AI is robbing students of intellectual confidence

According to a colleague from Imperial College London, postgraduate applicants are increasingly asking to be assessed online rather than in person. Why? The clue is in their habit of holding the interviewer’s gaze during the interview questions before glancing fractionally to the side and delivering polished, structured, fluent answers.

Too fluent. In several cases it became clear to my colleague that they were reading artificial intelligence-generated responses.

She was determined to get to the bottom of this. The master’s programme she runs attracts senior people, many already in industry. She was sure they were not simply cheating – and, sure enough, when she invited them to attend in person, the same candidates were different: less smooth, certainly, but still capable and thoughtful. So she chose to be generous. She set the Zoom interview aside and admitted them anyway.

I have seen a version of this dynamic, too. A mentoring client of mine, a Russell Group student and an excellent candidate, sent me a piece of work recently before a submission. I knew at a glance that the tone was not his. I know his language and voice well, and they are his strengths. Why on earth would he do something that could jeopardise his application? When I asked him, he said: “Obviously, I wanted it to be perfect. AI knows so much more than me.”

He was not being cynical. He was not exactly farming out thinking. He was simply anxious about his perceived lack of intellectual authority. His confidence in his own voice has been eroded by the machine.

This is not, I think, cheating in any sense we have previously meant by the word. It is something newer, subtler – but also more dangerous.

The scale of the shift is now clear. The Higher Education Policy Institute’s Student Generative AI Survey 2026, published in March, found that 95 per cent of UK undergraduates use generative AI in some form, and 94 per cent use it for assessed work. The proportion who include AI-generated text directly in submissions has risen from 3 per cent in 2024 to 12 per cent today, a fourfold increase in two academic years.

Another student said to me recently: “AI will always correct what I write, and I begin to feel I am not as good a writer as I thought I was.” She is, in fact, an excellent writer. But in any essay there are things that can be improved upon or that another reader, human or machine, will see differently. The question is whether the writer can hold her ground and say: “This is my voice and I stand by it.”

This is particularly relevant in the humanities and social sciences, although certainly not exclusively so. A US colleague, Jennifer Horan from Bryant University, has told me that her students are at times “paralysed” by the fear that whatever they have to say is somehow not good enough compared with what an AI could generate.

Systematic data on postgraduate AI use is, oddly, much thinner than the data on undergraduates. But a 2025 survey of 75 doctoral candidates at 19 institutions by academics at the University of Southampton found that most had used generative AI in their research. They framed it most commonly as a time-saver, an editor and, tellingly, a colleague. The paper takes its title from one respondent’s description of AI as “a rather stupid but always available brainstorming partner”. The casualness is the point. Even at doctoral level, where intellectual autonomy is supposed to be the qualifying virtue, students are already locating part of their thinking in the machine.

Schools and universities reward correctness over experimentation, fluency over hesitation, and finished sentences over emerging ideas. Generative AI produces all three on demand. It is, in a sense, the perfect graduate of the system we built. When students compare their own thinking, tentative and broken off, with the seamless prose of a chatbot, the comparison is predictably humiliating. Their own voice begins to sound, to them, like the rough draft.

The wider context is also part of the explanation. AI is already becoming core infrastructure in the workplace; according to a recent IBM study, 76 per cent of organisations now have a chief AI officer, up from 26 per cent the year before. Students have noticed. They are adjusting to a world in which the unmediated human voice is increasingly rare.

The solution certainly isn’t to prolong the war on AI in admissions. Detection is unreliable, prohibition is unenforceable, and the underlying problem is not technological. The problem is that we have produced a generation who believe their own thinking is not good enough unless it sounds like a machine. That is a pedagogical failure, not a software one.

Universities can address it, but only by valuing what AI cannot easily produce: real-time reasoning, productive uncertainty, the capacity to ask a bold question, to develop an idea rather than just present one. It means teaching students that hesitation is not failure. It is the start of thought.

It is fine to use AI as a thinking partner, but doing so requires us to draw a boundary, just as we would in any other relationship. The ability to think is the last bastion of our human autonomy. We simply cannot surrender it to machines that will always appear to know better.

Agnieszka Piotrowska is an academic, film-maker and psychoanalytic life coach. She supervises PhD students at Oxford Brookes University and the University of Staffordshire, is a TEDx speaker on AI intimacy and is author of the book AI Intimacy and Psychoanalysis (Routledge).