We Can’t Afford to Turn Away From Computing Degrees

Fewer students are choosing to study computing. BCS’ Steve Pettifer explains why this trend, and the misconceptions driving it, should concern us all.

A dangerous idea is beginning to take hold among teenagers choosing their university subjects, and I suspect it has spread to plenty of parents too: there is little point in studying computing any more because generative AI can now do everything for us.

It is an easy conclusion to understand. AI tools can generate code, explain errors, suggest fixes and produce functioning prototypes in seconds. To someone deciding whether to spend three or four years studying computer science, that can make the discipline look as though it is being automated just as they are preparing to enter it.

But that conclusion confuses the visible part of computing with the valuable part.

On results day, UCAS figures showed that the number of students accepting places on computing degrees at UK universities had fallen for the second year in a row, including a seven per cent drop among UK 18-year-olds.

We also know that the number of people taking dedicated degrees in AI has been increasing, but these students still represent only a very small proportion of all computer science graduates.

My professional body, BCS, The Chartered Institute for IT, is urging government, employers and my own sector, education, to take this trend extremely seriously.

This is not simply a question of protecting university departments or preserving a traditional route into technology. It is about whether we are preparing enough people to understand, challenge and improve the digital systems on which almost every part of modern life now depends.

What a computing qualification really means

Why? Not simply because of the risk of expecting whole areas of critical infrastructure to operate safely on vibe-coded software, without expert technologists who understand what is needed to maintain it.

The deeper reason is that the trend is based on a misunderstanding of what a computing qualification has ever actually been for.

A computer science degree was never simply about learning a particular programming language, framework or paradigm, any more than chemistry has ever just been about mixing substances in test tubes. The subject has always been about understanding the world around us well enough to create new things that work, while improving the things that do not.

In computing, that understanding has a name: computational thinking. It is the ability to take a messy real-world problem and work out how to make a machine solve it. Languages and frameworks may come and go. That underlying skill does not.

That distinction matters more in an AI-driven world, not less. If a machine can produce code on command, then the human contribution shifts further towards deciding what should be built, how the problem should be represented, which trade-offs are acceptable and whether the resulting system is actually doing what was intended.

Those are not peripheral skills. They are the heart of computing.

I suspect this myth comes from an analogy that appears reasonable at first, but is fundamentally wrong. You can type a sentence into an image model and receive a perfectly passable picture for a humorous birthday card, so it can feel as though you should also be able to type a sentence or two into another AI chat tool and get a fully functioning system of any complexity at the other end.

You cannot. Building, debugging, securing and maintaining genuine software systems is nothing like ordering a cartoon.

A piece of software does not merely need to look plausible. It has to work under pressure, deal with unexpected inputs, protect data, interact with other systems, survive updates and remain understandable to the people who inherit it later. The consequences of getting those things wrong range from mild inconvenience to serious financial, operational and safety failures.

That is why expertise still matters.

AI changes the work, but not the need for understanding

Generative AI is a genuinely powerful addition to the toolbox. It can help you think, save time and point you towards problems you might otherwise have missed. But it remains a tool alongside the others, not a replacement for understanding what you are doing.

Used well, these systems can remove friction from routine work. They can draft boilerplate, suggest alternative approaches, explain unfamiliar code and accelerate early experimentation. For experienced technologists, that can be enormously useful.

But greater speed does not eliminate the need for judgement. In many cases it makes judgement more important, because more material can now be generated more quickly and with an air of confidence that is not always deserved.

The ability to ask whether an answer is correct, secure, efficient, ethical or appropriate is becoming more valuable as the cost of producing an answer falls.

Evolving at pace

There is something particular about this technology, however, that makes the current anxiety understandable.

The past 50 years have hardly been short of disruptive technologies, and each one has still left computing as a solid career choice. Earlier technologies, though, were at least relatively straightforward to explain.

The computer: a machine that follows simple instructions mindlessly, but does so very quickly. The internet: a way for computers to communicate with one another. The web: a way to publish and connect content.

Each of those explanations fits neatly into a short sentence. We also watched those technologies develop from their earliest and most basic forms into what we recognise today. Generative AI, by contrast, appeared to arrive almost from nowhere in a surprisingly mature state.

That suddenness matters. People did not experience decades of gradual improvement before encountering tools capable of producing fluent prose, realistic images and usable code. For many, the first encounter was with something that already seemed remarkably capable.

It continues to evolve quickly and remains frustratingly resistant to simple explanation. I have spent more than a year trying, alongside colleagues who understand far more about the subject than I do, to find a way of explaining how it works that is both accurate enough and clear enough to be genuinely useful to a general audience.

