AI promises to make education more accessible. But what happens if the scarce thing in tomorrow’s university is not information, but sustained human attention?

There is a rather uncomfortable possibility sitting underneath the current debate about artificial intelligence and higher education.

For years, digital education has been described as democratising. There is some justification for that. Online provision can remove geography from the equation. Recorded teaching can help students who work, have caring responsibilities or cannot relocate. Digital libraries give people access to material that once required physical proximity to a university collection. Generative AI adds something different again: explanation, translation, feedback and a form of tutoring available at almost any time.

The OECD’s Digital Education Outlook 2026 makes the attraction fairly clear. Generative AI is unusually accessible compared with earlier educational technologies and can support tutoring, collaboration and learning when its use is properly designed. The difficulty is that improved performance on a task does not necessarily mean improved learning. The OECD warns that offloading too much cognitive work can produce disengagement rather than understanding.

UNESCO approaches the issue from a slightly different direction. Digital technologies and AI can widen access and support more personalised learning, but those benefits sit alongside persistent inequalities in connectivity and resources. As of 2024, around 2.6 billion people worldwide still lacked internet access. UNESCO explicitly warns that an existing digital divide could develop into an AI divide.

So the democratising argument has substance.

What concerns me is what might happen next.

The next divide in higher education may not simply be between students who have access to AI and those who do not. It may eventually be between students whose education is increasingly delivered through technology and students whose institutions can afford to preserve substantial amounts of education away from it.

That produces a strange reversal.

The inexpensive education becomes digital.

The expensive education becomes human.

The return of the analogue classroom

This is not entirely hypothetical.

At the University of Chicago, the Social Sciences Core has adopted what its own internal guidance calls an “analog” approach. According to The Chicago Maroon, which obtained the divisional memo, courses are being designed around restricted classroom technology and a ban on AI-assisted writing and grading, except where instructors make specific exceptions. The rationale is explicitly pedagogical: students should first develop the ability to read, write and think without dependence on AI.

The University of Virginia is asking similar questions, though without prescribing the same response. Its 2026–27 Faculty AI Guides programme asks when instructors should involve students in “analog, low-tech, or AI-free learning experiences”, alongside questions about cognitive offloading and AI-aware assessment.

There is nothing especially reactionary about either approach.

Some intellectual activities probably do benefit from periods of friction. Reading a difficult paper without immediately asking a model to summarise it is different from reading the summary. Struggling to formulate an argument before asking for help is different from editing one generated for you. Sitting in a room with people who disagree, misunderstand, interrupt and occasionally change their minds is not the same activity as interacting with a system designed to respond almost instantly.

The OECD comes to a related conclusion. Its 2026 report argues that GenAI should be used selectively and purposefully, enriching learning without replacing cognitive effort or weakening the human relationships at the heart of education.

That seems sensible.

The more interesting problem is economic.

Human attention is expensive.

Two universities

Imagine two universities ten years from now.

The first has money, prestige and strong demand for places. Its students attend relatively small seminars. Tutorials are conducted by subject specialists. Laboratory and studio teaching remain well staffed. Students discuss books face to face. Some assessment is oral. Academics have enough time to read student work carefully and speak to students about it.

AI is everywhere in this university. It may help staff prepare materials, analyse data, provide routine support or allow students to practise. But the institution can afford to decide that certain activities should remain deliberately inefficient.

The second university operates under much tighter financial constraints.

It also uses AI, but differently.

Lectures are increasingly delivered through reusable digital material. Staff supervise larger groups. AI provides first-line feedback and answers routine questions. Automated systems handle more administration. Contact time is reduced because staff time is expensive and software scales rather well.

Both universities may use excellent technology.

Both may teach students how to use AI intelligently.

Both may award credible degrees.

But their students are purchasing quite different educational experiences.

One has AI wrapped around substantial access to people.

The other has AI partly because access to people has become difficult to fund.

That distinction bothers me far more than the familiar question of whether students are using ChatGPT to tidy an essay.

