The AI problem we feared is not the AI problem we got
Folk memories can be powerful things. Perhaps the most visceral shared image we have of technology’s impact on jobs involves the Luddites of 19thcentury England: skilled workers abruptly turfed out of their jobs, who returned under cover of night to destroy the machines that displaced them. Little wonder that, as our latest machine age has dawned, the question on many people’s lips has been “Will AI steal my job?” But the Luddites’ story might have led us to worry about the wrong thing.
In many other (less famous) historical examples of automation, the important driver of change was not exit, but entry. Occupations threatened by technology shrank over time because fewer young people were hired into them, rather than because existing workers were ejected from them. This was true for Victorian boot-making, for example, where research by academic Hillary Vipond shows the arrival of boot-sewing machines led to a “collapse of entry into artisanal boot-making” rather than the mass displacement of artisans.
This is also largely how technology helped to hollow out routine manual and cognitive jobs over the past half century. A study of labour market flows in the US between 1976 and 2018 found that, outside of recessions, outflow rates from routine employment were pretty unremarkable. “Instead, declining routine employment overall is largely due to declining inflow rates for those who used to find employment in these occupations,” the authors wrote.
As for today? Despite fears of a jobs “bloodbath” and the occasional eyecatching example, AI has not led to many lay-offs. Even in the US, where employers tend to be more trigger-happy about firing people, the lay-off rate remains very low by historical standards.
Yet in occupations most exposed to AI, such as software development, there are clear signs that displacement is playing out, yet again, via entry rather than exit: entry-level hiring is falling even as demand for seniors remains firm.
This is not just a US story. A new paper from Denmark, which draws on rich company-level data, finds that businesses that adopt AI depart from the employment path they were previously following. The effect is concentrated in smaller firms, which continue to increase their headcount, just not as fast as before. As the researchers put it: “the adjustment runs through reduced hiring, not lay-offs. The stock of recently hired workers shrinks steadily, while long-tenured workers are retained, if anything, for longer.”
The same trend is evident if you look under the hood of the recent decline in global call centre employment. A study by Revelio of data from 12 middle-income countries concluded the industry was shrinking “because hiring slowed, not because separations picked up”.
It is still early days. But if these trends continue, and the primary margin of adjustment in the AI age does prove to be entry rather than exit, what does that imply for policymakers?
Attention will need to shift from data on lay-offs to longitudinal studies that reveal whether young people find other jobs instead, and whether those paths prove better or worse. There is no need to be fatalistic about this: historically young people have been adaptable, particularly when it comes to exploring the opportunities of new technology.
But there are risks too, especially while education systems continue to lag behind employers’ demands. The worstcase scenario would be that some young people fail to find a foothold in the labour market at all. This, too, has historical precedent. The paper on the erosion of routine employment in the US found that the decline in inflows to those occupations also accounted for a substantial proportion of the rise in nonparticipation in the US labour market over the previous two decades.
What policies would help? Not necessarily the ones we once thought. I have written admiringly about Sweden’s furlough scheme for life-long learning, for example, which is designed to help workers retrain and transition into different work. That is still a good policy. But it is reserved for people who are between ages 27 and 63, and therefore does nothing for the very people who — it now appears — are at the sharpest end of AI disruption.
They will need other interventions: better links between education and employers, for example, and more security and support for those who want to start their own businesses.
Of course, a recession could well trigger a more abrupt and widespread jobs shake-out. But for now, it is time to rec- ognise that the problem we feared is not the problem we’ve got. And if we had paid more attention to history, that needn’t have been a surprise.
Comments
Post a Comment