Multimodal Generative AI and Foundation Models for Behavioural Health in Online Gambling

TL;DR

This narrative review addresses the pressing public health challenge of problem gambling in the online era by exploring how multimodal generative AI and foundation models can support prevention, early detection, and harm reduction. It outlines six concrete use cases—synthetic data, responsible marketing, personalized interventions, gamified recovery tools, AI-assisted counselor training, and scenario modeling for policy—alongside ethical and technical challenges such as privacy, fairness, and governance. The work emphasizes data privacy, cross-platform validation, and stakeholder collaboration as essential for responsible deployment that aligns with public health goals. Overall, it provides a practical roadmap for integrating advanced AI technologies into health-focused strategies to mitigate gambling-related harms across actors including researchers, clinicians, policymakers, and platform operators.

Abstract

Online gambling platforms have transformed the gambling landscape, offering unprecedented accessibility and personalized experiences. However, these same characteristics have increased the risk of gambling-related harm, affecting individuals, families, and communities. Structural factors, including targeted marketing, shifting social norms, and gaps in regulation, further complicate the challenge. This narrative review examines how artificial intelligence, particularly multimodal generative models and foundation technologies, can address these issues by supporting prevention, early identification, and harm-reduction efforts. We detail applications such as synthetic data generation to overcome research barriers, customized interventions to guide safer behaviors, gamified tools to support recovery, and scenario modeling to inform effective policies. Throughout, we emphasize the importance of safeguarding privacy and ensuring that technological advances are responsibly aligned with public health objectives.

Figures (5)

• Figure 1: Environmental determinants of online gambling and associated data sources for research and prevention. The determinants are classified into accessibility, socioeconomic factors, regulatory environments, and commercial influences. These determinants interact to shape online gambling behaviors and outcomes. The figure also maps relevant data sources, including demographic surveys, public health records, and behavioral datasets, to support research on these influences and inform interventions aimed at reducing gambling-related harms.
• Figure 2: Technological interventions in online gambling prevention. This figure highlights a population health approach with three tiers: primary prevention focuses on risk factor analysis and public health promotion; secondary prevention includes early detection of gambling behaviors and risk modeling; and tertiary prevention addresses treatment and recovery with in-game interventions and relapse prevention technologies.
• Figure 3: UpSet plot illustrating the intersections of requirement vectors across AI application categories. The vectors are defined as follows: Detection and Prediction ($V_{\mathrm{DP}} = \{1, 0, 1, 0, 0, 1, 0, 1\}$), Behavioral Analysis ($V_{\mathrm{BA}} = \{1, 0, 1, 0, 0, 1, 0, 1\}$), Intervention and Monitoring ($V_{\mathrm{IM}} = \{1, 0, 1, 0, 0, 0, 0, 0\}$), and Prevention ($V_{\mathrm{P}} = \{1, 0, 1, 0, 0, 0, 0, 0\}$). A size of two for intersections indicates that two requirements co-occur across two categories. A size of zero means that no category combination includes those specific requirements simultaneously. The plot highlights the consistent inclusion of demographic ($r_1$) and behavioral determinants ($r_3$), while systemic factors like policy engagement ($r_7$) and fairness ($r_8$) are underrepresented, particularly in Intervention and Monitoring and Prevention.
• Figure 4: Generative AI-Driven Analysis of Marketing Strategies in Online Gambling Advertisement. An online sports-betting advertisement analyzed using generative AI (GPT-4O). Marketing strategies were detected from visual and textual elements, annotated numerically, and listed below. Model parameters: temperature = 0.2, max tokens = 950, top-p = 0.9, frequency penalty = 0.2, presence penalty = 0.0.
• Figure 5: Multimodal Foundation Model for Problem Gambling Counseling. This figure illustrates the integration of multimodal data ($\mathcal{S} = \{s_1, \dots, s_m\}$), including player behaviors, marketing influences, and conversational cues, into a foundation model ($r_{\Theta}$). The model processes these inputs to generate actionable insights ($\mathcal{F}$) that guide counselors ($\mathcal{C}$) in providing tailored interventions. The counselor ($c_1 \in \mathcal{C}$) interacts with the model’s recommendations to address specific client concerns, such as identifying risk factors (e.g., financial stress, high deposit frequency) and proposing strategies like deposit limits or behavioral exercises.