AI-written academic paper submitted to peer review

A Swedish researcher tasked an AI algorithm to write an academic paper about itself. The paper is now undergoing a peer-review process.

Almira Osmanovic Thunstrom has said she “stood in awe” as OpenAI’s artificial intelligence algorithm, GPT-3, started generating a text for a 500-word thesis about itself, complete with scientific references and citations.

“It looked like any other introduction to a fairly good scientific publication,” she said in an editorial piece published by Scientific American. Thunstrom then asked her adviser at the University of Gothenburg, Steinn Steingrimsson, whether she should take the experiment further and try to complete and submit the paper to a peer-reviewed journal.

Steingrimsson jumped on board and now serves as one of the paper’s three authors, alongside Thunstrom and GPT-3. With minimal command provided by its human co-authors, the algorithm finished a paper in just two hours. While this seemed straightforward, the legal and ethical questions that followed were not.

“We had to invent a whole new way of presenting a paper that we technically did not write,” Thunstrom said. This included asking the algorithm whether it consented to be published – it did, and whether it had any conflict of interest as a researcher – it did not.

Thunstrom said she had both “panicked for a second” in anticipation of the answer, as she could not have proceeded further had the AI said it did not consent to be published, and “laughed” at the idea of having to treat GPT-3 as a sentient being, even though it was not.

“Beyond the details of authorship, the existence of such an article throws the notion of a traditional linearity of a scientific paper right out the window,” Thunstrom said, adding that, ultimately, it all comes down to whether AI will be treated as a partner or a tool in the future.

“All we know is, we opened a gate. We just hope we didn’t open a Pandora’s box,” Thunstrom said after submitting the thesis to an academic journal she did not name, where it was assigned an editor. It was also published by a French pre-print server HAL.