Automated Scientific Revision Through Iterative AI Reviewing

Scientific publishing is an iterative process: reviewers identify weaknesses, and authors revise the work before it is evaluated again. Addressing reviews requires more than just rewriting. Authors need to run experiments, add baselines, and test whether new evidence answers the points in reviews. AI scientists can conduct research and prepare manuscripts while an AI reviewer can generate detailed critiques. However, neither establishes that repeatedly acting on AI reviews improves the underlying research. We ask whether an AI reviewer can guide code, experiment, and manuscript revisions of existing papers without replacing their central contributions.