(Washington) A debate is tearing the world of research apart: should the use of software like ChatGPT be allowed to facilitate the writing of scientific articles? Thursday at a press conference, the editor of the journal ScienceHolden Thorp, crossed swords with the head of artificial intelligence (AI) ethics at IBM, Francesca Rossi.
“We took the decision at the end of January to prohibit the use of ChatGPT for the writing of studies that we publish,” recalled Mr. Thorp. 20 years ago, we didn’t react when Photoshop made it possible to edit photos in a realistic way. As a result, we now spend a lot of time arbitrating between researchers who accuse each other of image manipulation. We decided to be more careful with the AI. »
For his part, M.me Rossi, who heads the Association for the Advancement of Artificial Intelligence, defended his decision to allow journals in that association to accept the use of ChatGPT. “It better be transparent,” said M.me Rossi. Anyway, soon it will be difficult to detect. We have drawn the line to include ChatGPT as a co-author. »
PHOTO FROM IBM SITE
The director of the Association for the Advancement of Artificial Intelligence, Francesca Rossi
Because Nature revealed at the end of January that four studies published that month named the software as a co-author: two American studies on the use of ChatGPT in medical and nursing schools, a Hong Kong literature review on the use of an immunosuppressant for increasing human longevity, and a study by neurobiologists on ChatGPT’s ability to write about itself.
Sources
For the moment, ChatGPT does not allow to produce a scientific publication because the software does not give its sources. But Microsoft is working on a version that will provide them, according to Mme Rossi.
On the sidelines of the press conference, The Press asked Mr. Thorp if a source version of ChatGPT could produce the literature review that usually forms the first section of a study reporting experimental results. “I don’t see what would prevent it,” replied Mr. Thorp. But it’s very undesirable, in my opinion, because we don’t know what the biases of the software will be. It is important to me that the human remains at the heart of the literature review. »
Mr. Thorp thinks ChatGPT could also handle the last two sections of a typical scientific paper, the discussion of the results and the conclusion summarizing the main lessons from the experimental data.
“That’s exactly what the software was designed to do,” Thorp said. But the human must be the one who interprets the results. »
PHOTO FROM SCIENCE SITE
The editor of the journal ScienceHolden Thorp
Would they accept a researcher asking ChatGPT to improve an already written post? “If we open the door to that, we will no longer have any control,” he replies. Science is based on trust [honor system]. If someone is found to have used ChatGPT, it would be a case of scientific fraud, such as Photoshop’s significant manipulation of images. »
Francesca Rossi argued during the press conference that researchers whose first language is not English could benefit from ChatGPT. “It could remove some barriers for researchers from diverse backgrounds,” she said.
To test ChatGPT’s capabilities, Mr. Thorp used it to write an email to the software’s inventors about his concern for scientific research. “It really is a fantastic tool,” said Mr. Thorp, who did not get a response.
bananarama
One thing is certain, scholarly journals are full of essays on the subject. And there are also studies created with ChatGPT without him being named as a co-author. For example, Irish business professors asked ChatGPT to write an empirical study project on cryptocurrencies. ChatGPT suggested studying the link between investor enthusiasm for cryptocurrencies and stock market performance.
The Irish study, published in the Finance Research Letters, concludes that ChatGPT has strengths and weaknesses with respect to the different stages of scientific research. She was titled The Bananarama hypothesisin honor of a song by the girl group from the 1980s. The title of this song, It Ain’t What You Do (It’s the Way That You Do It)a cover of a jazz classic, echoes the fact that a tool can be used in a positive or negative way depending on the circumstances, believe the Irish authors.
Other congress news
Light nets
Attaching lights to fishing nets can cut catches of sharks and other rays in half, according to a study by Arizona State University. Data from the catch of over 13,000 fish off Baja California, Mexico found that orange lights are the most effective, and do not reduce the catch of commercial fish. The lights on the nets had also been considered to reduce incidental catches of turtles, but this did not work, either because the latter are too slow or because they are myopic.
Rich, poor and windmills
The richer a person is, the less favorable they are to solar and wind farms on land, shows a study by Carnegie Mellon University. The wealthy prefer to reduce greenhouse gas emissions by storing CO2 emitted by factories and by means of offshore wind turbines, solutions that are not very popular with the poorest. The study stems from a survey of 500 Americans.
Vasectomy instead of abortion
The end of abortion protections by the U.S. Supreme Court last year led to a jump in vasectomy claims, according to researchers at Arizona State University. This increase ended a decline in the popularity of vasectomy in the United States that had begun at the turn of the millennium, the causes of which are the subject of much speculation. Vasectomy is much less popular in the United States than in Canada, 20% versus 10%, according to a 2015 UN study, because forced vasectomies performed among African Americans have given this method of birth control a bad reputation.
Learn more
-
- 2 months
- Time it took for ChatGPT to reach 100 million users
Source : Time
- 9 months
- Time it took for TikTok to reach 100 million users, in 2017
Source : Time
-
- 30 months
- Time it took for Instagram to reach 100 million users, in 2010
Source : Time