Analysis of Human-Human and Human-Computer Agent Interactions from the Viewpoint of Design of and Attribution to a Partner
Analysis of Human-Human and Human-Computer Agent Interactions from the Viewpoint of Design of and Attribution to a Partner
复制标题
从合作伙伴设计和归因的角度分析人与人、人与计算机代理交互
DOI：
--
复制
发表时间：
2006
期刊：
影响因子：
--
通讯作者：
H. Terai
中科院分区：
文献类型：
--
作者：
K. Miwa;H. Terai
从设计和归因的观点对人类和人类计算机的相互作用的分析和伴侣Kazuhisa miwa（miwa@is.nagoya-u.ac.jp）hitoshi terai（terai@cog.human.nagoya-u）。 AC.JP）Nagoya University Nagoya信息科学研究生院，464-8601日本摘要一些有趣的研究结果表明了归因和设计观点的重要性，包括像Eliza这样的传统例子，即使人类与人类交谈时，人类也与一个简单的计算机程序交谈，即使该程序仅产生非常肤浅的响应（Weizenbaum，1966） 。在《怪异山谷》的问题中，在Android Science中指出，机器人的外观非常针对人类，对机器人的熟悉程度相当大得多，而奇异的山谷出现了（请参阅COGSCI2005研讨会网站：http：http：// www。 AndroidScience.com/）。这些结果表明，在研究HAI的性质时，应分别处理两个因素，两个因素，设计和归因于计算机代理。在前面的研究中，我们将两个不同的因素（归因和设计）一起操作，或者处理一个因素。在这项研究中，我们提出了“幻觉实验范式”，在那里我们独立控制它们。这项研究的目的是根据这两个关键因素阐明HHI和HAI的性质：对计算机代理的设计和归因。此外，在媒体方程式范式中进行的实验主要集中在理解与计算机代理人的社会关系上，通过通过问卷调查对受试者对代理的印象。在这项研究中，我们在控制归因和设计作为独立变量的同时将主题解决行为作为因变量来衡量。通过测量可观察到的行为相互作用的性质，例如可观察到的，例如解决问题的绩效，除了主观估计外，我们希望基于更确定的经验证据讨论HHI和HAI的性质。最近，不仅人类互动（HHI），而且还出现了人类和人类的互动（HAI），其中人类和诸如计算机代理等人类的认知文物都出现了。要研究这种互动的性质，分别处理两个观点很重要：对计算机代理的设计和归因。归因的观点是人类归因于计算机代理的属性，而设计的观点是计算机代理的实际设计方式。我们提出了“幻觉实验范式”，我们可以独立控制这两个因素。进行了两个实验，其中一对受试者协同解决了一个简单的推理任务。我们分析了他们的假设形成和测试行为如何受这两个因素的影响。实验结果基本上表明，问题解决行为仅受设计因素的影响，而作为一种代表性社会行为的互惠行为也受设计和归因因素的影响。关键字：解决问题，；合作;相互作用。简介我们通过与外部环境的交互来生成新值，同时修改，处理和生成信息。最近，不仅人类的互动（HHI），而且还出现了人类和人类的互动（HAI），在这些互动（HAI）中已经出现了人类和认知文物（例如计算机代理人）的合作；许多研究人员已经开始表现出对这种互动性质的兴趣。在HHI和HAI研究的交集中出现的一个研究范式是媒体方程框架（Reeves＆Nass，1996）。在框架中获得的发现是，人类通常与与其他人类有关的计算机或电视节目有关。这种观点还开始为设计计算机代理提供新的原则（Dryer，1999）。从媒体方程研究中浮出水面的一个重要建议是，在研究HAI时，要分别处理两种观点很重要：对计算机代理的设计和归因。归因的观点包括人类对计算机代理的属性：即，他/她承认合作伙伴是人类或其他人工代理，而设计的视角是计算机代理的设计方式：即代理在多大程度上实际上像人类一样精致。任务2-4-6在这项研究中，我们将Wason的2-4-6任务用作实验任务（Wason，1960），因为它已被用作人类发现研究的标准实验任务，并且其性质很好理解（Newstead＆Evans，1995）。 2-4-6任务的标准过程如下。所有受试者都必须找到三个数字之间的关系规则。受试者发现的规则称为目标规则。在最受欢迎的情况下，在初始阶段向主题提出了三个数字'2、4、6'的数字。这
Analysis of Human-Human and Human-Computer Agent Interactions from the Viewpoint of Design of and Attribution to a Partner Kazuhisa Miwa (miwa@is.nagoya-u.ac.jp) Hitoshi Terai (terai@cog.human.nagoya-u.ac.jp) Graduate School of Information Science, Nagoya University Nagoya, 464-8601 JAPAN Abstract Some interesting findings have indicated the importance of the perspectives of attribution and design, including such a traditional example as ELIZA where humans converse with a simple computer program called ELIZA as they talk with humans, even though the program generates only very superficial responses (Weizenbaum, 1966). In the issue of the uncanny valley, pointed out in android science where the appearance of robots is very closely designed to humans, familiarity to robots decreases rather extremely, and the uncanny valley emerges (see a CogSci2005 workshop site: http://www.androidscience.com/). These results imply that in an investigation of the nature of HAI, two factors, design of and attribution to computer agents, should be dealt with separately. In the preceding studies, our two different factors, attribution and design, were manipulated either together or a single factor was dealt with. In this study, we propose the ''illusion experiment paradigm'' where we control them independently. The objective of this study is to clarify the nature of HHI and HAI based on these two crucial factors: design of and attribution to computer agents. Additionally, the experiments conducted in the Media Equation paradigm have mainly