Session 1

AI and Innovations in Direct Social Work Practice

Moderator: Dr. Xiayu Summer Chen (Assistant Professor, University of Central Florida)

O1.1  Can AI Chatbots Provide Therapeutic Support Comparable to Human Therapists? A Qualitative Exploration of the Paradox of Agential Status

*Xiaolu Dai¹,2, Ling Li Leng³, Yiyan Liu⁴, Yu-Te Huang⁵, Daniel Fu Keung Wong⁴

¹Beijing Normal-Hong Kong Baptist University, China; ²Xingguang Social Work Agency, China; ³Department of Sociology, Zhejiang University, Zhejiang, China; ⁴Department of Social Work, Hong Kong Baptist University, Kowloon Tong, Hong Kong, China; ⁵Department of Social Work and Social Administration, The University of Hong Kong, Pokfulam, Hong Kong, China

Abstract

Introduction: The rapid advancement of Large Language Models has sparked heated debate over whether Generative Artificial Intelligence (AI) chatbots can serve as “digital therapists” capable of providing psychotherapeutic support to individuals with mental health concerns. While discussions are intense and some research has examined ChatGPT’s ability to handle various psychology-related tasks, less is known about clients’ actual experiences in real-world settings.

Objectives: This study aims to explore whether AI chatbots can offer mental health support comparable to that of human therapists from the clients’ perspective. Particularly, we seek to understand how the non-agent nature of AI chatbots may impact its role in mental health care.

Methods: We conducted semi-structured interviews with 16 Chinese adults who have sought mental health support from both human therapists and ChatGPT, inviting them to compare the similarities and differences between the two experiences.

Results: Thematic analysis identified three themes: (1) promoting open and authentic selfdisclosure with limited deep exploration; (2) the myth of relationship: caring, acceptance, and understanding; (3) fostering therapeutic change: the promise and pitfalls of data-driven solutions. Particularly, as a non-agent operating on algorithms, ChatGPT lacks autonomous behaviors as well as subjective experiences, motives, and values. These non-agential characteristics influenced users’ experiences both positively and negatively, shaping their perception of ChatGPT’s strengths and limitations in supporting them to address mental distress.

Conclusions: AI chatbots supports individuals with mental health needs in unique ways that differ from those of human therapists. This distinctiveness seems closely related to its non-agential features, bringing both advantages and disadvantages. Therefore, rather than striving to make AI chatbots more human-like, we should leverage the unique strengths of their non-agential nature to complement human-delivered psychotherapy. At the same time, human therapists should maximize the benefits of their agent-related qualities while being mindful of potential negative consequences inherent to these traits.

O1.2  “Social Work as a “Thermostat”: The Caregiving Crisis for Families with Elderly and Mentally Disabled Members and the Trust Mediation Framework of Social Work Empowered by AI”

*Xuehua Zhao

¹Department of sociology, Peking University, Beijing, China

 Abstract

In the era of high-speed intelligence, AI technology is rapidly permeating everyday life, yet elderly parents in urban subsidized housing who provide long-term, independent care for their middle-aged children with mental disabilities remain a group with extremely low AI accessibility and adaptability. This study adopts a qualitative approach, conducting in-depth interviews and participatory observations with over 20 “elderly care for disabled” families in subsidized housing communities of a large city in China, with a focus on 8 typical cases for detailed analysis. The study identifies three significant challenges: (1) under the “dual aging” of elderly caregivers and middle-aged care recipients, the decline in caregiving ability and the increase in care needs form a “scissors gap”; (2) the long-term fragile balance of family livelihoods, combined with rigid policies, brings families the risk of a “welfare cliff”; (3) parents experience anticipatory grief regarding the placement of their children after their own passing, which is predictable but difficult to accept. Based on this, the author proposes and preliminarily tests a “trust-first AI empowerment framework”: with social workers acting as “variable-temperature trust intermediaries,” AI technology is transformed into a perceivable and acceptable care resource for families with mentally disabled members through the integrated model of “interpersonal presence—digital inclusion—respite services.” Social workers play the role of a “thermostat”: at the policy level, they turn digital warnings in welfare governance into warm, proactive care; at the technical level, they translate medical professional barriers into practical daily knowledge for the elderly; at the social level, they use AI co-creation to counter stigma and cultivate a community integration foundation. The study emphasizes that the effectiveness of AI-enabled social work depends on “interpersonal presence” rather than “technical presence,” and this trust-first stance provides a localized ethical pathway and theoretical reference for the integration of AI and social work in the field of mental health

O1.3  A Comprehensive Methodological and Theoretical System for Palliative Care Social Work Practice Empowered by an AI-Based Digital Intelligence Platform

*Longtao He¹, Xiangshu Deng¹

¹Southwestern University of Finance and Economics, Sichuan, China

 Abstract

Background and Purpose: Existing debates on artificial intelligence in social work largely focus on literacy, ethics, or implementation challenges. Less attention has been paid to whether social structures still require human agency in decision-making processes once algorithmic systems become institutionally embedded. This paper shifts the analytical focus from technological adoption to structural transformation.

