Premium

Japanese startup Sakana’s AI-generated research paper passes peer review

The AI startup submitted three research papers generated by its AI system for review at the ICLR 2025 workshop.

The success of the AI-authored paper shows how briskly AI is progressing in the world of science. (Image: Sakana)

So far the world has witnessed AI systems that are capable of holding human-like conversations, but ones that could produce scientific research papers was a far-fetched idea. Sakana, a Japan-based AI startup, has claimed that its AI system has successfully generated scientific papers that passed peer review. The development comes at a time when the debate around the rising significance of AI in accelerating scientific progress is gaining steam.

The AI system named AI Scientist-v2 generated three papers. It created the hypotheses, experimental code, visualisations, data analyses, and text for all three papers. And all of this was reportedly performed without any human modification. A submission generated by the system was accepted at the International Conference on Learning Representations (ICLR) 2025 workshop with an average reviewer score of 6.3, essentially ranking much higher than several human-written papers. For the uninitiated, the ICLR workshop is intended as a forum for those in the global machine learning community who wish to help tackle climate change.

Also Read | Who will win the race to Artificial General Intelligence?

The Japanese AI startup also pointed out a few caveats that include the AI making citation errors and workshop acceptance rates being higher than typical conference tracks. The company also said that the paper did not meet its internal bar for ICLR conference papers; however, it displayed early signs of progress.

Story continues below this ad

When it came to the evaluation, the company said that it worked with ICLR workshop organisers and agreed to submit three AI-generated papers at the workshop for peer review. The reviewers were informed about the likelihood of the papers under review being AI-generated (3 out of 43 papers), but not if the papers assigned to them would actually be AI-generated or not.

Sakana AI said that the AI-generated papers it submitted were entirely generated end-to-end by AI without any modifications from humans. The AI Scientist-v2 wrote every word of the entire scientific manuscript, from title to the final sentence. It also placed the figures and carried out the formatting. “We, as the humans overseeing this research, merely gave it the broad topic to perform research on (because the topic should be relevant to the workshop we submitted to) and picked 3 AI-generated papers to submit. We chose this number following discussions with the workshop organisers to avoid overburdening reviewers,” the company stated on its official website.

Also Read | LinkedIn Engineering head Mohak Shroff: There will be many AI agents… you will need an AI chief of staff

Meanwhile, earlier this month, another AI startup Autoscience introduced Carl, an AI system that successfully authored a research paper accepted at the International Conference on Learning Representations (ICLR). This achievement highlighted Carl’s capability to contribute to academic research by producing work meeting peer-review standards.

Introducing Carl, the first AI system to create a research paper that passes peer review. Carl’s work was just accepted at an @ICLR_conf workshop on the Tiny Papers track. Carl forms new research hypotheses, tests them & writes up results. Learn more: https://t.co/lBE0SRIOm3

— Autoscience Institute (@AutoScienceAI) March 3, 2025

Even though this is a breakthrough, there is more that needs to be studied in this domain. This feat shows the growing significance of AI in academic research processes. Earlier this year, Google released its AI co-scientist, a virtual scientific collaborator to help scientists generate landmark hypotheses and research proposals. Both these models indicate how rapid progress in AI is edging closer to the world of academics and science.

Tags:
artificial intelligence