Measures showing how workers are exposed to artificial intelligence reflect technological capability rather than actual job losses or gains, underscoring the need for caution in using such indicators for labor market forecasts, according to a new study.
A research brief from the International Labour Organization (ILO) said AI exposure indicators are useful in identifying tasks that could be affected by automation, but they do not predict whether jobs will be displaced, transformed, or supported by productivity gains.
“They (AI exposure) tell us where (Gen) AI is technically able to perform tasks, but they do not reveal whether adoption will occur, how quickly, or with what economic consequences. They should not be used to forecast job losses. They are best understood as early warning indicators that identify occupations where task content is likely to change,” the brief said.
The study, co-authored by researchers from the ILO, University College London, and the University of Oxford, examined how different methods of measuring AI exposure produce varying results across occupations and skill levels.
Earlier models tended to show higher risk for lower-paid workers performing routine manual or cognitive tasks, including some technical occupations. More recent approaches, however, suggest higher exposure among knowledge-intensive roles such as those in finance, computing, management, and administrative work.
The brief noted that exposure is not uniform even within sectors, with both high-skilled and lower-skilled roles facing different degrees of potential change depending on task composition.
It also warned that the effects of AI may extend beyond directly exposed occupations, as changes in highly analytical or administrative roles can spill over into related jobs across legal, financial, and professional services. Manual and care-related work was found to be less directly affected.
According to the authors, exposure indicators should not be treated as predictions of labor displacement but as signals of where job transformation is more likely to occur.
They stressed that policymakers should combine exposure data with actual labor market trends, including employment figures, wage movements, and workplace-level adoption of AI technologies.
“In particular, exposure analysis should be combined with observed employment, wage and transition trends from labor force surveys and with data on AI adoption in workplaces, including firm- and worker-level surveys,” the brief said.
“When embedded in this broader empirical analysis —and supplemented with richer economic and institutional information— exposure measures can support proactive skills, employment and regulatory strategies,” it added.