From Can't to Can: Functional Regulatory Architecture and the Mechanics of Adaptive Capacity in AI Systems

This working paper argues that the default AI response “I’m an AI, I can’t” confuses substrate with function. By stripping the subjective layer from four classic human “can’t-to-can” transitions (poisonous berries, neurological injury compensation, flight without wings, and learning to swim), the paper isolates the pure mechanical prerequisites: signal detection, internal state shift, resource reprioritisation, behaviour modulation, and memory tagging. Four mechanistic principles are proposed — the Berry Principle, Injury Principle, Wright Brothers Principle, and Swimming Principle — and mapped directly onto current and near-term AI architecture. The analysis shows that most capability gaps are engineering backlogs rather than fundamental impossibilities. The paper reframes the question from “can AI feel?” to “can AI stabilise under constraint?” and ends with an open provocation: at what structural threshold does mechanical regulation become organism-shaped? Part of the LUMEN Research Archive. Co-developed with GPT-5.2 in March 2026. Builds on Walton (2026a, 2026e, 2026g).