What happens when computers can do almost everything?
About fifteen years ago, Gareth Jones invited me to an event he was running where one of the speakers, a futurologist, said something that has stuck with me ever since. Within twenty years, he argued, there would only be two kinds of jobs left: jobs you couldn’t get a computer to do, and jobs it simply wasn’t worth getting a computer to do.
I remember thinking at the time that he was probably right, and fifteen years later I think he was remarkably close. Give it another five years and I suspect we will be living in something very much like the world he described, not because literally every job will have disappeared, but because the economic logic underneath a huge number of them will have changed. Once a computer can do something faster, more cheaply and increasingly well, the interesting question is no longer whether it can technically replace the person doing it, but why somebody would continue paying a person to do it.
That sounds frightening if your starting assumption is that the objective is to preserve roughly the same structure of employment we have now. I’m not sure it is.
Most of the conversation about AI and work still seems to begin with questions like which skills people should learn, which professions are safest and how we protect jobs. Those are reasonable questions, but they assume that the thing we are trying to preserve is the existing shape of work, when perhaps the more interesting question is what becomes valuable once the ability to perform large amounts of useful work becomes abundant.
One answer is breadth. Some people are naturally inclined towards being polymaths. I certainly am, but I also think it is something almost anyone can cultivate to some degree. Learning across disciplines, understanding technology and people, business and culture, design and psychology, and becoming able to move between different professional languages becomes unusually valuable in a world full of AI because the ability to perform individual tasks is becoming cheaper.
Somebody still has to understand what problem is being solved, recognise what matters, connect information from different domains and decide whether the thing produced is actually any good. The value begins to migrate away from performing every part of the work yourself and towards understanding how the parts fit together.
That is why being multilingual, both literally and professionally, becomes such an advantage. It is not simply that you know more things. It is that you can direct more kinds of intelligence, which means AI gives the generalist far more leverage than they had before.
The apparently opposite response is to go very deep: become exceptionally good at one thing, not simply competent or employable, and not just the person who knows the software slightly better than everyone else, but genuinely difficult to replace.
My friend Paolo Caldato is one of the people I think about here. There are people whose expertise is so deep that what you are really paying for is not the individual act they perform, but decades of accumulated pattern recognition. They have seen enough unusual situations that they recognise something before everybody else even realises there is something to recognise.
AI can absorb knowledge remarkably quickly, but knowledge and judgement are not quite the same thing. Judgement is built through encounters with reality, including all the awkward cases where the obvious answer turns out to be wrong, and a brilliant specialist who suddenly has a machine capable of carrying out huge amounts of supporting work may therefore become more valuable rather than less. The expertise remains concentrated in one person, while the amount they can do with it expands dramatically.
This creates a slightly odd shape to the future labour market. At one end you have people who are broad enough to direct across disciplines. At the other you have people whose expertise is so deep that their judgement remains scarce. Both are using AI, but in very different ways, and the uncomfortable territory may be the middle.
For a long time, a large proportion of professional employment has consisted of being reasonably good at performing repeatable knowledge work. You knew enough to do the job reliably, the organisation needed enough people doing it, and automating it was too expensive or too difficult to be worthwhile. There was nothing wrong with that arrangement, and it supported millions of perfectly useful careers, but it is precisely the territory AI is becoming very good at occupying.
That does not mean the middle disappears overnight, and it does not mean every generalist or specialist is suddenly safe. It means the source of value starts to shift. If execution becomes cheap, then being paid mainly for execution becomes a more fragile position, while breadth gives you leverage over systems and depth gives you judgement that remains scarce.
There is another route which sits slightly outside that argument altogether, which is to do something where the human interaction is part of the thing being bought.
Waiting tables, caring for people, hospitality, teaching, childcare, working with older people, coaching, or simply looking after somebody who is frightened, confused or lonely all have this quality to some extent. A robot may eventually be perfectly capable of carrying a plate from a kitchen to my table, but I am not convinced that means I want to eat in restaurants staffed entirely by robots, because part of the pleasure of going out is being around people. The exchange with the waiter is not an inefficiency in the system. It is part of the experience.
The same is even more obviously true of care. There are tasks within caring professions that machines will undoubtedly perform better than humans, and monitoring, lifting, scheduling, diagnostics and administration are all obvious candidates, but the person sitting beside you when you are scared is not simply an inefficient medical device. Human presence has value in itself, and I suspect it becomes more valuable rather than less as more of our administrative, commercial and professional interactions happen with machines.
These jobs may not necessarily be the highest paid jobs in the future, which is a separate problem and one we may eventually have to confront quite aggressively if the work we value most socially remains the work we reward least economically. But they produce something machines cannot simply manufacture more cheaply, because the human being is part of the product.
Then there is money
Of course, none of those answers solve the larger economic problem. If computers genuinely become capable of producing an increasing proportion of the economic value in society, then at some point we have to confront the fact that distributing purchasing power primarily through employment stops making much sense. That is where Universal Basic Income comes in.
I do not really think of UBI as an AI policy. I think of it as an economic response to abundance. If productivity and employment begin to separate, then income and employment probably have to separate to some extent as well, otherwise we arrive at the slightly absurd situation where society becomes extraordinarily good at producing things while an increasing number of people cannot afford to buy them because they were not required in the production process.
There are enormous questions around exactly how UBI works, what level it should sit at and how it interacts with taxation, housing and existing welfare systems, and I am not pretending those questions are simple. But the underlying problem seems difficult to avoid. If human labour stops being scarce, we cannot continue behaving indefinitely as though access to money must always be conditional on selling it.
There is another response which is much smaller, much less political and immediately available to almost everyone, which is simply to learn to need less money. I do not mean austerity as a virtue, and I am not especially interested in romanticising poverty or telling people to cancel Netflix and grow potatoes. I mean frugality as resilience.
I have not bought vegetable stock for years because I keep vegetable peelings and make it myself. It costs essentially nothing and is normally better than the stuff I would buy, and while that is obviously a tiny example, there is a broader principle underneath it. Every pound you discover you do not actually need to spend reduces the amount of your life you have to exchange for money.
That gives you room to manoeuvre. It makes it easier to take risks, retrain, work fewer hours, survive periods of uncertainty, choose interesting work over highly paid work, or simply refuse situations that make you miserable. In a stable labour market that is useful. In a period where the economic value of entire categories of work may change extraordinarily quickly, it becomes a form of insulation.
What work is for
I do not think any of this means people stop working. I think it may force us to untangle several things that we currently bundle together under the word ‘job’.
Employment provides money, but we also expect it to provide identity, status, purpose, social contact, structure, achievement and a sense that we are contributing something useful. There is no particular reason all of those things have to come from the same place, and historically they often have not.
Perhaps some people become extraordinary specialists and spend a relatively small amount of their time doing extremely valuable work, while others become generalists orchestrating machines capable of doing things that previously required entire teams. Perhaps far more of us spend time caring, teaching, making things, cooking, entertaining, volunteering, looking after children or older relatives, or simply being available to the people around us, without feeling that every useful thing we do needs to be converted into a job title before it counts.
That is the part of the AI conversation I find most interesting, because it is much bigger than productivity. If machines take over more of the work we once had to do, that does not automatically tell us what humans should do instead. It simply gives us the possibility of choosing.
The futurologist I heard at Gareth’s conference fifteen years ago was talking about jobs, but what interests me now is the question underneath that prediction. If computers eventually become capable of doing almost everything we currently consider economically useful, then the challenge is no longer simply finding things humans can still do.
It is deciding what we actually want humans to spend their time doing.
