본 연구는 AI 미디어를 매개로 한 AI 음성 대화형 에이전트(AI conversational agent)와의 커뮤니케이션 과정을 이해하기 위해 의인화적 관점에서 접근한다. 인간과 대화하는 사회적 로봇으로서의 AI 대화형 에이전트가 목소리나 대화내용 등에 있어 인간과 유사하게 설계되었다는 점에 주목한다. AI 대화형 에이전트에 대한 인지된 의인화를 구성하는 요소들은 무엇이며, 어떠한 문항들로 측정될 수 있는지 탐색하였다. 이러한 과정을 통해 개발된 척도의 신뢰도와 타당도를 다면적으로 검증하였다. 결과적으로 AI 대화형 에이전트에 대한 인지된 의인화는 이성적 및 정서적 지지, 합리성, 친밀성, 예의성, 그리고 인지적 개방성이라는 5개 요인으로 구성되는 것으로 확인되었으며, 이를 측정하기 위한 문항은 19개로 도출되었다. 의인화 구성요인 가운데 합리성, 친밀성, 예의성, 인지적 개방성이 AI 스피커에 대한 이용만족에 정적인 영향을 주는 것으로 나타났다. 본 연구를 통해 개발된 AI 대화형 에이전트 의인화 평가척도는 사람들의 AI미디어 이용행위를 비롯한 커뮤니케이션 과정을 이해하는 데 도움을 줄 수 있을 것이다.

This study seeks to develop a scale that measures people’s perception of an artificial intelligence (AI) conversational agent that utilizes a voice-user interface (VUI), owing to the increase in communication with social robots due to technological advances in AI, big data, and machine learning. The study focuses on the concept of anthropomorphism because an AI conversational agent, one of the social robots, is designed to simulate human voice and dialogue. Focusing on how people perceive the level of anthropomorphism of conversational agents while communicating with them, this study explores the dimension of perceived anthropomorphism and the questions to measure them. For this purpose, psychometric scale development procedures were followed and a large pool of items were rigorously tested for their reliability and validity. First, anthropomorphic factors were explored by researching the literature on anthropomorphism and interviewing experts in this area, and 35 measurement items were generated. The initial item pool consists of questions related to civility, rationality, refinement, intimacy, individuality, rational support, emotional support, and cognitive openness. In order to test the internal consistency among the initial question items, 81 university students were surveyed as preliminary research, and 35 measurement items were finalized. Another survey was conducted to examine whether the scale adequately reflected the proposed dimensional structure of the construct. Considering gender, age, and residence, 484 people who had experience using smart speakers within the last six months were surveyed, and their responses were analyzed. As a result, the perceived anthropomorphism of an AI conversational agent was identified as being composed of five factors- rational and emotional support, rationality, intimacy, civility, and cognitive openness-and 19 questions were developed to measure them. In particular, of the variables for perceived anthropomorphism, rationality, intimacy, civility, and cognitive openness had a positive effect on satisfaction with smart speakers. Reliability and validity were verified in multiple ways to ultimately develop a perceived anthropomorphic evaluation scale for an AI conversational agent. The scale developed by this study can be used to measure and assess subjective awareness of how human an AI conversational agent feels. This study is also expected to help predict users’ attitudes or behaviors regarding AI media that use conversational agents. As communication with agents in AI media increases, interactive AI agents’ quality increases, and they are producing increasingly better results. Consequently, this study is expected to contribute to understanding the process of communication with non-human actors and their communication strategies.