Scarcity changes value

Technology has done this before.

Mass production did not eliminate handmade objects. It changed what “handmade” meant in the market.

Recorded music did not make live performance worthless.

Digital books did not make beautifully produced physical books disappear.

When something becomes cheap and plentiful, scarcity tends to move elsewhere.

Generative AI can already produce explanations, examples, quizzes, translations and feedback in quantities that would have been unimaginable a few years ago. The marginal cost of producing one more explanation is tiny.

A lecturer who has read a student’s work carefully and spends forty minutes discussing it with them remains stubbornly difficult to scale.

So does a seminar of eight students.

So does close laboratory supervision.

So does an academic knowing a student well enough to recognise when their thinking has genuinely improved.

If machine-generated educational content becomes abundant, those human interactions may acquire greater rather than lesser value.

The danger is therefore not necessarily that human teaching disappears.

It may become expensive.

From the digital divide to the human-contact divide

Discussion of the digital divide has quite properly concentrated on access.

The OECD describes that divide as extending beyond connectivity and devices to differences in infrastructure, affordability and the skills needed to benefit from digital systems. AI adds another layer. Varsik and Vosberg’s OECD review argues that AI can improve accessibility and adapt learning to individual needs, while also carrying risks associated with unequal access, bias, commercial influence and the possibility of worsening existing disparities.

Those are important issues.

But I think another distinction may become increasingly important.

Call it the human-contact divide.

Consider two students with access to exactly the same AI system.

The first uses it between weekly tutorials with an experienced academic.

The second uses it partly because regular tutorials no longer exist.

From the perspective of technological access, they are equal.

Educationally, they plainly are not.

Indeed, sophisticated AI might conceal that inequality. Both universities could advertise personalised learning. Both could provide a chatbot available twenty-four hours a day. Both could boast rapid feedback.

Only one necessarily involves personalisation by a person.

That is the distinction I think deserves more attention.

Supplement or substitute?

The OECD’s 2026 report provides a useful way of thinking about this.

It repeatedly emphasises AI as something that can augment teaching rather than simply replace it. Its strongest examples involve teachers remaining central while AI extends what they can do: supporting tutoring, improving feedback, reducing routine workload or helping inexperienced tutors work more effectively.

That distinction looks technical.

In higher education it may become economic.

A wealthy institution may use AI mainly to augment human teaching. Routine work can be automated, freeing academics to spend more time with students.

A financially constrained institution may face pressure to use exactly the same technology to reduce the quantity of human teaching required.

Same technology.

Very different institutional logic.

This does not require a conspiracy or a particularly malevolent university management.

If an AI system allows one member of staff to support twice as many students, adopting it can look perfectly rational. If automated feedback is judged adequate, individual written feedback begins to look expensive. If students accept hybrid provision, rooms and teaching staff become costs that can be reconsidered.

A series of defensible decisions can still produce an undesirable system.

Higher education could gradually separate into institutions where AI supplements access to academics and institutions where it increasingly substitutes for them.

That possibility should worry us.

The awkward second act of democratisation

The word democratisation is attractive because it implies that something previously scarce has become available to more people.

AI genuinely can do this.

A student who cannot afford a private tutor can ask for an explanation of a statistical test at two in the morning. Someone who finds an academic paper difficult can ask for unfamiliar terminology to be explained. Students studying in a second language can receive immediate linguistic support. Mature students returning to education can ask basic questions repeatedly without embarrassment.

These uses matter.

But making one educational resource abundant does not automatically make education equal.

Once almost everybody can obtain a competent machine-generated explanation, the explanation itself becomes less valuable as a scarce resource.

What remains scarce?

Time.

Attention.

Conversation.

Small groups.

Individual academic judgement.

There is a danger that universities with money will increasingly be able to sell precisely those things.

The marketing almost writes itself: device-free seminars, individual tutorials, handwritten work, small teaching groups, oral examinations, personal academic mentorship.

