where are all the job losses?

This is an onsite version of our new AI Shift newsletter, which is free to read today. Sign up here if you are a Premium subscriber and get the newsletter sent directly to your inbox each Thursday. Standard subscribers can upgrade to Premium to keep receiving the AI Shift and enjoy more than 15 other exclusive newsletters. Explore all FT newsletters here.

Welcome to The AI Shift, a new weekly newsletter written by us: Sarah O’Connor and John Burn-Murdoch. Each week, we’ll pose one question about AI and the labour market, then answer it as rigorously as we can. We’re also hoping to learn from you about how AI is changing your own jobs and industries.

Got questions? We’re doing a special live Q&A at 1pm today over on FT.com — you can reach us here. Thanks to all those who’ve written in so far.

This week: is AI taking jobs, or not?

If you’ve been reading the news on this question, you could be forgiven for being . . . pretty confused. Recent headlines include: “The age of AI layoffs is already here”, “AI is not killing jobs” and “For Some Recent Graduates, the A.I. Job Apocalypse May Already Be Here”. How are people coming to such wildly different conclusions? Over to John for a dive into the data…

John writes

This really comes down to what scale of disruption we’re talking about. If the question is whether we are seeing widespread AI-driven displacement of roles across a range of sectors and countries, then the answer is “no”. Dozens of analyses — including from yours truly — have explored this over recent months using labour force survey data from the US, UK and western Europe and have consistently found no clear relationship between exposure to AI and trends in employment.

But if the question is whether there is any indication that generative AI may be displacing some types of work from specific slivers of the economy, then the answer ranges from “maybe” to “yes”.

The “yes” refers to online gig work, where we’ve had strong evidence for more than two years that freelance graphic designers and copywriters have seen reductions in both work volumes and rates of pay since the arrival of OpenAI’s ChatGPT. Gig workers are often left out of the labour market conversation, but this is a real and important finding.

The “maybe” is junior coding jobs in the US, where a fine-grained analysis by researchers at Stanford’s Digital Economy Lab found employment for young software developers is down by nearly 20 per cent vs late 2022. Studies from Harvard and King’s College London also found dips in junior hiring at companies that have been adopting generative AI, again with the strongest evidence in the tech sector. None of these analyses prove LLMs are responsible — junior tech hiring is noisy at the best of times, and the huge pandemic-era hiring binge in the tech sector began to unwind when the US started tightening monetary policy in March 2022, not when ChatGPT came out eight months later — but they are consistent with that potentially being the case.

Some content could not load. Check your internet connection or browser settings.

If the combination of nos, maybes and yeses is confusing, that’s partly because they aren’t just answering different questions, they’re using different data. The nos come from labour force surveys, which sample tens of thousands of people out of the whole economy, making them well-suited to spotting big macro shifts but too sparse to discern what’s happening in one subset of one sector. The maybes and yeses come from more detailed databases of millions of payroll and job posting records that hold a microscope to the labour market.

So Sarah, I’d love to get a better sense from you on: what are people actually saying about the role AI is playing? And what is happening outside the US?

Sarah writes

You’re right, John — it’s striking how US-focused the reporting on this question has been. I decided to look somewhere else: Sweden. As a small, open, tech-friendly economy, it’s the sort of place I would expect to see rapid AI adoption (fun fact: Stockholm is second only to California’s Bay Area in terms of unicorns per capita).

If you want to know what’s happening in the Swedish labour market, the unions are a good place to start. Unlike in the US or the UK, their reach is huge (about 70 per cent of workers are members) and they have access to high-quality information because they have seats on company boards. Unionen, the biggest Swedish union for white-collar workers, has about 700,000 members and calls itself “the largest white-collar trade union in the world”. When I got in touch with Tobias Brännemo, its chief economist, I was in luck. They had just surveyed their members about this exact question, and gave us an exclusive preview of what they found.

The results tell us three interesting things.

First, AI has indeed spread rapidly through the Swedish economy. Two-thirds of respondents said their company had concretely implemented AI in their operations (rising to 87 per cent of companies in ICT), up from 29 per cent in 2023 (the figures exclude those who say “don’t know”). The free-text part of the survey reveals lots of detail, from obvious use cases (creating marketing copy; generating code; translation) to the less obvious (analysing incident information to predict areas of accident risk; predicting electricity market needs; creating complete travel itineraries for clients.)

Second, in spite of all this activity, the net impact on employment seems to have been . . . nil. Among those who said AI had been implemented, about 80 per cent said there has been no impact on staffing levels as a result, while about 10 per cent reported an increase, and 10 per cent a decrease. While it’s only a snapshot survey, this data is useful because it asks about causation. As you say, John, many macro-level analyses rely on correlations.

That said, Unionen’s survey hints that the next phase of corporate AI adoption might have a bigger effect on jobs. Of the 56 per cent who said their companies had concrete plans for further AI implementations, the proportion expecting a reduction in staff doubled, to 20 per cent. It makes sense to me that after an experimentation phase, the next phase might be more focused on driving efficiencies. “Maybe we are in some kind of turning point, but to this date, we haven’t seen big effects,” Brännemo summed up.

The survey results are also a reminder that job disruption can happen outside the country in which a company is based. Some respondents (mainly in ICT) noted that AI had affected jobs within the company, but abroad, or that AI had replaced Swedish jobs which would otherwise have been offshored. The effect of AI on outsourcing is a really interesting question to which we’ll return in a future newsletter.

So what have we learned? Sarah

For now, I don’t think there’s strong evidence for AI-induced job losses, at least outside of certain contract types (freelancers) or roles (junior programmers). That’s not to say it’s a great time to be looking for a job, especially if you’re young. But correlation is not causation. There are many other factors impacting labour markets, from trade wars to the end of the era of cheap money. On that note, it’s worth raising an eyebrow when companies announcing lay-offs link them in vague terms to becoming “AI ready”, but without any corroborating detail. Let’s face it, it sounds a lot more dynamic than just saying the business isn’t doing very well.

John

The strongest “AI is displacing human workers” story you can tell from the data is that generative AI is displacing tasks, not jobs. The more a job consists of clearly-defined tasks, the more vulnerable it is. Freelancers are at the bottom of the ladder: a task is the job. Write some ad copy. Draw an image. Plus nobody has to get HR involved: the commissions just stop coming. Junior tech workers are a rung or two further up: tasks are well-specified and often self-contained, hiring and firing especially volatile. But most jobs — including less junior roles in tech — are not like these. They involve defining and refining tasks as well as performing them, all while considering the particular context of a project or firm, and going back and forth with people who have their own perspectives and priorities. Here AI isn’t displacing people, it’s assisting them.

Does your experience tally with what we’ve found? Next week, we’ll be doing a deep-dive into AI’s impact on productivity. If you have thoughts on that, or any other questions you would like us to explore, you can reach us at aishift@ft.com.

where are all the job losses?

Tags: