Keeping it "organized and logical": after-action review for AI (AAR/AI)
Keeping it "organized and logical": after-action review for AI (AAR/AI)
复制标题
保持“有条理、有逻辑”：人工智能事后审查 (AAR/AI)
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Alan Fern
中科院分区：
文献类型：
--
作者：
Theresa Mai;Roli Khanna;Jonathan Dodge;Jed Irvine;Kin;Zhengxian Lin;Nicholas Kiddle;E. Newman;Sai Raja;Caleb R. Matthews;Christopher Perdriau;M. Burnett;Alan Fern
随着人工智能在现代社会的普及，可解释的人工智能（XAI）正变得越来越重要，但很少有人研究XAI如何直接支持人们试图评估人工智能代理。如果没有严格的过程，人们可能会以特别的方式进行评估——导致同一代理的评估可能存在很大的差异，这仅仅是因为他们的过程不同。AAR，即行动后评估，是一些军事组织用来评估人类特工的一种方法，它已经在许多领域得到了验证。根据这一策略，我们为人工智能导出了一个AAR，以组织人们在顺序决策环境中评估强化学习（RL）代理的方式。我们的定性研究结果揭示了AAR/AI过程的几个优点和缺点，以及其中的解释。
Explainable AI (XAI) is growing in importance as AI pervades modern society, but few have studied how XAI can directly support people trying to assess an AI agent. Without a rigorous process, people may approach assessment in ad hoc ways---leading to the possibility of wide variations in assessment of the same agent due only to variations in their processes. AAR, or After-Action Review, is a method some military organizations use to assess human agents, and it has been validated in many domains. Drawing upon this strategy, we derived an AAR for AI, to organize ways people assess reinforcement learning (RL) agents in a sequential decision-making environment. The results of our qualitative study revealed several strengths and weaknesses of the AAR/AI process and the explanations embedded within it.