What if the problem with AI in education isn't that students use it to cheat—but that they use it to avoid the very thinking that produces learning?

Generative AI can explain, summarize, organize, connect, and reason through course material with remarkable ease. Unfortunately, those are also many of the cognitive activities students need to perform themselves in order to learn.

This presentation explores the risks of cognitive offloading and pseudoskilling—and asks a seemingly paradoxical question: Can we design AI-powered assignments that make students think more, rather than less?

Topics Discussed

• How deep processing, association, and chunking contribute to learning

• Why generative AI creates new opportunities for cognitive offloading

• Pseudoskilling: apparent competence without the cognitive structures required for independent performance

• Factors that can encourage overreliance on AI, including cognitive overload, time pressure, low motivation, and inconsistent guidance

• Designing AI activities around a human-first principle: students think before AI intervenes

• The AI Concept Mechanic for diagnosing and repairing misconceptions

• The AI Trainer for retrieval practice and metacognitive calibration

• The Error Detective for evaluating rather than automatically trusting AI output

• Teach the AI activities that require students to explain and elaborate

• Summative ALEXs that use AI-powered scenarios to promote integration, application, and transfer

• Why the ultimate goal of introductory instruction should be machine-independent knowledge, not merely better machine-assisted performance