Nurses Judge AI Advice in Made‑Up Patient Scenarios

A group of sixty‑eight health workers took part in a small test. Most of them were registered nurses – fifty‑nine to be exact – and the rest were physicians. They looked at pretend patient situations that were created for the study. The study aimed to see how much faith they placed in computer‑generated advice.\n\nEach person saw twenty‑one of these vignettes. They belonged to one of two sets, and each set had ten correct AI suggestions and eleven wrong ones. Before being told if the AI was right, they decided whether to follow the advice and answered three quick questions about how much they trusted the suggestion, how clear it seemed, and how likely they were to act on it. They recorded their choices right away, before any feedback about correctness was given.\n\nThe results showed that the workers accepted about half of the inc

orrect recommendations – three hundred seventy‑four out of seven hundred forty‑eight. They rejected only a small share of the correct ones – one hundred twelve out of six hundred eighty. Trust scores were lower when the AI was wrong, and the trust level slipped a little as they moved through the vignettes, especially in the first set of cases. Overall, the pattern suggested that trust followed the quality of the AI’s advice, though the drop was modest.\n\nTo understand the patterns, the researchers built a mixed‑effects model that accounted for differences between people and between vignettes, with the case series as a side note. They also found that a stronger prior interest in using AI was linked to higher trust scores. This modeling approach helped isolate the influence of trust from other factors like order or personal traits.

actions