OpenAI study points to faster scientific research
Nearly 98 per cent of scientists surveyed in an OpenAI study expect AI to accelerate discoveries in their fields. More than one in five believe research that would typically take a decade could potentially be completed within three years with AI assistance. Researchers estimated that the technology helps them complete tasks at comparable quality levels in nearly half the time.
The findings draw on 580,186 scientific messages exchanged by 1,819 researchers and a survey of 251 scientists enrolled in OpenAI’s ChatGPT for Academic Researchers programme. Applications included literature reviews, writing, evidence analysis, software development and generating research ideas. Reading and writing accounted for more than half of the messages analysed. Scientists also used AI to identify research questions and refine hypotheses, suggesting potential to explore a broader range of investigations.
Human oversight remained important: researchers sought clarification, verification or revisions following around 32 per cent of AI-generated responses. Resource constraints also persisted. Around 51.8 per cent of respondents identified limited time or personnel as a major bottleneck, while 43.8 per cent cited inadequate funding. Only 6.4 per cent reported no significant research bottlenecks. OpenAI cautioned that translating efficiency gains into faster breakthroughs also depends on sufficient funding, research personnel and the capacity to conduct experiments.