The AI job apocalypse is coming. Or is it?

Open this photo in gallery:

Is artificial intelligence the salvation of mankind, or its downfall? Are markets justified in bidding up AI stocks as high as they have, or is it all a bubble? Will we look back on the hundreds of billions of dollars the AI titans are currently investing in capacity as having laid the foundations for a new age of prosperity, or as an especially destructive example of the herd instinct, this time among tech billionaires?

At the moment, the best answer to questions like these is: Yes. Nobody knows anything, but everyone is certain about everything. Oceans of hype wash across social media, whether hailing the coming revolution or bewailing it, sometimes written by the same people.

There is one thing, however, on which there is widespread, if not universal agreement: AI will lead to massive, extinction-event-level loss of jobs, especially in the knowledge economy – precisely the university-educated, service-industry, symbolic-analyst professions that were supposed to inherit the Earth, insulated from the fluctuations of inventory, immune to globalization.

And not only them. In the more extreme formulations, AI will kill every kind of job, not only the virtual but the physical as well – a world, literally, without work. AI will be smarter than we are, quicker, stronger – better at everything, at a fraction of the cost, without need of vacation, rest or employee health benefits. It will create great wealth for those who control it – for as long as they do – but will cost everyone else their livelihoods, and as important, a sense of purpose. Society will be consumed with redistributing income; individuals, with finding meaning in their lives.

And it will all happen so fast! This is the other thing on which there is widespread agreement. The world might have endured previous industrial revolutions, but they were neither so universal – every job made obsolete, not just some – nor so instantaneous: millions upon millions laid off, more or less at the same time. Even if an answer to the jobs problem could be found in the long run, the short run will be so painful, and costly, as to be overwhelming.

Mind you, there’s very little evidence of this happening, as yet. Unemployment across the OECD remains at about 5 per cent, where it has been for the last three years – and lower than at any time in the previous two decades. Still, there are reports, anecdotes, warning signs: a big law firm laying off hundreds of support staff here, a chemicals giant there.

The past couple of weeks have seemed especially portentous, with the release of upgraded versions of the two leading AI chatbots, OpenAI’s ChatGPT and Anthropic’s Claude, together with variants dedicated to writing computer code (called, respectively, Codex and Claude Code). Indeed, OpenAI boasts that its latest update was itself written largely by AI: The model, it says, “was instrumental in creating itself.” The news caused a massive selloff in software stocks, similar to those that have hit the wealth management, data services, tax preparation and other sectors.

The situation has been likened to the eerie sense of disquiet we all felt in the early weeks of 2020: First you heard about those people falling ill with some mysterious disease in China, then you thought maybe you should cancel that trip to Europe, the next moment we were up to our ears in it. So why wait till more evidence comes in now? No one likes to be caught unaware, or to be accused of complacency. And besides, there can be a certain, shall we say, convenience to it. A firm that is laying off a fifth of its staff firm may find it easier to attribute it to “getting ahead of the AI curve” than to declining demand for its product – or the mistakes of management.

Anthropic’s release of AI tools for lawyers prompts massive legal-tech sell-off

There are benefits to playing Cassandra, not least to people in the AI business themselves, who have taken to issuing ever direr warnings of the potentially apocalyptic effects of their wares, even as they are developing them at full speed. And why not? Not only do you display a fetching concern for the public good, but it’s a great way to advertise the power of your product – in the tradition of those movie posters warning “Not suitable for the faint of heart!”

Let’s leave the broader fears about AI – everything from how it might be used by terrorists or other bad actors, to the potential for catastrophe in seemingly innocent prompts (the quickest way to “find a solution for climate change” might be to eliminate humanity), all the way to the full-on, open-the-pod-bay-doors-HAL, superintelligence nightmare scenarios – to one side for the moment: Not because they’re not serious, but because they’re so unknowable.

But what about the mass unemployment storyline? Should we be preparing for that particular disaster? Are we really likely to see, as claimed, half of all entry-level white-collar jobs wiped out within five years? A global job market collapse by next year? A future in which “no one will work”?

