To test their hypothesis, the researchers enlisted participants for a scenario-based survey experiment via Mechanical Turk (MTurk), a popular crowdsourcing platform. They gathered data from a total of 655 participants. Bansal, Matta, and Diaz-Ordonez employed attention checks and follow-up inquiries to ensure the accuracy and reliability of their findings.
The study revealed that transparency plays a crucial role in the patient-provider dynamic concerning the use of artificial intelligence (AI). This finding supported their primary hypothesis: when AI is used transparently, it fosters greater trust in both the healthcare provider and the AI technology they employ. Essentially, the researchers observed that trust in the provider translates into trust in the tools they utilize.
A surprising aspect of their results was related to the perceived accuracy of AI diagnoses and its effects on the patient-provider relationship. While increases in transparency were linked to heightened trust—as anticipated—greater accuracy in AI diagnoses appeared to have an adverse effect, either diminishing trust or causing it to plateau. Bansal proposed several explanations for this unexpected outcome.
He noted, “People fear that if AI becomes too accurate, doctors may rely solely on it, neglecting their own critical judgment, especially in primary care. This anxiety is likely reflected in our findings.”
Matta emphasized that these results could fundamentally alter our understanding of trust in relation to artificial intelligence. “What makes this significant is that it contradicts the assumption that accuracy enhances trust,” he remarked. “The implications are profound. Without this insight, there would be widespread concern about doctors being replaced by AI. However, our research suggests otherwise.”
The findings were showcased at the Midwest Association for Information Systems (MWAIS) Conference in May 2025, hosted in Oklahoma, with the subsequent MWAIS 2026 conference taking place at Ohio University. The study is currently accessible through conference proceedings, and Bansal and Matta are in the process of preparing it for journal publication.
These results are gaining traction, as Matta and Bansal have noted a growing number of recent studies arguing that outputs generated by AI may be viewed as less valuable than those produced by humans. Bansal emphasized the importance of careful interpretation of their findings, as these outcomes are contextualized by transparency and accuracy. “We must not generalize and suggest that accuracy lacks importance in AI healthcare. Within the framework of transparency and accuracy, the former takes precedence, reducing the significance of accuracy when transparency is established.”
Furthermore, Bansal and Matta clarified that the diminished importance of accuracy in primary care does not imply it is irrelevant in other healthcare contexts. They intend to broaden their research to include specialized medical fields to compare these findings across different areas of healthcare.
“One study alone isn't sufficient,” Bansal concluded. “Our discovery offers counterintuitive insights, underscoring the need for further investigation.”
