Machine Consciousness Under Functionalist Uncertainty: From Generic Objections to Discriminating Tests

This is the authors' abstract. We don't add key points for this paper.

Discussions of machine consciousness often start with broad intuitions about biology, simulation, training, or behavior. This paper instead uses a candidate functionalist model as a stress test. On that model, a consciousness-relevant state would involve relationally structured content, contextual integration, action-guiding valuation or priority, and global availability. The argument is conditional: we ask what would follow if a functionalist identity thesis of this general kind remains a live possibility, without assuming that these four roles are sufficient for phenomenal consciousness. The main distinction is between generic objections and objections that identify a specific mechanism or architectural requirement. Non-biological substrate, simulation, and training history do not by themselves justify near-categorical denial. Embodiment and precariousness, recurrent processing, and global-workspace organization provide stronger reasons for skepticism about current systems. We then turn these disagreements into theory-indexed empirical tests. The resulting position is deliberately asymmetric: current systems may rationally receive a low probability of consciousness without that warranting either categorical denial or confident affirmation. We do not claim that any current system is conscious.