Anything’s possible, I suppose. But there are plenty of reasons to doubt it. There will obviously be some jobs lost, possibly a great many, in certain sectors. But the number of jobs lost may not be anything like what has been imagined, and they may well be offset by the numbers of jobs created – not just in the long term, but even in the short.

We should be aware, first, of the assumptions embedded in the typical disaster scenario. It assumes, first, that AI continues on the kind of exponential curve of improvement on which it has seemed to be until now – or even accelerates, as the idea of AI building itself suggests. There are knowledgeable people who contest this, arguing that the gains from scale have already been exhausted (you can only download the entire internet once) and the gains from algorithmic improvement are harder to extract.

Opinion: AI won’t take your job, but the AI bubble might

Second, it assumes that companies are quick to invest in AI, and quick to find ways to capitalize on their investment. History suggests that is unlikely. Businesses are often loath to discard previous technologies and ways of doing things, until they absolutely have to. The old technologies are proven; the new are uncertain. It’s costly, disruptive and risky. Even after they have made the jump to a new technology, it often takes a long time to figure out how to use it to full advantage. It took years before electrification produced widespread productivity gains; the same applied to computers, or the internet.

But all right, presumably the layoffs follow, eventually, as AI is fully phased in. Why should we believe these will not be as calamitous as feared? The simplest reason – and the one that has been observed in every previous wave of technological change – is rooted in the productivity gains to which the introduction of AI, and the jobs it eliminates, gives rise.

Higher productivity means lower costs; lower costs lead to lower prices; lower prices (along with higher wages, which tend to track productivity) mean higher real incomes; higher real incomes can be spent on other things, creating jobs in other industries. Which things? What industries? Who knows – but there’s never any shortage of things for people to spend their money on. Consumer wants are limitless, as is the ingenuity of entrepreneurs in filling those wants. Who knew they needed an iPhone 20 years ago?

But AI will be better than us at everything! Surely this time is different: The job loss, surely, will be across the board, rather than sectoral. Not necessarily. AI’s technological advances may not cost workers their jobs, but enhance them, allowing each to produce more with the same effort. If each worker is more productive with AI, it doesn’t automatically follow that a company will try to produce the same amount with fewer workers: It may try to get its existing work force to produce more. It may even – stay with me here – hire more workers.

How AI is both hurting and helping the next generation of workers

The phenomenon is known as the “Jevons paradox,” after the eminent 19th-century British economist William Stanley Jevons. Technological change had made steam engines more efficient in their use of coal. And yet, he noticed, this had not led to a reduction in total coal consumption, as had been assumed, but rather to an increase in it. Why? Because as steam power became cheaper, people found new uses for it that would have been beyond their reach before.

The same has been observed in all sorts of modern settings. Perhaps the most famous example: The advent of the spreadsheet app did not eliminate the need for accountants. Rather, it made them capable of doing all sorts of tasks they could not have done previously.

Before, they were employed keeping track of columns of numbers on paper ledgers; any change to one required hours of labour to update the rest. After, they could perform much more sophisticated forecasts and complex financial analysis.

And not just for the sort of large corporation that might have been able to afford such things in the past, but for small businesses as well. Result: Total employment in the profession expanded, rather than contracting.

There are other reasons to believe that human labour will survive the advent of AI. One is a kind of comparative-advantage argument: Even if AI is better than we are at everything, in absolute terms, it will still pay for it to specialize in the things that it is “most best” at, and leave the rest to us. Or to combine our efforts, in ways that best harness the relative strengths of each.

Another is simply that we have a particular interest in our own species. It has been 30 years since a computer first beat the greatest living human chess player. Yet the marquee event in chess is still between two unaided humans; competitions between computers remain curiosities. Likewise, we still want to watch a field of humans in the 100-metre dash, notwithstanding the invention of the automobile. While we also watch humans racing in cars, will we be as interested in watching a race between driverless cars? I doubt it.

We’ve had machines that can make everything we need for centuries. Yet we still want, and are willing to pay for, things that have been “made by hand,” whether a piece of furniture, an article of clothing, or a meal. The camera did not eliminate painting, nor did the movies kill the theatre. So long as humans survive AI, so will human labour. I think.