AI for Social Good and Society is structured around researchers co-creating tools and studies with communities while reshaping how technology itself is built and governed in order to benefit society. By pairing rigorous science with community partnership, these projects demonstrate how AI can drive equitable health outcomes and strengthen social trust.
PRISM-Capabilities: Integrating AI into Community-Engaged Research
SIG researchers introduced the PRISM-Capabilities model to guide the ethical and equitable use of artificial intelligence in community-based implementation science. Building on the Practical, Robust Implementation and Sustainability Model (PRISM) and the Capabilities Approach, this framework ensures that AI tools are designed with—and for—communities, emphasizing transparency, fairness, and sustainability.
Drawing on rich qualitative and quantitative data from the HEALing Communities Study (HCS), the team demonstrated how this model can help public health initiatives harness AI while protecting equity and community voice. The work offers a blueprint for using AI to strengthen decision-making and accelerate evidence-based interventions without reproducing bias or undermining trust.
The full study is available to read in Implementation Science.
Human-Centered Benchmarks for Evaluating AI Chatbot Equity
To ensure AI chatbots are safe and effective for LGBTQ+ communities, SIG researchers developed a human-driven benchmarking tool to evaluate outputs from popular chatbots. An expert panel assessed responses across five domains: validity, reliability, usability, bias, and safety, revealing significant equity gaps that automated tools often miss.
Results show that human reviewers are essential for identifying cultural blind spots and potential harms, especially for stigmatized groups. This approach, which advances a more ethical and community-informed standard for evaluating AI tools in health and social care, was presented at the 2025 Society for Prevention Research Annual Conference.
Structuring AI Prompts to Support LGBTQ+ Health Equity
Researchers tested how different prompt structures can shape AI chatbot responses to better support LGBTQ+ populations. Using the RISEN framework: Role, Instructions, Steps, End-goal, and Narrowing, they examined how prompts influenced the quality and inclusivity of responses from leading chatbots like ChatGPT, Claude, and Gemini.
In a use case focused on mental health strategies for queer BIPOC youth, more structured prompts yielded practical, actionable guidance, while open-ended ones allowed for richer cultural nuance. The findings suggest that prompt engineering is a powerful tool for frontline professionals seeking culturally responsive and equitable AI support.
AI-Enhanced WINGS: Expanding Support for Women Experiencing Gender-Based Violence
SIG researchers are integrating generative AI into the evidence-based Women Initiating New Goals of Safety (WINGS) intervention to expand its reach and impact. WINGS provides intimate partner violence (IPV) screening, personalized safety planning, and service referrals (SBIRT). The AI-powered version will offer 24/7 automated support, tailored safety reminders, and customized referrals for women from key affected populations—including those living in poverty, women who use drugs, women who engage in sex work, and women living with HIV.
Machine-learning algorithms will continuously refine intervention strategies using real-world data, while a multi-stakeholder Community Advisory and Accountability Board (CAAB)—including AI experts, IPV service providers, community leaders, and women with lived experience—will guide design and ensure ethical oversight. The project will pilot-test the tool with women from these communities to assess feasibility, safety, and accuracy.
By combining AI-enhanced implementation science with community engagement and equity-focused methods, this work aims to broaden access to WINGS, strengthen referrals to effective services, and reduce IPV rates. The project represents a new model for human-centered, ethically governed AI that advances health equity for women at highest risk of gender-based violence.
AI and Stigma in Addiction: Finding Insights from HCS Coalitions
In a companion project, SIG researchers used natural language processing and large language models to examine hundreds of HEALing Communities Study coalition meeting notes and transcripts, revealing how stigma shapes local decisions about overdose prevention strategies. The analysis showed that stigma was most often discussed when coalitions weighed evidence-based practices like medications for opioid use disorder, and that conversations about racial equity increased the likelihood that stigma would surface.
By demonstrating how AI can illuminate hidden patterns in community deliberations, this work highlights both the promise and responsibility of AI in addiction research. It shows how real-time, AI-driven analysis could help coalitions track stigma and equity concerns as they emerge, ensuring more inclusive and responsive public health interventions.
The full study was published in the Journal of Addiction Medicine.