Getting Playful with Explainable AI: Games with a Purpose to Improve Human Understanding of AI
Getting Playful with Explainable AI: Games with a Purpose to Improve Human Understanding of AI
复制标题
玩转可解释的人工智能：旨在提高人类对人工智能理解的游戏
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Adam Perer
中科院分区：
文献类型：
--
作者：
Laura Beth Fulton;Qian Wang;Jessica Hammer;Ja Young Lee;Zhendong Yuan;Adam Perer
可解释人工智能 (XAI) 是机器学习 (ML) 领域的一个新兴主题，旨在让人类了解人工智能系统如何做出决策。 XAI 在为医学和刑事司法等领域带来透明度方面变得越来越重要，在这些领域，人工智能可以为高后果决策提供信息。尽管已经提出了许多 XAI 技术，但除了轶事证据之外，很少有其他技术得到过评估。我们的研究提供了一种新的方法来评估人类如何解释人工智能的解释；我们通过将 XAI 与有目的的游戏 (GWAP) 集成来探索这一点。 XAI 需要大规模的人工评估，而 GWAP 可用于通过轮次游戏呈现的 XAI 任务。本文概述了 GWAP 对 XAI 的好处，并通过我们创建的多人 GWAP 演示了应用程序，该 GWAP 侧重于解释为图像识别训练的深度学习模型。通过我们的游戏，我们试图了解人类如何选择和解释图像识别系统中使用的解释，并为 XAI 的 GWAP 设计的有效性提供经验证据。
Explainable Artificial Intelligence (XAI) is an emerging topic in Machine Learning (ML) that aims to give humans visibility into how AI systems make decisions. XAI is increasingly important in bringing transparency to fields such as medicine and criminal justice where AI informs high consequence decisions. While many XAI techniques have been proposed, few have been evaluated beyond anecdotal evidence. Our research offers a novel approach to assess how humans interpret AI explanations; we explore this by integrating XAI with Games with a Purpose (GWAP). XAI requires human evaluation at scale, and GWAP can be used for XAI tasks which are presented through rounds of play. This paper outlines the benefits of GWAP for XAI, and demonstrates application through our creation of a multi-player GWAP that focuses on explaining deep learning models trained for image recognition. Through our game, we seek to understand how humans select and interpret explanations used in image recognition systems, and bring empirical evidence on the validity of GWAP designs for XAI.