AI System Generates Expert-Level Scientific Software for Research Tasks

Google DeepMind researchers introduced Empirical Research Assistance, an AI system that generates expert-level scientific software to support computational experiments. The system combines a large language model with tree search to systematically improve quality metrics and explore solution spaces.

ERA achieved expert-level performance on several research tasks. In bioinformatics, the system discovered 40 novel methods for single-cell data analysis that outperformed top human-developed methods on a public leaderboard. In epidemiology, ERA produced 14 models that outperformed the CDC ensemble and all other individual models for forecasting COVID-19 hospitalizations.

The system also generated expert-level software for geospatial analysis, neural activity prediction in zebrafish, and numerical solution of integrals. ERA created a novel rule-based construction for time series forecasting. The research was published in Nature on May 19, 2026.

The paper lists 37 authors from Google DeepMind, Google Research, Google Platforms and Devices, MIT, Harvard University, McGill University, and Caltech. "An AI system to help scientists write expert-level empirical software" is the title of the Nature paper.

The work addresses bottlenecks in scientific discovery caused by slow manual creation of software for computational experiments.