Understanding the mechanism of human–AI interaction: a distributed cognition perspective

Purpose This conceptual paper develops a framework for explaining how human–AI interaction unfolds in generative artificial intelligence (GenAI) environments and why the same interaction can support cognitive enhancement in some cases yet contribute to cognitive atrophy in others. Design/methodology/approach Drawing on a distributed cognition perspective, the paper treats human–AI interaction as a coupled cognitive system rather than an isolated individual act. It integrates distributed cognition, the gulf of envisioning and research on cognitive affordances to trace how representations move across the user, interface, AI and external representations. Findings The framework models interaction as a five-phase cycle: intent envisioning, representation externalization, generative reasoning, outcome assessment and cognitive update. It identifies three recurrent gaps – capability, instruction and intentionality – as structural points where representational coordination may break down. Divergent cognitive outcomes depend not only on output quality but also on whether the human-AI interaction supports the completion of the full cognitive cycle. Originality/value The contribution is not a new standalone theory, but a synthesis that re-specifies common interaction difficulties as system-level problems of representational coordination. This provides a clearer basis for analyzing design, evaluation and longer-term cognitive consequences in GenAI-mediated information interaction.