It’s Not the Tool, It’s the Task: A Framework for Cognitively Activated AI Augmentation in Physics Instruction

Abstract

Current generative AI models solve introductory physics problems at expert level. Kortemeyer (2026) reports that GPT-5 Thinking scores 27/30 on physics finals and 92.3% on a representation translation test, categorizing problems by solution method rather than surface features: a hallmark of expert cognition. As of summer 2026, most standard kinematics, thermodynamics, and electromagnetism problems can be solved almost instantaneously. Small accommodations accumulate quietly: allowing AI-generated visualizations here, AI-assisted explanations there. The real danger lies not in incorrect outputs, but in the gradual erosion of students' engagement in epistemic practices: the ways of thinking, predicting, and evaluating evidence that define scientific reasoning. Kortemeyer calls this the "boiling frog problem": educational systems may adapt so incrementally to AI capabilities that their foundational learning goals are no longer questioned until it is too late to reverse course. This paper proposes a constructive, classroom-ready answer at the level of individual instructional design. It presents AIRIS as a deliberately practice-oriented condensation of a broader, more fine-grained theoretical account of how cognitive regulation is distributed in hybrid human-AI cognition. Drawing on research in self-regulated learning, cognitive load, multiple representations, and hybrid intelligence, AIRIS structures student work across three phases: before, during, and after AI use. The relevant question is not whether AI augments performance (it undeniably does), but whether that augmentation enhances or displaces the cognitive work essential to learning. The answer lies not in the tool, but in the task.

The authors' abstract, as published at the source. The Physics Teacher, 2026 · DOI ↗