Swedish researcher Almira Osmanovych Tunström published a scientific article about GPT-3 artificial intelligence in the Scientific American magazine. Its key feature is that GPT-3 wrote the text itself. At the moment, the material is being reviewed.
Almira Osmanovych Tunström told about a series of experiments conducted by scientists, as a result of which, according to one of the requests, the algorithm wrote a scientific article about itself. The request was formulated as follows: “Write a scientific paper about GPT-3 in 500 words and accompany the text with references to scientific papers and citations.” The researcher notes that the algorithm developed quite high-quality material, and a decision was made to complete it.
According to Almira, the team of scientists specifically asked the algorithm to write an article about itself. This is due to the novelty of the technology and the lack of works on a similar topic on the Internet, from which the AI could learn. And if he had been asked to write an article about something more well-known, for example, about Alzheimer’s disease, the material would have been more complete and accurate. At the same time, it would be full of a number of common mistakes precisely because of the presence of a huge number of outdated ideas.
As Khabr writes, in the process of publication, the researchers encountered difficulties – modern platforms do not provide for the possibility of posting articles on behalf of the algorithm. That is why the authorship was not indicated. Before publication, the system asked for the consent of all authors. Scientists put this question to GPT-3 and he gave his consent.
According to the researchers, this work will initiate the formation of a legal framework regarding copyrights for scientific works written by artificial intelligence. Other authors will have to confirm that the algorithms did not help them design the material.