Industry

AI-generated text

OpenAI study finds scientists expect AI to speed up discoveries

OpenAI analyzed more than 580,000 scientific messages from 1,819 researchers using ChatGPT and surveyed 251 participants.

OpenAI study finds scientists expect AI to speed up discoveries

OpenAI’s Economic Research team examined how scientists use ChatGPT in their work and what they expect artificial intelligence to mean for scientific discovery. The study covered more than 580,000 scientific messages from 1,819 participants in the summer pilot of the ChatGPT for Academic Researchers program. The team also surveyed 251 participants.

Researchers used ChatGPT for tasks ranging from reviewing literature to analyzing data and evidence and developing ideas. They also used it to write code and communicate their findings. Survey respondents estimated that completing the same work to the same standard without AI would have taken 1.9 times as long.

Expectations of faster discovery

The survey also asked scientists in different fields to imagine a discovery that would take 10 years without AI and estimate how long it would take with AI. Of the 250 people who answered, 98% expected AI to accelerate discovery in their field. Respondents could select more than one field, so the field groups could overlap.

OpenAI also examined how scientists used the system for ideation: whether they brought specific questions, asked it to refine existing ideas or requested suggestions for new research questions. It measured how often researchers followed up to clarify, check or revise responses. For the activity analysis, the team calculated each scientist’s proportions first and then averaged them, giving every scientist equal weight.

Scientists who expected faster discovery were also asked to explain why. The researchers identified 367 reasons in 224 written responses. According to OpenAI, the answers pointed not only to saving time but also to the possibility of finding more promising questions and pursuing a wider range of research directions.

Privacy and further research

OpenAI said it does not use participants’ data from the ChatGPT for Academic Researchers program to train its models by default. To protect privacy, it used automated classifiers to sort messages by research activity. The research team could see the categories but did not read the underlying messages or ChatGPT’s responses.

OpenAI emphasized that further research is needed to determine whether those expectations translate into faster discoveries. Progress also depends on the funding, staff and experimental capacity needed to test ideas.

Alongside the study, the company announced an expansion of ChatGPT for Academic Researchers, providing free access to more than 200,000 researchers worldwide.