Most workers think AI will affect their jobs: Get the latest from Data Lab.

Linking generative AI exposure to worker sentiment: It’s complicated

June 10, 2024

Share this

AI, a cooling labor market, and remote work

This week in the ADP Research Institute Data Lab, we showed that most workers think AI will impact their job, but disagree on how. They’re evenly split on whether it be a net positive or a net negative.

This story of AI and worker sentiment unspools against the backdrop of a cooling labor market and uncertainty about the future of remote work. Recent research1Eloundou, Tyna, Sam Manning, Pamela Mishkin, and Daniel Rock. (2023). GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. Working paper accessed May 6, 2024, at 2303.10130 (arxiv.org). from OpenAI and the Wharton School of the University of Pennsylvania suggests that jobs with a heavy reliance on writing and computer programming skills — a sizable share of the knowledge labor market — have high exposure to the newest forms of generative AI.

In this macroeconomic context, you’d expect remote workers—many of whom are knowledge workers who do a lot of writing or programming—to be more nervous about AI than on-site or hybrid workers, and our earlier post showed some evidence for this.

Data from our global survey of nearly 35,000 private-sector workers in 18 countries showed that, among workers who think AI will affect their jobs, 51 percent of those who work remotely think AI will replace some or most of their existing functions. A slightly smaller share of on-site workers (50 percent) thinks the same; among hybrid workers it’s 47 percent.

These slight differences in sentiment between remote, on-site, and hybrid workers provide weak evidence for the nervous remote worker hypothesis. But these comparisons fail to account for how AI sentiment by work location varies depending on a person’s occupation.

So how does AI sentiment vary by the type of work people do?

AI sentiment and industry exposure to generative AI

Generative AI exposure research suggests that occupations relying heavily on computer programming and writing have higher exposure; those relying heavily on science and critical thinking have lower exposure.

Our global survey didn’t ask about occupation, but it did ask respondents to name the industry where they work, offering them 12 options.

We selected the eight industries for which we could make a clear case for high or low exposure to generative AI.2We omitted education, professional services, other, and travel and transportation. Although these ratings are subjective and falsely dichotomous, they help frame the problem in the absence of validated generative AI exposure measurements by industry. Another challenge is that the survey’s response options for industry don’t align with industry classification systems in pre-existing measures of industry AI exposure.

Table 1: Subjective assessment of industry exposure to generative AI

Generative AI exposure

Industry

Reasoning

High

Arts and culture

Heavy reliance on writing and other creative activities.

Financial services

Heavy reliance on both writing and computer programming.

IT and telecom

Heavy reliance on computer programming.

Sales, media, and marketing

Heavy reliance on writing and other creative activities, as well as computer programming, especially in digital marketing.

Low

Architecture, engineering, and building

Heavy reliance on science and critical thinking, balanced against reliance on computer programming among architects and engineers.

Manufacturing and utilities

Low reliance on writing or computer programming.

Retail, catering, and leisure

Low reliance on writing or computer programming.

Health care

Heavy reliance on science and critical thinking.

Surprisingly, workers in industries with high exposure to generative AI tend to be slightly more optimistic than those in low-exposure industries that AI will help them in the workplace. This result runs counter to the logic that knowledge workers with high generative AI exposure will feel more nervous about AI in a cooling labor market.

Take arts and culture workers, whose labor relies heavily on writing and other creative endeavors that generative AI tools emulate. Among arts and culture workers who think AI will affect their job, 56 percent think it will replace some or most of their existing functions. Among retail, catering, and leisure workers, who arguably have less exposure to generative AI, that number is 61 percent.

In another example, many IT and telecoms workers rely heavily on computer programming. Among those who think AI will affect their job, 48 percent believe it will replace their functions. Compare that to health care workers, many of whom are doctors or nurses with extensive science training and thus lower generative AI exposure. Among this group, 53 percent express worry about AI.

Regardless of AI exposure, these numbers mirror the topline results presented in our earlier post: Across industries, workers split evenly, to varying degrees, on whether AI will help or replace them in the workplace.

What about workers who think AI will have no impact, or have no opinion?

We’ve focused so far on those workers who think AI will affect their jobs in some way. But survey respondents had the choice of saying they don’t think AI will affect their job, or they don’t know enough to form an opinion.

You would expect workers in industries with higher exposure to generative AI to be more likely to think it will affect them somehow.

And that’s exactly what we saw.

How the relationship between work location and AI sentiment varies by industry

Workers in industries with high exposure to generative AI are more likely to think the technology will have some impact on their job, and they’re slightly more optimistic that AI will help them in the workplace. But does that result depend on where people work?

The chart below makes plain that the evidence is anything but clear-cut.

Remote workers in high-exposure industries such as financial services or IT and telecoms are less optimistic than hybrid or on-site workers in the same industry, but so too are remote workers in low-exposure industries such as health care or retail, catering, and leisure.

A curious difference appears, however, between two low-exposure industries: manufacturing and utilities on the one hand, and architecture, engineering, and building on the other.

Manufacturing and utilities workers who do their jobs on site are more likely than hybrid and remote workers in the same industry to believe AI will replace their existing functions. In contrast, remote workers in architecture, engineering, and building are more likely than hybrid and on-site workers in the same industry to think AI will replace them.

This result might be a fluke, or it could reveal the diverse ways people think about AI depending on the type of work they do.

A person working on site in manufacturing and utilities might think less about chatbots and more about robots when they ponder AI and their job. In contrast, when an architect or engineer thinks about AI, they might imagine technology that someday develops generative engineering and design capabilities. In a cooling knowledge labor market, this line of thinking could make remote design and engineering workers more nervous than hybrid or on-site workers.

The takeaway: Worker sentiment about AI is complicated

Research on occupational exposure to generative AI suggests that workers in industries with a heavier reliance on writing and computer programming will be more likely to think AI will affect their job in some way. Our survey supports that hypothesis.

Labor-market trends in knowledge work, along with ongoing compromises on work-from-home policies, suggest that remote workers in industries highly exposed to generative AI might be more nervous about how AI will affect their job. Yet we find that workers in those industries are, if anything, more optimistic about AI’s impact. What’s more, the pattern of worker sentiment by industry and work location gives no support to the nervous remote worker hypothesis.

This hit-and-miss speaks to the complexity of worker sentiment. More importantly, it highlights a need for better research methods and data to unravel the links between AI exposure and worker sentiment for colloquial definitions of AI, from chatbots to robots.

But our results also show that workers across industries and work locations are, for the most part, evenly divided on whether they think AI will help or replace them.