Clinical Report: Public Views on Response Speed and Human Oversight in AI-Assisted Care
Overview
In a cross-sectional survey of 1,185 US adults, preferences for AI-assisted patient-portal responses differed by message type. Respondents more often favored rapid AI replies for logistical and general messages, but preferred waiting for health-professional review for clinical concerns; acceptance also varied across demographic and other respondent characteristics.
Background
Generative AI may offer more timely support amid professional shortages and prolonged wait times, but human review can delay responses. This study assessed preferences for rapid AI-generated replies versus slower responses reviewed by a doctor or nurse across general, logistical, and medical portal messages. It also compared preferences for immediate unsupervised AI mental health support, therapist-supervised AI, and delayed access to a human therapist. The study was approved by Columbia University's Institutional Review Board, obtained written informed consent, and followed STROBE reporting guidance.
Data Highlights
Measure
Finding
Survey participants retained
1,185 of 1,237 respondents; 52 (4.2%) were excluded after failing attention checks
Women
599 (50.5%)
Preference for AI reply within 5 minutes: logistical messages
680 (57.4%)
Preference for AI reply within 5 minutes: general messages
600 (50.6%)
Preference for health-professional-reviewed reply, potentially delayed up to 48 hours: medical messages
748 (63.1%)
Difference in preferences by message category
P < .001
Key Findings
The cross-sectional online survey was conducted from June 17 to July 26, 2024, among English-reading US adults aged 18 years or older.
For logistical messages, such as appointment scheduling and prescription refills, 57.4% preferred an AI response within 5 minutes.
For general messages, including nonmedical questions and administrative requests, 50.6% preferred a rapid AI response.
For messages about symptoms, treatment, or other clinical concerns, 63.1% preferred waiting for a response reviewed by a health professional; preferences differed significantly across message categories (P < .001).
Black or African American participants were more likely to accept AI for general, logistical, and medical questions. Generation X and Millennial participants showed greater acceptance for general questions; higher numeracy was associated with acceptance for general and logistical questions, and higher health literacy with acceptance for general and medical questions.
Greater reported discrimination in medical settings was associated with acceptance of AI for medical questions. Women were less likely to accept AI for medical questions, while Generation Z participants were less likely to accept it for logistical questions.
Clinical Implications
The survey found different preferences for AI response speed and professional review across general, logistical, and medical portal messages. The reported subgroup associations indicate that acceptance also varied by respondent characteristics; the study did not establish effects on clinical outcomes or evaluate a deployed messaging system.
Conclusion
Among surveyed US adults, preferences for rapid AI responses versus health-professional review depended on the type of portal message. Most respondents favored professional review for clinical concerns, while rapid AI responses were more often preferred for logistical and general messages.
