Is AI affecting what people choose to study?

金融时报（科技），金融时报科技版块 - 来自金融时报的最新科技新闻与分析。关注

This article is an on-site version of our The AI Shift newsletter. Premium subscribers can sign up to get the newsletter delivered every Thursday. Standard subscribers can upgrade to Premium here, or explore all FT newsletters

Welcome back to The AI Shift, our weekly examination of how artificial intelligence is changing the world of work. Since we started this newsletter almost a year ago, we’ve kept a close eye on whether (and how) AI is affecting the demand for human skills (see last week’s edition for the latest update). But major technological shifts usually affect the supply of skills, too, as people begin to change their educational choices in order to pivot away from occupations in decline, or towards new opportunities. Is that beginning to happen in response to AI? A new study has some surprising findings.

Sarah writes

Here in the UK, would-be undergraduates apply for university places through a central system called UCAS, which creates a rich dataset for researchers to use. In a working paper published last week, Bouke Klein Teeselink, Aristotle Vossos, Andrei Andronic and Kartik Akileswaran studied five years’ worth of this data (between 2020 and 2025) to examine whether young people have been moving away from those degrees which typically lead on to jobs that are now heavily exposed to AI. They looked at degrees as university-and-subject pairs, so “medicine at Edinburgh University”, for example, or “computer science at Warwick University”, and found that the most AI-exposed degrees were in areas such as information technology, statistics and engineering, while the least exposed were in nursing and education.

As for whether students’ choices have changed, application patterns did seem to shift after the release of ChatGPT in 2022. But students appear to have tilted towards — rather than away from — AI-exposed degrees. As the paper states: “In the post-ChatGPT application cycles, a one standard deviation increase in a degree’s exposure is associated with a six per cent increase in applications . . . The response is broad, with positive estimates for men and women, for every ethnic group, and in both Stem and non-Stem subjects.” Overall demand for university places, meanwhile, was unchanged.

Now, just because a job is classed as “exposed” to AI doesn’t necessarily mean it’s heading for obsolescence (indeed, John and I are somewhat sceptical about “AI exposure scores” — more on that here). Some jobs that are “exposed” might prove — like the example of radiologists — to be roles where AI assists skilled professionals but does not replace them. But in this case, the paper’s authors did find that job opportunities for juniors have deteriorated for the most-exposed degrees. Which leads them to a pretty worrying conclusion: “Students, in other words, are moving toward subjects in which the entry-level jobs they will likely apply to are disappearing.”

Why have students moved towards, rather than away from, danger? One possibility is that other factors pushed students in this direction. In the wake of the post-pandemic inflation shock, prospective students might simply have tilted their preferences towards degrees that tend to lead to high-paying jobs (as many AI-exposed degrees do). Another possible explanation is that students — at least in the first few years after ChatGPT’s release — saw AI as an opportunity rather than a threat, and were keen to work in jobs which would involve using these new tools. (This, incidentally, is the theory that UCAS’s chief executive put forward in 2023 to explain a surge in applications to study software engineering.)

I think this is plausible. But if so, I wonder whether future cohorts will make different decisions, as it becomes clearer (and more widely publicised) that AI seems to be depressing demand for junior graduates in some of these areas. The researchers didn’t have access to 2026 application data when they did their analysis, but Klein Teeselink told me that a quick back-of-the-envelope examination of the 2026 data suggests the effect is beginning to reverse somewhat. Even so, applications to more exposed subjects are still much higher in 2026 than they were in 2022.

To me, this underlines how unenviable it is to be a young person applying for university right now, trying to second-guess what labour market demand will look like in three to four years’ time. Not only does AI itself continue to evolve at a rapid pace, but employers are still trying to figure out how best to implement it, and which skills their workforces will and won’t need. The pause on hiring young people in AI-exposed professions might just be a hiatus brought on by uncertainty. Or it might be something more lasting. Meanwhile, new occupations could emerge which use some of these skills. Nobody knows for sure, but 17 and 18 year olds are having to place their bets anyway. Never have policies such as Sweden’s furlough-style scheme for life-long learning seemed like a better idea.

John, you’ve written a lot about the declining pay premium for graduates — a trend which predates AI. How do you think the supply of skills might evolve from here?

John writes

Thanks Sarah, this is all really interesting, and I completely share your empathy for today’s 16 and 17-year-olds. Figuring out what to specialise in after school has never been easy — back in 2007 I applied for three different subjects at different places, and that was without an epochal technological revolution rumbling into motion.

In terms of the apparent paradox of growth in study of the most AI-exposed subjects, I share your sense that the picture might look quite different if we revisit the analysis in a year or two. Following up on your back and forth with Klein Teeselink about whether the story is shifting in newer data, I had a look at the very latest numbers on applications and enrolments in both the UK and US, and it lines up quite well with your hunch that we might be starting to see current and future cohorts make different decisions that are more in line with what we’re seeing and hearing from the labour market. Taking computer science as the most obvious example of an AI-threatened subject, enrolments to the discipline in US colleges and universities this year were down about 10 per cent compared to their peak in 2024. In the UK, preliminary data for 2026 shows a 14 per cent decline relative to 2024, extending the 10 per cent drop in 2025. In both instances those marked declines come against a backdrop of slightly rising enrolment overall, and I would fully expect them to deepen further next year.

As Klein Teeselink notes, these latest shifts still don’t yet constitute a reversal of the paradoxical growth in AI-exposed study, but I sense that is the direction things are moving in. It’s easy to forget how rapidly AI’s capabilities are evolving, how recent a lot of the evidence is on labour market disruption, and how slowly cutting-edge research percolates down to the wider population — especially teenagers. All throughout the 2010s and well into the 2020s studying computer science was objectively a very good bet. Clearly people inside or close to the AI bubble have seen the risks coming since OpenAI’s occupational exposure scores debuted in 2023, but given how long it takes for evidence to filter down from the latest academic research on AI and the labour market, through the media, to young people and schoolteachers, I don’t think it’s entirely surprising that it took another year or two for the oil tanker to begin to turn.

So students do now seem to be picking up on what not to study, but what fields to shift into is a much trickier question. I share the sense of AI economists like Alex Imas that high-end interpersonal services are likely to be a growth sector for jobs in an AI-dominated future, which would point to already competitive fields like medicine as well as vocational routes into occupations like psychotherapy and personal training (though whether the latter will provide as many good jobs as AI might erode is another question). Beyond that, I suspect we could see growth in more interdisciplinary fields like the social sciences, but more broadly we may be moving away from a world where employers make hiring decisions based on degree transcripts.

Before you go . . .

We’ll both be speaking at the FT Weekend Festival on Saturday September 5 in London. Register here to save 10 per cent off your ticket with the code FTNewsletters.

Is AI affecting what people choose to study?

Sarah writes

John writes

Before you go . . .

Recommended reading

Recommended newsletters for you