level: research
google research has introduced empirical research assistance (era), an ai tool that writes and refines scientific code. described in a nature paper, era uses a tree search approach to explore thousands of coding solutions for a given problem and success metric. it searches literature, writes code, combines techniques, and evaluates results. benchmarks across genomics, public health, satellite imagery, neuroscience, time-series forecasting, and mathematics show era achieves expert-level performance, potentially democratizing computational modeling.
over the past six months, google scientists and collaborators applied era to open scientific questions. they developed a model for epidemiological forecasting that predicts u.s. hospital admissions for flu, covid-19, and rsv up to four weeks ahead, ranking at or near the top of cdc leaderboards. era also created a seasonal runoff forecast for california's snow-fed river basins, outperforming the state's official outlook. other projects include mapping atmospheric co2 concentration every 10 minutes from weather satellite data, optimizing 3d solar panel designs, and improving retail sales forecasts using economic indicators and google trends data.
era is one of the systems behind computational discovery, a new experimental tool now available through a trusted tester program in google labs. built with era and alphaevolve, it aims to accelerate scientific discovery. alongside hypothesis generation and literature insights, these tools support different stages of the scientific method. google is gradually opening access, inviting scientists to register interest at labs.google/science.
why it matters: era can automate time-consuming coding tasks in research, enabling faster experimentation and potentially making expert-level computational science more accessible across disciplines.