Taxonomy of Consciousness Attribution Error Modes in Artificial Systems

Abstract

Contemporary debates over consciousness attribution in artificial systems are marked by persistent disagreement and limited convergence, despite rapid technical progress. This paper argues that much of this impasse arises not from incompatible theories of consciousness, but from recurring errors in attribution reasoning. In the absence of direct access to subjective awareness, researchers necessarily rely on indirect proxies such as behavior, linguistic report, internal complexity, social interaction, emergence narratives, or moral intuition. The central error diagnosed here is the treatment of such proxies as sufficient grounds for attribution, rather than as indicators operating under epistemic constraint. The paper develops a taxonomy of attribution error modes, showing how both attribution-positive and attribution-negative positions commit structurally similar mistakes by allowing confidence to outpace justification. It further argues that convergence across multiple proxies, while capable of undermining confident denial, does not resolve attribution and instead marks a transition to epistemic uncertainty. The analysis does not propose a theory of consciousness or new attribution criteria. Its aim is to clarify the limits within which attribution claims can be responsibly made, and to provide a shared vocabulary for diagnosing reasoning failures that currently fragment discourse across the field.