Some years ago, I came across Walter Moberley’s The Crisis in the University. In the years after the Second World War, universities faced a perfect storm: financial strain, shifting student demographics, and a society wrestling with lost values. Every generation has its reckoning. Universities don’t just mirror the societies they serve – they help define what those societies might become.
Today’s crisis looks very different. It isn’t about reconstruction or mass expansion. It’s about knowledge itself – how it is mediated and shaped in a world of artificial intelligence. The question is whether universities can hold on to their cultural distinctiveness once LLM-enabled workflows start to drive their daily operations.
The unwritten rules
Let’s be clear: universities are complicated beasts. Policies, frameworks and benchmarks provide a skeleton. But the flesh and blood of higher education live elsewhere – in the unwritten rules of culture.
Anyone who has sat through a validation panel, squinted at the spreadsheets for a TEF submission, or tried to navigate an approval workflow knows what I mean. Institutions don’t just run on paperwork; they run on tacit understandings, corridor conversations and half-spoken agreements.
These practices rarely make it into a handbook – nor should they – but they shape everything from governance to the student experience. And here’s the rub: large language models, however clever, can’t see what isn’t codified. Which means they can’t capture the very rules that make one university distinctive from another.
The limits of generic AI
AI is already embedded in the sector. We see it in student support chatbots, plagiarism detection, learning platforms, and back-office systems. But these tools are built on vast, generic datasets. They flatten nuance, reproduce bias and assume a one-size-fits-all worldview.
Drop them straight into higher education and the risk is obvious: universities start to look interchangeable. An algorithm might churn out a compliant REF impact statement. But it won’t explain why Institution A counts one case study as transformative while Institution B insists on another, or why quality assurance at one university winds its way through a labyrinth of committees while at another it barely leaves the Dean’s desk. This isn’t just a technical glitch. It’s a governance risk. Allow external platforms to hard-code the rules of engagement and higher education loses more than efficiency – it loses identity, and with it agency.
The temptation to automate is real. Universities are drowning in compliance. Office for Students returns, REF, KEF and TEF submissions, equality reporting, Freedom of Information requests, the Race Equality Charter, endless templates – the bureaucracy multiplies every year.
Staff are exhausted. Worse, these demands eat into time meant for teaching, research and supporting students. Ministers talk about “cutting red tape,” but in practice the load only increases. Automation looks like salvation. Drafting policies, preparing reports, filling forms – AI can do all this faster and more cheaply.
But higher education isn’t just about efficiency. It’s also about identity and purpose. If efficiency is pursued at the expense of culture, universities risk hollowing out the very things that make them distinctive.
Institutional memory matters
Universities are among the UK’s most enduring civic institutions, each with a long memory shaped by place. A faculty’s interpretation of QAA benchmarks, the way a board debates grade boundaries, the precedents that guide how policies are applied – all of this is institutional knowledge.
Very little of it is codified. Sit in a Senate meeting or a Council away-day and you quickly see how much depends on inherited understanding. When senior staff leave or processes shift, that memory can vanish – which is why universities so often feel like they are reinventing the wheel.
Here, human-assistive AI could play a role. Not by replacing people, but by capturing and transmitting tacit practices alongside the formal rulebook. Done well, that kind of LLM could preserve memory without erasing culture.
So, what does “different” look like? The Turing Institute recently urged the academy to think about AI in relation to the humanities, not just engineering. My own experiments – from the Bernie Grant Archive LLM to a Business Case LLM and a Curriculum Innovation LLM – point in the same direction.
The principles are clear. Systems should be co-designed with staff, reflecting how people actually work rather than imposing abstract process maps. They must be assistive, not directive – capable of producing drafts and suggestions but always requiring human oversight.
They need to embed cultural nuance: keeping tone, tradition and tacit practice alive alongside compliance. That way outputs reflect the character of the institution, reinforcing its USP rather than erasing it. They should preserve institutional knowledge by drawing on archives and precedents to create a living record of decision-making. And they must build in error prevention, using human feedback loops to catch hallucinations and conceptual drift.
Done this way, AI lightens the bureaucratic load without stripping out the culture and identity that make universities what they are.
The sector’s inflection point
So back to the existential question. It’s not whether to adopt AI – that ship has already sailed. The real issue is whether universities will let generic platforms reshape them in their image, or whether the sector can design tools that reflect its own values.
And the timing matters. We’re heading into a decade of constrained funding, student number caps, and rising ministerial scrutiny. Decisions about AI won’t just be about efficiency – they will go to the heart of what kind of universities survive and thrive in this environment.
If institutions want to preserve their distinctiveness, they cannot outsource AI wholesale. They must build and shape models that reflect their own ways of working – and collaborate across the sector to do so. Otherwise, the invisible knowledge that makes one university different from another will be drained away by automation.
That means getting specific. Is AI in higher education infrastructure, pedagogy, or governance? How do we balance efficiency with the preservation of tacit knowledge? Who owns institutional memory once it’s embedded in AI – the supplier, or the university? Caveat emptor matters here. And what happens if we automate quality assurance without accounting for cultural nuance?
These aren’t questions that can be answered in a single policy cycle. But they can’t be ducked either. The design choices being made now will shape not just efficiency, but the very fabric of universities for decades to come.
The zeitgeist of responsibility
Every wave of technology promises efficiency. Few pay attention to culture. Unless the sector intervenes, large language models will be no different.
This is, in short, a moment of responsibility. Universities can co-design AI that reflects their values, reduces bureaucracy and preserves identity. Or they can sit back and watch as generic platforms erode the lifeblood of the sector, automating away the subtle rules that make higher education what it is.
In 1989, at the start of my BBC career, I stood on the Berlin Wall and watched the world change before my eyes. Today, higher education faces a moment of similar magnitude. The choice is stark: be shapers and leaders, or followers and losers.
