A new study reveals that AI chatbots can write complex code to predict preterm births in a matter of minutes — a task that normally takes experienced programmers days or even months to complete. This could mean the wait for critical diagnostic tests and medical treatments may be drastically reduced.
Researchers at UC San Francisco and Wayne State University pitted human scientists against AI to see who could best predict preterm births using data from more than 1,000 pregnant women.
The findings, published in Cell Reports Medicine, showed that generative AI tools could perform “orders of magnitude faster” and sometimes better than human computer science teams who had spent months poring over the same data.
In one example, a junior research duo consisting of a master’s student and a high school student used AI assistance to generate working computer code in minutes. According to the researchers, this same task would have taken experienced programmers anywhere from a few hours to several days.
“These AI tools could relieve one of the biggest bottlenecks in data science: building our analysis pipelines,” said Marina Sirota, a professor of Pediatrics at UCSF and co-senior author of the study. “The speed-up couldn’t come sooner for patients who need help now”.
Accelerating the pace
The study highlights a massive shift in how medical research could be conducted in the future. The researchers utilised data from a crowdsourced competition called DREAM, in which over 100 groups worldwide spent three months developing machine learning algorithms to identify signs of preterm birth in vaginal microbiome data. Compiling and publishing these human-led results took nearly two years.
To test the AI, researchers instructed eight chatbots to build similar pregnancy assessment algorithms using the same data, with no human input beyond a carefully phrased, highly specialised natural-language prompt.
While only four of the eight AI tools produced usable code, those that did created prediction models that performed as well as — and sometimes outperformed — the human teams. The entire generative AI project, from inception to paper submission, took just six months.
Saving thousands of lives
The immediate real-world application of this research is profound. Preterm birth is the leading cause of newborn death, with around 1,000 babies born too soon in the US every day. By drastically accelerating the path from data to discovery, researchers hope to develop more reliable diagnostic tests that can warn mothers and doctors much earlier in a pregnancy.
While the researchers cautioned that scientists must remain on guard for misleading results, the technology’s true power lies in freeing up experts to tackle the actual disease, rather than the computer code.
“Thanks to generative AI, researchers with a limited background in data science won’t always need to form wide collaborations or spend hours debugging code,” said co-senior author Adi L. Tarca of Wayne State University. “They can focus on answering the right biomedical questions”.
