Session 8

AI Ethics, Digital Equity, and Responsible Innovation in Social Work

Moderator: Dr. Lixia Zhang (Assistant Professor, University of Louisville, USA)

O8.1  Advances in AI and Mental Healthcare Equity: Promise, Peril, and Pragmatic Considerations

*Emma M. Sterrett-Hong¹

¹University of Louisville, Kentucky, USA

Abstract

Unmet need, and inequitable access to mental health services, is a global crisis (WHO, 2025). For example, globally, only 9% of people experiencing depression receive adequate treatment (Santomauro et al., 2024). In addition, disparities in access between high-income countries and low-income countries is marked, with, for example 22% of people with depression in high-income countries receiving the minimally acceptable treatment compared to 2% to 3% of their counterparts in low-income countries (Moitra et al., 2022). Within high-income countries, disparities in access have also been documented, such as between housed and houseless people and between White and ethnic minorities in the United Kingdom (Lowther-Payne et al., 2023) and between White and Black, as well as Hispanic, people in the United States (Olfson et al., 2023). The potential for various AI technologies to extend the reach of mental health services has increasingly been studied in empirical literature and formal reports (e.g., Auf et al., 2025; Cross et al., 2024, Thakkar et al., 2024).

This talk will review the current state of the literature regarding the adoption of AI and its potential for promoting mental health equity. Findings regarding the use of AI in mental health assessment, predictive analytics, as an adjunctive tool to human-delivered mental health treatment, and as the sole mental health treatment provider will be summarized. Key benefits such as increased efficiency and access (e.g., McBain et al., 2025) as well as documented risks, such as stigmatizing language and encouragement of harmful behaviors (e.g., Moore et al., 2025), depending on the modality of AI use, will be discussed. Ethical considerations and implications, including in comparison to the National Association of Social Workers Code of Ethics (NASW, 2021) and the Healthcare Four Ethical Principles framework (Varkey, 2021), will be discussed. Limitations related to equitable access and culturally-responsive care will be highlighted. Finally, practical strategies for the harnessing of AI to safely and ethically improve availability of evidence-based mental health services to populations with unmet mental health needs will be described (e.g., Open & Breakspear, 2026).

O8.2  Being Divided Community by AI between Young and Aging Population: In-depth Interview in the US and Japan

*Tatsushi Hirono¹

¹Department of Social Work, Austin Peay State University, Tennessee, USA

Abstract

According to the researcher’s research result, our community was divided into two groups that were “AI accessible younger generations” and “AI Not-accessible aging population.” Moreover, our community was divided by AI accessibility including age, religion, and health status. The researcher found that individual AI accessibility / skills, age, religion and health status differences may affect their income and behavior regardless of their stance on AI. The researcher employed mixed methods as the research method. The researcher conducted an internet survey and collected 110 quantitative data. He also interviewed 30 Japanese individuals in the USA, and 30 Japanese individuals in Japan, and collected 60 qualitative data. By using the SPSS and MAXQDA software for analyzing the quantitative and qualitative data, he found five qualities in the USA and Japan: (1) AI is “evil” in some sects of Christians in the USA (they also believed that conspiracy theory and singularity); (2) People must know some basic knowledge of AI to keep their jobs and to live normal lives in Japan (there is no option); (3) Buddhist believers accept AI more positively than Christian believers both in the USA and Japan; (4) Communities were divided into two groups: AI accessible and Not-AI accessible populations in the USA and Japan. The researcher also found that (5) the people’s “fear” and “anger” would lead to communities being divided into multiple groups: Accepting AI, Not-Accepting AI and Neutral. Especially, some middle-aged workers, aged 40’s and 50’s, who had recently lost their jobs to AI tend to “hate” AI. For example, a mid-40’s Japanese office worker who was fired by the developing AI said, “I hate AI because they robbed my job, they can’t understand human beings, but they can have more cost-performance.” Or another mid-50’s Japanese worker said, “I felt worthless and depressed when the AI took my job.” Furthermore, the researcher found that Buddhist believers had less mental health obstacles to “copy & paste” what AI generated essays and arts. One of the Japanese middle school teachers who is a Buddhist said, “I am positive that my middle school’s students use AI because our students can create more innovative products based on the AI products. It’s not a copycat, but it’s inspired by AI.” However, the research found that the Japanese communities both in the USA and Japan had already divided by: the “AI haves” and the “AI have-nots.” To reunite our communities, we need to get rid of “fear” and “anger.” We need more proper education of AI. Furthermore, to remove this “AI Phobia,” mental health professionals, teachers, and religious leaders should cooperate and tell people about the “Scientific Evidence/Truth regarding AI.”

