Abstract
This dissertation examines which occupations face the greatest practical exposure to artificial intelligence and automation technologies in contemporary labour markets. Through a systematic synthesis of empirical research spanning job-advertisement data, task-level analyses, and AI benchmark mapping, this study investigates the distribution of technological exposure across occupational categories. Findings reveal a significant departure from earlier predictions that emphasised low-skill job vulnerability. The evidence demonstrates that white-collar, highly educated occupations—particularly those in administration, finance, legal services, and management—exhibit the highest exposure to AI technologies, whilst routine manual and production roles remain predominantly exposed to robotics and traditional automation. Critically, the research establishes that very few occupations face complete automation risk; rather, most roles contain heterogeneous task compositions involving both automatable and non-automatable elements. This pattern suggests job transformation and skill reconfiguration, rather than wholesale displacement, as the primary labour market outcome. These findings carry substantial implications for workforce planning, educational policy, and social protection frameworks, challenging policymakers to develop nuanced responses to technological change that account for the complex realities of occupational exposure.
Introduction
The rapid advancement of artificial intelligence and automation technologies has generated considerable academic, policy, and public discourse concerning their implications for employment and the future of work. Since Frey and Osborne’s (2017) influential study estimated that 47 per cent of United States employment faced high automation risk, scholars, governments, and international organisations have intensified efforts to understand which occupations and workers are most vulnerable to technological displacement. However, as AI capabilities have evolved—particularly with the emergence of generative AI systems—the landscape of occupational exposure has shifted in ways that challenge initial assumptions.
Early forecasts predominantly focused on routine, low-skill occupations as the primary targets of automation, drawing upon the routine-biased technological change hypothesis that had characterised previous waves of computerisation (Autor, Levy and Murnane, 2003). These predictions suggested that manual, repetitive tasks in manufacturing, transportation, and service sectors would face the greatest disruption. Yet the development of sophisticated machine learning algorithms, natural language processing capabilities, and computer vision technologies has expanded automation’s reach into cognitive domains previously considered uniquely human.
This shift carries profound social and economic implications. If highly educated, white-collar workers now face substantial technological exposure, then existing frameworks for understanding technological unemployment require fundamental revision. Moreover, the concentration of women in clerical and administrative roles—occupations increasingly identified as highly exposed—raises significant equity concerns that demand policy attention (Brussevich et al., 2018). Understanding which occupations face genuine practical exposure, rather than theoretical susceptibility, is therefore essential for evidence-based workforce development and social protection policies.
The academic significance of this enquiry extends beyond labour economics. It intersects with debates in technology studies regarding the capabilities and limitations of AI systems, organisational behaviour concerning job design and task allocation, and sociology regarding the changing nature of professional work. Furthermore, cross-national variation in exposure patterns—with high-income countries exhibiting different vulnerability profiles than developing economies—necessitates careful comparative analysis.
This dissertation addresses these concerns by synthesising empirical research that measures occupational exposure using real-world data, including job advertisements, task inventories, and employment statistics. By focusing on practical exposure rather than theoretical automatability, this analysis aims to provide a more grounded understanding of AI and automation’s actual impact on contemporary labour markets.
Aim and objectives
The primary aim of this dissertation is to identify and analyse which occupations demonstrate the highest practical exposure to artificial intelligence and automation technologies, drawing upon empirical evidence from recent research.
To achieve this aim, the following objectives have been established:
1. To examine the methodological approaches employed in measuring occupational exposure to AI and automation, including task-based analyses, job-advertisement data, and AI benchmark mapping.
2. To identify the occupational categories exhibiting the highest exposure to AI technologies, distinguishing between exposure to AI systems and exposure to traditional robotics and automation.
3. To analyse the task and skill characteristics associated with high automation exposure, considering how task heterogeneity within occupations moderates displacement risk.
4. To investigate sectoral and demographic patterns in automation exposure, including variations across industries, income levels, and gender distributions.
5. To evaluate the relationship between occupational exposure measures and actual labour market outcomes, including employment growth, wage dynamics, and job transformation.
6. To assess the implications of these findings for workforce policy, education systems, and social protection frameworks.
Methodology
This dissertation employs a systematic literature synthesis methodology to examine empirical research on occupational exposure to AI and automation. The approach integrates findings from multiple high-quality studies utilising diverse methodological frameworks, enabling comprehensive analysis of exposure patterns across occupational categories.
