From Answer Generation to Cognitive Partnership

Abstract

Generative artificial intelligence is frequently evaluated through outcomes such as efficiency, accuracy, productivity, and task performance. This conceptual paper shifts attention from what AI produces to how sustained interaction with generative systems may reorganize human thinking. The Cognitive Partnership Cycle is proposed as a recursive model of human–AI learning consisting of Human Question, AI Expansion, Human Reflection, Integration, Revision, and a New Question. Drawing on sociocultural theory, distributed and extended cognition, metacognition, cognitive flexibility, organizational learning, and systems thinking, the model distinguishes conversational iteration from cognitive iteration and positions human agency and epistemic responsibility as governing conditions across the cycle. Six theoretical propositions specify mechanisms through which cognitive partnership may occur, while four failure modes—cognitive offloading, automation bias, illusion of understanding, and recursive error amplification—identify conditions under which generative interaction may substitute for or distort thinking. The framework offers implications for higher education, professional learning, talent development, knowledge work, instructional design, and future empirical research focused on trajectories of human–AI cognition rather than final outputs alone.