A sociotechnical system perspective on AI.

Abstract

The aim of this special issue is to provide a sociotechnical system perspective on Artificial Intelligence (AI). Discussions on AI still often focus on the technology itself rather than on the broader systems in which it functions. This is even true when it comes to the social and ethical issues raised by AI. For example, people often talk about fair algorithms or explainable AI, as if fairness and explainability depend solely on the technology and not also on the broader sociotechnical systems in which AI technologies are embedded. Consider, for example, an AI algorithm that is used by a government agency to find potential cases of fraud with social welfare. Fairness in this case no doubt partly depends on the algorithm itself; for example does it meet certain fairness metrics as they have now been proposed in the literature (e.g., Mehrabi et al., 2019)? However, there are multiple and conflicting fairness metrics, and in order to decide on which of these to focus, one first needs to know more about the broader context: what are the main unfairness that might occur in this context? Which people are most vulnerable and dependent on governmental decisions and should be protected from unfair decisions?