The rapid development of General-Purpose AI Models (GPAI), capable of performing a wide range of tasks, has raised new challenges in terms of regulation, transparency, and safety. In response, the European Union has introduced the AI Act, a regulation aimed at establishing a harmonized legal framework for the responsible development and use of AI. This thesis specifically analyzes Chapter 5 of the AI Act, which defines the obligations for providers of GPAI models, with particular focus on technical documentation, model evaluation, and the role of the “Code of Practice.” It also examines the distinction between general-purpose models and those considered to pose systemic risks, as outlined in Article 51 and Annex XIII, based on technical parameters that include the number of computational operations used for training. The analysis aims to clarify the regulatory and operational implications of the AI Act and the obligation to implement the aforementioned measures. Finally, the thesis provides a critical and up-to-date overview of the evolution of European regulation in this field, assessing its potential consequences and opportunities for the future development of artificial intelligence within the European Union.
Il rapido sviluppo dei modelli di intelligenza artificiale di uso generale (General-Purpose AI Models), capaci di svolgere una vasta gamma di compiti, ha sollevato nuove sfide in termini di regolamentazione, trasparenza e sicurezza. In risposta, l'Unione Europea ha introdotto l'AI Act, un regolamento volto a stabilire un quadro giuridico armonizzato per lo sviluppo e l'utilizzo responsabile dell'IA. Questa tesi analizza specificamente il Capitolo 5 dell'AI Act, che definisce gli obblighi per i providers di modelli di GPAI, con particolare attenzione alla documentazione tecnica, alla valutazione del modello e al ruolo del “Code of Practice”. Viene inoltre esaminata la distinzione tra modelli a impatto generale e quelli a rischio sistemico, come indicato all'Articolo 51 e all'Allegato XIII, secondo parametri tecnici che includono il numero di operazioni di calcolo utilizzata per l'addestramento. L'analisi si prefigge l'obiettivo di chiarire le implicazioni normative e operative dell’AI Act e l'obbligo di implementare le suddette misure. Infine, la tesi offre una panoramica critica e aggiornata sull'evoluzione della normazione europea nella materia, valutandone le potenziali conseguenze e opportunità per lo sviluppo futuro dell'intelligenza artificiale nell'Unione Europea.
L’Artificial Intelligence Act: Regolamentazione dei sistemi di tipo GPAI e il ruolo del “Code of Practice”
FORTIN, ANTONIO
2024/2025
Abstract
The rapid development of General-Purpose AI Models (GPAI), capable of performing a wide range of tasks, has raised new challenges in terms of regulation, transparency, and safety. In response, the European Union has introduced the AI Act, a regulation aimed at establishing a harmonized legal framework for the responsible development and use of AI. This thesis specifically analyzes Chapter 5 of the AI Act, which defines the obligations for providers of GPAI models, with particular focus on technical documentation, model evaluation, and the role of the “Code of Practice.” It also examines the distinction between general-purpose models and those considered to pose systemic risks, as outlined in Article 51 and Annex XIII, based on technical parameters that include the number of computational operations used for training. The analysis aims to clarify the regulatory and operational implications of the AI Act and the obligation to implement the aforementioned measures. Finally, the thesis provides a critical and up-to-date overview of the evolution of European regulation in this field, assessing its potential consequences and opportunities for the future development of artificial intelligence within the European Union.| File | Dimensione | Formato | |
|---|---|---|---|
|
Fortin_Antonio_tesi.pdf
Accesso riservato
Dimensione
809.62 kB
Formato
Adobe PDF
|
809.62 kB | 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/20.500.12608/93253