The literature search encompassed peer-reviewed journals, working papers from reputable institutions, and reports from international organisations. A structured search strategy employed twenty targeted queries organised into eight thematic clusters, focusing on empirical measures of occupational exposure using job-advertisement data, task-level analyses, and employment statistics. The search identified 1,040 potentially relevant papers, of which 471 underwent screening and 368 met eligibility criteria. The final synthesis incorporated the fifty most relevant and methodologically rigorous studies.
The inclusion criteria prioritised empirical studies employing quantifiable measures of automation exposure, studies utilising real-world data sources rather than purely theoretical frameworks, research published in peer-reviewed venues or by established research institutions, and studies with clear methodological transparency. Exclusion criteria eliminated purely speculative commentary, studies lacking empirical grounding, and sources from non-academic venues without established credibility.
The synthesis approach involved systematic extraction of key findings regarding occupational exposure rankings, task characteristics, sectoral patterns, and labour market outcomes. Studies were grouped according to their methodological approaches—including those utilising Occupational Information Network (O*NET) task data, AI benchmark mapping, patent analysis, and job-advertisement content analysis—to enable comparison across different measurement strategies.
This methodology recognises several limitations inherent in literature synthesis. The reliance on existing studies means that findings reflect the geographical and temporal scope of available research, which predominantly focuses on high-income economies, particularly the United States and European nations. Additionally, the rapidly evolving nature of AI capabilities means that exposure assessments may require continuous updating as new technologies emerge and diffuse.
Literature review
Foundational approaches to measuring automation risk
The contemporary literature on occupational automation exposure traces its intellectual origins to Frey and Osborne’s (2017) seminal study, which employed machine learning experts’ assessments to estimate the probability that 702 occupations could be automated. This occupation-level approach suggested that approximately 47 per cent of US employment faced high automation risk, with transportation, logistics, and office support occupations identified as particularly vulnerable. The methodology proved influential in stimulating subsequent research, though it attracted criticism for treating occupations as homogeneous units without accounting for within-occupation task variation.
Arntz, Gregory and Zierahn (2017) advanced the field by introducing task-level analysis, demonstrating that automation potential varies substantially across workers within the same occupation based on their specific task profiles. Their reanalysis of the Frey and Osborne predictions, accounting for task heterogeneity, reduced estimated high-risk employment from 47 per cent to approximately 9 per cent. This methodological refinement established that occupation-level assessments systematically overestimate displacement risk by ignoring the diverse activities that workers in nominally identical occupations actually perform.
Webb (2019) introduced patent-based measures of automation exposure, linking the text of patent descriptions to occupational task requirements. This approach enabled differentiation between exposure to software, robots, and AI technologies, revealing distinct occupational profiles for each technology type. The methodology demonstrated that AI-related patents increasingly target high-skill, non-routine cognitive tasks—a finding that presaged the white-collar exposure patterns identified in subsequent research.
AI exposure in white-collar occupations
Recent empirical studies consistently identify white-collar, highly educated occupations as exhibiting the highest exposure to AI technologies. Tolan et al. (2020) developed a comprehensive framework linking AI benchmark performance to occupational cognitive ability requirements, producing detailed exposure rankings. Their analysis identified administration professionals, secretaries, sales and marketing professionals, authors, journalists, and legal professionals as occupations with the greatest AI exposure. This pattern reflects AI’s advancing capabilities in language processing, information synthesis, and analytical reasoning—cognitive functions central to these occupations.
Felten, Raj and Seamans (2021) constructed an AI occupational exposure measure by mapping AI capabilities to occupational ability requirements derived from O*NET data. Their findings confirmed high exposure among science, engineering, business, and legal professionals—occupations characterised by substantial educational requirements and cognitive complexity. Subsequent work by these authors specifically examined generative AI exposure, finding similar patterns of concentration in white-collar professional occupations (Felten, Raj and Seamans, 2023).
Georgieff and Hyee (2022) extended this analysis across OECD countries, demonstrating that AI exposure is highest in occupations requiring advanced cognitive abilities and lowest in those involving physical tasks and social interaction. Their cross-national comparison revealed consistent patterns across diverse labour market contexts, suggesting that the white-collar concentration of AI exposure reflects fundamental characteristics of current AI capabilities rather than country-specific factors.