Methods: Drawing on empirical data from AI-supported palliative care social work in China, the study examines how decision processes are gradually pre-structured by technological systems. While practitioners continue to act and bear responsibility, the generation of decisions increasingly occurs within system-defined parameters.

Results: The analysis suggests that AI integration produces a condition of structural decoupling in which agency remains operational but is partially externalized. Rather than collapsing professional roles, this transformation stabilizes service systems through proceduralization while simultaneously weakening the structural necessity of human judgment.

Conclusions and Implications: By moving beyond empowerment-versus-threat narratives, this paper proposes a structural framework for understanding how social work is re-embedded within algorithmically mediated care regimes.

O1.4  AI-Assisted Mental Health Support in Chinese Communities: Cultural Adaptation and Ethical Practice

*Min XUE¹

¹ School of Marxism, Zhejiang University, Zhejiang, China

 Abstract

Background and Purpose: The integration of artificial intelligence into mental health services offers significant opportunities for expanding access and personalizing care in the Asia-Pacific region. However, the deployment of AI-driven tools such as chatbots and predictive analytics within culturally distinct Chinese communities raises pressing questions about cultural responsiveness and ethical appropriateness. Shaped by Confucian relational ethics, collectivist values, and unique patterns of mental illness stigma, Chinese populations across Macau, Hong Kong, and mainland China require AI applications that are attuned to local norms rather than simply imported from Western contexts. This study examines the cultural and ethical prerequisites for implementing AI-assisted mental health support in Chinese-speaking communities and proposes an ethical framework to guide social work practitioners and educators.

Methods: The study employed a mixed-methods design. A systematic scoping review of peer-reviewed literature published between 2018 and 2025 identified 47 relevant studies on AI applications in mental health within Chinese cultural settings. Subsequently, semi-structured in-depth interviews were conducted with 24 key stakeholders, including social work educators, frontline practitioners, and service users with lived experience of mental health challenges, recruited from Macau, Hong Kong, and Guangdong Province. Thematic analysis was performed using NVivo 14, guided by a culturally adapted implementation framework.

Results: Preliminary findings indicate three central concerns. First, participants noted a cultural dissonance whereby AI systems trained predominantly on Western psychological constructs fail to recognize culturally specific expressions of distress and relational conceptions of well-being prevalent in Chinese societies. Second, a deficit in relational trust was evident, as help-seeking in Confucian-influenced contexts remains deeply embedded in interpersonal networks; AI-mediated interactions were widely perceived as lacking the human warmth essential for establishing therapeutic rapport. Third, stakeholders highlighted ethical precarity arising from the absence of clear regulatory frameworks governing AI use in community mental health, particularly regarding informed consent procedures for older adults, cross-jurisdictional data sharing within the Greater Bay Area, and the potential for algorithmic bias to exacerbate existing mental health disparities. Despite these concerns, participants acknowledged the potential utility of AI in triage, psychoeducation, and overcoming geographic barriers to care in underserved rural and island communities.

Conclusions and Implications: The study concludes that a culturally situated approach to AI ethics is necessary for social work practice, one that moves beyond universalist principles to embrace relational autonomy, collective well-being, and local moral contexts. A provisional ethical framework is proposed, organized around four dimensions: cultural attunement requiring co-design with community stakeholders, relational accountability ensuring AI supplements rather than replaces human connection, data sovereignty mandating community governance of sensitive information, and equity-oriented design to mitigate algorithmic harms. These findings carry implications for social work education, particularly the integration of critical AI literacy with culturally grounded practice wisdom, and for policy development aimed at cross-border regulatory coordination. The research offers actionable insights for leveraging AI innovation while preserving the cultural and ethical integrity of social work in Chinese communities across the Asia-Pacific region.

O1.5  Integrating Implementation Frameworks to Define and Measure Implementation Quality in Real-World Non-Pharmacological Dementia Interventions

*Yimin Wu¹, Qiuling An¹

¹School of Social Development, East China Normal University, China

Abstract

Background and Purpose: In community service systems, the effectiveness of non-pharmacological dementia interventions depends on implementation quality. However, prior research lacks a unified operational definition and measurement framework for implementation quality, and empirical evidence remains limited regarding how multilevel implementation determinants influence quality in practice. This study aims to fill this gap. First, it integrates the QIF and EPIS framework to construct a localized Integrated Quality Implementation Framework (IQIF), transforming the abstract concept of implementation quality into measurable and comparable operational indicators. Second, guided by the CFIR framework, the study systematically examines the predictive effects of multilevel implementation determinants on implementation quality, with the aim of identifying key domains and weaknesses affecting implementation within the China context.

