Transient Machine Consciousness: A Conditional Framework for Episodic Awareness, System Boundaries, and Physical Transformations

Abstract

Research on artificial consciousness often begins with a binary question that conflates architecture, function, persistence, agency, and phenomenality. This paper develops a conditional framework for asking whether a bounded artificial process could instantiate a transient phenomenal episode without thereby constituting a persistent person. H1a states only a conditional non-entailment: even a bridge-licensed phenomenal attribution would not by itself establish identity, autobiography, agency, moral patienthood, or personhood. H1b is the stronger, presently unestablished modal hypothesis that an admissible bridge permits its complete physical condition-set to be realized while declared additional properties are absent and any required larger-subject alternative is excluded; no complete artificial witness is supplied. H1c defines assessment eligibility by requiring a physical realizer, boundary, interval, grain, bridge, complete criterion-set, individuation rule where needed, evidential method, and prospective decision rule. Positive candidate classification additionally requires the locked evidence to support the bridge-relative condition-set while discriminating, to the extent claimed, against live non-phenomenal alternatives. The framework also separates five physical transformation families, matching descriptors from held-out endpoints, and trained models from passes, sessions, agents, and deployments. Theory-relative functional indicators and causal interventions can update architectural claims but do not directly measure experience. Present public evidence yields no phenomenal verdict, and absence of such a verdict does not establish impossibility. The contribution is a graded architecture of conditional claims and reproducible tests, not a demonstration of machine consciousness or a consciousness score.