諮商系所學生使用聊天機器人經驗初探

Counseling Students' First Impressions of Using Woebot: A Mental Health Chatbot

10.6251/BEP.202409_56(1).0003

葉寶玲(Pao-Ling Yeh)；郭文正(Wen-Cheng Kuo)；蔡佳容(Chia-Jung Tsai)

Woebot ； 使用者經驗 ； 諮商系所學生 ； 聊天機器人 ； Woebot ； user experience ； counseling department students ； chatbots

教育心理學報

56卷1期（2024 / 09 / 01）

45 - 72

繁體中文;英文

諮商與科技的結合是時代之未來趨勢。諮商實務工作者少有機會認識聊天機器人，且臺灣目前缺少實徵性研究支持，本研究旨在了解諮商系所學生使用心理健康聊天機器人前、使用過程、使用後的感受與想法，以及對後續研發的建議。本研究為質性研究，以目的取樣透過社交平臺招募到10位諮商系所學生為研究參與者，個人進行與聊天機器人Woebot每日一次共14天14次談話，並將談話內容截圖與接受個別訪談。訪談錄音檔與截圖的對話內容均謄打為逐字稿後，由三位作者進行編碼分析。結果發現研究參與者使用前對心理健康聊天機器人的態度觀點不一，使用過程的感受正負向皆有，也有産生改變經驗。心理健康聊天機器人具有長期與立即陪伴功能、適合孤單時使用，可作為真實諮商的輔助；影片等多元媒材能提高使用動機，引導建設性的思考。然而，研究參與者認為使用過程缺乏足夠的情緒支持、感受不到當個案的獨特性、出現語言與跨文化之障礙，也對於未來研發提出具體建議。本研究乃國內罕見的探討心理健康聊天機器人使用之研究，提供一個原創性的主題，呈現了新興科技與諮商連結的使用經驗，也對諮商實務應用及未來人工智慧聊天機器人研究提供建議。

