Responsible AI is the practice of developing and using AI systems in a way that benefits the humans, society, and environment, while minimising the risk of negative consequences. Various responsible AI principles have been released recently. However, those principles are very abstract and not practical enough. Further, significant efforts have been put on algorithm-level solutions which are usually confined to a narrow set of principles (such as fairness and privacy). To bridge the gap, we adopt a pattern-oriented approach and build a responsible AI pattern catalogue for operationalising responsible AI from a system perspective. In this article, we first summarise the major challenges in operationalising responsible AI at scale and introduce how we use responsible AI pattern catalogue to address those challenges. Then, we discuss the case study we have conducted using the chatbot development use case to evaluate the usefulness of the pattern catalogue.

本文提出通过模式驱动的负责任的 AI 工程方法和构建一种能从系统角度实现负责任 AI 的垫片目录来填补负责任 AI 的操作化的空白，研究了 chatbot 开发过程中的风险，并提供了垫片驱动的缓解措施。

金融服务的负责任聊天机器人开发: 一种面向模式的负责任人工智能工程方法