The Past and the Future of Explainable AI Techniques

Abstract

Machine learning models of high predictive performance, such as deep neural networks and ensemble models, now play a central role in the current artificial intelligence technologies and have started to be applied to the problems related to our health or properties. However, one of the primary obstacles here is the opacity of such high-performance models. So far, dozens of techniques for reducing the opacity have been explored, and form a research field called “explainable aritificial intelligence (XAI).” In this paper, I review the past literature on XAI, organize key concepts and techniques in the current XAI research, and discuss the future direction of XAI.