On Design and Evaluation of Human-centered Explainable AI systems
On Design and Evaluation of Human-centered Explainable AI systems
复制标题
以人为中心的可解释人工智能系统的设计和评估
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Upol Ehsan
中科院分区：
文献类型：
--
作者：
Upol Ehsan
随着人工智能系统在我们的生活中变得无处不在，等式的人性化方面需要仔细研究。设计和评估“黑箱”人工智能系统的挑战主要取决于人类在循环中的角色。解释被视为事后可解释性的一种形式，可以帮助AI代理和用户之间建立融洽、信任和理解，特别是在理解失败和意外AI行为时。为了有效地设计和评估解释生成系统，我们需要更深入的端到端调查，将完全实现的人工智能代理和自动解释生成系统纳入用户研究。在本文中，我们提出了一个案例研究，重点是非专家用户如何感知不同风格的自动生成的理由由AI代理沿着的信心，人性化，充分的理由，和可理解性的尺寸。我们总结了我们的研究结果，并提供了一个迫切需要解决的研究问题。
AsAI systems become ubiquitous in our lives, the human side of the equation needs careful investigation. The challenges of designing and evaluating "black-boxed" AI systems depends crucially on who the human is in the loop. Explanations, viewed as a form of post-hoc interpretability, can help establish rapport, confidence, and understanding between the AI agent and the user, especially when it comes to understanding failures and unexpected AI behavior. To effectively design and evaluate explanation generation systems, we need deeper end-to-end investigations of incorporating fully-realized AI agents and automated explanation generation systems into user studies. In this paper, we present a case study that focuses on how non-expert users perceive different styles of automatically generated rationales by an AI agent along the dimensions of confidence, humanlike-ness, adequate justification, and understandability. We summarize our results and provide a desiderata of research questions yet to be addressed.