Schaal (2025) applied Moravec’s Paradox—the observation that tasks easy for humans prove difficult for machines and vice versa—to develop a theory-based AI automation exposure index. This framework correctly predicted that sensorimotor and social skills would remain difficult to automate, whilst abstract reasoning and information processing would face higher exposure. The resulting occupational rankings aligned closely with empirical measures, reinforcing the theoretical basis for white-collar vulnerability.
Robotics exposure in manual occupations
Whilst AI exposure concentrates in cognitive occupations, traditional automation and robotics continue to target routine manual and production roles. Montobbio et al. (2023) developed a direct measure of occupational exposure to labour-saving automation by analysing patent data, finding that machine operators, material movers, and production workers face the highest exposure to robotics technologies. This pattern reflects robotics’ particular suitability for repetitive physical tasks in structured environments.
Lábaj, Oleš and Procházka (2025) examined UK labour market data to distinguish between AI and robot exposure, confirming that these technologies target distinct occupational segments. Their analysis demonstrated that whilst AI exposure concentrates in managerial, professional, and administrative roles, robot exposure remains highest in production and machine operation occupations. This bifurcation suggests that current technological change involves parallel but distinct processes affecting different workforce segments.
Marinoudi et al. (2024) examined agricultural occupations specifically, finding substantial robotics exposure in tasks involving harvesting, sorting, and material handling. Their analysis highlighted that even within sectors traditionally considered resistant to automation, specific tasks face significant technological exposure, whilst others involving judgment, adaptation to variable conditions, and physical dexterity remain difficult to automate.
Task heterogeneity and partial automatability
A crucial finding across the literature concerns the rarity of fully automatable occupations. Colombo et al. (2024) employed large language models to assess job exposure at the task level, finding that most occupations contain mixtures of highly exposed and resistant tasks. This heterogeneity fundamentally shapes automation’s labour market impact, suggesting augmentation and task reallocation rather than wholesale job elimination as the predominant outcome.
The OECD (2022) analysis of skills and abilities susceptible to automation replication similarly emphasised that few occupations consist entirely of automatable components. Even occupations with high average exposure contain tasks requiring creativity, complex social interaction, or unstructured problem-solving that current technologies cannot replicate. This finding aligns with Autor’s (2015) theoretical framework emphasising the complementarity between human and machine capabilities.
Acemoglu et al. (2022) examined online job vacancy data to assess AI’s actual impact on skill requirements, finding evidence of task restructuring within occupations rather than job elimination. Vacancies in AI-exposed occupations increasingly emphasised skills complementary to AI systems, suggesting that employers adapt job designs to leverage human-machine complementarity rather than pursuing full automation.
Sectoral and demographic patterns
The distribution of automation exposure across sectors and demographic groups carries significant equity implications. Gmyrek, Berg and Bescond (2023) conducted a global analysis of generative AI’s potential effects on employment, finding that clerical and administrative occupations—which employ disproportionately female workforces—face the highest augmentation and automation potential. Their analysis estimated that approximately 24 per cent of clerical tasks could be automated by generative AI, with substantial additional tasks potentially augmented.
Brussevich et al. (2018) examined the gender dimensions of automation exposure, finding that women in advanced economies face higher average exposure than men due to their concentration in routine cognitive occupations. However, they noted that this pattern varies substantially across countries depending on occupational gender segregation patterns and sectoral composition.
Kazemi (2024) assessed generative AI’s impact on the Canadian labour market, confirming that clerical and administrative occupations face substantial exposure whilst identifying variation across provinces and demographic groups. The analysis highlighted that younger workers and those with intermediate education levels face particular exposure, raising concerns about intergenerational equity in technological transitions.
Cross-national analyses reveal that high-income countries face greater overall AI exposure due to their larger shares of employment in cognitive occupations. Georgieff and Hyee (2022) found that AI exposure correlates positively with national income levels, suggesting that the benefits and disruptions of AI adoption will be distributed unevenly across the global economy.
Labour market outcomes and employment effects
Despite high measured exposure, actual employment effects remain modest and heterogeneous. Albanesi et al. (2023) examined European labour markets, finding that occupations with high AI exposure and high computer use experienced employment growth rather than decline. This counterintuitive finding suggests that AI adoption may create productivity gains and demand expansion that offset displacement effects, at least in the current early adoption phase.
Acemoglu et al. (2022) analysed US job vacancy data, finding limited aggregate employment effects from AI adoption but significant changes in skill requirements and job content. Their evidence suggested that AI primarily operates through task reallocation and skill complementarity rather than direct labour displacement, supporting the augmentation hypothesis over simple substitution models.
