AI chatbots have outpaced human research teams in analysing complex medical datasets, delivering results in months rather than years and, in some cases, surpassing expert-built models.
Scientists at the University of California, San Francisco and Wayne State University staged a direct comparison. They asked human-only teams and AI-assisted teams to solve the same problem: predict preterm birth using microbiome and clinical data from more than 1,000 pregnant women.
The outcome challenges assumptions about who holds the analytical edge.
Junior Researchers, Senior-Level Output
A UCSF master’s student, Reuben Sarwal, partnered with high school student Victor Tarca. With generative AI support, they built functional prediction models in minutes. Experienced programmers often require hours or days to write comparable analytical code.
Generative AI’s advantage lies in translating highly technical prompts into working code. Not every tool delivered. Only four of the eight tested systems generated usable outputs. Yet those that succeeded operated without large expert teams guiding every step.
Freed from time-intensive coding, the junior researchers ran experiments, validated results and submitted their findings to a scientific journal within months.
“These AI tools could relieve one of the biggest bottlenecks in data science: building our analysis pipelines,” said Marina Sirota, PhD, professor of Pediatrics and interim director of the Bakar Computational Health Sciences Institute at UCSF. “The speed-up couldn’t come sooner for patients who need help now.”
For healthcare leaders, the implication is clear. If AI compresses development cycles, diagnostic tools could reach clinics faster. In a field where delays cost lives, speed is not cosmetic. It is strategic.
Why Preterm Birth Demands Faster Answers
Preterm birth remains the leading cause of newborn death and a major driver of long-term motor and cognitive disabilities. In the United States, roughly 1,000 babies are born prematurely each day.
Despite decades of research, scientists still struggle to pinpoint its triggers.
Sirota’s team aggregated microbiome data from approximately 1,200 pregnant women across nine studies, tracking outcomes through delivery. The scale created opportunity and complexity in equal measure.
“This kind of work is only possible with open data sharing, pooling the experiences of many women and the expertise of many researchers,” said Tomiko T. Oskotsky MD, co-director of the March of Dimes Preterm Birth Data Repository and associate professor at UCSF.
The dataset’s volume made analysis slow and technically demanding. To accelerate progress, the researchers turned to DREAM, a global competition inviting teams to build machine learning models capable of identifying patterns linked to preterm birth.
More than 100 groups participated. Most completed their models within three months. Publishing and consolidating the findings took nearly two years.
That lag reflects a familiar challenge across industries. Teams gather vast data, then stall while building and debugging analytical pipelines. What if that bottleneck disappears?
Putting AI To The Test
To find out, Sirota partnered with Adi L. Tarca, PhD, professor at Wayne State University, who had co-led related DREAM challenges focused on estimating pregnancy stage.
The joint team tasked eight AI systems with independently building predictive algorithms from the same datasets used in the original competitions. Researchers provided carefully structured natural language prompts. The systems received no human-written code.
The objective mirrored the competition:
• Analyse vaginal microbiome data for signals of preterm birth
• Use blood and placental samples to estimate gestational age
Pregnancy dating shapes medical decisions. When clinicians misjudge gestational age, they may mistime interventions or labour preparation. Accuracy influences outcomes.
After executing the AI-generated code, researchers found that four of the eight systems produced models that matched the performance of top DREAM teams. In some cases, the AI models performed better.
The entire AI-driven project, from concept to journal submission, concluded in six months.
Human Oversight Still Matters
The researchers emphasise that generative AI can produce flawed or misleading outputs. It does not replace scientific judgement.
However, it changes how researchers allocate time.
“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,” Tarca said. “They can focus on answering the right biomedical questions.”
That shift mirrors changes in other sectors. When automation reduced manual spreadsheet work in finance, analysts redirected effort towards strategy and risk modelling. Here, scientists could spend less time troubleshooting syntax and more time interpreting biological meaning.
The broader question sits beyond preterm birth. If AI can independently construct high-performing biomedical models, how will research teams evolve? Will institutions prioritise prompt engineering alongside traditional coding? Will grant timelines compress?
Generative AI has not replaced expertise. It has rebalanced it. Researchers still define hypotheses, validate outputs and interpret findings. Yet the machinery of analysis now moves at digital speed.
For families confronting the risks of premature birth, faster answers could translate into earlier interventions. For research institutions, the lesson is sharper: the competitive advantage may belong to teams that learn how to work with AI, not around it.
Author: George Nathan Dulnuan