Universities have spent the last three or four years trying to figure out what to do about AI, and a clear direction of travel has emerged.

UNESCO, the World Economic Forum, and university strategies from the US, UK, Australia, Europe and Asia, are all suggesting that as AI makes knowledge and analysis more accessible, universities should place greater emphasis on helping students develop critical thinking skills and good judgement.

This is often shortened, either in writing, or just in people’s minds, to something like:

“Universities should pivot away from knowledge delivery, towards judgement and critical thinking.”

I think this idea contains a big mistake.

I’m not saying we must cling to the past.

The old model of standing at the front of a lecture theatre, transmitting information at students and hoping some of it sticks, was inadequate long before ChatGPT arrived.

We are absolutely right to be creating more active learning, more authentic assessment, more project-based and interdisciplinary experiences, and much closer engagement with industry.

We absolutely should be helping our students develop the judgement and critical thinking skills necessary to survive in the AI-enabled world of 2026 and thrive as technology continues to advance at pace.

What concerns me is something different.

The big mistake in the emerging consensus is to treat knowledge and judgement as alternatives.

As though expert judgement is a transferable capability that can be developed independently of deep disciplinary knowledge.

And that’s not how expertise works.

To explain why I need to talk about pizza. (Keep reading. I promise it’ll make sense in a minute.)

I make pizza. And some days I’m pretty good at it. But when I moved to Malaysia, I had to leave behind the outdoor pizza oven I used in the UK.

So as I was trying to create the same thin, crisp pizzas in a domestic kitchen oven, I asked ChatGPT for advice.

After a little conversation it suggested confidently:

Get your dough hydration down to 31%. This is the level top chefs use to get maximum crunch.

As you read this, let me ask you: Do you think that’s a good idea?

Well… I knew instantly it was nonsense.

Anyone with experience of making bread knows that dough at 31% hydration won’t produce a pizza. It will produce a rock hard cracker. If you’re lucky.

I recognised immediately that ChatGPT was wrong. Not because of my critical thinking skills. But because I already knew what right looked like.

The judgement came from the knowledge.

The same principle applies at every level up to and including academic research.

Earlier this year, legendary computer scientist Donald Knuth published a paper describing how Claude and ChatGPT helped solve a difficult open problem in graph theory.

Knuth and his colleague Filip Stappers had made a good start, but couldn’t develop a full mathematical proof. So Stappers fed the problem to Claude.

Over the next hour the AI explored multiple approaches, hit dead ends and changed course. Eventually Claude generated a construction that enabled Knuth to write a rigorous proof for half the problem, from which Ho Boon Suan was able to use ChatGPT to solve it completely.

Speaking as mathematician interested in AI, the paper is extraordinary.

But the breakthrough did not happen because the AI replaced human expertise.

It did not happen because AI knows a lot and Knuth has good judgement and critical thinking skills.

It happened because world-class mathematicians possessed the knowledge required to evaluate, refine, reject, adapt and ultimately use the ideas that AI generated to create a rigorous proof.

AI didn’t replace the human expertise. It amplified it.

And without the human knowledge and expertise, there would have been no result and no impact.

So while I agree we must elevate judgement, we cannot just throw knowledge away, because it’s where judgement comes from!

Hubert and Stuart Dreyfus: researchers at the University of California, Berkeley; described how expertise develops. It builds through progressive stages. From novices following explicit rules to experts responding almost intuitively through years of accumulated knowledge and experience.

Experts are not guessing or relying on magical instincts unavailable to everyone else.

They have simply built such rich, tightly-woven mental repositories of knowledge in their domain that they immediately recognise patterns, anomalies and opportunities that novices simply cannot see.

Expert judgement is knowledge made fluent through experience.

This is why I worry about the emerging conversation around AI and higher education.

Yes. We must provide more opportunities for our students to develop transferable work-related capabilities, including judgement and critical thinking.

But we must remind ourselves every day, that knowledge is the foundation of all expert judgement.

We must use those active learning, problem-based, industry-involved approaches we’re developing to build both knowledge and critical thinking.

We must encourage our students to use AI to ask the questions they are embarrassed to ask in class, to explore alternative  explanations and to challenge their own assumptions. And we must help them do so in ways that build their internal knowledge base. Not replace it.

Universities have always been places where knowledge is created and shared.

However powerful AI becomes, we must not abandon that mission.