scope:

This guide provides a technological framework that aims to increase trustworthiness of AI systems using explainable artificial intelligence (XAI) technologies and methods. The document also provides measurable solutions to evaluate AI systems in terms of explainability. Specifically, the document illustrates the following aspects of XAI systems:

a) The requirements of providing human-understandable explanations for AI systems in different use-cases, e.g., healthcare and financial applications;

b) Approaches to offer a series of available tools for giving an AI model tenable explanations;

c) A set of measurable solutions to evaluate AI systems and corresponding performance, such as availability, resiliency, accuracy, safety, security, and privacy, of AI system under such status.

Purpose

Artificial Intelligence systems are expected to be trustworthy and adhere to human ethics principles including fairness, privacy and transparency etc. To achieve the trustworthiness requires underlying mechanisms of AI systems are transparent, and understandable to all stakeholders of AI systems. This mandate motivates the study of a variety of explainable artificial intelligence (XAI) technologies and methods. Given the urgent need and profound influence of XAI, this guide aims to provide a technological framework that facilitates the adoption of relevant methods, evaluation and comparison of different approaches, and showcases typical scenarios in which XAI can bring great values to AI system stakeholders and our society.