Abstract With the ability to approximate human decision-making and to parse through large and complex data, artificial intelligence (AI) technologies such as machine learning algorithms stand to play important roles in diagnosing, treating, and preventing society’s most pressing social, health, and academic challenges. However, some AI applications for these ends have resulted in biased and discriminatory outcomes, demonstrating poor accountability to the communities these technologies impact and serve. The foundational premise of this paper is that AI-augmented prevention science needs community engagement to mitigate these harms. Drawing on a decade of work conducted at the University of Southern California’s Center for AI in Society, we present a model of community-engaged AI-augmented prevention science. What sets our formalization apart from previous frameworks for community-engaged AI research is its prevention science orientation. We highlight potential roles for community and AI at each stage of the prevention science life cycle, from problem conceptualization to intervention implementation, and delineate when community input can compel moments of critical retreat in that life cycle to remain accountable to community needs and values. We then illustrate how we integrated AI and community-engaged methods in our work to correct racial biases in a predictive risk algorithm used to prioritize vulnerable people experiencing homelessness for housing interventions. Our case study highlights two moments of critical retreat—informed by the values, experiences, and contextual knowledge of our community partners—that led to more equitable and inclusive outcomes. We conclude with recommendations for how to advance community-engaged approaches in AI prevention science research.
Rice et al. (2025) studied this question.