The Last Generation of Experts

A couple of weeks ago I wrote a piece on the joy of being average, about how the demand to be exceptional is a standard most people are structurally required to miss. I’ve spent my life being pretty good at a lot of things and great at none of them, and I’ve come to think that it’s a pretty great way to live.

Whilst I stand by what I said, looking back, I think there was an important nuance I missed.

There is a difference between being exceptional, and being an expert. Let me explain…

Despite sounding similar, the words have completely different etymological roots. Expert comes from the Latin experiri, to try or to test, which is where we get experience and experiment. Exceptional comes from excipere, to take out. One of them describes what you have been through. The other describes being lifted out of the group and set above it.

Exceptional is a position in a ranking, and only one person can hold it. Expertise is a standard, and any number of people can meet it. A society needs a great many people who meet it, in fact, there is no version of a functioning society where the number is small.

An expert is somebody whose judgement holds up in a particular field. They can look at a situation and tell you which parts are important, and they will spot the mistake that everybody else in the room misses.

We live surrounded by them, you never really notice, but that’s kind of the point. You walk between tall buildings without wondering whether they’re going to topple over and crush you. You take the tablets the doctor gives you and assume they’ll make you better. You get on a plane, you drink from a tap, you cross a bridge walked by several thousand people each morning. Every one of these is you putting your trust in a complete stranger. A stranger doing work you couldn’t assess if you tried, and this system holds up so reliably that most of us go a lifetime without ever really thinking about it.

The irony is that the more the world fills up with confident, competent-looking answers, the more a society needs people who can tell which of those answers are wrong.

Nobody checks the work. I have no way of knowing whether my accountant is giving me sound advice, and no way of telling whether the person who wired my house did it correctly. Once a society is expansive enough that what you rely on outruns what you could ever check yourself, we have to develop proxies for knowing when to trust.

The guild was the first system built to do this job at scale. A medieval guild existed to protect the standing of a trade, and it did that by controlling who was allowed to practice within their profession. It set how many years you served before you could work on your own account, it inspected what came out of its members’ workshops, and it could stop a person trading if their work wasn’t up to standard. A customer had no way of judging a barrel or a horseshoe, so the mark on the work told him the guild had judged it for him, and the guild’s own good name rode on it having judged correctly.

The universities came out of a similar arrangement. Universitas was the word for the corporation of masters and students that ran a medieval school, and the University of Paris was governed by the masters’ guild. A degree began life as the licentia docendi, the licence to teach, meaning a licence to practise the trade. Doctoral training was an apprenticeship, and the rank of master walked straight in from the workshop.

So a degree certificate is the guild mark in another form. You can’t examine every doctor you meet on their knowledge of pharmacology, so a piece of paper saying they trained for years at a university and passed the examinations will have to do. None of us has any idea what they did during that time. We have decided, collectively, that this is a valid vehicle for our trust.

A qualification means something because getting one is hard, it says less about what you actually learned than most people assume. What it tells an employer is that you got through something that would have stopped somebody who couldn’t do the job. That is information they have no other way of getting, which is the reason the certificate is of any worth at all.

The value of that certificate has been falling for a decade. Gallup has been tracking American confidence in higher education since 2015, and it has dropped by a third over that period. Nearly half of Americans now expect AI to make a degree matter less than it used to.

Information is free now. Anybody can read the same material an expert is trained on, and anybody with a Chat GPT subscription can produce a clean memo, a confident diagnosis, or an analysis with the right structure and the right vocabulary, in seconds, on a subject they know nothing about. It comes out indistinguishable from the work of somebody who spent a decade earning the right to write it.

So the certificate has stopped separating people, and so has everything downstream of it. A take-home task, a writing sample, a portfolio: each of them rested on the assumption that somebody who could produce the work could do the job. That assumption held for a very long time and it doesn’t hold now.

Employers are starting to find they can’t tell who’s good. Managers approve work they have no way of assessing, from people whose capability they have never seen.

