2: AI Governance: anticipating the EU AI Act (Irene Niet, Céline Budding)
Artificial Intelligence (AI) is one of the most discussed technologies in 2023. The technology can be implemented in a wide range of sectors, with various goals and intents, prompting some to term it a ‘general purpose technology’ (1). The European Commission (EC) aims to harness the opportunities AI brings, for example in the twin transition (the dual sustainability and digital transition) (2). Simultaneously, the EC recognizes the potential challenges of AI. In an attempt of anticipatory governance and to guide the development and implementation of AI, the EU proposed the EU AI Act in April 2021; negotiations are ongoing (3).
In the draft act, the EC has opted for a risk-based approach when it comes to AI (4). They have formulated rules and regulations based on three levels of risk. AI with unacceptable risk is not allowed. High-risk AI can be implemented but is subjected to strict(er) rules, measures, and assessments. The details of these additional requirements have thus far not been explained (5). Finally, limited risk AI can be implemented with the fewest regulations, as long as it is transparent when AI is used and for what purpose (3). Stakeholders struggle with the practicalities of this proposed legislation. Partly, this is because the EC has identified the implementation of AI in many sectors (including energy, education, and employment) as high-risk. Consequences of this are unclear. Additionally, the EU AI Act requires assessment of AI, but is uncertain who is responsible for these assessments. In its proposed amendment, the Council of the European Union designated national authorities to deal with the assessments (6), although it is doubtful whether these institutions have the capacity to do so (5,7).
During this session, we therefore have two aims. First, to gain a broad understanding of the implementation implications of the EU AI Act. We do so through several presentations from different disciplines or implementation sectors. Second, we discuss what an ideal implementation of the AI Act entails. What governance is needed to implement the EU AI Act well? What is needed in the areas of explainable AI and risk-based assessments? And who should be responsible for the assessments? The session has two rounds: 1. A paper presentation round. In approximately 1 hour, 3 papers are presented concerning the implementation implications of the EU AI Act. These will be selected following a call for papers and a selection process. We are interested in papers discussing the implementation of the EU AI Act, what a risk-based assessment would entail for AI and who should carry out the assessments. 2. A discussion round. In this 30-minute second part of the session, we intend to have a lively discussion (following the format of a round table). The participants are those who have presented their papers and those who wish to join the session. The discussion will centre around the question: What does the ideal implementation of the AI Act entail and what governance is needed to create this?
Keywords: EU AI Act, governance, explainable AI, transparency, risk-based assessment