Social eXplainable AI (Social XAI): Towards Expanding the Social Benefits of XAI.

Abstract

The eXplainable Artificial Intelligence (XAI) research agenda is constantly expanding. From a mainly technical topic, it has now developed into a multidisciplinary area of research, supporting domain experts, policymakers, and lay users. However, humans working in domains, such as healthcare, media and policing still find themselves questioning the value of XAI and whether it is worth their time and effort. This hampers effective human–machine collaboration and results in various trust-related biases. An emerging field of social XAI introduces the need to operationalise XAI and overcome its potential user issues. For example, XAI has been shown to lead to over-reliance on AI, for some in contrast to algorithmic aversion or is simply ignored as a redundant feature. Based on current work in Human-Centered XAI, I outline a set of design features that could improve human-XAI interaction and ultimately expand the social benefits of XAI. I present a framework of three dimensions that shows how the current work in Human-Centered XAI is increasing the effectiveness of the human-XAI interaction.