Towards “Explorable” AI: Learning from ML Developers’ Sensemaking Practices

Abstract

In this note, we report on a qualitative design study in the field of machine learning (ML) and in particular on the sensemaking practices of ML developers as they interact with the interface of a novel adversarial AI method. This paper makes contributions to discourses on interpretable or explainable AI (XAI) systems through an empirical understanding of ML developers’ sensemaking practices. These findings make salient the concept of “explorability” as an alternative design metaphor for interactive AI systems – instead of a focus on explainability or interpretability as fixed qualities of AI systems, explorability focuses on emergent meanings and ways in which they might be enabled or constrained through practice.