But artificial intelligence seems poised to follow the opposite trend. AI’s ability to handle office tasks like copywriting and data analysis seems more likely to hurt white-collar jobs than jobs where practical skills remain paramount.
In a way, “AI is turning back the clock,” he says Bryan SeegmillerKellogg assistant professor of economics.
Research by Seegmiller and colleagues supports this prediction. They used large language models to analyze nearly 200 years of data on technological innovations and occupations. And they found that as the use of artificial intelligence evolves in the workplace, the relative demand for white-collar jobs declines, with the jobs attracting a larger share of workers.
The team included Huben Liu, a Kellogg PhD in finance. Dimitris PapanikolaouKellogg professor of economics. and Lawrence Schmidt at the MIT Sloan School of Management.
The researchers’ findings extend a more recent pattern, beginning in the late 20th century, where technology has increasingly replaced the cognitive tasks common to white-collar occupations. “This is a trend 50 years in the making,” says Seegmiller.
This does not mean that college students should immediately drop out of school and train to become plumbers. But they should zero in on the skills that AI can’t yet replace and take advantage of AI to perform better in other parts of their jobs, Seegmiller adds.
“[You need to] have a healthy appreciation and respect for technological forces,” he says, “and understand the forces that work for you and that work against you.”
Learning from the past
Seegmiller and his colleagues previously found that when artificial intelligence directly replaced most jobs for jobs from 2014 to 2023, it reduced the demand for human workers in that field. But if AI only took over a subset of an occupation’s tasks, the technology’s effect on employment was actually positive.
The researchers wondered if this pattern applied more broadly to other technologies throughout history. If so, it could provide a frame of reference to predict what might happen in the labor market as AI improves.
“We always want to know how things are going to turn out,” says Seegmiller. “The only thing we can learn from is the past.”
For the current study, the team collected data on technology patents from 1850 to 2024, covering everything from steel production methods to e-commerce systems. They also used a large language model to generate job descriptions for jobs listed in the US Census during the same period.
They then compared the text from the patents to the job descriptions in each decade using natural language processing techniques they had developed in previous research on the impact of technology on the labor market. Whenever the description of a patented technology and a job were closely aligned, that job was considered “exposed” to the technology. In other words, technology had the potential to replace a human worker for this task.
As you might expect, manual jobs were by far the most exposed to technology from the mid-19th century onwards. Then, around the middle of the 20th century, a major shift occurred: thanks to advances in computers and IT, cognitive jobs became steadily more technology-exposed, although manual jobs were still the most exposed overall.
In an analysis from 1910 onwards, researchers found that greater exposure to technology generally harmed the employment growth of a job. However, if exposure was limited to a small subset of tasks, it actually benefited employment growth for a given job because it allowed workers to focus their attention on other tasks.
Overall, the results confirmed that the historical forces driving the effects of technological change on jobs have important parallels to how AI is likely to affect labor markets. As Seegmiller notes, “While AI’s capabilities may be unprecedented, they are similar to earlier technologies in some important ways: the ability to reallocate effort toward remaining unexposed tasks within a job always helps reduce negative impact, and interpersonal skills always appear less exposed.”
A reversal of trends
The team then estimated the net effect of exposure to technology and innovation across different groups of occupation types.
They found that, throughout the 20th century, higher-paying jobs generally benefited from technology—as did occupations that required more training and jobs that employed more women, such as clerical and administrative jobs. In contrast, lower-paid “middle-skilled” jobs in trade, such as those in transportation and manufacturing, which typically employed more men, suffered.
However, when the researchers used a mathematical model to simulate how artificial intelligence would affect different professions over the next 5 to 10 years, the resulting picture aligned with a newer trajectory. He predicted that AI will reduction the relative demand for jobs. In this sense, their model anticipates “a reversal of previous trends,” Seegmiller says.
Specifically, it predicted that demand would decrease for jobs requiring a high level of education compared to jobs requiring less education—while demand for jobs requiring an average level of education would decrease more. It also predicted a drop in demand for traditionally higher-paying jobs, such as clerical, technical and managerial jobs. Finally, the demand for jobs that tend to have a higher percentage of women will decrease compared to jobs with a higher percentage of men.
This does not mean that the total number of jobs will necessarily increase. Instead, this prediction suggests that, of the jobs that remain in a post-AI world, a larger portion of them will likely fall into this category.
This AI-driven reversal of past trends reflects an essential difference between AI and other technological changes in the past. “While previous technologies have always exposed a mix of manual and cognitive work, AI technologies are really lasering in on knowledge-focused work,” says Seegmiller.
No easy answers
While the response of some employees may be to consider moving to a profession less exposed to AI, Seegmiller advises people to approach the study’s findings with caution.
The model’s predictions are like “painting with a broad brush,” he says. And even in professions hit hard by AI, “there are ways to differentiate yourself.”
For example, as AI automates some tasks, workers could focus on building skills that require human contact, such as interpersonal communication, collaboration and creative thinking. Throughout the study, very few interpersonal tasks have been displaced by technology, and Seegmiller believes this will continue to be the case in general, “even if ChatGPT can pretend to be a therapist.”
Workers can also boost their productivity by using artificial intelligence to perform simpler tasks for them. This could allow workers to “focus on where our added value is,” he says. “Maybe it’s using these tools to be more productive in generating ideas or seeing the big picture.”
However, the researchers acknowledge that there are no easy answers to how workers should deal with the seismic changes ahead.
“There’s no golden ticket here,” says Seegmiller. But by carefully considering how technology can hurt and help, you can “make informed decisions about how you invest in yourself.”
