Learning to Think With AI: Task-Level Frameworks for Human-AI Collaboration in Education

As generative AI becomes embedded in how students learn, the central question shifts from whether AI belongs in education to how its use can be structured to build rather than erode critical thinking. This review synthesizes research on AI tutoring outcomes, cognitive risks of AI over-reliance, equity gaps in AI access, and emerging frameworks for productive human-AI collaboration developed in professional contexts. Drawing on this synthesis, it proposes a four-mode framework, delegation, dialogue, critique, and coaching, for classifying AI use by the cognitive role AI plays in a task, and argues that assessment design may be a particularly actionable lever for protecting learning outcomes, alongside addressing unequal access to AI itself.