Students Should Use AI Only for Tasks They Already Do Well

Summary

Daniel Willingham argues that students should use AI only for tasks they already know how to do well. His starting point is that school assignments are usually not meant to produce useful products; they are exercises designed to make students perform mental work that supports learning. Current district policies often allow AI for brainstorming, research, editing, summarizing, or study support while asking students to remain the lead and disclose AI contributions, but Willingham says they offer little guidance about when those uses help or undermine learning. Writing illustrates the problem because composing a substantial paper requires integrating sources, building an argument, reasoning about an audience, and expressing ideas precisely, all of which can be displaced by an AI-generated product. He proposes a litmus test: AI should not substitute for a task when doing that task is itself the learning objective, much as students should not use advanced calculator functions before mastering the underlying mathematics. He is skeptical that independent LLM use reliably scaffolds novices, who may lack the knowledge to ask useful questions and may simply accept polished suggestions. He also argues that expecting students to choose the harder learning path is motivationally unrealistic. Once a student is competent, AI may replace routine work that merely supports the real objective, but teachers must decide whether the task remains part of what students are meant to practice.