A commentary in npj Digital Medicine argues that science needs new reporting standards because AI is now the first interpreter of much biomedical research.
Published August 24, the piece by Nicholas Peoples, Ken Milne, Bing Luo and Lijing L. Yan describes how large language models increasingly summarize studies, extract evidence and shape how clinicians apply findings to patient care. At the same time, research reaches wider non-specialist audiences through AI-generated summaries on apps and search engines.
The problem: interpretive errors and generalization bias in machine-written summaries can skew clinical decisions and endanger public health. Models trained on messy, heterogeneous literature may carry forward study limitations, conflate correlation with causation, or overstate effects when compressing complex methods into plain language.
Rather than waiting for more capable models, the authors contend science itself can adapt. They call for modernized reporting standards that make papers easier for machines to interpret accurately, such as structured declarations of scope, population and uncertainty, and machine-readable statements that limit overreach.
The commentary lands amid growing evidence that clinicians and patients increasingly rely on AI tools to digest medical literature. Guardrails built into the publication process itself, the authors argue, would improve both human reading and machine interpretation at once.
