The EU AI act in development practice: a pro-justice approach

Abstract With the adoption of the EU AI Act, organisations must address compliance obligations, including with ambiguous aspects of the Act, such as risk management and fundamental rights impact assessments. In this paper we show how feminist and justice-oriented ethical frameworks can be leveraged to clarify this ambiguity and to help organisations translate AI ethics theory into practice, in the form of a pro-justice compliance toolkit. The toolkit introduced, “High Risk EU AI Act Toolkit”, or “HEAT” was the result of a cross-sector and multidisciplinary collaboration engaging experts in law, technology, design, justice, and other disciplines from academia, government and industry. Herein we describe how a pro-justice approach contributed practical and concrete techniques, methods, and guidelines for everyday implementation of AI Act Compliance. Specifically, we describe: (1) how feminist understanding of expertise and acknowledgment of partiality of knowledge shaped the HEAT method for engaging stakeholders, including minoritized groups; (2) how our pro-justice approach expands on the Act’s minimal requirements in relation to accessibility through the incorporation of disability and design justice; (3) how we encourage HEAT users to engage with feminist data science and (4) how our pro-justice interpretation of the Act presents environmental concerns as inseparable from other concerns about the social impact of AI.