AI Writes Better Science Code Than Human Experts

Scientists just got a turbo boost for their research, and it could change how we solve everything from disease outbreaks to climate change.

A team from Harvard and Google DeepMind created an AI system called Empirical Research Assistance (ERA) that writes scientific software better than human programmers. Published in Nature, the breakthrough automates work that typically takes researchers months or even years to complete.

Here's what makes this special. Modern science runs on custom software built for specific tasks like predicting weather patterns or modeling how diseases spread. Creating these programs requires endless testing and refinement by human experts, which becomes a massive bottleneck for research progress.

ERA changes that equation entirely. The system uses Google's Gemini language model combined with a search strategy similar to the one that powers AlphaGo, the famous game-playing AI. Starting with basic code aimed at a problem, ERA proposes thousands of modifications and improvements, testing which combinations work best far faster than any human could.

The results speak for themselves. When tasked with predicting COVID-19 hospitalizations, ERA generated 14 models that beat the best systems used by the CDC during the pandemic. In another test, it discovered four new methods for analyzing genetic data that outperformed top approaches designed by human scientists.

Harvard Ph.D. student Qian-Ze Zhu put ERA to work modeling the activity of over 70,000 neurons in a zebrafish brain. What would have taken him weeks of learning new software packages, ERA completed automatically in hours. "This new system is going to accelerate scientific discovery by allowing you to explore a lot of ideas at the same time," Zhu explained.

The AI doesn't work in isolation. Researchers can guide it by feeding in ideas from scientific papers or textbooks, and ERA incorporates those concepts into improved versions of the code. This ability to combine research ideas helps the system find creative solutions that human researchers might never think to test.

Michael Brenner, the Harvard professor who co-led the project, calls this a "needle-in-a-haystack" capability. The system explores combinations of approaches at a scale impossible for individual scientists.

The Ripple Effect

This technology could democratize cutting-edge research tools across every scientific field. Smaller labs without armies of programmers could suddenly access the same quality software as major institutions. Climate researchers could test more models for predicting environmental changes. Medical researchers could accelerate drug discovery and disease tracking.

By compressing months of software development into days or hours, ERA frees scientists to focus on the creative work only humans can do: asking the right questions, designing meaningful experiments, and deciding which challenges matter most for society.

The system already supports work that led to three recent Nobel Prizes in chemistry. Now that power becomes available to researchers everywhere, potentially accelerating the pace of discovery across every scientific domain.

Scientific breakthroughs just got a whole lot faster.