Google has launched its AI tool Empirical Research Assistance (ERA) aimed at improving scientific coding. Detailed in an article published in the journal Nature, ERA is integral to the Computational Discovery prototype, which is available through a trusted tester program in Google Labs.
ERA employs Gemini technology to enhance the efficiency of testing and refining computational experiments. This tool allows users to search scientific literature, write code, explore solutions, and combine techniques. The approach involves a tree search method that optimizes generated code toward achieving specified goals.
In extensive testing, ERA demonstrated expert-level performance across a range of benchmark problems, including genomics, public health, and neuroscience. Google Research has collaborated with scientists over the last six months and presented four projects utilizing ERA to tackle significant scientific challenges.
As of late April 2026, Google has produced eight manuscripts highlighting ERA’s application to diverse scientific issues, five of which were newly released. On the same day as the publication, Google announced gradual access to the Computational Discovery tool, developed using ERA and AlphaEvolve.
New experimental capabilities introduced alongside ERA include Hypothesis Generation and Literature Insights, which support various stages of the scientific method. Interested individuals can register at labs.google/science.
The development of ERA involved collaboration from numerous scientists, with key contributions from John Platt and Michael Brenner. The algorithm’s design was led by Eser Aygun, Gheorghe Comanici, and Shibl Mourad, among others.
