AI wrote a scientific paper that passed peer review

Researchers led by Jeff Clune developed the "AI Scientist," an AI system capable of generating a complete scientific paper from a general topic prompt without human involvement. This system, which orchestrates existing foundation models, surveyed literature, generated hypotheses, planned experiments, analyzed data, and wrote the paper, even conducting an internal peer review. One of the three papers generated by the AI Scientist was accepted at the mediocre-bar ICBINB workshop at the 2025 International Conference on Learning Representations (ICLR). While the accepted paper was deemed mediocre, suffering from issues like hallucinated references and lack of rigor, its creation marks a turning point where AI moves from assisting science to attempting to be the scientist. The speed and low cost (estimated at $140 in 15 hours) contrast sharply with human effort, raising concerns about flooding the peer-review system with automated submissions. Top venues are setting limits on purely AI-written papers, but experts predict this technology will only improve, potentially leading to an era where AI drives rapid scientific advances, though some argue human curation and interaction will remain vital.

(Source：Scientific American)