It is now widely recognized that artificial intelligence can reshape work, the economy, and their future trajectories. While public debates have mainly focused on whether AI substitutes for human labour, its effects extend beyond, influencing labour market structures, working conditions, the distribution of economic power between firms and workers, and required abilities and skills. This PhD thesis adopts this broader perspective and analyses the impact of artificial intelligence on labour markets in Europe and the United States along three complementary dimensions. First, a structural “objective” dimension, concerning changes in occupational composition and employment shares associated with exposure to AI. Second, an “objective” institutional and distributive dimension, focusing on transformations in firms’ market power on the labour demand side, measured through monopsony. Third, a “subjective” dimension, centred on changes in job content—namely abilities, skills, and knowledge—and how these evolve with advances in AI. Each chapter addresses one of these dimensions through a distinct yet conceptually integrated empirical analysis. The first chapter investigates the relationship between occupational exposure to AI and changes in the structure of the European labour market across 25 countries from 2011 to 2021, using AI exposure measures from the literature and data from the European Labour Force Survey (LFS-EU). Results show that employment shares increased in occupations more exposed to AI, suggesting a complementary rather than substitutive effect of technology during the period considered. Stratification by education reveals employment growth in low- and high-skill occupations and a decline in medium-skill jobs, consistent with theories of Routine Biased Technological Change and job polarization. Younger workers experience relative employment gains, potentially reflecting greater adaptability to AI-related skills. Country-level analyses highlight substantial heterogeneity, pointing to the role of national institutions, education systems, and the pace of AI diffusion. The second chapter, co-authored with Michele Cantarella and Chiara Strozzi, examines the relationship between monopsony power and occupational exposure to AI in 26 European countries between 2011 and 2020. Using LFS data on wages and employment and Google Patents to construct an AI exposure index, monopsony power is measured through the wage elasticity of labour supply. Results show a marked decline in labour supply elasticity over time, indicating increasing monopsony power. This trend appears largely independent of AI exposure, which plays a limited role. Stratified analyses indicate that monopsony power is strongest in low-wage occupations, followed by high-wage ones, while middle-wage occupations exhibit weaker monopsony. A rising trend is also observed in low-education occupations. The third chapter, co-authored with Michele Cantarella and Chiara Strozzi, focuses on how AI reshapes the subjective content of work rather than task structures. Using longitudinal O*NET data from 2011 to 2025 and two novel AI exposure measures—based on AI research topics and LLM benchmark performance—we construct exposure indices at the occupation–requirement–year level. Results show that within occupations, AI exposure is positively associated with increases in importance-weighted requirement levels across Abilities, Skills, and Knowledge. At the occupation level, however, higher AI exposure is associated with a decline in overall requirement levels, driven mainly by reductions in Abilities. This divergence suggests that AI strengthens specific complementary requirements while contributing to an erosion of average occupational requirements.
È ormai ampiamente riconosciuto che l’intelligenza artificiale possa rimodellare la struttura del mercato del lavoro. Questa tesi di dottorato analizza l’impatto dell’intelligenza artificiale sui mercati del lavoro in Europa e negli Stati Uniti lungo tre dimensioni complementari. La prima è una dimensione strutturale “oggettiva”, relativa ai cambiamenti nella composizione occupazionale associata all’esposizione all’IA. La seconda è una dimensione “oggettiva” di tipo istituzionale e distributivo, che riguarda le trasformazioni del potere di monopsonio delle imprese. La terza è una dimensione “soggettiva”, incentrata sui cambiamenti nel contenuto del lavoro — in termini di abilità, skill e conoscenze — e sul modo in cui questi evolvono con i progressi dell’IA. Il primo capitolo analizza la relazione tra l’esposizione occupazionale all’IA e i cambiamenti nella struttura del mercato del lavoro europeo in 25 paesi nel periodo 2011–2021, utilizzando dati della Labour Force Survey europea (LFS-EU). I risultati mostrano che le quote di occupazione sono aumentate nelle occupazioni più esposte all’IA, suggerendo un effetto complementare, piuttosto che sostitutivo, della tecnologia. La stratificazione per livello di istruzione evidenzia una crescita dell’occupazione nelle professioni a bassa e alta qualificazione e una riduzione in quelle a qualificazione intermedia, in linea con le teorie della job polarization. I lavoratori più giovani registrano guadagni occupazionali relativi, probabilmente grazie a una maggiore adattabilità all’acquisizione di competenze compatibili con l’IA. Le analisi a livello nazionale mettono in luce una forte eterogeneità. Il secondo capitolo, scritto in collaborazione con Michele Cantarella e Chiara Strozzi, esamina la relazione tra il potere di monopsonio e l’esposizione occupazionale all’IA in 26 paesi europei tra il 2011 e il 2020. Utilizzando dati LFS su salari e occupazione e dati di Google Patents per costruire un indice di esposizione all’IA, il potere di monopsonio viene misurato attraverso