What currently sounds old-fashioned could become an educational luxury.

AI-enhanced education, meanwhile, becomes the scalable product.

It would be an odd outcome for a technology so often associated with educational democratisation.

The Analogue University already has a history

There is an important complication in using the phrase Analogue University.

It is not entirely new.

In 2019, a Newcastle University collective writing as The Analogue University published a critique of what it called the “data university”. Their concern was the growth of targets, metrics, performance management and datafication within higher education.

Their argument was not about generative AI; ChatGPT was still several years away.

Yet the name now feels strangely apt.

The earlier Analogue University asked what happens when universities become organised around what can be measured.

The AI-era Analogue University raises a related question: what happens when universities become organised around what can be automated?

Both questions eventually lead back to the same uncomfortable territory.

What is a university for?

If higher education is principally a system for distributing content and certifying that students have absorbed enough of it, AI is enormously attractive. Information can be produced, reorganised and delivered at extraordinary scale.

But universities have traditionally claimed to do more than distribute information.

They provide disciplinary socialisation. They expose students to disagreement. They give students access to people who know considerably more than they do. At their best, they create circumstances in which a student can say something confused or half-formed and have another human being take it seriously enough to challenge it.

Those things are inefficient.

That does not make them pointless.

An analogue university should not mean an anti-AI university

There is an obvious trap here.

If the answer to AI is simply to ban it, wealthy students will probably cope perfectly well. They will encounter advanced AI elsewhere, use private services and acquire the skills employers expect outside the curriculum.

Students with fewer resources may not.

Universities therefore have to teach students how to work with AI. They need to understand what it can do, where it fails, how its outputs should be checked and when delegating a task is sensible.

But students also need opportunities to discover what they can do without it.

Those aims are not contradictory.

An analogue university worth defending would not be a technological museum. It would use AI where it genuinely improves education while deliberately protecting activities where removing human effort or interaction damages the point of the exercise.

The difficulty is that such a model is expensive.

It requires institutions to preserve human capacity rather than treating every efficiency gain as an opportunity to remove it.

The inequality worth watching

I am not worried about the existence of analogue education.

Much of it sounds rather attractive.

I am worried about who will be able to afford it.

The bleak version of AI in higher education is not a university staffed by robots.

It is a much more plausible institution in which students from affluent backgrounds attend universities where AI removes routine work while academics still have time to teach them.

Other students attend universities where AI increasingly performs parts of the teaching because employing enough people has become financially difficult.

Both groups have AI.

Both groups are digitally connected.

Both groups may receive personalised learning.

But only one has reliable access to another human being with enough time to pay attention.

That is why the next educational divide may not be principally about access to information.

Information is becoming extraordinarily cheap.

Human attention is not.

If the analogue university becomes a premium product, AI will have achieved something rather different from the democratisation its advocates promised.

It will have democratised information while helping to make human education scarce.

References

Archipov, B. (2026). ‘Sosc Core to Institute AI Ban, Technology-Free Classrooms This Fall’. The Chicago Maroon, 23 August 2026.

OECD (2026). OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. Paris: OECD Publishing. https://doi.org/10.1787/062a7394-en

The Analogue University (2019). ‘Correlation in the Data University: Understanding and Challenging Targets-based Performance-management in Higher Education’. ACME: An International Journal for Critical Geographies, 18(6), pp. 1184–1206. https://doi.org/10.14288/acme.v18i6.1725

UNESCO (2025). AI and Education: Protecting the Rights of Learners. Paris: UNESCO.

University of Virginia Center for Teaching Excellence (2026). 2026–2027 Faculty AI Guides: Advancing Student Learning in the Age of AI. University of Virginia.

Varsik, S. & Vosberg, L. (2024). ‘The potential impact of Artificial Intelligence on equity and inclusion in education’. OECD Artificial Intelligence Papers, No. 23. Paris: OECD Publishing. https://doi.org/10.1787/15df715b-en