ChatGPT in Science: Expected Gains Still to be Confirmed

OpenAI examined the usage of ChatGPT among 1,819 scientists participating in its academic pilot program. The researchers surveyed anticipate faster discoveries, although this acceleration has yet to be demonstrated.
Researchers Hope to Accelerate Their Discoveries
Verifying that scientists' expectations indeed lead to faster discoveries will require further investigation, warns OpenAI. The company also highlights the role of funding, staffing, and experimental equipment necessary to test ideas.
Among the participants who responded regarding discovery timelines, 98% believe that AI will shorten these timelines in their field. This question, which received 250 responses, asked participants to estimate how long a discovery that would typically take 10 years without AI would take with it. Multiple disciplines could be selected, so the groups overlap, and the rounded percentages do not necessarily total 100%.
When invited to explain their optimism, the scientists could cite several reasons. OpenAI identified 367 reasons across 224 written responses: 52% pertain to faster execution of work, 23% to more promising scientific questions, 14% to a broader range of questions, and 11% to other reasons.
Regarding the work accomplished, the surveyed participants estimate that an equivalent achievement, of the same quality, would have taken 1.9 times longer without AI. This is a participant assessment, distinct from the verification expected on the pace of discoveries.
ChatGPT Used for Idea Development by 68% of Scientists
According to OpenAI, 68% of the scientists studied have used ChatGPT to develop ideas. Among the messages dedicated to this use, 36% sought to refine an idea provided by the researcher, 35% submitted a specific question, and 23% requested a suggestion for a scientific question. The remaining 6% did not fall into any of these categories.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
The observed practices also include literature reviews, evidence examination, programming, and communication of results. The team also measured subsequent exchanges aimed at clarifying, verifying, or modifying the chatbot's responses, depending on the scientific activity. The chosen indicator is the average proportion of requests followed by such exchanges for each scientist.
OpenAI Expands Free Access to Over 200,000 Researchers
OpenAI is simultaneously expanding its academic program, providing free access to over 200,000 researchers worldwide. The company claims it aims to support their AI research and links this goal, along with support for advancements beneficial to society, to its stated mission of ensuring humanity benefits from AGI.
To analyze the usage of the pilot program, OpenAI's economic research team reviewed over 580,000 scientific messages and conducted interviews with 251 participants. OpenAI states that, by default, the information provided by participants in the academic program is not used to train its models.
The company indicates it has protected confidentiality by using automated classifiers: these categorized messages by activities, including literature review, programming, or data analysis. The team could consult these categories but did not read the messages themselves or the chatbot's responses, assures OpenAI.
For the distribution of activities, the shares were calculated individually and then averaged, giving equal weight to each scientist. The percentages are rounded. The activities grouped in the residual category represent less than 1%.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.