AI as a thinking partner: rethinking human cognitive processes in the age of generative AI

Generative AI (GAI) tools are increasingly integrated into education. However, existing research has primarily examined their effects through performance outcomes, offering limited insight into how they change learners’ cognitive processes. Grounded in information processing theory, this study investigates how students distribute cognitive processes between themselves and GAI during programming tasks and how such distributions relate to cognitive load, attention allocation, and programming performance. Using fine-grained cognitive event data from a laboratory-based experiment of 150 undergraduate students, we identified three distinct cognitive modes through clustering and sequential pattern analyses: AI-reliant, AI-independent, and human–AI hybrid. The AI-reliant mode involved extensive cognitive offloading to GAI; the AI-independent mode relied primarily on internal processing; and the human–AI hybrid mode integrated GAI outputs with learners’ own reasoning. While cognitive load did not differ significantly across modes, attention allocation did: The AI-independent mode focused more on the programming IDE, whereas the AI-reliant and hybrid modes focused more on the GAI. Notably, the hybrid mode exhibited more frequent attention shifts between GAI and IDE. After controlling prior knowledge, the AI-reliant mode achieved significantly higher programming performance. This study extends information processing theory by conceptualizing cognition as a distributed procedure across human and AI agents.