The Human Role in Thinking With AI
Human thought has always depended on resources beyond the individual mind, including what other people know and how they reason, the language through which ideas are formed and expressed, the institutions that organize, preserve, and transmit accumulated understanding, and the tools that extend human cognitive capacities. Generative AI introduces a new form of participation in cognition by contributing to the explanations, interpretations, arguments, and judgments through which people come to understand, believe, and decide. This can make it increasingly difficult to recognize how cognitive work has been allocated between human and AI, creating what we call opacity in cognitive authorship. Such opacity may distort both self-assessment and social judgment, making it harder to determine what the person actually understands, can do independently, or can reasonably be credited with or held responsible for.
We introduce cognitive sovereignty as a psychological construct describing the capacity to govern the allocation of epistemically significant cognitive work between oneself and external systems while remaining accurate about that allocation, meaningfully connected to the reasoning that matters, and able to adjust or reclaim delegated work when needed. The construct comprises three capacities, which we term cognitive source attribution, reasoning connection, and delegation control.
This framework distinguishes AI-assisted from independent capability, constructing from evaluating reasoning, and felt understanding from the ability to reconstruct and apply what one has understood. It distinguishes cognitive sovereignty from related constructs and develops testable propositions and behavioral measures. The paper does not claim that extensive AI use is inherently harmful. Rather, it argues that meaningful human authorship depends on whether the allocation of cognitive work remains visible, understood, and governable. By making this otherwise hidden allocation empirically tractable, the framework aims to distinguish when AI extends human capability while preserving meaningful authorship from when it obscures what the human contributed, understands, or remains able to do.
