From explanations to shared understandings of AI

Abstract

A key challenge in the design of AI systems is how to support people in understanding them. We address this challenge by positioning explanations in everyday life, within ongoing relations between people and artificial agents. By reorienting ex-plainability through more-than-human design, we call for a new approach that consid-ers both people and artificial agents as active participants in constructing understand-ings. To articulate such an approach, we first review the assumptions underpinning the premise of explaining AI. We then conceptualize a shift from explanations to shared understandings, which we characterize as situated, dynamic, and performative. We conclude by proposing two design strategies to support shared understandings, i.e. looking across AI and exposing AI failures. We argue that these strategies can help designers reveal the hidden complexity of AI (e.g., positionality and infrastructures), and thus support people in understanding agents' capabilities and limitations in the context of their own lives.