Politics of transparency in Artificial Intelligence Policy - Evidence from public consultation on the European Union Artificial Intelligence Act

Transparency plays a central role in the governance of artificial intelligence (AI). In current scholarly and policy debates the need for transparency is juxtaposed against the opacity of black-box algorithmic decision making systems, which gives transparency a technical or sociotechnical framing. Yet the sociotechnical framings of transparency beg a deeper question: what political goals does transparency serve in AI policy? Furthermore, do the proposed solutions match with the stated political goals in AI policy? These questions are far from trivial, because transparency and openness are already serving a variety of sometimes incompatible political goals in other fields of policy. To answer these questions, this paper analyses the understanding of transparency in the submissions to a public consultation on the European Union AI act conducted in 2021 (N=304). The paper draws from critical policy analysis literature, using qualitative content analysis to identify both problem framings and proposed solutions in the submissions. The paper fills a gap in public policy research on AI, which has focused on analysing AI policy documents or individual regulatory proposals, whereas current paper presents empirical evidence on the policy debate itself. The findings have empirical value both as evidence on the debate on AI Act itself and as a large international sample of submissions on AI policy.