Governing AI : En studie av problemrepresentationer och styrningslogiker i EU:s AI-förordning

Sammanfattning

The purpose of this study is to examine how the EU AI Act (2024) has been adopted as a regulatory framework, with the principal focus being on how “problems” are represented. The thesis examines how the AI Act represents and addresses specific problems associated with AI, such as disinformation and deepfakes, biometric mass surveillance, discrimination and bias, privacy risks, cybersecurity, and robustness of the digital infrastructure. The thesis also examines which risks are accentuated, which are marginalized and “unspoken”, and how knowledge and power shape the governing rationality of the European Union regarding AI regulation. The analysis is conducted through the application of Carol Bacchi’s What’s the Problem Represented to be? (WPR) method in combination with a Governmentality-perspective. Through the use of qualitative document analysis, the European AI Act is analyzed to identify dominant discourses, underlying problem assumptions, and the effects produced by these. Our findings indicate that the regulation mainly represents AI-related risks as technical and manageable through “risk-based” classification, compliance mechanisms and certain transparency obligations. Although certain issues, such as discrimination and surveillance are recognized, they are primarily addressed through technical precautions rather than social, political and economic interventions at a structural level. Other risks, such as power disparities within the AI industry and broader democratic implications such as public- and institutional trust - remain largely under-problematized and “forgotten”. Finally, the thesis concludes that the EU AI Act serves, not only as a judicial instrument, but also as a tool for governance, which shapes how risks regarding AI are understood, represented and addressed. Through a combination of risk rationalities and habitual values, the European Union also accomplishes a certain indirect governance by defining acceptable practices, responsibilities and legitimate knowledge surrounding AI.