I have updated and improved my website, in case anyone is interested. Lots of material in there!

Revisiting the occupational impact of AI in the generative AI era Generative AI is reshaping what artificial intelligence can do in the workplace, calling into question pre-GenAI assessments of which workers and tasks are most exposed. In this paper we trace the evo...

New paper out! With P. Casas, F. Martínez-Plumed, E. Gómez, I. González-Vázquez & S. Salotti, we revisit the occupational impact of AI in the GenAI era, tracking exposure from 2008 to 2024 across 127 occupations in Europe. Thread 1/13 #EconSky #sociology

JRC Working Papers on Labour, Education and Technology (LET) The JRC Working Papers on Labour, Education and Technology (LET) address economic, social and policy questions related to the changing nature of work, skills and education as a result of technological...

This paper is part of a broader JRC effort on understanding the implications of the digital transformation on work. For more papers on this topic, check out our Labour, Education and Technology working paper series. 13/13 END

The policy challenge is clear: AI is a rising tide that reaches all boats. Whether it lifts them or sinks them depends on education, skills policies, collective bargaining, and how we choose to implement these technologies. 12/13

We remain agnostic about whether exposure means automation or augmentation. The same exposure can translate into complementarity for some workers and substitution for others, depending on work organisation, bargaining power, and institutional context. 11/13

A methodological aside: we used LLMs to help annotate benchmarks and tasks (alongside human experts). Interestingly, AI agents worked well as complements to humans but failed when used as substitutes — they couldn't learn from each other in a Delphi process. 10/13

An interesting quirk of our measure: simple arithmetic or form-filling get LOW exposure scores. Not because AI can't do them — quite the opposite. There's no research on them because they were "solved" long ago. Our metric is forward-looking, not backward. 9/13

AI exposure and number of abilities required by specific types of task content

At the task level, the most exposed activities are things like writing reports, reading professional documents, learning from others, and analytical reasoning. The least exposed: physical strength, dexterity, body coordination. Social tasks sit in between. 8/13

AI exposure across cognitive abilities. AS = Attention and search; CE = Comprehension and expression; CL = Conceptualisation, learning and abstraction; QL = Quantitative and logical reasoning; MP = Memory processes; PA = Planning, sequential decision-making and acting; CO = Communication; EC = Emotion and self-control; MS = Mind modelling and social interaction; MC = Metacognition and confidence assessment; VP = Visual processing; AP = Auditory processing; SI = Sensorimotor interaction; NV = Navigation.

The abilities where AI is still weak? Social cognition (emotion, mind modelling, metacognition) and embodied interaction (navigation, sensorimotor). These remain hard frontiers for current AI systems. 7/13

Why is exposure reaching everyone? Because the cognitive abilities where AI research has advanced most are "ideas-related": attention & search, comprehension & expression, logical reasoning, conceptualisation. These are needed in virtually ALL jobs, not just the fancy ones. 6/13

AI exposure and average income decile for specific occupations, showing a strong positive correlation

Exposure is reaching everyone, but it´s not uniform. High-skilled occupations (professionals, technicians) remain the most exposed, and the gap with elementary occupations has widened, not narrowed. The correlation between AI exposure and income is 0.85. 5/13

Average scores of our measure of occupational AI exposure for specific occupations and broad occupational families at 1-digit level

Our measure of AI exposure shows a steep increase across ALL occupational categories since ~2015, accelerating sharply in recent years. In relative terms: elementary occupations in 2024 score higher than professionals did in 2018 — a sign of how fast the frontier is moving. 4/13

What makes our measure different? Most AI exposure metrics rely on expert opinions or rubrics. Ours is grounded in actual AI research output: we count papers across 352 benchmarks over time. This captures where AI is heading, not just where it is — and lets us track the acceleration. 3/13

How? We link 352 AI benchmarks → 14 cognitive abilities → 108 work tasks → 127 occupations. Key contribution: instead of a static snapshot, we use the intensity of AI research over time to build a dynamic measure of occupational exposure to AI. 2/13

Revisiting the occupational impact of AI in the generative AI era Generative AI is reshaping what artificial intelligence can do in the workplace, calling into question pre-GenAI assessments of which workers and tasks are most exposed. In this paper we trace the evo...

New paper out! With P. Casas, F. Martínez-Plumed, E. Gómez, I. González-Vázquez & S. Salotti, we revisit the occupational impact of AI in the GenAI era, tracking exposure from 2008 to 2024 across 127 occupations in Europe. Thread 1/13 #EconSky #sociology

The war in Iran could trigger a global food shock
• More than 1.1mn tonnes of fertiliser and fertiliser inputs is currently stuck in the Gulf
• Shortages are starting during northern hemisphere’s planting season, raising the risk of lower harvests for staples like rice
www.ft.com/content/1054...

