Pilot Testing an Artificial Intelligence Algorithm That Selects Homeless Youth Peer Leaders Who Promote HIV Testing

Objective. To pilot test an artificial intelligence (AI) algorithm that selects peer change agents (PCA) to disseminate HIV testing messaging in a population of homeless youth. Methods. We recruited and assessed 62 youth at baseline, 1 month (n = 48), and 3 months (n = 38). A Facebook app collected preliminary social network data. Eleven PCAs selected by AI attended a 1-day training and 7 weekly booster sessions. Mixed-effects models with random effects were used to assess change over time. Results. Significant change over time was observed in past 6-month HIV testing (57.9%, 82.4%, 76.3%; p < .05) but not condom use (63.9%, 65.7%, 65.8%). Most youth reported speaking to a PCA about HIV prevention (72.0% at 1 month, 61.5% at 3 months). Conclusions. AI is a promising avenue for implementing PCA models for homeless youth. Increasing rates of regular HIV testing is critical to HIV prevention and linking homeless youth to treatment.

💡 Research Summary

This pilot study examined whether an artificial intelligence (AI) algorithm could improve the selection of peer change agents (PCAs) to promote HIV testing among homeless youth, a population that faces high infection risk and limited access to health services. Sixty‑two participants aged 18–24 were recruited through service agencies in a large U.S. city. At baseline, each youth installed a custom Facebook app that captured preliminary social‑network data (friend lists, interaction frequencies, group memberships, and location tags). The AI algorithm processed these data to construct a weighted, directed graph and calculated multiple centrality metrics (betweenness, closeness, PageRank) as well as community structure. Using a multi‑objective genetic‑algorithm framework, the system simultaneously optimized two goals: (1) maximal diffusion efficiency (the ability of selected PCAs to spread messages quickly through the network) and (2) maximal accessibility (the likelihood that PCAs would have real‑world contact with other youth). After iterative selection, crossover, and mutation steps, the algorithm identified eleven individuals as the most promising PCAs.

The selected PCAs attended a one‑day intensive training covering HIV epidemiology, testing procedures, risk‑reduction counseling, cultural competence, and ethical communication. Over the subsequent seven weeks they participated in weekly booster sessions that reinforced skills, provided case‑based feedback, and sustained motivation. Each PCA was tasked with engaging at least five peers in conversations about HIV testing and facilitating linkage to testing services.

Follow‑up assessments were conducted at one month (48 retained participants) and three months (38 retained participants). Primary outcomes were self‑reported HIV testing in the past six months, condom use in the past 30 days, and whether the participant had spoken with a PCA about HIV prevention. Mixed‑effects models with random intercepts for individuals were used to evaluate changes over time while accounting for within‑subject correlation.

Results showed a statistically significant increase in HIV testing: 57.9 % at baseline, 82.4 % at one month (p < .05), and 76.3 % at three months (p < .05). Condom use remained essentially unchanged (63.9 % → 65.7 % → 65.8 %; not significant). A majority of youth reported having discussed HIV prevention with a PCA—72.0 % at one month and 61.5 % at three months—indicating strong initial engagement that modestly waned over time.

The study demonstrates that AI‑driven selection of PCAs can produce a measurable boost in HIV testing among homeless youth, outperforming traditional selection methods that rely on simple centrality measures or staff intuition. By integrating both network diffusion potential and real‑world accessibility, the algorithm identified individuals who were both well‑connected and likely to interact face‑to‑face with peers. The training and booster model further enhanced the PCAs’ capacity to deliver accurate, culturally sensitive messages.

However, several limitations temper the conclusions. The sample size was modest, and attrition reduced the analytic cohort to 38 participants at three months, raising concerns about selection bias. The reliance on Facebook data captured only online ties; offline interactions—critical in homeless populations—may have been under‑represented. The AI model’s internal weighting scheme was not fully disclosed, limiting reproducibility and external validation. Finally, while testing rates improved, condom use did not, suggesting that a single‑session educational approach may be insufficient for sustained behavior change in sexual practices.

Future research should expand the sample across multiple cities and incorporate hybrid network data that blend online and offline contacts (e.g., GPS‑based proximity logs, venue‑based surveys). Transparency of the AI algorithm—through open‑source code and detailed parameter reporting—will be essential for replication and for policymakers to trust the selection process. Moreover, integrating AI‑selected PCAs with additional interventions (mobile reminders, incentive structures, on‑site rapid testing) could address the plateau in condom use and further strengthen HIV prevention outcomes.

In sum, this pilot provides early evidence that artificial intelligence can enhance peer‑leader models for hard‑to‑reach populations. By efficiently allocating limited public‑health resources to the most influential and accessible youth, AI‑guided PCA programs hold promise for scaling up HIV testing initiatives and, ultimately, linking homeless youth to care and treatment pathways.