The warning signs are piling up. The International Monetary Fund is worried about a fall in AI-related stocks, and the great American data centre build-out is starting to crack. What if the bubble bursts?
You might think that would at least be good news for employment: less money for AI, less automation. Economists, reported this week by American public radio NPR, suggest it could be exactly the opposite.
What past crises tell us
When the economy is doing well, companies hire, and they often hang on to fairly inefficient ways of working, because they have no urgent reason to change them.
When a crisis hits, everything changes. Sales fall, money becomes scarce, costs have to be cut. That is when companies make the choices they had been putting off: cutting jobs, reorganising, and replacing certain tasks with machines or software.
Work by American economists has shown this for several recessions. So-called "routine" jobs, made up of repetitive tasks, whether in the office or on the shop floor, disappear mainly during crises. And once the recovery comes, they do not come back. Economists talk of "jobless recoveries": activity picks up again, but the jobs that were cut are not recreated.
Another study observed that after the 2008 crisis, in the American cities hardest hit, companies had durably raised the skills requirements in their job adverts. The crisis had acted as an accelerator for changes that were already in the pipeline.
In 1942, the Austrian economist Joseph Schumpeter popularised a phrase that became famous: "creative destruction". In his view, capitalism progresses by constantly destroying old ways of producing and replacing them with new ones. Crises are the moments when this movement speeds up: the least efficient companies disappear, the others reorganise. It is an engine of progress over the long term. But for those whose jobs disappear, the destruction is immediate, and the creation often arrives elsewhere, later, for other people.
The "double whammy" scenario
Applied to AI, the reasoning is chilling. A crash would doubtless reduce investment in new data centres and ever bigger models. But existing AI tools would not disappear. They would even become cheaper.
Companies under pressure could then use them en masse to cut their headcount. For employees, it would be a double whammy: the economic crisis, and automation accelerated by the crisis.
Young people would be on the front line. They already are: in France, permanent-contract hires among 21-29 year-olds have fallen sharply in the occupations most exposed to AI.
Nothing is set in stone
Caution is needed. This is only a scenario, drawn from observing past crises, which did not all resemble the one that might be coming. Nobody knows whether an AI bubble will burst, or when, or with what force.
But this scenario has one merit: it is a reminder that the time to prepare workers, through training and solid protections, is before the crisis, not during it.
Key takeaways
An AI crash would not put an end to AI. It could even accelerate its arrival in companies, at the worst possible moment for employees.
The history of crises teaches a simple lesson: the great transformations of work do not happen gently, when all is well. They happen when all is not.