l’elasticità dell’offerta di lavoro rispetto al salario. I risultati mostrano una marcata diminuzione di tale elasticità nel tempo, indicando un aumento del potere di monopsonio delle imprese. Questo andamento risulta in larga parte indipendente dall’esposizione all’IA, che svolge un ruolo limitato. Le analisi stratificate indicano che il potere di monopsonio è più forte nelle occupazioni a basso salario, seguito da quelle ad alto salario, mentre le occupazioni a salario medio presentano un monopsonio più debole. Si osserva inoltre un trend crescente di monopsonio nelle occupazioni a basso livello di istruzione. Il terzo capitolo, scritto in collaborazione con Michele Cantarella e Chiara Strozzi, si concentra su come l’IA stia rimodellando il contenuto soggettivo del lavoro piuttosto che le strutture delle mansioni. Utilizzando dati longitudinali O*NET dal 2011 al 2025 e due nuove misure di esposizione all’IA — basate sui temi della ricerca in IA e sulle prestazioni dei LLM nei benchmark — costruiamo indici di esposizione a livello occupazione– abilità/skill/conoscenza–anno. I risultati mostrano che, all’interno delle occupazioni, l’esposizione all’IA è positivamente associata a un aumento dei livelli dei requisiti ponderati per importanza nelle dimensioni di abilità, skill e conoscenze. A livello occupazionale, tuttavia, una maggiore esposizione all’IA è associata a una riduzione dei livelli complessivi delle caratteristiche, trainata principalmente da una diminuzione delle abilità. Questa divergenza suggerisce che l’IA rafforza requisiti specifici e complementari, contribuendo al contempo a un’erosione di abilità, skill e conoscenze medie delle occupazioni.
Intelligenza Artificiale e cambiamenti nella struttura del mercato del lavoro
MOLINARI, GIUSEPPE
2026
Abstract
It is now widely recognized that artificial intelligence can reshape work, the economy, and their future trajectories. While public debates have mainly focused on whether AI substitutes for human labour, its effects extend beyond, influencing labour market structures, working conditions, the distribution of economic power between firms and workers, and required abilities and skills. This PhD thesis adopts this broader perspective and analyses the impact of artificial intelligence on labour markets in Europe and the United States along three complementary dimensions. First, a structural “objective” dimension, concerning changes in occupational composition and employment shares associated with exposure to AI. Second, an “objective” institutional and distributive dimension, focusing on transformations in firms’ market power on the labour demand side, measured through monopsony. Third, a “subjective” dimension, centred on changes in job content—namely abilities, skills, and knowledge—and how these evolve with advances in AI. Each chapter addresses one of these dimensions through a distinct yet conceptually integrated empirical analysis. The first chapter investigates the relationship between occupational exposure to AI and changes in the structure of the European labour market across 25 countries from 2011 to 2021, using AI exposure measures from the literature and data from the European Labour Force Survey (LFS-EU). Results show that employment shares increased in occupations more exposed to AI, suggesting a complementary rather than substitutive effect of technology during the period considered. Stratification by education reveals employment growth in low- and high-skill occupations and a decline in medium-skill jobs, consistent with theories of Routine Biased Technological Change and job polarization. Younger workers experience relative employment gains, potentially reflecting greater adaptability to AI-related skills. Country-level analyses highlight substantial heterogeneity, pointing to the role of national institutions, education systems, and the pace of AI diffusion. The second chapter, co-authored with Michele Cantarella and Chiara Strozzi, examines the relationship between monopsony power and occupational exposure to AI in 26 European countries between 2011 and 2020. Using LFS data on wages and employment and Google Patents to construct an AI exposure index, monopsony power is measured through the wage elasticity of labour supply. Results show a marked decline in labour supply elasticity over time, indicating increasing monopsony power. This trend appears largely independent of AI exposure, which plays a limited role. Stratified analyses indicate that monopsony power is strongest in low-wage occupations, followed by high-wage ones, while middle-wage occupations exhibit weaker monopsony. A rising trend is also observed in low-education occupations. The third chapter, co-authored with Michele Cantarella and Chiara Strozzi, focuses on how AI reshapes the subjective content of work rather than task structures. Using longitudinal O*NET data from 2011 to 2025 and two novel AI exposure measures—based on AI research topics and LLM benchmark performance—we construct exposure indices at the occupation–requirement–year level. Results show that within occupations, AI exposure is positively associated with increases in importance-weighted requirement levels across Abilities, Skills, and Knowledge. At the occupation level, however, higher AI exposure is associated with a decline in overall requirement levels, driven mainly by reductions in Abilities. This divergence suggests that AI strengthens specific complementary requirements while contributing to an erosion of average occupational requirements.| File | Dimensione | Formato | |
|---|---|---|---|
|
Molinari.pdf
accesso aperto
Licenza:
Tutti i diritti riservati
Dimensione
1.47 MB
Formato
Adobe PDF
|
1.47 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/360791
URN:NBN:IT:UNIMORE-360791