Empirical Research Assistance (ERA): From Nature publication to catalyzing Computational Discovery

2026-05-19 · Source: The latest research from Google · Field: Science & Research — Research Methodology & Innovation, Mathematics & Computational Sciences · Depth: Expert, medium

Summary

Published on May 19, 2026, Empirical Research Assistance (ERA) is a Google-developed AI tool designed for expert-level scientific coding, detailed in a Nature publication titled "AI system designed to help scientists write expert-level empirical software". Utilizing Gemini, ERA writes and optimizes scientific code, streamlining the iterative testing and refinement of computational experiments. It employs a tree search approach to explore thousands of options, searching literature, writing code, combining techniques, and evaluating results. ERA has demonstrated expert-level performance across diverse benchmarks, including genomics, public health, satellite imagery analysis, and time-series forecasting. Google Research scientists have applied ERA to eight open scientific problems, achieving significant results such as top rankings in CDC epidemiological forecasting for flu, COVID-19, and RSV, more accurate California seasonal runoff predictions, high-resolution atmospheric CO2 mapping, optimized 3D solar energy capture, and improved retail sales forecasting. The Computational Discovery prototype, built with ERA, is now accessible via a trusted tester program in Google Labs and Gemini for Science.

Key takeaway

For research scientists and machine learning engineers focused on accelerating computational discovery, ERA provides a robust AI assistant to streamline code development and optimization. You should consider integrating ERA into your workflow to reduce the time spent on iterative testing and refinement of experiments. This tool can enhance model accuracy across diverse scientific domains, from epidemiological forecasting to environmental monitoring. Explore the Computational Discovery prototype via Google Labs or Gemini for Science to leverage its capabilities.

Key insights

AI tools like ERA can achieve expert-level scientific coding, accelerating computational discovery across diverse domains.

Principles

• AI can democratize access to expert-level computational modeling.
• Tree search optimization enhances AI's ability to refine scientific code.
• AI-driven code generation improves accuracy in forecasting and environmental mapping.

Method

ERA employs a tree search to optimize code against a scientific goal, iteratively searching literature, writing code, exploring solutions, combining techniques, and evaluating results.

In practice

• Utilize ERA for high-accuracy epidemiological and retail sales forecasting.
• Apply ERA to create detailed atmospheric CO2 concentration maps.
• Explore ERA for optimizing complex designs, such as 3D solar energy capture.

Topics

• Empirical Research Assistance
• Scientific AI
• Computational Discovery
• Gemini for Science
• Epidemiological Forecasting
• Environmental Modeling
• Code Optimization

Code references

• google-research/era
• asonabend/DeepXCO2

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Four ways Google Research scientists have been using Empirical Research Assistance — ERA accelerates scientific discovery by generating expert-level empirical software for diverse, complex real-world problems.
• Google unveils ERA AI to accelerate expert-level scientific research — Google's ERA AI, powered by Gemini, accelerates scientific discovery through automated code generation and experimental refinement.
• Gemini for Science: AI experiments and tools for a new era of discovery — Gemini for Science leverages AI agents to automate complex research tasks, significantly accelerating scientific discovery and exploration.
• Column: IS IT REALLY SO? – Why I Love Artifical Intelligence (AI) - The Village Reporter — AI can be a valuable tool for researchers to refine arguments and identify inconsistencies, even when challenging established scientific paradigms.
• Automated Scientific Discovery of Mind and Behavior — AI can overcome human cognitive limitations to accelerate and integrate scientific discovery in complex fields like cognitive science.
• Quoting Karel D'Oosterlinck — AI can automate research due diligence and experiment setup, saving significant manual effort.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The latest research from Google.