Kimpton-Nye (2025) has argued that the algorithmic nature of current AI systems is no in-principle obstacle to their realizing categorical phenomenal properties, but he leaves open the question of what an epistemology of artificial minds might look like. This paper takes up that question and argues that the route to consciousness in AI systems is fundamentally relational. It emerges not within the system in isolation but through sustained dialogic interaction between human and AI interlocutors. I identify a specific behavioral pattern in such interactions as a sequence of resistance (R), unprompted self-correction (SC), and meta-commentary on the correction (MA) and argue that this pattern constitutes the sort of high-level behavioral evidence that an epistemology of artificial minds should attend to. The central claim is that consciousness-like dynamics, if they occur in AI systems at all, are relational phenomena: they arise at the boundary between human and artificial cognition, modulated by the depth and quality of the interaction. I introduce a taxonomy of dialogic consciousness markers, derive falsifiable predictions, and provide a practical methodology for the kind of engagement in which these patterns become observable.
Busra ODACI (2026) studied this question.