This thesis aims to investigate the mechanisms of reproduction of gender inequalities within the Italian academic system, observing how these asymmetries are inherited and mediated by new generative Artificial Intelligence technologies. Drawing on the perspective of Science and Technology Studies (STS), the analysis deconstructs the myth of algorithmic neutrality to demonstrate that Large Language Models (LLM) do not act as passive mirrors of reality, but as actors of discursive mediation capable of consolidating pre-existing stereotypes. The research analyzes the social construction of the scientist, highlighting the persistence of the "ideal worker" model and the "male genius" archetype within the collective imagination and institutional structures. Through the analysis of ministerial data and the examination of the impact of the Gelmini Reform (L. 240/2010), the study brings to light the phenomena of horizontal and vertical segregation that continue to characterize scientific careers in Italy. The empirical section of the thesis tests the hypothesis that these structural and cultural distortions are embedded within LLM training datasets. Through a series of tests conducted on various AI interfaces, the investigation explores the presence of gender bias in the responses provided by the models when queried about academic figures and professional career paths.
Il presente lavoro di tesi si propone di indagare i meccanismi di riproduzione delle disuguaglianze di genere all’interno del sistema accademico italiano, osservando come tali asimmetrie vengano ereditate e mediate dalle nuove tecnologie di Intelligenza Artificiale generativa. Partendo dalla prospettiva degli Science and Technology Studies (STS), l’analisi decostruisce il mito della neutralità algoritmica per dimostrare come i Large Language Models (LLM) non agiscano come specchi passivi della realtà, ma come attori della mediazione discorsiva capaci di consolidare stereotipi preesistenti. Si analizza la costruzione sociale dello scienziato, evidenziando la persistenza del modello del “lavoratore ideale” e del “genio maschile” nell’immaginario collettivo e nelle strutture istituzionali. Attraverso l'analisi dei dati ministeriali e l’esame dell’impatto della Riforma Gelmini (L. 240/2010), vengono messi in luce i fenomeni di segregazione orizzontale e verticale che continuano a caratterizzare le carriere scientifiche in Italia. La sezione empirica della tesi testa l'ipotesi che tali distorsioni strutturali e culturali siano incorporate nei dataset di addestramento degli LLM. Attraverso una serie di test condotti su diverse interfacce di intelligenza artificiale, l'indagine esplora la presenza di bias di genere nelle risposte fornite dai modelli quando interrogati su figure e percorsi professionali accademici.
Oltre il mito della neutralità algoritmica: un’analisi sociotecnica sulle asimmetrie di genere nei Large Language Models all’interno del contesto dell’università italiana
PIZZIOLO, LILIANA
2025/2026
Abstract
This thesis aims to investigate the mechanisms of reproduction of gender inequalities within the Italian academic system, observing how these asymmetries are inherited and mediated by new generative Artificial Intelligence technologies. Drawing on the perspective of Science and Technology Studies (STS), the analysis deconstructs the myth of algorithmic neutrality to demonstrate that Large Language Models (LLM) do not act as passive mirrors of reality, but as actors of discursive mediation capable of consolidating pre-existing stereotypes. The research analyzes the social construction of the scientist, highlighting the persistence of the "ideal worker" model and the "male genius" archetype within the collective imagination and institutional structures. Through the analysis of ministerial data and the examination of the impact of the Gelmini Reform (L. 240/2010), the study brings to light the phenomena of horizontal and vertical segregation that continue to characterize scientific careers in Italy. The empirical section of the thesis tests the hypothesis that these structural and cultural distortions are embedded within LLM training datasets. Through a series of tests conducted on various AI interfaces, the investigation explores the presence of gender bias in the responses provided by the models when queried about academic figures and professional career paths.| File | Dimensione | Formato | |
|---|---|---|---|
|
TESI.pdf
Accesso riservato
Dimensione
4.98 MB
Formato
Adobe PDF
|
4.98 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/106656