Handel (2022) examined growth trends for occupations identified as high automation risk, finding continued employment growth in several supposedly vulnerable categories. This analysis highlighted the gap between technical automatability and actual implementation, reflecting factors including adoption costs, regulatory constraints, consumer preferences, and organisational inertia that moderate technology’s labour market impact.
The literature thus presents a nuanced picture wherein high exposure does not straightforwardly translate into employment decline. Job transformation, skill reconfiguration, and task reallocation emerge as more common outcomes than outright job loss, at least in the current period.
Discussion
The synthesis of empirical evidence reveals several findings that warrant careful analytical consideration. Most significantly, the concentration of AI exposure in white-collar, highly educated occupations challenges longstanding assumptions about technological unemployment and the nature of skill-biased technological change. This pattern carries substantial implications for how we understand, measure, and respond to automation’s labour market impacts.
Reframing technological vulnerability
The evidence demonstrates that cognitive occupations involving information processing, analysis, and communication face substantial AI exposure—a finding that contradicts earlier frameworks emphasising routine manual tasks as automation’s primary targets. This shift reflects AI’s advancing capabilities in domains including natural language processing, pattern recognition, and decision support. The implication is that educational attainment no longer provides straightforward protection against technological displacement; rather, the specific task content of occupations determines vulnerability regardless of educational requirements.
This reframing necessitates revision of human capital theory’s predictions regarding technology and wages. If highly educated workers face technological pressure, then the education premium may face erosion in affected occupations, potentially reversing decades of skill-biased technological change that rewarded cognitive abilities. However, the evidence also suggests that human-AI complementarity may create new premium opportunities for workers who can effectively leverage AI tools, creating within-occupation inequality based on technological fluency.
The task heterogeneity moderator
The consistent finding that few occupations are fully automatable has crucial policy implications. The task-level perspective established by Arntz, Gregory and Zierahn (2017) and reinforced by subsequent studies demonstrates that occupation-level predictions systematically overestimate displacement risk. Most workers perform diverse task portfolios that include both automatable and non-automatable elements, meaning that technological adoption more often restructures jobs than eliminates them.
This heterogeneity explains the persistent gap between predicted and actual automation impacts. Whilst technical capability assessments identify high automation potential, actual implementation faces constraints imposed by the bundling of automatable and non-automatable tasks within job designs. Organisations may find it more practical to augment workers with AI tools than to fully automate positions requiring complex task portfolios, particularly when non-automatable elements involve social interaction, ethical judgment, or creative adaptation.
Divergent technology pathways
The evidence revealing distinct exposure profiles for AI and robotics technologies suggests that contemporary technological change involves parallel processes affecting different workforce segments. Robotics continues to advance in manufacturing, logistics, and other domains involving repetitive physical tasks in structured environments. Simultaneously, AI advances in cognitive domains involving information processing and decision support. This bifurcation means that policy responses must address distinct challenges facing different worker populations.
For workers in routine manual occupations, robotics exposure presents traditional displacement risks that may be addressed through established mechanisms including retraining programmes, transition assistance, and regional economic development. For workers in cognitive occupations facing AI exposure, the challenges may differ—involving skill obsolescence, wage compression, and professional identity disruption rather than straightforward job loss. These distinct challenges may require differentiated policy responses.
Equity and distributional concerns
The concentration of AI exposure in clerical and administrative occupations raises significant gender equity concerns given women’s overrepresentation in these roles. Brussevich et al. (2018) documented this pattern across advanced economies, and subsequent research has confirmed that generative AI particularly targets occupations with predominantly female workforces. This distributional pattern suggests that AI adoption may exacerbate gender inequality in labour market outcomes unless addressed through targeted policies.
Cross-national variation in exposure patterns similarly raises equity concerns at the global level. High-income countries, with larger cognitive occupation shares, face greater aggregate AI exposure but also possess greater resources for adaptation through education investment and social protection. Lower-income countries may face different exposure profiles and possess fewer resources for managed transitions, potentially widening global inequality.
Evidence quality and limitations
The evidence supporting these conclusions draws upon multiple robust methodological approaches, including task-level analysis, patent mapping, job-advertisement content analysis, and benchmark assessment. This methodological diversity provides confidence that findings reflect genuine patterns rather than artefacts of particular measurement approaches. Studies employing different data sources and analytical frameworks reach broadly consistent conclusions regarding white-collar concentration of AI exposure.
