Yleiskäyttöisiä tekoälyjärjestelmiä ja niiden hallintoa koskevan sidosryhmien välisen puheen kehitys
Year: 2025
Type: dissertation
Abstract: This thesis examines discourse on general-purpose artificial intelligence (GPAI) and its governance in European policy contexts between 2020 and 2024. This period was marked by ChatGPT's release and the development of the European AI Act (AIA). The research adopts a mixed-methods approach, combining computational NLP techniques with qualitative inquiry. The analyzed corpus includes EU policy documents, industry communications, multi-stakeholder feedback on the AIA proposal, and media coverage spanning the AIA negotiation period and mainstream adoption of GPAI. The thesis views language as constitutive of technology governance. Discourse shapes which aspects of technology are considered governable, which interventions are legitimized, and whose expertise is recognized as authoritative. Findings indicate that policy language shifted from anticipatory preparation to contemporary considerations following ChatGPT's release. Scale of adoption emerged as a dominant framing through which GPAI capabilities and risks were understood, serving as shared vocabulary across stakeholders despite divergent interpretations. Security discourse intensified, emphasizing cybersecurity vulnerabilities and computational sovereignty while exposing tensions between EU regulatory ambitions and technological dependencies. Users were increasingly reconceptualized from passive subjects requiring protection to active agents and potential sources of risk, moving governance focus from provider accountability toward AI literacy and education. Industry communications paired references to technical complexity and infrastructure challenges with calls for collaborative governance, positioning specialized expertise as a prerequisite for effective intervention. Multi-stakeholder consultations exhibited a participation paradox: procedural inclusion occurred without substantive discursive diversification. Across sectors, stakeholders drew upon similar justificatory frameworks, invoking rights, efficiency, and innovation, yet advocated divergent implementations. Analysis of consultation feedback and subsequent legislative amendments revealed correlations between stakeholder concerns and regulatory changes, reflecting anticipation of distinct regulatory requirements for GPAI models and systems before their broader diffusion. Methodologically, the thesis demonstrates how natural language inference (NLI) models can operate as computational sampling tools, enabling large-scale corpus analysis while retaining interpretive depth. Overall, the findings shed light on how emerging technologies are presented during periods of rapid change, with implications for anticipatory governance and participation in technology regulation. This research was conducted as part of the Civic Agency in AI (CAAI) project, funded by the Kone Foundation and the Research Council of Finland.
Source: Aaltodoc (Aalto University)
Author: Kajava, Kaisla
Institution: Aalto University
Language: English
Subfield: Law
Field: Social Sciences
Domain: Social Sciences
Keywords: general-purpose artificial intelligence, AI governance, EU AI Act, ChatGPT, stakeholder dialogue, anticipatory governance, multi-stakeholder consultation, AI literacy, AI literacy education, security discourse, technology dependence, user agency, industry communication, natural language inference, technology regulation
Open Access status: green
OpenAlex ID: W7160687903
