New ILO brief warns AI exposure indicators are signals, not predictions of job losses

A new research brief from the International Labour Organisation (ILO) has shed light on how artificial intelligence (AI) exposure indicators are being used to assess the potential impact of AI on jobs, while warning against treating them as forecasts of employment losses.

As generative AI (GenAI) adoption accelerates globally, exposure indicators have become a popular tool for estimating which occupations and tasks may be automated or transformed. However, the ILO cautions that these measures only show technological susceptibility, not actual labour market outcomes.

“Exposure indicators should be interpreted as early signals of possible change, not predictions of job displacement,” the report notes.

The findings highlight a key shift in how researchers assess AI risk. Earlier automation models tended to flag lower-skilled, routine manual and cognitive jobs as most vulnerable.

The report also stresses that exposure is not confined to directly affected roles. Highly exposed occupations often sit at the centre of skill networks and career pathways, meaning changes in these jobs could create ripple effects across related roles and industries.

To build these indicators, researchers commonly draw on datasets such as the U.S. Occupational Information Network (O*NET), alongside surveys and cross-country labour frameworks. But differences in job structures, skills, and economic conditions across countries limit the accuracy of global comparisons.

The brief also outlines three major methodological approaches used in AI exposure research:

• Expert judgement models, which rely on human assessments of task automation potential

• Patent-based analysis, which links technological innovation to job tasks

• GenAI-based self-assessment models, where AI systems evaluate their own capabilities against occupational tasks

Each approach, the ILO says, captures different dimensions of technological change but also introduces bias, from expert assumptions to model overconfidence or incomplete real-world understanding.

Recent GenAI-based methods, for example, can rapidly analyse large datasets but may overestimate AI capabilities or overlook physical, tacit, and context-dependent work.

Across all methods, the ILO underscores a consistent message: exposure does not equal displacement. Instead, it reflects potential transformation of work, which must be interpreted alongside real-world data on employment trends, wages, productivity, and policy environments.

By clarifying both the strengths and limitations of AI exposure indicators, the ILO aims to help policymakers avoid overreaction and instead focus on evidence-based strategies that support adaptation, reskilling, and inclusive labour market transitions.