focused on understanding the human social relationship with computer agents by measuring subjects' impressions of the agents with questionnaires. In this study we measure subject problem solving behavior as a dependent variable while controlling attribution and design as independent variables. By measuring the nature of interactions emerging in behavior that are objectively observable, such as problem solving performance in addition to subjective estimations, we expect to discuss the nature of HHI and HAI based on more established empirical evidence. Recently, not only Human-Human Interaction (HHI) but also Human-Agent Interaction (HAI) where humans and cognitive artifacts such as computer agents collaborate have emerged. To investigate the nature of such interaction it is important to deal with two perspectives separately: design of and attribution to computer agents. The perspective of attribution is what a human attributes to a computer agent whereas the perspective of design is how a computer agent is actually designed. We propose the ''illusion experiment paradigm'' where we can control independently these two factors. Two experiments were performed in which a pair of subjects solved a simple reasoning task collaboratively. We analyzed how their hypothesis formation and testing behavior were influenced by these two factors. Experimental results basically indicated that subject problem solving behavior was only influenced by the factor of design, whereas their reciprocity behavior as one representative social behavior was influenced by both design and attribution factors. Keywords: Problem solving,; Collaboration; Interaction. Introduction We generate new values through interactions with external environments while revising, processing, and generating information. Recently, not only Human-Human Interaction (HHI) but also Human-Agent Interaction (HAI) where humans and cognitive artifacts such as computer agents collaborate have emerged; many researchers have begun to show interest in the nature of such interactions. One research paradigm emerging at the intersection of the HHI and HAI studies is the Media Equation framework (Reeves & Nass, 1996). A finding obtained in the framework is that human beings often relate to computer or television programs as they relate to other human beings. This view has also begun to provide new principles for designing computer agents (Dryer, 1999). An important suggestion that surfaced from Media Equation studies is that when studying HAI it is important to deal with two perspectives separately: design of and attribution to computer agents. The perspective of attribution includes what a human attributes to a computer agent: i.e., he/she recognizes a partner as a human or another artificial agent whereas the perspective of design is how a computer agent is designed: i.e., to what degree the agent is actually constructed as sophisticatedly as humans behave. Task 2-4-6 task In this study, we use Wason's 2-4-6 task as an experimental task (Wason, 1960) because it has been used as a standard experimental task in studies on human discovery, and its nature is well understood (Newstead & Evans, 1995). The standard procedure of the 2-4-6 task is as follows. All subjects are required to find a rule of a relationship among three numerals. The rule that subjects find is called a target rule. In the most popular situation, a set of three numerals, ''2, 4, 6,'' is presented to subjects in the initial stage. The