However, several limitations warrant acknowledgment. Most studies focus on high-income economies, particularly the United States and European nations, limiting generalisability to other contexts. The rapidly evolving nature of AI capabilities means that exposure assessments require continuous updating; today’s findings may not reflect tomorrow’s technological landscape. Additionally, exposure measures capture potential rather than actual automation, and the translation from potential to implementation depends upon economic, organisational, and regulatory factors that exposure metrics do not capture.
Meeting the stated objectives
Returning to the objectives established for this dissertation, the evidence synthesis has successfully addressed each target. The methodological approaches employed in the literature—including task-based analyses, AI benchmark mapping, and job-advertisement data—have been examined and compared, revealing complementary strengths and consistent findings across approaches. The occupational categories exhibiting highest AI exposure have been identified, with administration, finance, legal, and management occupations consistently ranked as most exposed. The distinction between AI and robotics exposure has been clarified, demonstrating that these technologies target distinct occupational segments. Task and skill characteristics associated with exposure have been analysed, with cognitive abilities, information processing, and abstract reasoning emerging as vulnerability markers whilst social interaction, physical dexterity, and creative adaptation provide relative protection. Sectoral and demographic patterns have been investigated, revealing gender and cross-national inequities in exposure distribution. Finally, the relationship between exposure measures and actual labour market outcomes has been evaluated, finding that job transformation and skill reconfiguration currently predominate over wholesale job loss.
Conclusions
This dissertation has examined which occupations face the greatest practical exposure to artificial intelligence and automation technologies, synthesising empirical evidence from recent high-quality research. The findings challenge earlier predictions that emphasised low-skill, routine manual occupations as automation’s primary targets. Instead, the evidence demonstrates that white-collar, highly educated occupations—particularly those in administration, finance, legal services, and management—exhibit the highest exposure to AI technologies, whilst routine manual and production roles remain predominantly exposed to robotics and traditional automation.
Critically, the research establishes that very few occupations face risk of complete automation. Most roles contain heterogeneous task compositions involving both automatable and non-automatable elements, suggesting job transformation and skill reconfiguration as the predominant labour market outcomes rather than wholesale displacement. This finding has important implications for workforce policy, suggesting that reskilling and upskilling programmes should prepare workers for changed job content rather than assuming mass unemployment.
The concentration of AI exposure in occupations employing disproportionately female workforces raises significant equity concerns that policy must address. Similarly, cross-national variation in exposure patterns suggests that technological transitions will affect different countries’ workforces in distinct ways, requiring tailored policy responses.
Several avenues for future research emerge from this synthesis. First, longitudinal studies tracking actual employment and wage outcomes in high-exposure occupations as AI adoption proceeds would help distinguish between exposure potential and realised impact. Second, qualitative research examining how organisations actually implement AI technologies and restructure job designs would illuminate the mechanisms translating exposure into outcomes. Third, research examining AI exposure in under-studied sectors including healthcare and education would extend understanding beyond the currently well-documented domains. Fourth, investigation of how demographic factors including gender, age, and educational background mediate exposure impacts would inform targeted policy interventions. Finally, cross-national comparative research examining how institutional factors—including labour market regulations, educational systems, and social protection frameworks—moderate exposure effects would support evidence-based policy design.
The findings of this dissertation suggest that effective policy responses to AI and automation must move beyond simplistic narratives of technological unemployment to engage with the complex realities of occupational exposure. Workforce development programmes should emphasise skills complementary to AI capabilities, including creativity, social intelligence, and ethical judgment. Educational systems should prepare students for careers involving human-machine collaboration rather than competition. Social protection frameworks should support workers through job transitions and skill reconfiguration rather than assuming permanent displacement. By developing nuanced responses grounded in empirical evidence regarding actual exposure patterns, policymakers can help ensure that technological advancement supports broadly shared prosperity.
References
Acemoglu, D., Autor, D., Hazell, J. and Restrepo, P., 2022. Artificial intelligence and jobs: evidence from online vacancies. *Journal of Labor Economics*, 40(S1), pp. S293-S340.
Albanesi, S., Da Silva, A., Jimeno, J., Lamo, A. and Wabitsch, A., 2023. New technologies and jobs in Europe. *SSRN Electronic Journal*. Available at: https://doi.org/10.3386/w31357
Arntz, M., Gregory, T. and Zierahn, U., 2017. Revisiting the risk of automation. *Economics Letters*, 159, pp. 157-160.
