Don’t fear the AI ‘jobpocalypse’ – Financial Times

Welcome back. Fears of an AI jobs apocalypse are growing. At Davos, IMF managing director Kristalina Georgieva said the technology would hit labour markets like a “tsunami”. London mayor Sadiq Khan warned last month of a “new era of mass unemployment”.

The current state of the jobs market in advanced economies adds to the rising anxiety. Joblessness is edging up, entry-level positions are getting harder to find and companies are reporting a slew of tech-related redundancies.

So for this week’s contrarian take, I deliberately pick holes in the gloomy narrative — and hopefully give workers and recent graduates a lift in the process.

I’ll begin with the impact of AI on labour markets so far.

It is true that labour markets in services- and knowledge-based economies have cooled since ChatGPT was released in November 2022. But correlation isn’t causation.

Focusing on the US, where AI investment and adoption have been most prominent, the rise in the S&P 500 and corresponding drop in job openings from around 2023 has been cited as evidence of a tech-linked hit to workers.

But if we zoom in, that assessment falls apart. We can see that the drop in vacancies predates the release of OpenAI’s large language model (LLM). Instead, the fall in job postings dovetails with a period in which the Federal Reserve pressed the brakes on the US economy by raising interest rates by 5 percentage points.

What’s more, because of the post-pandemic surge in openings and hiring, the subsequent reduction appears to reflect a normalisation.

These broad trends in hiring and central bank rate increases hold in many G7 nations during the same period.

There are non-AI drivers of labour market weakness, too. For instance, in Britain the rise in youth unemployment is, in part, linked to the Labour government’s policies, such as payroll tax increases.

Entry-level employment has also weakened in recent years. But, again, this could be largely cyclical, notes Vicky Redwood, senior economic adviser at Capital Economics.

“Inexperienced workers are often first to suffer during a hiring slowdown, and factors such as degree ‘inflation’ could be affecting graduate jobs,” she says.

In the euro area, a significant rise in the share of twenty-somethings with university education between 2019 and 2024 may in part explain why graduate unemployment there has risen more than overall joblessness.

“A jobs slowdown in the IT sector is partly a correction of the hiring boom following the pandemic,” adds Redwood. “But beyond some very specific adverse impacts in occupations like software programmers, the impact of AI on graduates is much smaller than is made out.”

Likewise, corporate headlines of tech-linked job losses should be read with caution. In the US, a report by Challenger, Gray & Christmas counted 54,836 job lay-off plans last year that were linked to AI. That only amounts to 4.5 per cent of total job-cut announcements.

“Linking job losses to increased AI usage rather than other negative factors like weak demand or excessive hiring in the past conveys a more positive message to investors,” points out Ben May, director of global macro research at Oxford Economics.

Finally, though some studies point to weakening job prospects in AI-exposed sectors, the sectoral evidence remains mixed (more on why below).

A recent academic paper drawing on monthly US vacancies data found no significant effect on job openings or total jobs in sectors more exposed to AI since the technology emerged.

Reflecting this, US analysis by the Yale Budget Lab finds that the overall jobs composition is changing more quickly than during previous periods of technological change, but not markedly so.

Indeed, in the US and euro area, employment in white-collar roles — including professional, management and office positions, which are often considered to be at risk of AI-driven redundancies — has increased overall since ChatGPT was released.

Of course, it is still early days. Even so, there are a few counterpoints to the pessimism around the future impact of AI on jobs at all levels.

David Deming, a labour economist and professor at Harvard University, notes that historically, young people have tended to fare best during periods of technological change.

“Over the last century, disruptive innovation has generally favoured the young and the well-educated,” he says. “Today, young people’s relative tech fluency and capacity to retrain mean they can adapt to new ways of doing things.”

Corroborating this, a 2021 US study by Péter Hudomiet and Robert Willis showed that older, less tech-savvy workers were hit hardest by the IT revolution that began in the 1980s. They faced pay cuts, earlier retirement and transfers to less intensive jobs.

Surveys suggest younger workers tend to use LLMs more than their older colleagues. Companies will need to tap the cohort’s expertise and build talent pipelines for senior positions.

Next, technological change has been the main driver of employment growth throughout history. A 2022 academic analysis, again based on US data, found that 60 per cent of workers today are employed in occupations that did not exist in 1940.

Tech-linked job creation occurs through several channels.

It can create new occupations directly, as did digital platforms, social media and video games. Today there is rising demand for roles focused on developing, training and governing AI systems. LinkedIn estimates that AI has generated 1.3mn new jobs globally between 2023 and 2025.

It can also create new jobs through specialisation, as is the case with robotics technicians and medical imagers. Lastly, it can boost discretionary demand in occupations, such as personal care, that emerge due to any income gains. (I recommend reading the latest edition of Sarah O’Connor and John Burn Murdoch’s newsletter The AI Shift on this topic.)

Sure, AI could be even more transformative than previous technologies, but that need not mean wide-scale redundancies either.

Most jobs in service-driven economies today are made up of a bundle of activities. Just as employment boomed in the computer era in part because machines automated repetitive elements of some roles and freed up worker time for more value-added tasks, AI can do likewise.

An analysis of millions of US job postings over the past five years by the Burning Glass Institute finds that even as AI automates certain skills, it is simultaneously elevating demand for others — often within the same roles.

“Occupations with high AI exposure tend to be knowledge work: roles rich in information processing, communication and analysis, where AI can both substitute for routine tasks and amplify complex ones,” notes the think-tank’s report Beyond the Binary. “AI isn’t sorting the labour market into winners and losers. It’s transforming job content across a broad swath of the economy.”

For instance, a project manager whose scheduling tasks are automated may now have wider strategic responsibilities. The financial analyst who no longer builds models from scratch might now interpret and pressure-test AI output.

The need in many jobs for human accountability, interpretation and interaction will remain.

This might help explain the mixed evidence of job losses in AI-exposed sectors so far.

The upshot is that the impact of AI on today’s jobs market isn’t as severe as the headlines make out. Nor is the future effect of the technology on employment destined to be negative.

This is not to suggest that AI won’t shrink or isn’t already shrinking some roles. Entry-level jobs, particularly those with a high administrative, clerical or routine component, are at risk.

Other positions with high automation exposure but little room for augmentation are too. For instance, in the US, computer programming employment has dropped since ChatGPT’s release, and the Bureau of Labor Statistics projects a further 6 per cent fall by 2034.

There will be disruption. The job losses may come before new roles are created, and there will be a shift in the emphasis of skills existing roles require.

But for workers and prospective labour market entrants, there is time to adapt. Companies cannot adopt and optimise AI immediately, and the technology is still developing.

AI-induced job losses are indeed a concern. But the bigger problem may be ensuring our education and skills systems enable workers to reap the openings that AI creates.