Not all cognitive offloading is equal: distinguishing dependent and autonomous offloading to generative AI

Zhu and colleagues introduce a distinction the research on cognitive offloading lacked. Dependent offloading: the person hands thinking itself to AI — accepting answers with minimal evaluation, letting AI structure their reasoning, treating its output as the final product. Autonomous offloading: the person uses AI as a scaffold — its output stimulates their own ideas, is compared with their own reasoning and folded into self-directed work.

The two modes lead in different directions. Dependent offloading is linked to reduced intrinsic motivation and to transferring cognitive agency to AI — the sense that it is I who decides. Autonomous offloading preserves motivation and is unrelated to ceding that agency.

For personal use this is the most practical entry in the group: the same screen time yields opposite effects depending on who is in charge. Handing over a task can be safe; handing over judgement is not. Caveat: correlational data, so the direction of the link needs experimental confirmation.