Interest in the ethics of artificial intelligence has grown rapidly in recent years and produced a broad variety of policies, frameworks, and academic work intended to make the development, usage, and governance of AI more trustworthy and ethical. Bringing these types of works down to the ground to actually make AI fair and trustworthy requires translating abstract normative concepts such as transparency, fairness, and accountability into practical requirements and design decisions. This keynote discusses this process of translating AI ethics principles into practice and the practical harms that can result if ethics is oversimplified or ignored. How can ethical principles concerning concepts like ‘bias’ and ‘fairness’ be operationalised for traditional AI systems, and how can attempts to make AI ‘fair’ inadvertently lead to more harmful systems in practice?

In this panel debate we discuss if algorithmic fairness is possible. We ask how effective are current EU and national policies in addressing bias, discrimination, and risk in AI systems? How do we translate principles into practice? 

There is increasing awareness of the possibility that bias or discrimination can occur where AI is used to make or support decisions. This talk will examine how the tests for direct and indirect discrimination under EU law apply to AI. The talk will then consider how such discrimination can be tackled through legal mechanisms, and whether current legal protections are adequate to ensure that AI discrimination can be detected and addressed.

The panel will focus on practical governance and regulatory approaches that help ensure AI and automated decision-making systems promote equality and prevent discrimination, highlighting what has proven effective in practice and how these approaches can be applied in both the public and private sectors.