Future work must understand this variability, and design explanation formats that provide more consistent responses.

Thanks for my co-authors: Elizabeth Bradburn, @yaringal.bsky.social, @profaris.bsky.social , J. Alison Noble.

Paper: www.nature.com/articles/s41...

We found that AI predictions significantly reduced clinician error.

But adding explanations gave no further significant reduction.

Explanations also had no significant effect on self-reported trust or reliance.

Critically, while some clinicians improved with explanations, others performed worse.