Individuation Before Indicators: A Structural Dependency in AI Consciousness Assessment

A method for assessing whether AI systems are conscious derives indicator properties from scientific theories of consciousness, checks candidate systems for those properties, and aggregates the findings into a credence. This paper argues that the method has an unrepresented stage prior to indicator assessment: the selection of the system to be assessed. The selection is logically antecedent to the indicators — each indicator is a property of an already-individuated system — and it must be made independently of them, on pain of circularity. It cannot be performed by the indicators or removed by accumulating more of them. That the selection is live and consequential, rather than a formal nicety, is shown from within the assessment literature: in the case of a reasoning model that works on a scratchpad, the method's own author finds the choice of system boundary decisive for whether the system counts as recurrent, and leaves it unresolved. Because the selection is antecedent, the indicator findings, and the credence built on them, are conditional on a choice the method neither states nor warrants. The conclusion is not that the method fails but that it is a conditional evaluator: it assesses a system once a system has been fixed, and what fixes the system lies outside what it assesses. A framework owes either an explicit criterion for fixing the system or an acknowledgment of the convention by which it is fixed — a requirement that holds whether or not there is a fact of the matter about system boundaries, and that bears directly on the use of these assessments in questions of moral status, comparative ranking, and AI governance.