Public Attitudes Toward and Acceptance of Artificial Intelligence in Health Care in a Multilingual Alpine Region: Population-Based Cross-Sectional Survey (Preprint)

BACKGROUND Italian law makes clinician oversight of artificial intelligence (AI) in health care a legal precondition. A recent scoping review of 35 Italian studies found acceptance of health care AI positive but conditional and identified two principal gaps: no general-population or patient study used a probabilistic sampling frame, and none compared Italy's language groups. OBJECTIVE We assessed attitudes toward and acceptance of AI in health care in the general adult population of South Tyrol, a trilingual autonomous province in northern Italy, quantified how acceptance varies across 16 application areas, and identified correlates of both. METHODS We surveyed "So denkt Südtirol" (So South Tyrol Thinks), a probabilistic panel of 1200 adults drawn from the population register by stratified random sampling, online between February 1 and March 6, 2026. The questionnaire assessed technological affinity, frequency of AI use, 5 attitude items, and acceptance of 16 applications. Items were adapted from a German population survey; no validated bilingual instrument existed. Analyses used poststratification weights. Correlates were examined with survey-weighted logistic regression (attitude items) and Poisson models (application counts). Because 60 coefficients were estimated, Benjamini-Hochberg false discovery rate (FDR) q values are reported. All regression analyses are exploratory. RESULTS Of 1200 panel members, 901 participated (participation rate 75.1%). Attitudes were positive but conditional on the application and on human oversight: 46.8% (95% CI 43.5-50.1) supported AI in their own care and 48.7% expected AI to improve population health, whereas 44.0% were concerned about losing human control. Acceptance ranged from 79.7% (95% CI 76.9-82.2) for automated medication reminders to 28.2% (95% CI 25.3-31.3) for support in urgent decisions. Uncertainty responses ranged from 14.1% to 39.1% and were more frequent among women (rate ratio [RR] 1.20, 95% CI 1.06-1.37) and rarer among daily AI users (RR 0.59). After FDR adjustment, only sex, technological affinity, and frequency of AI use remained associated with attitudes. Accepted applications increased stepwise across the 3 age groups (61-80 vs 18-40 years: RR 1.26, 95% CI 1.14-1.38) and with AI use (daily vs never: RR 1.64, 95% CI 1.42-1.89). Language group was largely unrelated to the outcomes. CONCLUSIONS This study provides the first population-based probabilistic evidence on attitudes toward health care AI in Italy and the first comparison across language groups. Attitudes were shaped by digital experience rather than by language group, and acceptance was strongly graded by application. The optimism-knowledge gap described among Italian clinicians has a population-level analogue: broad endorsement in principle alongside many respondents who know too little to take a position. Because Italian law already requires human oversight, the widespread concern about loss of human control aligns with the regulatory architecture and should be communicated as an existing safeguard. CLINICALTRIAL No