Such is the hype surrounding generative AI that the last few years has seen no shortage of attempts to understand the impact the technology might have on the labor market. The latest effort comes from Harvard, with researchers examining 100 years of technological disruption of the labor market to try and understand the levels of “occupational churn”, which is basically the share of the labor market each profession held at various points.
The analysis found that the digitization covered by the personal computer, internet, and smartphone era was actually remarkably stable in terms of jobs, with relatively little churn witnessed between 1990 and 2017. This has begun to change in recent years, however, with this driven, at least in part, by recent developments in AI.
A changing landscape
The paper, co-authored by OpenAI board member Larry Summers, highlights the “automation anxiety” many experienced in the 2000s and 2010s, driven in large part by Oxford University’s Frey and Osborne’s influential paper suggesting that up to half of jobs could be automated in the next decade.
This disruption didn’t really occur, with that era instead seeing a slow pace of churn in the labor market. The researchers show that this has been changing from 2019 onwards, however.
“We find that—contrary to popular imagination—the pace of labor market disruption has slowed in recent decades,” the researchers explain. “The changes in the structure of US employment at the end of the nineteenth century were greater than in any decade of the digital era, including the most recent one.”
Key trends
They identify four key trends that capture the changing nature of disruption since the generative AI boom took off. The first revolves around the polarization of the job market, with a general hollowing out of middle-income work. The research identifies growth in jobs at both the bottom and the top of the wage distribution, with particular growth in well-paid, high-skilled jobs in recent years.
“The trend people were worried about in the 2000s was the downward ramp,” the researchers say. “That meant low-paid jobs were growing but middle- and high-paid jobs were not. It’s only in the late 2010s that we see an upward ramp, with mostly high-paid jobs that are growing.”
A lot of this growth was seen in STEM-related fields. For instance, whereas STEM jobs represented 6.5% of the labor force in 2010, this had grown to around 10% by 2024, which is a growth of around 50% in a relatively short space of time.
Despite a general growth in low-end jobs, the data also shows, however, that low-paid service work has flatlined. It’s a sector that grew considerably from the 80s onwards, but it’s declined considerably since 2019. The researchers believe AI is one possible explanation, but the tighter job market and higher minimum wage legislation are also likely to be at play too.
“It doesn’t look like many of these jobs are coming back,” the researchers explain. “The ones that have returned are in food service, personal services like manicurists and hairdressers, medical assistants, and some cleaning jobs.”
There was also a large decline in retail sales jobs, with technology the biggest factor in this fall. The researchers show that retail sales jobs were around 7.5% of the labor market in 2013, but just 5.7% in 2023, which is a fall of around a quarter. This was largely a continuation of a trend that was already underway due t the rise in e-commerce, but the introduction of predictive AI has played an accelerating role.
A key place
While previous fear around AI wasn’t justified, the researchers believe that AI is something that we should all be thinking about today, regardless of what we do for a living. While it can be empowering and augment many tasks, they believe it will also mean that many activities will simply be done by machines rather than humans in the future.
For knowledge workers, the researchers believe that it is most likely that a short-term boost to productivity will be seen, with displacement both a longer-term phenomenon and one that sees workers using technology usurping those less adept at doing so.
“The impact of AI is likely to be widespread and long-lasting, fitting the mold of past GPTs,” the researchers explain. “However, history teaches us that even if AI disrupts the labor market, its impact will unfold over many decades.”?
This change is likely to be exacerbated during any upcoming recession, where employers may start to demand a lot more from knowledge workers and be more demanding in their expectations. With some economists predicting recessions in much of the western world in 2025, that may offer scant consolation for knowledge workers everywhere.
