Chapter 1 - Are AI models explainable, interpretable, and understandable?

Abstract

Artificial intelligence (AI) has been evolved for the last 50 years. The recent AI techniques, machine learning (ML) and deep learning, have been successfully established in many fields such as recommendation, computer vision, machine translation, social media, and system diagnostics. The highly performing AI models, such as ensembles and neural networks, overwhelmed the prediction performance of traditional AI techniques such as symbolic and logic-based expert systems, but instead they turned to be increasingly difficult for humans to interpret and understand the model behavior. Yet, the explain ability of AI and the interpretability of ML are requisite when the AI-based systems are to be adopted for critical decision making in real world. How can you trust them for important decisions, if you cannot understand how AI works? Unfortunately, the systematic methods for the explainable AI (XAI) are not yet mature even in academia. This chapter will introduce the meaning of XAI, as well as the category of AI explanation and interpretation methods. In addition, the applications of XAI and the related important issues will be discussed.