AI is ending older workers’ careers early, and it is coming for the well-paid ones first

The debate about AI and jobs has focused on graduates. New research suggests it should also be looking at people in their late fifties, CNBC reports.

Workers aged 55 and over in AI-exposed occupations are now exiting work at higher rates than before ChatGPT launched. The finding comes from Geoffrey Sanzenbacher at Boston College’s Center for Retirement Research.

The crucial detail is where they land. The increase shows up as unemployment, not retirement, meaning these people are out of work and still looking.

How the research works

The study pairs US Current Population Survey data with an AI exposure index from Tufts University’s Digital Planet Initiative. Exposure measures how well AI can perform a job’s tasks, not simply whether the job is doomed.

The scores are revealing. Web designers, web developers, database architects, computer programmers, and data scientists sit at the very top, while mining operators, orderlies, and painters sit at the bottom.

The design compares the period before ChatGPT’s November 2022 launch with the period after. That matters, because exposed and unexposed jobs already behaved differently.

The reversal

Here is what makes the result striking. Before ChatGPT, older workers in AI-exposed jobs were significantly less likely to leave employment than their peers.

That was the expected pattern. Desk jobs are less physically punishing, better paid, and easier to stay in for longer.

Since the launch, that advantage has largely been wiped out. The exit rate for exposed workers rose enough to offset the head start they used to have.

The numbers, job by job

The model predicts how much each occupation’s exit rate shifted. Painters, at the low-exposure end, saw a rise of roughly 2%.

Computer programmers saw an increase of more than 25%, with predicted exits climbing from 8.7% to 11.1%. Accountants and auditors rose about 22%, from 9.9% to 12.1%.

So the damage is landing on the educated and well-paid. Workers in exposed jobs are more likely to be white, more than twice as likely to hold a degree, and earn around $1,410 a week against $869 for the unexposed.

That inverts the usual story about automation. This wave is not starting on the factory floor.

Three ways it happens

Sanzenbacher sets out the mechanisms. Automation can simply replace the worker, pushing them into unemployment or out of the labour force.

Pressure to adopt the technology can also drive people out. Some workers would rather leave than learn a new system late in a career, a pattern researchers observed when personal computers spread.

The third path runs the other way. Higher productivity could raise wages and let workers focus on more engaging tasks, extending careers rather than ending them.

The adoption gap suggests which way older workers are leaning. Around 18% of workers aged 50 to 64 use generative AI, far below rates among people in their thirties and forties, and children are picking it up three times faster than adults.

Attitudes track that. An AARP survey found 18% of over-55s saw AI purely as an opportunity, while 28% saw it purely as a threat.

What the research does not say

The effects are modest in absolute terms, and the author is careful about it. He warns that this is an early analysis and that caution is in order.

Two confounds could cut either way. Government research cuts may have hit AI-exposed jobs hard, exaggerating the effect, while the current AI startup boom may be propping up demand for data scientists and understating what happens when that boom stops.

And exposed jobs are still the safer ones overall. Even after the increases, programmers exit work less often than painters do, because the job is less physical and better paid.

The honest reading is narrowing, not collapse. AI is closing the gap in career length between well-paid and low-paid work, from the top down.

Why it matters more than it looks

Losing a job at 58 is not the same as losing one at 28. Involuntary separation late in a career frequently becomes an early retirement nobody chose, on a pension nobody planned.

The policy collision is direct. Governments keep proposing that people work longer to shore up retirement systems, while the technology may be quietly pushing them the other way.

This also complicates the industry’s reassurances, including Sam Altman’s view that an AI jobs apocalypse is unlikely. An apocalypse is not required to wreck a specific 59-year-old accountant’s last working decade.

Employers are already restructuring around the technology, as when GM cut 600 IT workers in an AI skills swap. Experience is a poor defence when the skill being swapped in is one you are not expected to have.

TNW has written about the people left behind by tech layoffs and AI hype, and this study puts numbers under that. It is also why proposals like giving the public a stake in the AI companies keep gaining ground.

The people most exposed did everything the last forty years told them to do. They got the degree, took the desk job, and avoided the work that breaks your back, and it turns out that is precisely the work the machines learned first.