AI Could Be Undermining Teachers’ Most Important Skill

Students Don’t Trust AI Feedback. Their Teachers May Have Even More Reason to Worry.

Every time a professor reads a student’s essay and decides what to say, something more than grading is happening. A judgment is being sharpened, a relationship is being built, and a professional skill is being practiced. New research suggests the rise of AI may end up eroding this essential element of academia. As AI tools take over more of that work, educators may gradually lose the very abilities that make their feedback worth having.

A paper published in Assessment & Evaluation in Higher Education argues that the rush to automate academic feedback carries a hidden cost that seldom gets discussed. Seven researchers from universities in the UK, Australia, and Denmark warn that if colleges begin outsourcing feedback to generative AI, they may not be merely changing how students receive information. They may be quietly hollowing out the professional craft of the educators left behind.

It is a concern that cuts against the dominant narrative around AI in education, which tends to focus almost entirely on whether AI feedback is good enough to replace a professor’s comments. In one cited survey of Australian students, nine in ten rated educator feedback as somewhat or very trustworthy. For AI-generated feedback, that figure dropped to just six in ten, with students also ranking human feedback higher on quality, helpfulness, and clarity. But those numbers alone miss a deeper problem on the educators’ side of the exchange.

What AI Feedback Takes Away From Educators

Reading student work and crafting a thoughtful response is not a passive clerical task. One instructor studied by researchers adjusted her comments based on a student’s personal circumstances, prioritizing encouragement over correction. As one ethnographic study of marking practices found, educators’ “readings of texts were tangled up with ‘readings’ of students.” That attentiveness develops over years of practice, and it is exactly the kind of skill that quietly erodes when the work gets handed off.

When AI generates the first draft of feedback and educators simply review and approve it, that process of professional refinement gets cut short. Educators engage less deeply with student work, exercise less independent judgment, and see fewer of the patterns that make their responses genuinely useful. Over time, the authors warn, this produces what they describe as a gradual erosion of feedback expertise, driven not by laziness but by delegation. Educators may eventually start to see student work through AI’s framing rather than their own.

The Limits of AI Feedback

Much of what makes a professor’s feedback valuable is precisely what AI cannot access. Good judgment is built from years in a discipline, shaped by knowing individual students, and informed by context no algorithm can retrieve. A student struggling with foundational concepts needs very different feedback than one who is technically proficient but intellectually coasting. An experienced educator reads for both. AI tools cannot fully account for either.

Most commercial AI feedback tools are also built around an outdated model: feedback as a one-way transfer of information from provider to receiver. When a system generates comments without any ongoing relationship with the student, those comments risk becoming what one earlier scholar called “dangling data,” technically present but practically ignored. Even accurate AI feedback may go unused if students lack the trust or connection to act on it.

When AI Feedback Actually Helps

None of this amounts to a blanket case against AI in education. For students who find human feedback intimidating, one research team has described AI as a “pedagogical sandbox,” a lower-stakes space to test ideas and get reactions without the pressure of a professor’s judgment. That is especially relevant for students who feel too intimidated to approach their professors directly, and for whom a safe first step matters.

AI also holds real promise for tracking student development over time, building a sustained, broader view of how skills and thinking evolve across a course of study rather than delivering isolated comments on individual assignments. That kind of ongoing analysis is something most educators lack the bandwidth to provide alone, and it represents a genuinely smarter use of the technology than automating grading.

Still, students who interacted repeatedly with AI tools described something important missing. Each session started from scratch. A focus group study found that students noted “each interaction [with GenAI] feels like a new beginning,” contrasting it with human relationships that build “a continuous timeline that builds gradually over time, allowing for personalized guidance and support.” That continuity matters. It is what turns feedback into an ongoing conversation rather than a series of disconnected drops.

A Warning About Who Gets Left Behind

An equity concern threads through the paper as well. Students who already know how to evaluate and act on feedback will get more from AI tools than those who do not. For students without that foundation, AI feedback may be harder to interpret and easier to dismiss. The authors suggest this dynamic could widen long-standing achievement gaps, echoing what education researchers call a “Matthew Effect,” in which those already ahead pull further ahead.

Protecting against that outcome means treating feedback as a practice worth investing in, not a cost to be minimized. It means giving educators time to read student work closely, respond as full professionals, and develop the judgment that no approval queue can replace.

What is at stake, the paper argues, goes beyond whether AI feedback is accurate enough. It is whether the educators overseeing it will still know how to do better.

---