Paper Number

ECIS2026-1929

Paper Type

SP

Abstract

AI-driven diagnostic support is increasingly integrated into clinical workflows, yet clinicians continue to exhibit reluctance to rely on algorithmic advice. This research-in-progress study investigates how clinicians evaluate diagnostic advice from human experts, rule-based AI, and generative AI (genAI), and whether per-case confidence disclosures shape credibility and adoption intentions. Drawing on algorithm aversion research, we conduct a 3×2 vignette-based experiment manipulating advisor type and stated diagnostic confidence. Preliminary results show a persistent credibility premium for human advisors: identical advice is rated as more trustworthy and competent when attributed to clinicians than to either AI system. Clinicians also treat rule-based AI and genAI similarly, indicating a simple “AI label” heuristic that obscures meaningful transparency-performance trade-offs. Confidence cues exert no effect on trust, competence or willingness to use. These findings challenge assumptions about transparency interventions and highlight the need for institutional trust scaffolds to support clinical human-AI collaboration.

Share

COinS
 
Jun 14th, 12:00 AM

Why Clinicians Still Trust Humans Over (Gen)AI Systems: The Anatomy Of Transparency and Algorithm Aversion In Healthcare

AI-driven diagnostic support is increasingly integrated into clinical workflows, yet clinicians continue to exhibit reluctance to rely on algorithmic advice. This research-in-progress study investigates how clinicians evaluate diagnostic advice from human experts, rule-based AI, and generative AI (genAI), and whether per-case confidence disclosures shape credibility and adoption intentions. Drawing on algorithm aversion research, we conduct a 3×2 vignette-based experiment manipulating advisor type and stated diagnostic confidence. Preliminary results show a persistent credibility premium for human advisors: identical advice is rated as more trustworthy and competent when attributed to clinicians than to either AI system. Clinicians also treat rule-based AI and genAI similarly, indicating a simple “AI label” heuristic that obscures meaningful transparency-performance trade-offs. Confidence cues exert no effect on trust, competence or willingness to use. These findings challenge assumptions about transparency interventions and highlight the need for institutional trust scaffolds to support clinical human-AI collaboration.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.