Advancements in artificial intelligence have led to the development of numerous operational robots; thus, increased research attention has been paid to human-robot interactions. The integration of technology into counseling is an emerging trend. However, practitioners have few opportunities to familiarize themselves with chatbots. Additionally, the lack of supporting empirical evidence and the challenges involved in selecting an appropriate chatbot from abundant options pose considerable difficulties that must be overcome. Various robotics science theories have been proposed to describe human interactions with chatbots, particularly mental health-care robots. These robots mainly involve the establishment of caregiving relationships through virtual avatars, data collection, care planning, and health information provision. A widely applied theory related to chatbot interaction is the technology acceptance model, which posits that understanding the reasons why users accept or reject new information systems can explain their intention to use such systems. User experience, which emphasizes a user-centered approach to interactions with products, is influenced by the interactions between the user's internal state, system design, and external environment. The present study explored the emotional, cognitive, and behavioral dimensions of chatbot use from the perspective of user experience. In recent decades, smart technology has been increasingly applied to clinical psychology. Most currently available chatbots focus on psychoeducation or symptom monitoring, serving only as self-help tools for psychotherapy. Woebot is a widely used mental health chatbot that was developed by psychologist Alison Darcy in 2016 on the basis of the principles of cognitive behavioral therapy. It can be downloaded to computers and smartphones. The present study used Woebot to explore the experiences of chatbot users. Compared with face-to-face services provided by human therapists, chatbots offer several advantages that have promoted great interest in their development and use. Chatbots are becoming increasingly accessible; they can be accessed through not only computers but also smartphones, allowing therapy recipients to circumvent the stigma associated with psychotherapy. Chatbots do not take breaks or require appointments to be scheduled, are unaffected by geographical and transportation constraints, and are generally free or low-cost, making them an affordable mental health resource. However, studies have identified several differences and limitations associated with chatbot services relative to services provided by human therapists. The present qualitative study conducted semi-structured interviews and analyzed the collected data by applying coding in grounded theory, which is suitable for discerning processes and inner experiences. This theory is applicable with our objective of examining experiences of using Woebot and the contextual relationships underlying such experiences. Because of the lack of research on the appropriateness of chatbots developed on the basis of psychotherapy from user perspectives, the present study aimed to provide a clearer understanding of the experiences of Woebot users from various backgrounds and to identify the differences between them. The author, who has a Ph.D. in counseling psychology, is experienced in using Woebot and conducting qualitative research. In the present study, convenience and purposive sampling was applied to recruit 10 participants-3 men and 7 women- through social media platforms, social media applications, and recruitment posters. The participants used Woebot for 14 days, completing a total of 14 sessions. The participants were aged 20-31 years (mean age, 26 years) and comprised three undergraduate students and seven graduate students. The research instruments comprised a demographic information form, interview outlines, recording devices, and Woebot. The demographic information form was used to collect information such as the participants' name, gender, age, total hours of individual counseling received, age of first exposure to the Internet, and duration of daily smartphone use (measured in hours). The interview questions focused on unique experiences, feelings, and thoughts formed during interactions with Woebot. Differences in the ending conversations with Woebot compared with face-to-face counseling were determined. Moreover, this study provided overall impressions and suggestions regarding the use of chatbots in counseling. The research process was as follows. In the preparation phase, the research objective and questions and the types of data to be collected were determined. Individuals who met the inclusion criteria were recruited, briefed on the study's purpose and procedures, and asked to sign informed consent forms. The study design involved the daily use of Woebot for 2 weeks. During the data collection phase, screenshots of the participants' interactions with Woebot, verbatim transcripts of these screenshots, and individual interview transcripts were collected. These were transcribed into Microsoft Word documents to facilitate data retrieval and reference. Data analysis was performed on the basis of coding in grounded theory, which involved examining and reading the collected data to derive preliminary concepts. Each sentence was analyzed individually to segment the data and to derive categories and subcategories. Axial coding was then performed to link these categories and subcategories to form themes. The titles for the coding results were generated on the basis of individual interview transcripts. Subsequently, relevant conversation screenshots and transcript content were extracted and used as evidence. To avoid introducing bias into the coding process, the author invited a coanalyst to conduct data analysis. During the analysis, multiple data sources were used for triangulation. Each interview transcript was coded individually, and the results of the study participants from different backgrounds were verified and compared using selected screenshots depicting the usage of Woebot. The study results indicated that prior to using mental health chatbots, the participants exhibited different attitudes toward them. During chatbot use, positive feedback, negative feedback, and changing opinions were observed among the participants. The results also highlighted various problems encountered by the participants, such as predefined options limiting the participants' options and expression, the awkwardness associated with not knowing when the conversation had ended, discrepancies in information on coronavirus disease 2019 between the United States and Taiwan, logical but unrealistic interactions, and meditation activities being temporarily inaccessible because of misunderstandings. The participants reported both positive and negative thoughts and opinions about the use of Woebot. They were positive about the chatbot's ability to provide immediate and long-term companionship, usefulness as a real counseling aid, provision of engaging videos that increase usage motivation, and provision of guidance and constructive suggestions about current events. They were negative about the lack of sufficient emotional support and the limited ability of the chatbot to understand topics due to language and cultural differences. The participants indicated that using Woebot led to beneficial cognitive, behavioral, and emotional changes. Cognitively, the participants reported changes in their thoughts, more opportunities to think, and more rational thoughts under the guidance of the chatbot. Behaviorally, Woebot assisted the participants in practicing mindfulness. Emotionally, the participants felt that the most substantial benefit of using the chatbot was emotional improvement, and they did not always clearly recognize the changes in their cognition. For Woebot-related research and development, the participants suggested increasing the alignment of Woebot with the real-life counseling experiences, habits, and emotional needs of users; providing more than just predefined options; enabling more coherent and continuous interactions; and adding voice functions. With these improvements, the chatbot can be more effectively integrated into counseling services, thus providing a more realistic user experience. In summary, the present study revealed that although most participants perceived Woebot as a novel and entertaining tool, their experience was associated with various problems that affected their mood and conversation outcomes, such as the lack of emotional empathy, the feeling that their uniqueness as a client was not acknowledged, and language and cultural barriers. These challenges also highlighted that both chatbots and human counseling are limited by the strengths and weaknesses of their underlying theoretical frameworks. Users expect chatbots to more closely align with the standards of human counseling and to use their native languages, emphasizing the need for establishing a database that considers cultural differences. We suggest that current chatbots should be used only as a support tool for counseling to help foster rational thought patterns and habits, which can aid in transforming behaviors and alleviating negative emotions. For patients who require considerable emotional venting, the theoretical fit should be considered during the selection of a chatbot or other intervention methods. Future studies should investigate chatbot developers or participants from diverse groups to understand their objectives, philosophies, and experiences with chatbot development. Future consent forms for chatbot studies should include provisions prohibiting chatbot manipulation to reduce the time and effort required to address its disruptive effects. For the future development or enhancement of chatbots, advancements in voice recognition technology should be incorporated into chatbots. Moreover, button-based chatbot text conversations should be expanded to include voice-based chatbots, thereby enabling more humanized interactions. Additionally, chatbots based on various therapeutic approaches can be developed to meet the needs of various types of patients. Databases can also be expanded to incorporate diverse cultural contexts to align more closely with the cultures and life experiences of users. This expansion aims at enhancing the positive feelings of users and the sense that they are understood by a chatbot while using it.