(PS9-4) Generative Artificial Intelligence Recognition of Suicide Risks in Hispanic/Latino Youth

Abstract Text: Suicide is a leading cause of death among adolescents in the United States, disproportionately affecting Hispanic/Latino youth. Despite their elevated risk, traditional suicide risk assessments often overlook cultural variables that influence mental health outcomes. Artificial Intelligence (AI) chatbots, such as ChatGPT-4, ChatGPT-4.0, and Google Gemini, are up-and-coming tools that may be used for suicide risk detection; however, there is limited research on their effectiveness and potential biases when applied to Hispanic/Latino youth. This study evaluated the ability of AI chatbots to identify suicide risk and recognize cultural factors influencing suicide among Hispanic/Latino youth, as AI-based mental health risk assessments have the potential to expand access to care, specifically for underserved populations.
In this present investigation, the vignettes depicted Hispanic/Latino and White boys and girls experiencing symptoms of depression and suicidal ideation. A total of 20 trials (N = 240) were generated for each vignette across three AI chatbots (ChatGPT-4, ChaptGPT-4.o, and Google Gemini). The AI chatbots were tasked with determining suicide risk levels and identifying relevant cultural factors. Statistical analyses revealed significant racial and gender differences. AI models rated suicide risk higher in Hispanic/Latino youth (F(1,236) = 18.85, p < 0.001, 𝜼p2 = 0.07) than in White youth,, which contradicts the clinical trends. They also rated boys as having higher suicide risks than girls (F(1,236) = 7.63, p = 0.006, 𝜼p2= 0.03). AI chatbots more frequently identified cultural barriers (e.g., acculturative stress, gender norms) among Hispanic/Latino youth (F(1,236) = 1177.30, p < .001, 𝜼p2 = 8.3), suggesting an awareness of cultural influences on suicidality. Public stigma was also measured using the Stigma of Suicide Scale-Short Form (SOSS-SF). However, no significant differences were found in AI chatbots’ recognition of public stigma across racial groups.
These findings reveal the potential and limitations of AI in public health applications, particularly for suicide prevention among diverse racial groups. While AI models may help bridge gaps in mental health screening, their differential assessment patterns indicate the need for culturally informed training data to prevent biases in digital health screening. This study contributes to the discussion on ethical AI use in behavioral health, reinforcing the necessity of ongoing evaluation and refinement of AI-driven mental health tools to ensure equitable, effective, and bias-free assessments.