The interrelation between data and AI ethics in the context of impact assessments
- PDF / 842,266 Bytes
- 7 Pages / 595.276 x 790.866 pts Page_size
- 4 Downloads / 176 Views
ORIGINAL RESEARCH
The interrelation between data and AI ethics in the context of impact assessments Emre Kazim1 · Adriano Koshiyama1 Received: 9 July 2020 / Accepted: 2 November 2020 © The Author(s) 2020
Abstract In the growing literature on artificial intelligence (AI) impact assessments, the literature on data protection impact assessments is heavily referenced. Given the relative maturity of the data protection debate and that it has translated into legal codification, it is indeed a natural place to start for AI. In this article, we anticipate directions in what we believe will become a dominant and impactful forthcoming debate, namely, how to conceptualise the relationship between data protection and AI impact. We begin by discussing the value canvas i.e. the ethical principles that underpin data and AI ethics, and discuss how these are instantiated in the context of value trade-offs when the ethics are applied. Following this, we map three kinds of relationships that can be envisioned between data and AI ethics, and then close with a discussion of asymmetry in value trade-offs when privacy and fairness are concerned. Keywords Artificial intelligence · Machine learning · Data · Data protection · Ethics · Audit · Impact assessment
1 Introduction In the growing literature on artificial intelligence (AI) impact assessments, which includes technological auditing of metrics such as privacy, fairness and performance, and human rights, social and environmental impact assessments [1], the literature on data protection impact assessments (DPIA) is heavily referenced and drawn upon [2–5]. Given the relative maturity of the data protection debate and that it has translated into legal codification (most explicitly in the general data protection regulation (GDPR)) [6–9], drawing upon it is a natural place to start for AI. Indeed, when legality is referenced, the GDPR legislation is often mapped on to discussions regarding compliance of AI systems [3–5]. A paradigmatic example of this can be found in the UK’s Information Commissioner’s Office ‘Guidance on AI and data protection’ [5]. In this article, we anticipate directions in what we believe will become a dominant and impactful debate, namely how to conceptualise the relationship between data protection (which we read mainly as an expression of the value of
* Emre Kazim [email protected] 1
University College London, London, UK
privacy) and AI impact (which we read predominantly as an expression of the value of fairness). We begin by discussing the value canvas i.e. the principles that underpin data and AI ethics, and discuss how these are instantiated in the context of value trade-offs when the ethics are applied. Following this, we map a triad of potential relationships between data and AI ethics: • AI and data as pyramidal, with data protection being the
foundation;
• consideration of AI impact will cause a renegotiation of
data protection concerns and the two will be integrated in this process; and • AI impact assessments should be independent of data pr
Data Loading...