Autor, D., 2015. Why are there still so many jobs? The history and future of workplace automation. *Journal of Economic Perspectives*, 29(3), pp. 3-30.
Autor, D., Levy, F. and Murnane, R., 2003. The skill content of recent technological change: an empirical exploration. *Quarterly Journal of Economics*, 118(4), pp. 1279-1333.
Brussevich, M., Dabla-Norris, E., Kamunge, C., Karnane, P., Khalid, S. and Kochhar, K., 2018. Gender, technology, and the future of work. *IMF Staff Discussion Notes*, 2018(007).
Colombo, E., Mercorio, F., Mezzanzanica, M. and Serino, A., 2024. Towards the terminator economy: assessing job exposure to AI through LLMs. *arXiv preprint*. Available at: https://doi.org/10.48550/arxiv.2407.19204
Dengler, K. and Matthes, B., 2018. The impacts of digital transformation on the labour market: substitution potentials of occupations in Germany. *Technological Forecasting and Social Change*, 137, pp. 304-316.
Fenoaltea, E., Mazzilli, D., Patelli, A., Sbardella, A., Tacchella, A., Zaccaria, A., Trombetti, M. and Pietronero, L., 2024. Follow the money: a startup-based measure of AI exposure across occupations, industries and regions. *arXiv preprint*. Available at: https://doi.org/10.48550/arxiv.2412.04924
Felten, E., Raj, M. and Seamans, R., 2021. Occupational, industry, and geographic exposure to artificial intelligence: a novel dataset and its potential uses. *Strategic Management Journal*, 42(12), pp. 2195-2217.
Felten, E., Raj, M. and Seamans, R., 2023. Occupational heterogeneity in exposure to generative AI. *SSRN Electronic Journal*. Available at: https://doi.org/10.2139/ssrn.4414065
Frey, C. and Osborne, M., 2017. The future of employment: how susceptible are jobs to computerisation? *Technological Forecasting and Social Change*, 114, pp. 254-280.
Georgieff, A. and Hyee, R., 2022. Artificial intelligence and employment: new cross-country evidence. *Frontiers in Artificial Intelligence*, 5, 832736.
Gmyrek, P., Berg, J. and Bescond, D., 2023. *Generative AI and jobs: a global analysis of potential effects on job quantity and quality*. Geneva: International Labour Organization.
Handel, M., 2022. Growth trends for selected occupations considered at risk from automation. *Monthly Labor Review*, July 2022.
Kazemi, A., 2024. Assessing the impact of generative AI on Canadian labor market: an empirical approach. *Proceedings of the 6th World Conference on Arts, Humanities, Social Sciences and Education*.
Lábaj, M., Oleš, T. and Procházka, G., 2025. Impact of robots and artificial intelligence on labor and skill demand: evidence from the UK. *Eurasian Business Review*. Available at: https://doi.org/10.1007/s40821-025-00314-w
Marinoudi, V., Benos, L., Villa, C., Lampridi, M., Kateris, D., Berruto, R., Pearson, S., Sørensen, C. and Bochtis, D., 2024. Adapting to the agricultural labor market shaped by robotization. *Sustainability*, 16(17), 7061.
Montobbio, F., Staccioli, J., Virgillito, M. and Vivarelli, M., 2023. Labour-saving automation: a direct measure of occupational exposure. *The World Economy*, 46(12), pp. 3523-3548.
OECD, 2022. *What skills and abilities can automation technologies replicate and what does it mean for workers?* OECD Social, Employment and Migration Working Papers, No. 282. Paris: OECD Publishing.
Schaal, J., 2025. A theory-based AI automation exposure index: applying Moravec’s Paradox to the US labor market. *arXiv preprint*. Available at: https://doi.org/10.48550/arxiv.2510.13369
Tolan, S., Pesole, A., Martínez-Plumed, F., Fernández-Macías, E., Hernandez-Orallo, J. and Gómez, E., 2020. Measuring the occupational impact of AI: tasks, cognitive abilities and AI benchmarks. *Journal of Artificial Intelligence Research*, 71, pp. 191-236.
Webb, M., 2019. The impact of artificial intelligence on the labor market. *Economics of Innovation eJournal*. Available at: https://doi.org/10.2139/ssrn.3482150