O8.3  From Algorithmic Seduction to Digital Exploitation: An Eco-Systemic Social Work Intervention Study on Adolescent Exposure to Online Soft Pornography

*Hongxin Huang¹,², Mengmei Liu², Jiaxin Dai²

¹ Sun Yat-sen University, Guangzhou, China; ² Shenzhen Wenxin Social Work Service Center, China

Abstract

Background and Purpose: In the era of algorithmic recommendation, adolescents’ exposure to online soft pornography has evolved from passive exposure into systematic “digital exploitation.” Existing interventions are largely post-hoc remedies and lack an eco-systemic perspective. Drawing on eco-systemic theory, this study aims to: (1) examine the current status, pathways, and influencing factors of adolescents’ exposure to online soft pornography; (2) identify the deficiencies of current social work interventions; and (3) propose an eco-systemic intervention framework.

Methods: A mixed-methods design was employed. A survey was conducted among 523 middle school students in City S, measuring exposure frequency, platforms, attitudes, and the current state of online safety education. Semi-structured interviews were carried out with senior forensic social workers from a detention center and a district procuratorate. Thematic analysis was applied to interview transcripts, and descriptive statistics were generated from the survey data.

Results: First, the algorithmic inducement effect of AI-era platforms is significant. Adolescents’ daily time spent online is one of the strongest predictors of exposure frequency, indicating that platform recommendation algorithms extend exposure time and reinforce contact. Second, peer transmission acts as an amplifier of adolescents’ exposure to soft pornography. The effect size of peer discussion frequency (OR = 2.44) exceeds that of gender and time spent online, suggesting that soft pornographic content easily spreads virally among adolescent groups through social sharing. Third, existing education has a protective effect but is insufficient to counteract the combined influence of algorithms and peers; current educational content lags behind and its format needs to be upgraded. Fourth, the absence of family digital supervision is a fundamental shortcoming. The survey shows that only 30% of students have received parental guidance on this issue. Combined with interview data, the lack of family support constitutes the weakest link in the entire eco-system.

Conclusions and Implications: At the micro level, algorithmic inducement is the “invisible driver” of individual adolescents’ exposure to soft pornography. At the meso level, peer transmission serves as an “amplifier” of exposure diffusion. The gap between meso and macro levels results in the limited protective effect of existing education, which cannot offset the superimposed influence of algorithms and peers. Finally, the absence of macro-level systems—specifically the lack of family digital supervision—is the weakest link in the entire eco-system, and the “youth mode” on platforms is virtually ineffective. From an eco-systemic theory perspective, current social work intervention for this adolescent group suggests the following: at the micro level, media-empowerment case work can be implemented to enhance adolescents’ critical media literacy as well as their risk identification and avoidance abilities. At the meso level, courses should be developed to empower families and strengthen parental digital supervision capacity; simultaneously, school-based prevention groups should be established through school-based channels to block peer transmission. At the macro level, policy recommendations should be proposed for supervisory authorities to advance platform algorithmic ethics reviews and improve minors’ digital rights policies. From these three dimensions, a youth online safety eco-network should be established, promoting a paradigm shift in the current intervention system from “post-hoc remedy” to “proactive ecological protection.”

O8.4  Confidence as the Missing Mechanism: A Bounded Rationality Perspective on Social Work Ethics Education

*Xiao LI¹

¹Department of Social Work, East China University of Science and Technology, Shanghai, China

Abstract

Background and Purpose: This study examines how social work students build decision confidence in suspected child maltreatment cases and proposes reconceptualizing ethics education from procedural training to confidence cultivation.

Method: Thirty social work students received instruction on ethical theories, bounded rationality, and child maltreatment research. Using directed content analysis grounded in Simon’s bounded rationality theory, we analyzed students’ structured reflections and group discussions.

Results: Decision confidence was shaped by information uncertainty, ethical value conflicts, and cognitive biases. Two pathways of confidence construction emerged. External anchoring: justifying decisions through legal compliance, professional assessment, practical feasibility, and ethical reasoning. Internal anchoring: coherence between decisions and professional identity, sustained by ethical commitment and acceptance of “satisficing solutions.” Both anchors were teachable.

Implication: The procedural paradigm is incomplete because it ignores confidence—the condition that translates knowledge into action under normative uncertainty. We argue that ethics education should shift from finding “correct answers” to cultivating defensible choices and the confidence to stand behind them. Bounded rationality offers a normative framework for this reorientation.

Conclusion: Confidence is not a personality trait but a teachable mechanism. By designing education around external and internal anchoring, social work programs can prepare students for the ethical complexity of child protection practice.

O8.5  From Justice to Care: Rethinking AI Ethics in Social Work with the Adaptive-Relational Care Assessment Matrix

*Charles Tong-Lit Leung¹

¹Hong Kong Shue Yan University, Hong Kong, China

Abstract

Background and Purpose: The rapid integration of Artificial Intelligence (AI) into social work practice presents profound ethical challenges. Current AI ethics frameworks rely heavily on an "Ethics of Justice"—focusing on data privacy, algorithmic accuracy, and compliance. While necessary, this paradigm risks overlooking the relational core of social work. Furthermore, diverse welfare subvention models across the Greater Bay Area (GBA)—such as Hong Kong’s Lump Sum Grant (LSG), Macao’s Social Welfare Bureau (IAS) subventions, and Guangdong’s Government-Purchased Services—place immense administrative pressure on agencies, risking the commodification of therapeutic labour through AI-driven efficiency. To address this theoretical and practical gap, this conceptual paper introduces the Adaptive-Relational Care Assessment Matrix (ARCAM), a novel theoretical framework and practical blueprint designed to shift AI evaluation from technological compliance to relational safeguarding.

