Key considerations for addressing critical blindspots in European AI governance

Since 2018 several European policy efforts have aimed to incorporate ethics principles and European fundamental rights into the governance of artificial intelligence (AI). One of the major multi-stakeholder initiatives was the establishment of the High Level Expert Group on Artificial Intelligence (AI HLEG). This has led to the EU “Proposal for a regulation on a European approach for AI” (AI Act) in April 2021. However a closer analysis demonstrates that the dominant EU trustworthy and risk-based approach to governing AI entails five critical blindspots in relation to the sectoral specificities and the socio-technical consequences of AI in the media and communication (MC) sector. These refer to the lack of attention to infrastructures of surveillance capitalism, increasing power asymmetries in MC sector, ignoring the important cultural, social and democratic role of media, the lack of understanding how AI sorts and targets people in opaque ways, and the push by digital platform companies for self-governance with minimal oversight. In that way European AI policies largely neglect relevant issues that make AI worthy of citizen trust.

In this contribution we identify the main considerations for addressing these blindspots in European AI governance in the MC sector, based on a systematic analysis of the meaning and impact of AI on media and communication. We propose a multi-level framework based on desk research and multi-stakeholder expert discussions. We identify three areas where additional steps need to be taken for better governing AI: (1) addressing data power asymmetry, (2) empowerment by design for mitigating risks, and (3) cooperative responsibility through stakeholder engagement.

We generally find that much more research and regulatory efforts are needed on these three key areas in order to influence the development, deployment, use and governance of AI systems in the MC sector. These efforts need to involve a range of disciplines and be cognisant of the existing European policy context and sectoral specifics. The first related to the need to address data power asymmetries by rethinking informed consent in the context of machine learning powered AI, and in the deployment and use of predictive analytics by a complex ecosystem of companies. This will also require new systems of transparency and oversight for the use of automated decision making and the dominance of a small number of commercial companies in many of these technologies. The second relates to the need to develop systems that empower citizens and that operate in the public interest, from algorithmic registries to algorithmic auditing, and from skill development to the right to redress. Finally, we consider it necessary to move away from multi-stakeholder initiatives towards new systems of cooperative responsibility and accountability to protect citizen and worker rights. While the goal of a principles based and value-driven AI governance framework across the EU is to be commended, too few voices and sectors currently dominate the policy agenda.