Public defence in Computer Science, M.A. Kaisla Kajava

Title of the thesis: Evolving multi-stakeholder discourse on general-purpose artificial intelligence systems and their governance

Thesis defender: Kaisla Kajava
Opponent: Associate Professor Simone Natale, University of Turin, Italy
Custos: Professor Petri Vuorimaa, Aalto University School of Science

This PhD thesis examines how discourse on general-purpose artificial intelligence (GPAI) and its governance evolved in the European policy context between 2020 and 2024, a period marked by the release of ChatGPT and the negotiation of the EU's AI Act (AIA), the world's first comprehensive legal framework for artificial intelligence. 

GPAI refers to systems, such as large language models, that can be applied across a broad range of uses rather than a single task. The study draws on EU policy documents, public communications from major AI providers, multi-stakeholder feedback from the AIA consultation process, and policy media coverage. It applies a mixed-methods research design, leveraging Natural Language Inference (NLI) as an assistive tool for scanning large corpora. 

The thesis follows the view that language is not a neutral tool for communicating about technology; discourse shapes what aspects of AI are considered governable, which interventions are seen as legitimate, and whose expertise is recognized as authoritative. 

Three themes organize the findings. First, AI governance discourse has long framed AI as a future prospect. The ChatGPT release collapsed that distance, shifting discourse from anticipation to immediacy and, in doing so, accelerating the framing of AI regulation as security infrastructure. Second, scale, from both technical and social perspectives, became a central lens through which GPAI risks and governance needs were understood. In policy discourse, scale justified cybersecurity and sovereignty concerns; in industry communications, it was intertwined with AI complexity, and used to argue that governance required specialized expertise, positioning companies as indispensable regulatory partners. Third, the thesis examines how agency was distributed through language, including how AI was constructed as an autonomous actor, users cast as active risk agents, and expertise with regard to technical complexity presented as a prerequisite for legitimate governance participation, with consequences for which voices carry authority in regulatory debates. 

Most broadly, the thesis advances the view that language is one of the primary mechanisms through which governance is constituted. As GPAI becomes further embedded across sectors and regulatory frameworks continue to evolve, attending to the language through which AI is understood, contested, and legitimized remains significant for researchers, practitioners, and citizens alike.

Keywords: AI discourse, general-purpose AI, AI governance, natural language processing, mixed methods

Thesis available for public display 7 days prior to the defence at Aalto University's public display page.