AI can now review scientific papers and explain its reasoning

Researchers built an AI system that can review scientific papers and provide specific, traceable feedback. This system, DeepReviewer 2.0, identifies problems, cites evidence, and suggests follow-up actions, outperforming other AI models and often human reviewers in blind comparisons.

Scientific peer review is slow, expensive, and often inconsistent. This paper shows that AI can now perform a core part of that process, not just by generating text, but by providing auditable judgments. This means that the bottleneck of human review could be significantly reduced, speeding up the pace of scientific publication and potentially improving quality by catching more major issues.