“OF COURSE AI DISCRIMINATES”: IDENTIFYING COMMUNICATION GAPS AND CROSS-DISCIPLINARY TRANSLATION CHALLENGES

The paper argues for the need to foreground anti-discrimination as a distinct lens when assessing the social impacts of AI design, development and deployment in real-life situations. The argument is based on a survey of around 200 organisations in the public sector in Norway as well as 19 in-depth interviews and presents the challenges of translating ‘discrimination' as a socially relevant concept across disciplines and discursive contexts. The paper presents six discursive responses to the risk of discrimination in our study to foreground how focusing on discrimination allows one to address unique challenges that other concepts such as bias and privacy cannot address. By distinguishing concerns around discrimination from other ancillary concerns, such as bias and privacy, we present the need to ground our critical understanding of AI design, development and deployment in actual practices and situations and urge AI developers to actively adopt an anti-discriminatory lens in their practices without replacing it with ancillary concepts such as bias, differentiation, privacy, or other mainstream concepts such as justice or ethics.