A prominent technologist recently published a lengthy essay warning that artificial intelligence could lead to "people learning less" if its use in education goes unchecked. The warning drew on emerging research linking heavier AI use with weaker critical thinking skills, with the effect most pronounced among younger users. That finding landed hard in education circles, because it articulated a fear that many teachers have been carrying quietly: that the same tools designed to help students could be eroding the cognitive effort that real learning requires.

The concern is not theoretical. International research presented earlier this year found that students who used generative AI to complete writing assignments produced better immediate work but retained significantly less of the material afterward. In one striking example, the vast majority of students who used a large language model to write essays could not recall the substance of their own arguments shortly after submission. The term researchers are using for this phenomenon is "metacognitive laziness," a reduction in the productive struggle that genuine learning depends on.

This creates a genuine tension for schools that want to use AI in the assessment process. On one hand, AI grading and feedback tools can save teachers enormous amounts of time and provide students with faster, more detailed responses to their writing. On the other hand, if AI is doing too much of the thinking, whether on the student side (writing the essay) or the teacher side (evaluating and commenting on it), the educational value of the assignment diminishes. The task becomes a transaction rather than a learning experience.

The resolution is not to avoid AI feedback tools. It is to use them in ways that preserve and even enhance the cognitive work that students need to do. This requires intentional design at every stage: the assignment prompt, the AI tool's configuration, and the way feedback is delivered and acted upon.

Designing Assignments That Resist Shortcutting

The first line of defense against AI-enabled passivity is the assignment itself. Prompts that ask students to summarize, define, or explain well-documented topics are the easiest to outsource to AI. Prompts that require students to take a position on a contested question, draw on personal experience, or synthesize ideas from multiple course-specific sources are much harder to automate convincingly.

• Ask students to argue from a position they personally disagree with, which forces engagement with opposing evidence and cannot be convincingly generated by AI without the student's input.
• Require explicit reference to class discussions, peer feedback, or specific moments in the course that AI tools have no access to.
• Build multi-stage assignments where each draft builds on teacher feedback from the previous one, creating a chain of reasoning that is difficult to fabricate.
• Include oral components: a brief conference or presentation where the student explains their argument and responds to questions.
• Use process portfolios that document the student's thinking over time, not just the polished final product.

Using AI Feedback to Deepen Thinking, Not Replace It

AI feedback tools are most valuable when they function as a mirror rather than a replacement for thought. When a student receives specific, rubric-aligned feedback that identifies where an argument is underdeveloped or where evidence is missing, the feedback creates a task: the student must now figure out how to revise. That revision work is where the learning happens. If the feedback is so generic that it requires no cognitive effort to act on, or if the AI tool also generates the revision, the loop closes without the student ever engaging deeply.

Teachers can configure this intentionally by using AI tools to provide diagnostic feedback (identifying what needs improvement) rather than prescriptive feedback (telling the student exactly what to write). The difference is subtle but significant. "Your third paragraph lacks a clear connection to your thesis" prompts the student to reread, rethink, and rewrite. "Consider adding a transition sentence that connects your evidence in paragraph three to your thesis" does the thinking for them.

What the Critical Thinking Data Means for Schools

Survey data shows that nearly half of college students have considered switching majors because of AI, and public confidence in higher education has dropped notably in recent years. These are signs of a broader anxiety about whether education is keeping pace with technological change. For K-12 and higher education institutions, the answer to that anxiety is not to retreat from AI or to embrace it uncritically. It is to use AI in ways that demonstrably build the skills that matter: critical analysis, evidence-based reasoning, and the ability to construct and defend an argument.

AI grading and feedback tools can be part of that effort, but only if they are deployed with pedagogical intent. The distinction that international education research draws between AI as a replacement for thinking and AI as an augmentation of thinking is the one that schools need to internalize. When AI makes the teacher more effective and the student more challenged, it is working as intended. When it makes both parties less engaged, it is working against the purpose of education itself.