AssociationAI / AI Literacy
AssociationAI team Published

Article

Beginner

What Government Labor Researchers Say AI Is Doing to Jobs, and How to Read "Exposure" Before You Brief Members

BLS economists explained in February 2025 how they build AI into employment projections, using information up to June 2024. An ILO working paper from May 2025 scored occupations for generative AI exposure. Neither says exposure means job loss. Here is how an association should read both before it tells members anything.

AI Literacy Workforce Careers Member Communications

Picture this illustrative case: a trade association’s career center is planning a members-only webinar called “Will AI take my job?” The program committee has collected headlines that disagree with each other, and a board member has forwarded a chart claiming a quarter of all jobs are at risk. The staff lead has two weeks to decide what the association will say on the record. This is a teaching example, not a case study, but associations that run career centers, credentials and workforce programs are asked this question all the time, and the answer they give carries their name.

Two public research sources are a better starting point than headlines. One is from the U.S. Bureau of Labor Statistics, the agency that publishes the government’s official employment projections. The other is a working paper from the International Labour Organization, the United Nations agency for work. They measure different things, and the most common mistake is to read one as if it were the other.

What BLS actually projects

In the February 2025 issue of the Monthly Labor Review, three BLS economists explained how the agency builds AI into its projections. The starting point is undramatic: “The U.S. Bureau of Labor Statistics (BLS) Employment Projections (EP) program approaches AI in the same way as it does other technologies.” And “BLS projection methods are designed to measure and reflect structural technological changes, and these changes and their employment impacts tend to occur gradually.”

They add a line worth reading twice before any webinar: “There have been many claims about new technologies displacing jobs, and although such displacement has occurred in the past, it tends to take longer than technologists typically expect.”

Two limits matter. First, the date. “The employment projections outlined below reflect research based on EP’s interpretation of information available in June 2024, when the program was finalizing its 2023–33 projections.” Anything that has happened with AI tools since mid-2024 is not in these numbers. Second, the uncertainty: “The timing and scale of many potential impacts of GenAI are too uncertain to be reflected in BLS projections.”

Where BLS did project an effect, it was narrow. Its conclusion says: “Over the 2023–33 projections period, AI is expected to primarily affect occupations whose core tasks can be most easily replicated by GenAI in its current form. These occupations include medical transcriptionists and customer service representatives, whose employment is projected to decline by 4.7 and 5.0 percent, respectively, through 2033.” Those are BLS projections for 2023 to 2033, made with June 2024 information, for two occupations. They are not a forecast for every job, and they are not a measure of what has already happened.

Tasks change before jobs do

The most useful idea in the BLS article for association staff is about tasks. “Although new technologies may change the composition or weighting of tasks performed by workers in an occupation, sometimes dramatically, they may still have no employment impacts.”

The case studies show what that looks like. For budget analysts, BLS writes that “the communication and customer service tasks of budget analysts (e.g., discussing the nuances and alternative paths of proposed budgets) will likely continue to require conversations between humans and cannot be easily replaced by AI”. For legal work, AI output “will still require detailed review by lawyers because it may contain errors or biases”.

For members, this is the more honest message: parts of their work may change, sometimes a lot, and the work of checking, explaining and deciding is likely to stay with people.

What the ILO’s exposure index measures

The ILO paper asks a different question. It is ILO Working Paper 140, published in May 2025 by Paweł Gmyrek, Janine Berg and co-authors. It is a working paper, not a peer-reviewed article, and it scores how exposed occupations are to generative AI around the world.

Its headline finding is the one most likely to end up on a slide: “Globally, one in four workers are in an occupation with some GenAI exposure.” It adds that “Clerical occupations continue to have the highest exposure levels.” At the top end, “3.3% of global employment falls into the highest exposure category, albeit with significant differences between female (4.7%) and male employment (2.4%).” (ILO, 2025, p.4).

The method matters for how hard you lean on those numbers. The authors write that “we collect 52,558 data points regarding perceive potential of automation for 2,861 tasks” (the wording is the paper’s). The scores start from a survey of workers in Poland, who rated how automatable a sample of tasks from the Polish occupational classification seemed; the authors refined those results with a survey and Delphi-style discussions among international experts, then built an AI assistant to score the tasks in the ISCO-08 international classification. They are not observations of jobs that changed or disappeared.

Exposure is not job loss

This is the point the board member’s chart misses. Exposure means some tasks in an occupation could be affected. The ILO itself draws the conclusion: “As most occupations consist of tasks that require human input, transformation of jobs is the most likely impact of GenAI.” That is the same direction as the BLS view that tasks can shift without employment falling.

So “one in four workers are in an occupation with some GenAI exposure” does not mean “a quarter of jobs are at risk.” Writing it that way turns an ILO finding about tasks into a claim the ILO does not make.

Four questions to ask of any AI jobs figure

Before a number goes on a slide, the two sources suggest four checks. Is it a projection of employment, like the BLS figures, or a share of workers exposed, like the ILO’s? Is it for the United States or for the world? The ILO’s figures are global. What date does the information stop at? And what kind of source is it: an official statistical program, or a working paper still open to revision? A figure that cannot answer all four does not belong in a member briefing.

What the career center says on the record

For the webinar in our example, the two sources support a short, defensible message. BLS, using information to June 2024, projects gradual effects and names only a few occupations, such as customer service representatives, where it expects AI to reduce employment by 2033. The ILO’s 2025 working paper finds that many workers have some exposure, concentrated in clerical work, and expects jobs to change more than disappear. Both say the task level is where change shows up first.

The practical next step for members is to look at their own work. Our same-day tutorial on reading the BLS outlook for your own job shows how to find the official baseline for an occupation, and our earlier post on the split evidence for early-career members covers what newer research says about entry-level roles.

Three cautions belong in the speaker notes. Say the dates: BLS’s information runs to June 2024 and the ILO paper is from May 2025. Say what kind of source each is: an official projection method on one side, a working paper on the other. And never convert an exposure share into a count of lost jobs.

Sources (2)