If you're uncertain about your future and you want to do something about it, that's what the paid guides are for. Each one takes a single career decision and works it all the way through, with real numbers in it, and you come out the other end with something practical.

The signals that tell you somebody is an expert are losing their meaning, and at the same time the process by which a person becomes an expert is being taken apart.

To understand why the supply is being cut, we need to look at the different categories of knowledge. The first is declarative knowledge, the facts and frameworks you can write down and revise for an exam, which is exactly what a machine can be trained on. The second is tacit knowledge, which only arrives after you’ve done something hundreds of times, people often mistake this for intuition. Somebody looks at a problem you’ve been stuck on for weeks and tells you what happens if you go left and what happens if you go right, and you have no idea how they know.

Expertise is mostly the second kind. It accumulates in a person who has tried things over and over which is why it takes years and why nobody has worked out how to neatly package it into a lecture.

Every profession had a standard way of manufacturing it. It handed the newest person work that was tedious and necessary. Junior lawyers read leases. Junior analysts built the model nobody wanted to build, junior developers fixed small bugs, and junior designers resized things for a fortnight. The firm needed the work done, and the person doing it came out the other end knowing how the whole thing fits together.

Paul Saunders, a partner at Stewart McKelvey in Halifax, calls what’s happening to this the AI training conundrum. He remembers reviewing leases for due diligence as a junior and learning how contracts function by grinding through them line by line. Now, as he puts it, you can upload hundreds of leases into one of these platforms. Stanford’s David Freeman Engstrom has warned that firms will need to invent a new apprenticeship system or end up with lawyers who can supervise AI outputs without having built the judgement to know when those outputs are wrong.

For evidence that isolates the technology, the cleanest work is from Stanford, where Erik Brynjolfsson, Bharat Chandar and Ruyu Chen have been tracking American payroll records covering millions of workers. Their August update finds no sign of widespread job displacement across the economy, and finds employment of 22 to 25 year olds in the most AI-exposed occupations sitting 19% below where it would have been had it kept pace with their less-exposed peers, with no comparable gap for experienced workers.

As a result, the work that allowed people to become experts is disappearing from the bottom of every profession. The value of that work was always in what it did to the person doing it.

Tacit knowledge has one property that makes all of this dangerous. It lives in people, and when they leave, it goes with them.

The United States lost the ability to manufacture a component of its own nuclear arsenal, and nobody noticed for twenty years. It made a classified material for its W76 warheads at Oak Ridge in Tennessee, then shut the facility down once the last of those warheads was built. When the refurbishment programme needed the material again, the Government Accountability Office found that the agency had lost the knowledge of how to make it, because few records of the process had been kept and almost all the staff with expertise on production had retired or left. It cost sixty-nine million dollars in over-runs and delayed the programme by more than a year. The GAO’s own conclusion is the line worth carrying: assumptions such as “we did it before so we can do it again” are often wrong.

My guess is that expertise reorganises itself around liability.

A model can produce the work. It can’t be struck off, sued, disbarred, or have its name on the certificate when the thing falls down. Answerability is the part of expertise that can’t be automated at any price.

Which would mean the licensed professions get stronger while the general degree keeps weakening, because a professional body can remove somebody from its list and a university can’t take back a degree. Putting your name to something becomes the valuable act, because it is the only part of the job that still costs you when you get it wrong.

So the question worth asking has changed. What can you do has a cheap answer now, which is going to get cheaper still. What you are willing to be answerable for is the question I would put to somebody choosing subjects at sixteen, and to anybody sitting in a job wondering what to build next.

Every profession is going to have to decide how it makes its successors, and what they decide may turn out to be the question that shapes the future of work.

What to read next:

Working out what to aim at has always been the hardest part of a working life, and the obvious ways in are closing. That’s the problem my co-founder and I built Rumbo for. If you’re trying to work out where you fit, have a look here.