IEEE Approved Draft Guide for an Architectural Framework for Explainable Artificial Intelligence

Dramatic success in machine learning has led to a new wave of artificial intelligence applications that offer extensive benefits to our daily lives. The loss of explainability during this transition, however, means vulnerability to vicious data, poor model structure design, and suspicion of stakeholders and the general public -- all with a range of legal implications. The dilemma has called for the study of explainable AI (XAI) which is an active research field that aims to make AI systems results more understandable to humans. This is a field with great hopes for improving the trust and transparency of AI-based systems and is considered a necessary route for AI to move forward. This guide provides a technological blueprint for building, deploying and managing machine learning models while meeting the requirements of transparent and trustworthy AI by adopting a variety of XAI methodologies. It defines the architectural framework and application guidelines for explainable AI, including: 1) description and definition of XAI, 2) the types of XAI methods and the application scenarios to which each type applies, 3) performance evaluation of XAI.