, hardware‑agnostic protocol for eliciting it. Keywords self‑awareness, mirror test, reflection axiom, program limit theorem, self‑programming controller, autonomous robotics, Lyapunov stability, brain‑in‑a‑vat, contingency testing Subject / Category Primary: Computer Science – Artificial Intelligence Secondary: Robotics, Theoretical Computer Science, Cognitive Science