A Swedish researcher, Almira Osmanovic Thunström, ordered the AI bot GPT-3 to write a 500-word academic thesis about itself with scientific citations and references inside the text. Thunström commented that the bot generated a "fairly good" research introduction. GPT-3 is a deep-learning algorithm that analyzes a vast stream of information to create human-like text.
(Photo : Scott Graham) Person holding pencil near laptop
AI Bot Writes an Academic Paper
In her post on the Scientific American website, Thunström wrote that it was not her first time experimenting with AI (artificial intelligence) or GPT-3. Yet, she stood in awe when GPT-3 started to generate text after receiving instruction.
"There I was, staring at the screen in amazement. The algorithm was writing an academic paper about itself," Thunström said.
The researcher did not have any high expectations as she provided vague instructions.
GPT-3 was asked to write a paper about itself because there was not enough research about the AI bot. The researcher also would not necessarily be spreading AI-generated misinformation in the effort to publish if it made mistakes. Instead, the error would be a result of the experimental command to the paper. The researcher wanted to prove that although GPT-3 made mistakes when writing about itself, it doesn't mean it can't still write about itself.
Thunström shared the process she went through when she reached the legal section. There were several questions to be answered, including the author's consent to publishing the manuscript. To ensure that the AI bot allowed the publication, the researcher asked GPT-3 via a prompt if it agreed to be the manuscript's first author, together with Thunström and her advisor, Steinn Steingrimsson. The AI bot answered yes.
GPT-3's manuscript is under review by an editor. It is available on HAL Open Science's website, a pre-print server. The researcher thought the prolonged assessment of the manuscript was due to its authorship, which was a non-human author. Yet, when it is available for publication, it may open a new way of producing academic papers. They could also stop basing the number of articles they publish on the financial security and grants available. There is also a possibility that they could produce one study per day.
Although the result is promising, there are still questions to answer when the document gets approved. Questions such as, "Will the demand to provide evidence that authors have not used the assistance of GPT-3 or another algorithm be necessary?" and, "Do they have to co-author it if they have?"
The researcher said that the attempts to complete the paper and the process of getting it published opened up a series of legal and ethical questions about publishing. She wrote that philosophical arguments were raised about nonhuman authorship.
Future academic publishing may need to make room for AI-driven articles. The value of their publication record may shift. If anything nonsentient can claim credit for some of a human researcher's work. The author said that the research opened the gate, hoping it was not Pandora's box.
