A Study on Interactive, Contrastive Explanation in Explainable AI

From the Philosophical Perspectives of Explanation and Causation

Abstract

While the development of artificial intelligence (AI) has been remarkable, the black-box nature of the underlying machine learning, especially deep learning, has been an obstacle to its implementation in society with respect to trust and responsibility. To solve these black-box problems, not only technical efforts to implement transparency and accountability have rapidly been made in explainable AI community, but in recent years, some research has also begun to address philosophical questions about the nature of explanation. One of the existing research is Mittelstadt et al. (2019), which calls for the development of explainable AI to provide interactive, contrastive explanations, based on the analysis of the concept of explanation by Miller (2019). In this paper, first, we illustrate the need for explainable AI with pneumonia risk prediction system case, next, review Mittelstadt et al. (2019) and then, discuss utilities of the interactive, contrastive explanation which is proposed in it.