Theoretical Framework: The ARCAM framework synthesizes Complexity Theory and the feminist Ethics of Care. Rather than viewing AI merely as an administrative tool, Complexity Theory frames AI as an active "systemic agent" that alters practice environments and creates fuzzy boundaries within the human-client therapeutic triad. The Ethics of Care provides the normative compass, prioritizing human vulnerability, mutual responsiveness, and contextual narratives over abstract rules. To balance regional scalability (etic) with deep cultural relevance (emic), ARCAM utilizes a "Core + Module" architecture. The "Core" establishes universal ethical metrics across the region, such as the Empathic Monopoly Check and the Care Labor De-Commodification Index. The "Modules" comprise hyper-localized simulation vignettes that adapt the tool to distinct socio-political realities.

Proposed Application: To operationalize this theory, the paper outlines a proposed six-month Participatory Action Research (PAR) methodology for social work education. Rather than passively receiving tech-compliance training, students act as co-designers and "Ethical Auditors." The framework guides educators to deploy localized modules tailored to specific GBA contexts. For instance, a module for Hong Kong might critique predictive AI used for youth in subdivided flats, while a Macao-specific module would evaluate AI case-planning for cross-border casino shift-workers. This modular approach highlights unique local risks, such as client re-identification within tight-knit Kaifong (neighbourhood) networks, and the threat of centralizing e-governance (e.g., the Macao One Account or HK’s iAM Smart) without analog safety nets for digitally marginalized populations.

Conclusions and Implications: As welfare agencies increasingly adopt AI, social work education requires robust, context-sensitive ethical paradigms. This theoretical paper provides a critical intervention by offering the GBA educators and practitioners of social work a unified, adaptable practical blueprint. By embedding Complexity Theory and Care Ethics into the curriculum, this conceptual model lays the groundwork for future cross-border empirical research. Ultimately, it aims to transform students from passive technology consumers into proactive "algorithmic justice advocates" equipped to utilize AI to enhance, rather than replace, authentic human connections. The expected outcomes could be utilized not only in Hong Kong but also in Macao and the inland cities of the Greater Bay Area and beyond for scaling up.

O8.6  From Online Grooming to Criminal Exploitation: How Family and Digital System Failures Create a Victim-to-Perpetrator Cycle in Child Sexual Abuse

*Xiayu Wang¹

¹China University of Political Science and Law, China

Abstract

Background and Purpose: With the increasing prevalence of digital social platforms, minors are increasingly exposed to the risks of online sexual exploitation and grooming. This study examines a real-life case from Henan Province, China, where a young girl was introduced to a high-risk peer group through social media, subsequently becoming a victim of sexual exploitation and later engaging in criminal activities. The analysis highlights the critical role of digital platforms in facilitating exploitation and examines how the absence of effective family, legal, and social interventions contributes to the transformation of victims into perpetrators.

Methods: A case study approach was adopted to explore the trajectory of a girl who dropped out of school at age 9, was introduced to an online social circle at 11, and entered a prostitution ring at 13. Content analysis was used to examine how social media platforms enabled high-risk interactions and influenced victims’ perceptions. Additionally, legal and policy analysis was conducted to evaluate the gaps in child protection systems and to propose policy recommendations based on international best practices.

Results: 1.The Role of Social Media: Algorithmic recommendations, voice chat rooms, and rapid friend-matching features accelerated minors’ exposure to high-risk groups, making them more vulnerable to exploitation. 2.Cognitive Manipulation and Psychological Framing: “Peer group culture” shaped minors’ moral judgments, leading them to misinterpret sexual assault as romantic relationships and view violence as a form of friendship. 3.Failure of Legal and Social Support Systems: Many perpetrators in the case were minors themselves, making legal accountability difficult. Victims lacked adequate family supervision and psychological support, exacerbating their vulnerability. 4.Victim-to-Perpetrator Transformation: Prolonged trauma and social exclusion led victims to seek identity validation within their peer circles, ultimately adopting violent behaviors and engaging in criminal activities.

Conclusion: This study reveals how minors can be “stolen” into the world of online exploitation and criminality through social media. The interplay of digital platform mechanisms, legal loopholes, and failed parental supervision creates an ecosystem of child sexual exploitation. Strengthening platform regulations (age verification, restrictions on social features), legal reforms (enhanced accountability for minor perpetrators, improved victim protection), and family and social interventions (psychological support, digital literacy education) is critical for building a more comprehensive online child protection framework.