{"title":"Customizability in Conversational Agents and Their Impact on Health Engagement (Stage 2)","authors":"Stephen C. Paul, Nina Bartmann, Jenna L. Clark","doi":"10.1155/2024/5015913","DOIUrl":null,"url":null,"abstract":"<p>Conversational agents (CAs) are effective tools for health behavior change, yet little research investigates the mechanisms through which they work. Following the Computer as Social Actors (CASA) paradigm, we suggest that agents are perceived as human-like actors and hence influence behavior much as human coaches might. As such, agents should be designed to resemble ideal interaction patterns, for example, by resembling their users. In this registered report, we evaluated this paradigm by testing the impact of customization on similarity and reciprocity, which in turn were hypothesized to improve perceptions of the agent and compliance with the agent’s recommendations to complete a cognitive training exercise. In an online study, 2437 participants were randomly assigned to one of two surface-level CA customization conditions (present/absent) and to one of two deep-level CA customization conditions (present/absent) in a between-subject experimental design. As part of a conversation flow with a CA, participants assigned to the present surface- and/or deep-level customization conditions were able to choose their preferred CA based on the four personality summaries and/or choose their CA’s gender (male/female/agender robotic), avatar (choice between seven avatars corresponding to the chosen gender), and name. While the ability to customize increased similarity to the user and the perceptions of customizability, our findings show that customization did not impact experience or compliance. However, the perceived customizability of the agent was linked to increases in the likeability and usefulness of the agent. We conclude that our work finds no negative effects of customization; yet, its impact on the relationship between the agent and its user is complex and can benefit from more research as merited by its applicability to public health. As aging and ill populations increase the burden on health systems worldwide, CAs have the potential to transform the landscape of accessible care.</p>","PeriodicalId":36408,"journal":{"name":"Human Behavior and Emerging Technologies","volume":null,"pages":null},"PeriodicalIF":4.3000,"publicationDate":"2024-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1155/2024/5015913","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Human Behavior and Emerging Technologies","FirstCategoryId":"1085","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1155/2024/5015913","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"PSYCHOLOGY, MULTIDISCIPLINARY","Score":null,"Total":0}
引用次数: 0
Abstract
Conversational agents (CAs) are effective tools for health behavior change, yet little research investigates the mechanisms through which they work. Following the Computer as Social Actors (CASA) paradigm, we suggest that agents are perceived as human-like actors and hence influence behavior much as human coaches might. As such, agents should be designed to resemble ideal interaction patterns, for example, by resembling their users. In this registered report, we evaluated this paradigm by testing the impact of customization on similarity and reciprocity, which in turn were hypothesized to improve perceptions of the agent and compliance with the agent’s recommendations to complete a cognitive training exercise. In an online study, 2437 participants were randomly assigned to one of two surface-level CA customization conditions (present/absent) and to one of two deep-level CA customization conditions (present/absent) in a between-subject experimental design. As part of a conversation flow with a CA, participants assigned to the present surface- and/or deep-level customization conditions were able to choose their preferred CA based on the four personality summaries and/or choose their CA’s gender (male/female/agender robotic), avatar (choice between seven avatars corresponding to the chosen gender), and name. While the ability to customize increased similarity to the user and the perceptions of customizability, our findings show that customization did not impact experience or compliance. However, the perceived customizability of the agent was linked to increases in the likeability and usefulness of the agent. We conclude that our work finds no negative effects of customization; yet, its impact on the relationship between the agent and its user is complex and can benefit from more research as merited by its applicability to public health. As aging and ill populations increase the burden on health systems worldwide, CAs have the potential to transform the landscape of accessible care.
对话式代理(CA)是改变健康行为的有效工具,但对其作用机制的研究却很少。根据 "计算机作为社会行动者"(CASA)范式,我们认为,代理被视为类似人类的行动者,因此会像人类教练一样影响人们的行为。因此,应将代理设计成理想的交互模式,例如,通过与用户相似来实现。在这份注册报告中,我们通过测试定制对相似性和互惠性的影响来评估这一范式,并假设这反过来会提高对代理的感知,以及对代理建议完成认知训练的依从性。在一项在线研究中,2437 名参与者被随机分配到两种表层 CA 定制条件(存在/不存在)和两种深层 CA 定制条件(存在/不存在)中的一种。作为与 CA 对话流程的一部分,被分配到表面和/或深层次定制条件下的参与者可以根据四种个性总结选择他们喜欢的 CA,和/或选择他们 CA 的性别(男性/女性/两性机器人)、头像(在与所选性别相对应的七个头像中选择)和名字。虽然定制能力增加了与用户的相似度和对可定制性的感知,但我们的研究结果表明,定制并不影响体验或服从性。然而,人们对代理的可定制性的感知与代理的可喜欢性和有用性的增加有关。我们的结论是,我们的工作没有发现定制化的负面影响;然而,定制化对代理及其用户之间关系的影响是复杂的,可以从更多的研究中获益,因为它适用于公共卫生。随着老龄化和患病人口的增加,全球医疗系统的负担加重,CA 有可能改变无障碍医疗的面貌。
期刊介绍:
Human Behavior and Emerging Technologies is an interdisciplinary journal dedicated to publishing high-impact research that enhances understanding of the complex interactions between diverse human behavior and emerging digital technologies.