Methods: This study adopted a community participation-oriented mixed-method design (January–December 2024) and was conducted in five phases. (1) Theoretical synthesis: Core components shared by QIF and EPIS were identified to establish three foundational domains within the IQIF. (2) Sample inclusion: 37 non-pharmacological dementia intervention projects implemented by 16 organizations across 10 administrative districts in Shanghai were included. (3) Framework refinement: Through multiple rounds of stakeholder interviews and contextual screening, 24 specific IQIF indicators were revised and finalized. (4) Data collection: Implementation quality was operationalized using the IQIF. Two independent researchers conducted double-blind coding and calibration to ensure inter-rater reliability. Guided by CFIR, 27 constructs were selected and measured via Likert-scale surveys to assess implementation determinants. (5) Data analysis: Implementation determinant scores were treated as independent variables and IQIF scores as the dependent variable. Descriptive statistics, correlation analyses, and hierarchical regression models were conducted to test the predictive effects of multilevel determinants and to examine the applicability of the IQIF.

Results: Application of the IQIF across 37 projects demonstrated a significant positive association between overall implementation determinants and implementation quality, supporting the framework’s contextual applicability. Descriptive findings indicated that overall implementation quality was at a moderately high level, though significant disparities emerged across three dimensions, with rigor in outcome evaluation identified as the weakest component. Relative deficit domains included External Policy and Incentives, Access to Knowledge and Information, and Reflecting and Evaluating. Domain-level and construct-level regression analyses showed that the Characteristics of Individuals domain exerted stable and independent predictive power. Key positive predictors included Leadership Engagement, Available Resources, and Planning. In contrast, Organizational Incentives and Rewards demonstrated a consistent negative association with implementation quality.

Conclusions and Implications: The IQIF provides an operational tool for defining and measuring implementation quality in real-world settings and can be aligned with CFIR to identify key predictive domains influencing implementation performance. Findings suggest that improving the implementation quality of non-pharmacological dementia interventions in China requires prioritizing workforce capacity building, leadership engagement, resource assurance, and systematic implementation planning. Targeted strategies should also address weaknesses in evaluation rigor and insufficient external policy support through structured capacity development and institutional reform.

O1.6  Artificial intelligence within the world of social work of Consortium Area A/5, Irpinia, Italy

*Carmine De Blasio¹

¹Consorzio dei servizi sociali A5, Atripalda, ITALIA

Abstract

Backgrounds: AI has been introduced into the work of social workers through two main solutions: 1) Magic Notes, which transforms conversations into structured notes and assessments in seconds. It allows workers to securely record and convert interviews and meetings into organized documentation. 2) Magic Reports, which combines different sources of information—documents, audio recordings, handwritten notes, and videos—to create high-quality professional reports quickly and accurately.

Project Data: The project launched on October 6, 2025. Initial progress was partly impacted by the Christmas period, during which services slowed down and staff availability was reduced. The pilot project kicked off with a full week of in-person training.

Results: The main results observed in the first three months are: 1) 1,200 total Magic Notes recordings, 2) 120 Magic Reports completed, 3) 221 total hours of Magic Notes recordings, and 4) 31 out of 42 active users, including social workers, educators, and psychologists. Average user rating: 4.1 out of 5. Before the introduction of Magic Notes, staff spent an average of 22 hours per week on administrative tasks. The adoption of the platform has already begun to significantly reduce this burden, allowing operators to dedicate more time to direct citizen support. The feedback from operators has been very positive. Operators who use Magic Notes the most report that the platform has significantly improved their daily work, reducing the time spent on documentation and increasing the time spent directly interacting with citizens. The overall average feedback score is 4.1 out of 5.

Key updates on the project's progress: In the first three months, the highest usage was recorded among the 23 social workers, who together conducted over 500 recordings with residents. The average session duration was approximately 15 minutes, with an average of more than 25 recordings per worker. Adoption proved more challenging for psychologists, primarily due to concerns that the tool's presence might make patients uncomfortable and reduce their willingness to open up during consultations. Educators, however, noted benefits during home visits.

Future Outlook: As 2026 begins, several strategic priorities have been defined to increase adoption and maximize the project's impact. A mandatory refresher webinar will be held for all staff, with the aim of ensuring that each operator feels confident using Magic Notes and can derive maximum benefit from it. More intensive individual training sessions will also be implemented for team members experiencing greater difficulties.

Finally, the artificial intelligence assistant that records, transcribes, and suggests interventions will be further customized based on the A5 Consortium's specific services, regulations, and training programs. This will be achieved by integrating the Consortium's own information into the system, allowing the AI to reference local guidelines and suggest interventions consistent with adopted practices.