AI system learns to write peer reviews by looking at figures, not just text

Researchers built an AI that generates academic peer reviews by training itself to use not just the manuscript text but also figures, tables, and external information — something most review-writing systems ignore. This means automated reviews could actually evaluate whether a paper's claims match its data, not just summarize the words.

Academic peer review is already slow and expensive; most journals struggle to find reviewers. If an AI can generate competent reviews that account for visual evidence and external context, it removes a real bottleneck in scientific publishing. The catch is whether reviewers and journals actually trust these outputs enough to use them — this is a lab result showing the system works better than baseline models, not evidence that it works well enough to deploy at scale.