Occupational Exposure to Generative Artificial Intelligence: A Structural Perspective on Labour Market Transformation
- Dr Fernando Pinto Hernández
- Jan 9
- 5 min read
The rapid diffusion of generative artificial intelligence has reignited longstanding concerns about the future of work. Public debate has largely revolved around the prospect of job destruction and technological unemployment, often framed in binary terms. However, such a perspective is analytically insufficient. From an economic standpoint, the central issue is not whether jobs will disappear, but how artificial intelligence interacts with the internal task structure of occupations and how this interaction reshapes the content of work.
This task-based perspective is well established in the economic literature. Seminal work by Autor, Levy, and Murnane demonstrated more than two decades ago that technological change does not affect jobs as homogeneous units, but rather operates through tasks, selectively automating routine activities while complementing non-routine cognitive and interpersonal tasks. This insight remains central to understanding the labour market implications of generative artificial intelligence.
Recent research conducted by the International Labour Organization provides a rigorous empirical framework to operationalise this intuition. In 2023, and subsequently refined in 2025, the ILO developed a task-based methodology to assess the potential exposure of existing occupations to generative artificial intelligence. The analysis covers 436 occupations defined under the ISCO 08 classification and decomposes each occupation into its constituent tasks. Each task is then assigned a score between zero and one, where zero indicates no potential for automation using current generative AI technologies, and one indicates full automobility.
The occupational exposure score is computed as the average of task-level scores, while the associated standard deviation captures the dispersion of exposure across tasks within the same occupation. This distinction is crucial. A high average score indicates substantial exposure, but the standard deviation reveals whether this exposure is uniform across tasks or concentrated in specific activities. The resulting indicators do not measure realised impacts on employment, but rather the maximum potential exposure under full technological adoption.
Aggregating these occupational scores across ISCO major groups provides a synthetic and informative representation of how generative AI interacts with labour market structure.
Figure 1. Exposure to Artificial Intelligence by ISCO Major Groups
Source: Author’s calculations based on the International Labour Organization.
The figure highlights clear and economically meaningful patterns. At the top of the exposure distribution stand clerical support workers. This group exhibits by far the highest average exposure to generative artificial intelligence. Clerical occupations include administrative assistants, office clerks, data entry workers, payroll staff, call centre operators, and related roles. The core tasks of these roles revolve around information processing, record keeping, standardised communication, and routine administrative procedures. These activities align closely with the current capabilities of generative AI systems, which excel in text generation, document handling, classification, and structured information management.
Moreover, the relatively low dispersion of exposure scores within this group suggests that the potential applicability of AI is not limited to a narrow subset of tasks. Instead, it spans a large share of the occupational task bundle. This does not imply that clerical jobs will disappear. Rather, it indicates that the nature of these jobs is likely to change substantially, with artificial intelligence increasingly embedded as a productivity-enhancing tool that alters task composition and work organisation.
A second tier of exposure comprises professionals and managers. These groups display significant, though clearly lower, average exposure than clerical occupations. They encompass a wide range of highly educated roles in finance, information technology, research, law, engineering, management, and public administration. In these occupations, many tasks, such as drafting, coding, data analysis, and report preparation, are amenable to generative AI. At the same time, these jobs include activities that rely heavily on judgement, strategic decision-making, accountability, and interpersonal interaction.
This pattern is consistent with more recent theoretical and empirical work by Acemoglu and Restrepo, who emphasise that artificial intelligence can generate both displacement and productivity effects, depending on whether it substitutes for existing tasks or creates new ones. In high-skilled occupations, the dominant channel is often task reallocation and augmentation rather than outright replacement.
Technicians and associate professionals occupy an intermediate position in the exposure distribution. Their task structure combines procedural, technical, and interactive elements, leading to moderate average exposure accompanied by relatively high variability across tasks. This suggests that the impact of generative AI within this group will be uneven and highly occupation-specific.
By contrast, service and sales workers, plant and machine operators, craft and related trades workers, skilled agricultural workers, and elementary occupations all exhibit markedly lower average exposure. These groups are characterised by tasks that require physical presence, manual dexterity, situational awareness, or direct human interaction. Such tasks remain difficult to automate using current generative AI technologies. This pattern challenges simplistic narratives that equate vulnerability to automation with low skill levels. In fact, many occupations with relatively low formal educational requirements remain less exposed precisely because their task content is not easily codified.
The ILO framework explicitly cautions against interpreting exposure as realised employment impact. Exposure represents an upper bound. Actual outcomes will depend on infrastructure, access to digital technologies, organisational capacity, skill availability, regulatory conditions, and economic incentives. Global estimates indicate that approximately one quarter of total employment worldwide falls within one of the four exposure gradients identified by the ILO, with substantially higher shares in high-income countries. Gender differences are also pronounced, as women are disproportionately represented in clerical and administrative occupations concentrated in the higher exposure categories.
The most plausible aggregate effect of generative artificial intelligence is therefore not widespread job destruction, but gradual and uneven job transformation. Managing this transformation requires anticipatory policies, continuous skills development, and effective social dialogue. Linking exposure indices to national labour force survey microdata, as proposed by the ILO, provides a basis for targeted policy responses rather than generic interventions.
From a structural perspective, the distribution of exposure across occupational groups reveals a clear economic logic. Generative AI interacts most intensively with the informational core of modern economies. Occupations centred on organising, processing, and transmitting information are at the frontier of technological transformation, while those grounded in physical or relational tasks remain less directly affected.
This has important policy implications. Adjustment costs are likely to be concentrated in specific segments of the labour market, particularly administrative and support roles. Broad-based policies aimed at the entire workforce may be less effective than targeted measures focused on reskilling, internal job mobility, and task reallocation within affected occupations.
Ultimately, the task-based approach developed by the International Labour Organization reframes the debate on artificial intelligence and employment. The central question is not whether artificial intelligence will replace jobs, but how institutions, firms, and workers adapt to a reconfiguration of tasks within occupations. Economies that manage this transition proactively will be better positioned to translate technological progress into productivity gains and inclusive growth.
References
Autor, D. H., Levy, F., & Murnane, R. J. (2003). The skill content of recent technological change: An empirical exploration. Quarterly Journal of Economics, 118(4), 1279–1333.
Acemoglu, D., & Restrepo, P. (2020). AI and jobs: Evidence from online vacancies. Journal of Political Economy, 128(8), 2962–3018.
International Labour Organization. (2023). Generative AI and Jobs: A Global Analysis of Potential Effects on Job Quantity and Quality. ILO Working Paper No. 140. Geneva.
International Labour Organization. (2025). Generative AI and Jobs: Methodological Update and Extended Results. ILO Working Paper No. 140, revised edition. Geneva.
