After Scarcity: Rethinking the University in the Age of AI
This article is part of a four-part Frontiering series exploring what happens to higher education when intelligence, explanation and content generation become abundant. Rather than asking simply how universities should use AI, After Scarcity asks a more fundamental set of questions: who retains power when knowledge is commercialised, what happens to originality when everyone has access to similar intelligence, why judgement and foundational knowledge may become more important rather than less, and ultimately what a university is for when answers are no longer scarce. Across the four essays, the argument moves from power → originality → judgement → purpose.
- Why Are They Renting It Back? — Power
- When Intelligence Becomes Abundant, Originality Becomes Scarce — Originality
- AI Can Give You the Answer. But Do You Know Enough to Know When It Is Wrong? — Judgement
- The University in an Age of Abundant Intelligence — Purpose
Part 3: AI Can Give You the Answer. But Do You Know Enough to Know When It Is Wrong? - Judgement
There is an argument about education in the age of AI that sounds superficially compelling.
If artificial intelligence can retrieve facts, explain theories, solve equations, write code and provide expert-level answers on demand, perhaps students no longer need to learn as much foundational knowledge.
Why memorise something that a machine can tell you instantly?
Why spend years acquiring knowledge when increasingly sophisticated AI can sit beside you throughout your working life?
I think this gets the relationship between knowledge and AI almost exactly backwards.
The better AI becomes, the more important foundational knowledge may become.
Not because humans need to compete with machines at remembering things.
But because you need to know something before you can reliably judge whether the machine knows what it is talking about.
Fluency has become cheap
Generative AI possesses a remarkable characteristic.
It sounds convincing.
Ask a contemporary model to explain a legal case, analyse a business, diagnose a problem, critique a strategy or discuss a philosophical argument and it can generate something polished, structured and apparently authoritative within seconds.
This changes an important signal.
Historically, fluency was at least loosely correlated with expertise.
Someone capable of producing an articulate 1,500-word analysis of a complex issue probably knew something about that issue.
That relationship has weakened dramatically.
Today, extraordinary fluency can be produced with almost no underlying human knowledge.
Which makes shallow knowledge more dangerous rather than less.
Manu Kapur recently made precisely this argument in Times Higher Education: the student who gains most from AI is not simply the one capable of obtaining an answer, but the one capable of deciding whether that answer is any good (Kapur, 2026).
I think this distinction is crucial.
AI literacy is not the same thing as AI proficiency
Universities have understandably begun discussing AI literacy.
Students are being taught how to:
- prompt;
- select tools;
- generate content;
- automate tasks;
- use AI ethically;
- recognise bias; and
- verify outputs.
All of these matter.
UNESCO's AI competency work similarly combines technical understanding with critical thinking, human agency and ethical responsibility rather than treating AI literacy simply as tool operation.
But there is a danger that AI literacy becomes another form of software training.
Knowing how to operate AI is not the same as knowing how to think with AI.
And neither is sufficient if the user lacks the domain knowledge required to recognise a bad answer.
Consider medicine
A randomised clinical trial involving 50 physicians compared diagnostic reasoning by doctors using an LLM with doctors using conventional resources.
Access to the LLM did not significantly improve the physicians' diagnostic reasoning performance (Goh et al., 2024).
This is fascinating because the AI itself could perform well on many diagnostic tasks.
The bottleneck was not simply access to intelligence.
It was the ability to integrate that intelligence appropriately into human judgement.
Medicine provides a high-stakes example, but the underlying principle applies elsewhere.
An AI can draft a legal argument.
But someone needs enough legal knowledge to recognise the precedent it has misunderstood.
An AI can analyse a company's strategy.
But someone needs enough business knowledge to notice that the assumptions are implausible.
An AI can explain a statistical result.
But someone needs enough statistics to recognise that correlation has quietly become causation.
An AI can recommend a public policy.
But someone needs enough political, historical and ethical understanding to ask who benefits and who has disappeared from the analysis.
The AI can provide an answer.
Knowledge provides the means of interrogating it.
Judgement does not float in empty space
I have written previously that AI is making judgement increasingly important in education.
I still think that.