Job quality and the platformisation of regular work: A cross-country analysis of digital monitoring and algorithmic management in the EU Digitalisation and the increasing use of digital platforms for the coordination of work processes in an algorithmic way can have deep implications for the world of work. While digital monitoring and a...

New paper out! With Álvaro Mariscal de Gante and Ignacio González-Vázquez, we analyse how the platformisation of regular work affects job quality across the EU. We find negative effects for some types of platformisation but not others, and big country variation. 1/10 #EconSky #sociology

JRC Working Papers on Labour, Education and Technology (LET) The JRC Working Papers on Labour, Education and Technology (LET) address economic, social and policy questions related to the changing nature of work, skills and education as a result of technological...

This paper is part of a broader JRC effort on the digital transformation of work, building on the AIM-WORK survey. For more papers on this topic, check out our Labour, Education and Technology working paper series. 10/10 END

For policy: it's not black or white. Soft platformisation has no major consequences for job quality. But hard forms clearly shrink autonomy and intensify work — especially for blue-collar workers. The challenge is ensuring institutional safeguards, particularly where they are weakest. 9/10

Why such a stark divide? We think that labour market institutions are key. Collective bargaining coverage, worker consultation and co-determination are stronger in Western/Northern Europe. This likely mediates how platformisation is implemented and experienced by workers. 8/10

OLS estimates of job quality outcomes by type of platformisation of work in EU countries. The figure shows that most of the negative effects are concentrated in Central and Eastern Member States.

Key finding 3: there's a sharp East-West divide. Negative effects on job quality are almost exclusively concentrated in Central-Eastern EU countries. Western countries mostly show neutral outcomes. Look at this figure — the pattern is remarkably clear. 7/10

Key finding 2: not all platformised workers are equally affected. Physical and full platformisation show much worse outcomes than informational or partial. Manual routine tasks are easier to control algorithmically. White-collar workers retain more discretion even when platformised. 6/10

Key finding 1: harder forms of platformisation are associated with reduced worker autonomy and work intensification. Autonomy loss is the strongest effect — consistent with the idea that platforms can standardise and bureaucratise work in ways that constrain worker discretion. 5/10

Platformisation of work across some EU countries, showing more prevalence of platformisation in Poland, Romania and Spain and less prevalence in Germany, Finland and Sweden

Not all platformisation is the same. We distinguish: partial (mild forms), informational (tracking digital activity + algorithmic evaluation — typical white-collar), physical (GPS tracking + algorithmic direction — typical blue-collar), and full (all of the above, like gig economy riders). 4/10

This used to be a gig economy thing. But it's spreading to regular jobs across all sectors. According to AIMWORK, the largest EU survey on the topic (70K+ workers in all Member States), around 60% of EU workers using digital tools face some form of digital monitoring or algorithmic management. 3/10

What do we mean by 'platformisation of work'? The increasing use of digital platforms to coordinate work through data collection (digital monitoring) and algorithms (algorithmic management). GPS tracking of drivers, automated shift allocation, computer activity logging — that kind of thing. 2/10

Job quality and the platformisation of regular work: A cross-country analysis of digital monitoring and algorithmic management in the EU Digitalisation and the increasing use of digital platforms for the coordination of work processes in an algorithmic way can have deep implications for the world of work. While digital monitoring and a...

New paper out! With Álvaro Mariscal de Gante and Ignacio González-Vázquez, we analyse how the platformisation of regular work affects job quality across the EU. We find negative effects for some types of platformisation but not others, and big country variation. 1/10 #EconSky #sociology

Casi 200 años de voto en Europa por familias de partidos. El sistema nunca había estado tan fragmentado — y la extrema derecha vuelve a niveles de los años 30.

(por Daniele Caramani)

Nuevas tecnologías al rescate de ideas viejas: cómo los algoritmos están allanando el terreno a los ultras Celebrar y comprar las falsedades, medias verdades y trampas cognitivas que proliferan y premian los algoritmos a través de las redes sociales es siempre la opción más cómoda y fácil. En las cámaras d...

Escribo hoy en @elsaltodiario.com

Nuevas tecnologías al rescate de ideas viejas: cómo los algoritmos están allanando el terreno a los ultras | www.elsaltodiario.com/opinion/algo...

Digital Monitoring, Algorithmic Management and the Platformisation of Work in Europe This report presents new evidence on the platformisation of work in the European Union, examining the prevalence and potential impacts of digital tools, digital monitoring and algorithmic management. ...

You can find sectoral breakdowns here: publications.jrc.ec.europa.eu/repository/h.... Cheers!