I have failed.

When even practitioners struggle to explain what looks like magic, it is hardly surprising that the creative industries are anxious about what the technology means for them. Nor is it surprising that computing is asking itself exactly the same question.

The danger is that uncertainty gets translated into fatalism. If a technology looks mysterious and extraordinarily capable, it is tempting to assume it will simply replace the people who once performed the task. But computing has rarely evolved in such a clean, linear way.

Lessons from a history of disruption

We have been here before, even if previous examples have happened on a smaller scale.

What we ask computers to do has changed beyond recognition over the past few decades. COBOL code used to batch-process a payroll was remarkable in its day, but it is orders of magnitude simpler than the infrastructure sitting behind a single modern website. No sensible person would attempt to build all of that from scratch today.

As the work became more complicated and computers became more powerful, we repeatedly found ways to factor out and standardise the dull, repetitive elements. That allowed us to spend more time concentrating on the creative, research-led, entrepreneurial and business-critical work.

Higher-level programming languages replaced much lower-level work. Libraries removed the need to reinvent common functions. Frameworks standardised recurring patterns. Cloud computing allowed organisations to rent infrastructure that once required specialist teams and large capital investments.

None of those changes made computing irrelevant. They changed where the difficult and valuable problems were.

It would be unfair to describe generative AI as simply the latest example of that process, because it is more significant than that. But there is still a clear family resemblance.

It will not come up with the next world-changing idea for you. It can, however, help you build that idea, provided you understand what you are doing. It can help you critique it. It can help you debug it.

And when the apparently simple prompt “fix my code” stops producing the right answer, it is the person who genuinely understands digital systems who suddenly becomes indispensable.

The easy parts are being automated first

This is where the debate around jobs often becomes distorted.

If AI automates some routine programming tasks, that does not mean there is no longer a need for programmers or computer scientists. It means the balance of the job changes.

The same thing has happened repeatedly in technology. Tools tend to automate the tasks that are easiest to specify, repeat and verify. What remains is the harder work: deciding what problem matters, translating ambiguous requirements into a workable system, identifying hidden assumptions, dealing with edge cases and understanding consequences.

Those are exactly the kinds of abilities a good computing education is designed to develop.

Someone who knows only how to reproduce syntax may have more reason to worry about automation. Someone who understands algorithms, data, systems, networks, security and computational problem-solving is in a much stronger position to use AI productively rather than compete with it.

New paradigm, old skills

That is the key point.

The most valuable thing a computing education gives you is not fluency in language X or framework Y. It is the ability to reason about how real-world concepts can be mapped into the digital world, to think computationally and to understand how systems behave.

AI increases the value of that skill rather than diminishing it.

The labour market is already reflecting this. Demand for people who genuinely understand systems is rising even as some routine coding work becomes increasingly automated.

That should not be interpreted as an argument that computing degrees never need to change. Of course they do. Courses should respond to the technologies students will actually encounter. AI-assisted development, model evaluation, data governance, cybersecurity and responsible use of automated systems all deserve serious attention.

But updating a curriculum is very different from abandoning the discipline.

If anything, universities now have an opportunity to make the purpose of computing education clearer. Students should leave understanding not only how to build systems, but how to interrogate the tools increasingly used to build them.

Why this matters beyond the technology sector

There is another reason the decline should concern us.

Computing is no longer confined to software companies or IT departments. Digital systems now shape healthcare, finance, transport, education, manufacturing, media, science, public services and almost every other major sector.

That means a shortage of people who deeply understand computing does not remain a technology-sector problem for long.

As organisations adopt more AI, they need people capable of asking difficult questions about reliability, bias, security, privacy, resilience and accountability. They need people who can recognise when automation is appropriate and when human judgement needs to remain firmly in the loop.

If fewer young people develop those capabilities, we risk becoming more dependent on systems that fewer people are qualified to understand.

That is not technological progress. It is technological fragility.

The message to students needs to change

So the honest message to a 16-year-old considering their future is the exact opposite of the myth.

More and more people now need to understand enough about this discipline to use new tools effectively and recognise when those tools are leading them in the wrong direction.

And because AI and digitisation are reshaping almost every corner of society, we need more people going deeper into computing, not fewer. We also need a broader and more varied range of people entering the field, bringing different experiences, ideas and ways of thinking with them.

The question for prospective students should not be whether AI will remove the need to learn computing.

It should be whether they want to understand one of the most important technologies shaping the world around them, and whether they want the ability to do more than simply accept whatever answer a machine produces.

Generative AI will almost certainly change what tomorrow’s technologists spend their working days doing. That is exactly why strong foundations matter.

Writing code was always the easy part.