If AI Knows Everything, Why Am I Still Doing a PhD?

After announcing that I’d submitted my PhD thesis disposition paper, I received an interesting question.

Why would anyone waste time at a university these days?

After all, the greatest minds in technology apparently dropped out. ChatGPT and Claude know practically everything, and AI is now solving mathematical problems that have challenged humanity for decades, sometimes centuries.

So why bother?

Fair question. Although I’m not sure that pointing at a handful of billionaire dropouts is a particularly sound statistical argument.

First, I want the title.

There. I said it.

When I left Gymnasium, I was repeatedly told that I wouldn’t make much of a career.

Two decades in the technology industry later, pursuing a PhD has a certain personal appeal. Partly because I enjoy learning and research. Partly because I want to prove that the expectations people had of me were wrong.

I’m not going to dress that up as some solely noble academic mission.

But the title alone wouldn’t be worth the effort.

If someone offered me a fully written, scientifically sound PhD thesis tomorrow, with the promise that nobody would ever know, I wouldn’t take it.

Because I actually want to understand what I’m researching.

And that brings me to the more interesting part.

Not everyone needs a university.

I genuinely believe that.

Are you brilliant at plumbing? Can you build beautiful wooden furniture? Do you have a talent for something people genuinely need?

Go for it. There’s plenty of opportunity outside academia, and there’s no reason to spend years collecting degrees to validate skills that don’t require them.

The question isn’t whether university is universally good or bad.

It’s what you’re trying to achieve.

Learning a profession, discovering your talents, conducting original research, and buying yourself time to figure out what you want from life are different objectives.

They don’t all require the same path.

But isn’t knowledge becoming worthless?

This is where I disagree.

I’ve spent more than two decades working with technologies that eventually became fashionable. From big data and machine learning to real-time streaming and AI.

Being ahead of a few technological waves does something to your perspective.

I’ve watched technology move from research and experimentation into marketing presentations, customer demonstrations, and eventually production environments where somebody has to make it work.

AI is no exception.

These models are astonishingly good at combining information, identifying patterns, and increasingly generating genuinely novel solutions.

But the dots to connect don’t appear out of thin air in the first place.

Someone had to put them there.

And even when AI creates new dots, we still need to establish whether they’re correct, meaningful, and useful.

Consider the recent mathematical breakthroughs.

Claude’s formalization of Fermat’s Last Theorem is extraordinary. But the underlying theorem was proved decades ago, building on generations of mathematical research. Other AI-assisted discoveries are genuinely advancing the frontier of mathematics.

That’s progress worth celebrating.

It doesn’t demonstrate that human knowledge is worthless. It demonstrates what becomes possible when powerful technology builds upon it.

AI is making expertise easier to imitate.

Here’s something I’ve noticed professionally.

Technology rarely becomes disappointing when it leaves PowerPoint because the underlying technology is inherently bad.

It becomes disappointing because of how confidently people misuse it.

Incomplete data. Outdated data. Unrealistic workloads. Systems operating far outside their intended conditions.

And occasionally a MacBook running something that was never meant to become the backbone of an enterprise architecture.

But someone watched a TikTok explaining that adding „Act as a Senior Developer“ to a prompt solves everything.

Problem solved. Apparently.

The technology has made it remarkably easy to produce the appearance of expertise.

Unfortunately, it hasn’t made the consequences of missing expertise disappear.

And somehow, when things go wrong, I keep meeting the same small group of people cleaning up the mess.

I have one favourite question when someone presents an impressive new solution:

What are its technical limits? When does it stop working reliably, and how would you know?

The answers are revealing.

Some people explain their assumptions, limitations, and monitoring approach clearly.

Others produce a vocabulary-heavy fog, explain that I’m missing the point, or suggest the question is too obvious to deserve an answer.

I’ve met surprisingly few people who can give a clear, simple, honest response.

And that’s exactly why I don’t believe expertise has become obsolete.

So why research at a university?

I’m not going to pretend academia is perfect.

My own experience with academic supervision hasn’t exactly been an advertisement for frictionless university life. And studying remotely while maintaining a full-time professional career involves compromises.

But academia still offers something I value: a framework for investigating questions properly.

In industry, even interesting problems compete with budgets, customer commitments, deadlines, revenue targets, and P&L responsibilities.

You generally can’t disappear down a research rabbit hole indefinitely because you find a question intellectually fascinating.

Academic research gives you a different kind of challenge.

Find a question worth investigating. Study what’s already known. Develop a methodology. Run experiments. Test your assumptions.

And then explain what you’ve discovered in sufficient detail that other people can challenge it.

Not a polished three-slide presentation claiming to have found the universal solution.

Evidence.

Ideally reproducible evidence.

Peer review isn’t perfect either. But having people with substantial domain knowledge challenge your methodology, question your assumptions, and identify weaknesses is valuable. Buy

Sometimes uncomfortable. Occasionally frustrating.

And significantly more rewarding than winning a room with the fanciest PowerPoint.

Where does AI fit into my own research?

Pretty much everywhere it makes sense.

I use it to screen research papers, summarise lengthy publications, and decide which sources deserve closer attention.

I use it to improve grammar and wording. English isn’t my native language, and my ability to produce precise sentences varies considerably with sleep and coffee consumption.

I use AI to challenge preliminary ideas before consuming someone else’s time with questions that might have obvious answers.

And I use it to build quick proofs of concept to establish whether something is technically feasible.

That’s enormously valuable.

AI makes exploration faster and cheaper.

But I still have to understand the underlying research, verify the sources, evaluate the results, and take responsibility for the conclusions.

A convincing answer is not automatically a correct answer.

And an impressive demo is not evidence that a solution will survive the messy, ugly world beyond it.

There’s one more reason.

I don’t particularly want to outsource my ability to think and understand to a commercial technology provider.

I enjoy these tools. I use them constantly. I’m excited about what they’ll enable.

But if my entire professional competence depends on an AI subscription, I’m also dependent on someone else’s pricing, access policies, product decisions, and business interests.

That’s not a trade-off I’m particularly enthusiastic about.

If AI could complete my PhD thesis tomorrow, I’d still want to do the research.

Not because I enjoy unnecessary work.

Because I want to learn something I didn’t understand before.

And because when the next impressive solution lands on my desk, I want to be able to ask the uncomfortable questions without needing to ask the same system whether its own answers are trustworthy.

I’m perfectly happy to rent computing power.

I’d rather not rent my understanding.

Stay in the loop

Occasional, signal-focused insights on AI, data systems, and real-world execution. No noise. No spam..