Research on “AI Generation” was presented to this year’s ICLR, an AI-focused academic conference, which has caused controversy.
At least three AI labs – Sakana,,,,, exploreand car – Research that claims to have used AI to generate accepted ICLR workshops. At conferences like ICLR, seminar organizers usually review publishing research on the seminar track of the conference.
Sakana informed ICLR leaders before submitting an AI-generated paper and obtaining consent from peer reviewers. An ICLR spokesperson confirmed to TechCrunch that the other two laboratories (Intology and Autoscience) did not.
Several AI scholars entered social media to criticize primitive science and automatic science as an option for the scientific peer review process.
“All of these AI scientists’ papers will be peer-reviewed sites as their human buoyancy, but no one agrees to provide this free labor,” wrote Prithviraj Ammanabrolu, an assistant computer science professor at UC San Diego. A X post. “It made me lose respect for everyone involved, no matter how impressive the system is.
As critics point out, peer review is a time-consuming, labor-intensive and mostly volunteer ordeal. According to a recent natural survey40% of scholars spent two to four hours reviewing a study. This work has been upgrading. The number of papers submitted to the largest AI conference (Neurips) increased to 17,491 last year, a 41% increase from 12,345 in 2023.
There is already a problem of replication of AI generation in the academic community. An analysis Established Of the papers submitted to the AI conference in 2023, 6.5% to 16.9% may contain synthetic text. However, AI companies that use peer-reviewed effective benchmarks and promote their technology are relatively new events.
“[Intology’s] The paper received consistent positive reviews,” Intology at Post on X Touts its ICLR results. In the same article, the company went on to claim that the workshop reviewers praised one of its AI-generated research “smart ideas”[s]. ”
The academic community is not friendly to this.
Ashwinee Panda, a postdoctoral researcher at the University of Maryland, Say in X post Submitting AI-generated papers without granting workshop organizers the right to reject them, suggesting a “lack of respect for the time of human reviewers”.
Panda added: “Sakana […] I think submitting AI papers to the site without contacting [reviewers] not good. ”
Not all is gone, and many researchers are skeptical that AI-generated papers deserve peer review.
Sakana itself admit Its AI made an “embarrassing” citation error, and only one of the three AI-generated papers the company chose to submit would meet the criteria accepted by the conference. The company said Sakana withdraws its ICLR paper before it can be out of transparency and respect for the interests of the ICLR convention.
Alexander Doria, co-founder of AI startup Pleias, said the secretly synthesized ICLR submission raft pointed out that a “regulated company/public agency” is needed to conduct research evaluations of “high-quality” AI generation at a certain price.
“evals [should be] The time completed by the researchers fully compensates their time. ” series Posts On X. “There is no free outsourcing in the academic world [AI] EVALS. ”
