人工智慧在心理諮商與治療的應用：實務發展與趨勢探討

Applications of Artificial Intelligence in Counseling and Psychotherapy: Practical Developments and Trend Analysis

人工智慧近年蓬勃發展，並逐漸應用於諮商與心理治療領域。然而，相關發展趨勢尚未完全明朗，且實證研究仍然有限。本研究使用文獻計量分析，以 Scopus 學術文獻資料庫中的 480篇文獻為研究對象，透過表現分析與科學圖譜分析，探討人工智慧在諮商與心理治療領域之發展趨勢與重要主題。研究結果顯示，此領域目前正處於快速發展階段，其中美國為核心國家，多數研究為跨領域合作。早期研究集中於基礎技術發展與憂鬱情緒，中期聚焦於應用程式與產品開發及心理疾患等次領域，近年則關注倫理議題、多元技術整合與多情境應用。未來研究應致力於探索可解釋的人工智慧（Explainable AI），以提升案主的信任感與治療效能。基於研究發現，本研究提出「人機系統合作諮商模式」，作為人工智慧融入諮商實務的具體架構，以有效增進心理治療成效。

Artificial intelligence (AI) has influenced human worldviews and value systems (Michel-Villarreal et al., 2023) and has been applied as an assistive technology in counseling and psychotherapy (Shen, 2019; Wang, 2023). To date, AI has been used to provide basic emotional support (Inkster et al., 2018; Su et al., 2022), monitor mental health and deliver remote psychological services (Fiske et al., 2019; Sedlakova & Trachsel, 2022), offering more accessible and personalized options for mental health care. However, the implementation of AI has been associated with notable challenges, including those related to data security, privacy protection, and technological reliability (Aktan et al., 2022), all of which require further empirical investigation.

This study had the following objectives: (1) to explore the current state of AI research and applications in global counseling and psychotherapy, (2) to analyze thematic development trends in the field, and (3) to offer practical recommendations for counseling practice based on the findings.

This study employed bibliometric analysis to investigate the development and knowledge structure of AI in counseling and psychotherapy from both systematic and empirical perspectives (Hall, 2011). By using the “Bibliometrix” package in R (Aria & Cuccurullo, 2017), performance analysis and science mapping were conducted to evaluate citation patterns, publication volume, geographical distributions, and keyword co-occurrence networks to identify major trends and academic influences. Scopus was selected as the primary data source because of its comprehensive indexing of peer-reviewed literature across disciplines (Mongeon & Paul-Hus, 2016). Following the PRISMA protocol (Page et al., 2021), this study identified an initial pool of 3,398 articles, 480 of which met the inclusion criteria for research regarding AI applications in counseling and psychotherapy.

The analysis indicated that research in this field has advanced from an initial exploratory phase to a phase of rapid expansion. The United States, China, and the United Kingdom were the leading contributors, as indicated by their high publication output and extensive collaborative networks. Productive authors often engaged in interdisciplinary and industry-academia collaborations, underscoring the field’s practical relevance and applied value. The majority of the analyzed publications appeared in psychology and medical journals, indicating a primary focus on mental health. In addition, interdisciplinary connections with AI research continued to strengthen.

Regarding thematic developments, AI has been incorporated into various therapeutic models and technologies, including psychoanalysis, person-centered therapy, motivational interviewing, cognitive behavioral therapy (CBT), and virtual reality therapy. Among these, CBT was the most frequently applied and cited. Key AI technologies—including chatbots, natural language processing, and affective computing—were commonly employed to address mental health concerns such as suicidal ideation, depression, anxiety, autism spectrum disorder, attention-deficit/hyperactivity disorder, and posttraumatic stress disorder. Emerging research has increasingly focused on the democratization of digital mental health, the advancement of explainable AI, and the development of human-like AI therapists capable of forming empathetic therapeutic relationships.

Although notable progress has been made, the majority of studies in this area are theoretical or descriptive and lack empirical validation, particularly with respect to long-term treatment outcomes. Suicide risk prediction, in particular, remains an underexplored area. Although some studies have employed AI to predict suicidal behavior, the complexity of human psychology and the “black box” nature of AI systems limit interpretability and client trust in the systems. Therefore, this study recommends that future research develop dynamic case conceptualization models and real-time crisis alert systems that integrate AI with established mental health resources to create more comprehensive support networks. Furthermore, this study identified the concept of a “human-AI therapeutic alliance” as a promising yet underexplored research area. The majority of studies on this topic have been conceptual and provided limited empirical evidence. Future research should investigate the core components of the therapeutic alliance and incorporate them into the design of AI systems to enhance client trust, engagement, and clinical effectiveness.

On the basis of its findings, this study proposes a human-AI collaborative counseling model that integrates AI technologies with the professional expertise of psychotherapists to enhance mental health service delivery. This model comprises four stages: (1) initial assessment, (2) formal counseling sessions, (3) follow-up and personalized support, and (4) supervision and ethical oversight. This framework may improve the safety and effectiveness of AI-assisted psychotherapy.

In the initial assessment stage, AI serves as a supportive tool. Chatbots trained in empathetic communication conduct preliminary interactions with clients, assisting them in articulating their concerns, completing stress and emotional screenings, and providing their basic background information. This provides human therapists with an initial understanding of their clients’ psychological states and helps shorten intake sessions.

In the formal counseling stage, AI is used to provide real-time therapeutic support. It analyzes clients’ verbal expressions, emotional fluctuations, facial micro-expressions, and physiological responses, providing therapists with real-time insights into their clients’ needs. These inputs enhance therapists’ ability to detect subtle nonverbal cues, mitigate the effects of human bias and distraction, and improve the overall quality of therapeutic engagement.

In the follow-up and personalized support stage, AI is used to construct dynamic case conceptualization models and generate psychological digital twins of clients. These tools facilitate the prediction of psychological risks and enable real-time crisis detection. AI also delivers personalized interventions based on clients’ behavioral patterns and emotional states, incorporating CBT techniques and delivering psychoeducational content and behavioral prompts aligned with individual therapeutic goals.

In the supervision and ethical assurance stage, a rigorous regulatory framework is implemented to ensure data security and client confidentiality. These mechanisms are critical to mitigating ethical risks associated with the misuse or overreach of AI. Moreover, the study emphasizes the need to develop AI systems with strong cross-cultural adaptability to avoid misinterpretations and inappropriate interventions rooted in sociocultural differences.

This study advocates for additional cross-national and cross-cultural comparative research to evaluate the effectiveness of AI in diverse cultural contexts. Future investigations should examine specific AI technologies (e.g., machine learning, deep learning, chatbots, and natural language processing), therapeutic techniques (e.g., exposure therapy and cognitive restructuring), and mental health domains (e.g., suicide prevention and crisis management). Additionally, subfields such as therapist training, communication skill development, and clinical decision-making must be given additional attention to enhance the practical value of AI in mental health care.

In conclusion, this study highlights AI’s potential as a useful supplementary tool for enhancing counseling and psychotherapy. Future research should focus on identifying key intervention mechanisms and establishing a strong empirical foundation to support the integration of AI into clinical practice to improve treatment outcomes and enrich the client experience.