The AI Tidal Wave Doesn’t Have to Drown Workers

The effect of the latest generation of AI tools on the labor market still needs time to play out. But new research by Kellogg’s Bryan Seegmiller and Dimitris Papanikolaou hints at how this might unfold. The researchers, along with their colleagues, measured workers’ exposure to AI and its impact on employment across a wide range of industries.

They found that AI’s effect on the labor market in the past decade wasn’t as straightforward—or as dire—as one might initially assume.

On the one hand, the more exposure a job had to AI, the more likely it was that demand for that job would go down. On the other hand, workers in jobs with higher AI exposure were often able to make more adjustments, redirecting their attention to other relatively unexposed tasks and performing better in those areas. And if their company used AI heavily, the firm tended to increase its overall productivity and expand its workforce, similarly to how “a rising tide lifts all boats.”

For instance, because these opposing effects balanced each other out for highly paid positions, the change in demand for these occupations was nearly flat overall—even though these highly paid positions tended to have high direct exposure to AI.

“There are countervailing forces [with AI], some that work in your favor and some that work against you,” says Seegmiller, an assistant professor of finance.

To survive the AI boom, the findings suggest, workers may need to shift their responsibilities to tasks that complement AI’s growing role in their occupation. People might consider, for example, spending more time on big-picture thinking, communication, and collaboration.

As AI continues its forward march, having this flexibility is going to be important “to mitigate its negative effects,” Seegmiller says.

Painful adjustments

In previous research, Seegmiller and his colleagues studied how technological advances affected jobs in the late twentieth century. Occupations that paid mid-level wages got hit hard, they found, as robots, software, and information technology upended some industries and lowered demand for those workers.

While not everyone lost their jobs outright, some people struggled to adapt to the new requirements of their workplace, such as learning to use unfamiliar software. “This type of reallocation can be really painful,” Seegmiller says.

Seegmiller and Papanikolaou, a professor of finance at Kellogg, worked with Menaka Hampole, a Kellogg PhD graduate now at Yale University, and Lawrence D.W. Schmidt at MIT to investigate whether similar upheavals have occurred or might occur with the emergence of AI.

AI is “the next big thing that’s going to be shaping the labor market over the next couple of decades,” Seegmiller says.

The team tackled the question by analyzing about 58 million LinkedIn profiles collected by the database provider Revelio Labs, focusing on U.S. jobs held from 2014 to 2023. Based on information in the resumes, they were able to figure out how firms deployed AI to perform particular functions. For example, one J.P. Morgan employee’s job description mentioned using AI software to forecast risk and fraud in lending businesses.

Next, the researchers compared the tasks that AI performed with all the tasks described in O*NET, a database that provides comprehensive information about job duties and skill requirements for each occupation. If the description of an AI function was similar to a task typically performed by people, then that task was considered to be “exposed” to AI—that is, AI could likely replace a human worker for that task.

High pay, high exposure

The team found that AI exposure tended to be higher among higher-paying, white-collar jobs, peaking at the ninetieth percentile of income.

Highly exposed occupations included financial specialists, life-science technicians, chemical engineers, and credit analysts. In contrast, jobs that involved manual labor—such as bartenders, janitors, and cooks—were the least exposed.

Simple intuition might suggest that jobs with higher exposure to AI would be the most likely to get displaced. But the researchers’ model showed that the dynamic between AI and employment was much more nuanced than that.

Some jobs involve a wide range of tasks that have different levels of exposure to AI, or high “variance.” Having a high level of variance—meaning that the job involves both tasks with high AI exposure and tasks with low AI exposure—reduced the likelihood of being displaced because workers had enough wiggle room to adjust their responsibilities. For instance, with AI replacing one of their rote tasks, workers might be able to spend more time strategizing or forming important business relationships instead.

Seegmiller experienced this shift firsthand when he recently used AI to code an economic model, a task that normally would have taken him several hours without the help of AI. He wasn’t just left sitting around; his high-variance job as a professor gave him plenty of options for other work he could do instead, such as writing papers and thinking about how to explain his research effectively. The ability to “focus on things that I’m now more productive in, because I don’t have to spend time on other stuff, is actually good for me,” he says.

Furthermore, the extent to which a firm adopted AI also affected employment. Again, the team turned to the LinkedIn data to measure AI adoption by analyzing how often employees at each firm mentioned AI. They discovered that firms that were more AI-intensive and integrated AI into jobs more often saw a boost in overall productivity, allowing the company to expand its workforce.

Opposing forces

When the researchers took all these factors into account, they found that the net effect of AI on employment was close to zero, particularly for highly paid jobs.

Though some of these workers’ tasks were replaced by AI, many of the jobs also tended to have high variance, and so workers were able to easily shift their focus onto other tasks. In addition, they were more likely to work at firms that used AI often enough to ramp up overall productivity and employment growth. In fact, for jobs at the very top of the income scale, “employment share”—that is, the fraction of all jobs held at the time—saw a slight increase.

The effect of AI, “then, ultimately doesn’t just depend on whether some of your tasks get automated,” Seegmiller says. “It depends on the sum of these forces.”

Still, the net effect on employment share was negative for certain highly paid jobs, including in business, finance, and engineering. For instance, business and financial occupations saw a 1.9 percent decrease in employment share over five years. In architecture and engineering, there was a 2.6 percent drop.

The team also saw declines in employment share in some lower-paid, manual jobs. In food preparation and serving occupations, for example, most companies didn’t tend to use AI; so, though AI didn’t replace their workers, the companies also saw less AI-driven growth. Overall, this resulted in a 2 percent decrease in employment share in this field.

All in all, AI was a substantial driving force in changes to the labor market. When the team examined employment growth across all occupations during the study period, AI-related factors explained about 14 percent of the changes.

Embracing soft skills

If the latest advancements in large-language models like ChatGPT are any indication, AI’s role in society will continue to grow. What does that mean for the future of workers?

For one, jobs that involve sifting through large amounts of text, like legal occupations, might see more of their tasks automated, Seegmiller says. And software engineers might offload writing code to AI. The outcome for these kinds of jobs will ultimately depend on workers’ flexibility and how their companies decide to use AI. For example, engineers could adapt by shifting to more high-level strategy instead.

As workers brace for AI-driven changes, they may want to reevaluate how they think about their jobs. People should consider, Seegmiller says, how to work “in conjunction with AI rather than in competition with AI.”