If AI can produce the essay, report, strategy, presentation or analysis, the final artefact tells us less about what the student actually understands. Assessment therefore needs to capture the decisions, assumptions and reasoning that sit behind the artefact.
But Kapur's argument adds something important.
Judgement requires something to judge with.
We sometimes talk about critical thinking as though it were an entirely transferable skill.
Teach someone to “think critically” and they can apparently apply that skill to economics on Monday, immunology on Tuesday and constitutional law on Wednesday.
But meaningful critical thinking is heavily dependent upon what a person knows.
You cannot identify what is missing from an argument if you do not know enough about the field to know what should be there.
You cannot recognise an unlikely claim without some prior sense of what is likely.
You cannot reliably challenge a source unless you understand something about the evidence against which it should be assessed.
This is why foundational knowledge matters.
Not as an alternative to AI.
As the cognitive infrastructure required to use AI well.
Perhaps education needs a new hierarchy
The old model could sometimes be caricatured like this:
Know → remember → reproduce
AI makes that model increasingly difficult to defend.
But replacing it with:
Prompt → generate → submit
would be considerably worse.
A more useful model might be:
Know → question → generate → interrogate → verify → decide
AI sits inside that process.
It does not replace it.
Students still need knowledge.
But the purpose of knowledge changes.
We are no longer asking students to know something merely because information is difficult to retrieve.
We are asking them to know enough to reason, discriminate and exercise judgement in an environment where information — both excellent and terrible — is extraordinarily easy to generate.
This changes assessment
Perhaps universities should worry slightly less about whether students have used AI and somewhat more about whether students have demonstrated enough knowledge to supervise it.
That might mean asking students to:
- explain why they accepted one AI recommendation and rejected another;
- identify errors or omissions in an AI-generated answer;
- defend a conclusion orally;
- identify assumptions behind generated analysis;
- compare AI output with disciplinary evidence;
- improve a deliberately flawed AI response;
- provide a decision trail showing where their judgement entered the process; and
- demonstrate what they can still do when the AI is unavailable.
This is not a return to pre-AI education.
Nor is it an argument for endless closed-book examinations.
It is a recognition that verification itself is a knowledge-intensive activity.
The novice-expert divide may actually grow
This produces another uncomfortable possibility.
We sometimes assume AI will close the gap between novices and experts.
In some tasks, it undoubtedly will.
A novice with AI can produce work that looks dramatically more sophisticated than a novice working alone.
But appearance and capability are not identical.
An expert can ask AI a question and recognise when the answer is slightly wrong.
A novice may ask the same question and be impressed by exactly the same response.
Indeed, increasingly capable AI could make that problem harder because its errors become less obvious.
The badly written hallucination is relatively easy to dismiss.
The beautifully argued answer containing one subtle but consequential false assumption is much more dangerous.
AI therefore creates a peculiar paradox:
The less expertise required to produce an expert-looking answer, the more expertise may be required to evaluate one.
That should matter enormously to universities.
What, then, should we teach?
Certainly AI.
Students need to understand these tools and use them confidently.
But they also need disciplines.
History.
Economics.
Science.
Statistics.
Philosophy.
Law.
Management.
Literature.
Mathematics.
And all the specialised knowledge that gives professions their substance.
AI does not make those things obsolete.
It changes what they are for.
Perhaps foundational knowledge is no longer valuable primarily because it allows us to retrieve the answer.
It is valuable because it gives us the intellectual footing from which to question one.
The educational challenge of the AI era may therefore not be teaching students how to get answers from increasingly intelligent machines.
That part is becoming remarkably easy.
The challenge will be ensuring they know enough to ask:
Is this actually right?
And perhaps the even more important question:
How would I know?
References
Goh, E., Gallo, R., Hom, J., et al. (2024). Large language model influence on diagnostic reasoning: A randomized clinical trial. JAMA Network Open, 7(10), e2440969. doi:10.1001/jamanetworkopen.2024.40969.
Kapur, M. (2026, July 28). AI makes foundational knowledge more important than ever. Times Higher Education.
UNESCO. (2024). AI competency frameworks for students and teachers.
Valentini, A., & Blancas, A. (2025). The challenges of AI in higher education and institutional responses: Is there room for competency frameworks? UNESCO IESALC.