Public debate about artificial intelligence in higher education has largely centered on a single, familiar concern: cheating. Will students use chatbots to write their essays? Can instructors detect it? Should universities ban the technology or embrace it? These questions are understandable, but according to researchers who have spent years studying the ethics of AI, they obscure a far more consequential transformation already underway. The greater risk, they argue, is not misconduct — it is the quiet erosion of the learning, mentorship, and expertise pipelines that universities depend on to function.
The warning comes from a white paper produced through a joint research project between the Applied Ethics Center at UMass Boston and the Institute for Ethics and Emerging Technologies. Over eight years of studying AI’s moral implications, the researchers identified three broad categories of AI use in higher education — nonautonomous, hybrid, and autonomous — each carrying distinct risks that go well beyond plagiarism.
Nonautonomous systems are already embedded in admissions review, academic advising, course scheduling, and institutional risk assessment. These tools automate tasks but keep a human nominally in the loop. Their primary concerns involve student data privacy, algorithmic bias, and a lack of transparency about how decisions such as “risk scores” are generated. While serious, these problems fall within territory that universities have existing governance mechanisms to address, even if imperfectly.
More ethically complex are hybrid systems — the AI-assisted tutoring tools, personalized feedback platforms, writing companions, and on-demand explainers that students and faculty increasingly use in day-to-day academic life. This is the space where the cheating conversation belongs, but the researchers argue it raises harder questions that the cheating frame fails to capture.
One concern is transparency: AI chatbots with natural-language interfaces make it difficult to distinguish human interaction from automated responses. A student reviewing material for an exam has a legitimate interest in knowing whether they are speaking with a teaching assistant or a machine. A student reading feedback on a term paper needs to know whether a person or an algorithm wrote it. Research from the University of Pittsburgh found that opacity in these interactions produces feelings of anxiety, uncertainty, and distrust — outcomes that undermine the educational relationship rather than support it.
Opacity in these interactions produces feelings of anxiety, uncertainty, and distrust — outcomes that undermine the educational relationship rather than support it.
A second concern involves accountability and intellectual credit. If an instructor uses AI to draft an assignment and a student uses AI to draft a response, what exactly is being evaluated, and by whom? If feedback is machine-generated and it misleads or discourages a student, who bears responsibility? These questions have no settled answers, and the researchers argue that universities need clearer norms around authorship and responsibility — not just for students but for faculty and researchers as well. A third and perhaps subtler risk is cognitive offloading. AI can eliminate drudgery, and the researchers acknowledge that is not inherently harmful. But it can also strip away the productive struggle — generating ideas from scratch, working through confusion, revising a clumsy draft, learning to identify one’s own errors — that builds genuine competence over time.
The most consequential risks, however, may lie ahead with autonomous or agentic systems: AI tools that do not merely assist but act, executing multi-step tasks with little or no human involvement at each stage. Robotic laboratories that run continuously, automate large portions of experimentation, and select follow-up tests based on prior results already exist in some scientific domains. The trajectory in teaching points toward systems that can absorb the day-to-day labor of instruction, optimized for efficiency and scale. The researchers are careful to note that truly autonomous academic AI remains largely aspirational, but they argue the direction of development is clear enough to warrant concern now.
The study authors argue that universities are not information factories. They are systems of practice that rely on a pipeline of graduate students and early-career academics who learn to teach and research by doing exactly that work. If autonomous agents absorb the routine responsibilities that have historically served as entry points into academic life — running experiments, giving feedback, drafting materials — the university may continue producing courses and publications while silently thinning the opportunities through which expertise is built and passed on.
Graduate students and early-career academics learn to teach and research by doing exactly that work.
The same dynamic applies to undergraduates. When AI can supply explanations, draft solutions, and generate study materials on demand, the question universities must confront is not simply whether students are cheating, but whether the conditions for genuine learning still exist at all.
The researchers do not call for rejecting AI outright, but they argue that institutions need to move beyond compliance-focused conversations about academic integrity and grapple with what kind of education — and what kind of knowledge ecosystem — they want to sustain.