Google Research scientists Lizzie Dorfman and Michael Brenner have unveiled Empirical Research Assistance (ERA), an artificial intelligence tool designed to accelerate scientific discovery by streamlining computational experiments. Published today in the journal Nature, ERA leverages Google’s Gemini AI to write and refine scientific code, addressing a significant bottleneck in research workflows. This development is part of Google’s broader initiative to make advanced AI tools accessible, with ERA also contributing to the new Computational Discovery prototype now entering a trusted tester program via Google Labs. The immediate availability of this technology through Gemini for Science marks a pivotal moment for democratizing expert-level computational modeling and expanding research capabilities globally.

KEY DEVELOPMENTS

  • Empirical Research Assistance (ERA), an AI tool developed by Google Research, is now available, described in a paper published today in Nature.
  • ERA utilizes Gemini AI to generate and optimize scientific code, significantly reducing the time required for iterative computational experiment testing.
  • The tool has demonstrated expert-level performance across diverse scientific benchmarks, including genomics, public health, and neuroscience.
  • ERA played a foundational role in building the Computational Discovery prototype, which is launching today through a trusted tester program in Google Labs.
  • Google scientists and collaborators have already applied ERA to eight scientific problems, with five new manuscripts released, showcasing its immediate impact in areas like epidemiological forecasting.

WHAT HAPPENED

Today, Google Research announced the formal release of Empirical Research Assistance (ERA), an AI-powered system aimed at enhancing scientific coding efficiency. Developed by Lizzie Dorfman, Product Manager, and Michael Brenner, Research Scientist, ERA is detailed in the Nature publication titled “AI system designed to help scientists write expert-level empirical software.” This AI tool is specifically engineered to tackle the time-intensive process of iteratively testing and refining computational experiments, a common challenge in scientific research. ERA’s capabilities were first shared in a preprint last fall, outlining its ability to search literature, write code, explore solutions, combine techniques, and evaluate results based on a given scientific problem and success metric.

As part of wider science announcements at Google I/O, the company is making ERA accessible to scientists globally. The system employs a tree search approach to consider thousands of options, optimizing its output code against defined goals. ERA’s effectiveness has been rigorously tested across various disciplines, including genomics, public health, satellite imagery analysis, neuroscience prediction, time-series forecasting, and mathematics, consistently achieving expert-level performance. Furthermore, ERA is one of the core systems underpinning Computational Discovery, an experimental tool that is beginning to roll out more broadly today through Gemini for Science, marking a significant step in AI-assisted scientific research.

WHY IT MATTERS

The introduction of Empirical Research Assistance holds substantial implications for the scientific community, directly addressing one of the most resource-intensive aspects of modern research: computational experimentation. By automating and optimizing code generation, ERA can dramatically reduce the time scientists spend on iterative testing, allowing them to focus on higher-level problem-solving and analysis. This not only accelerates the pace of discovery but also lowers the barrier to entry for complex computational modeling, potentially enabling a broader range of researchers to engage with advanced scientific questions. The integration of ERA into Computational Discovery and its availability through Gemini for Science signals a strategic move by Google to embed AI deeply into the scientific workflow, promising to reshape how research is conducted globally.

8Manuscripts applying ERA to scientific problems

INDUSTRY IMPACT

ERA’s impact extends across numerous scientific domains, offering a versatile tool for researchers. Its proven ability to achieve expert-level performance in areas like genomics, public health, and neuroscience prediction suggests a broad applicability that could affect various industries. For instance, in public health, ERA has already been used to create epidemiological forecasting models that predict U.S. hospital admissions for flu, COVID-19, and RSV up to four weeks in advance. These forecasts have consistently ranked among the top performers on CDC leaderboards, demonstrating immediate public benefit and replicable techniques for other diseases and countries. Similarly, its application in forecasting seasonal runoff for California’s snow-fed river basins highlights its potential for environmental management and agricultural planning.

ANALYSIS

Google’s launch of Empirical Research Assistance represents a significant advancement in the application of AI to scientific endeavors. The strategic decision to make ERA publicly accessible, alongside its integration into the new Computational Discovery platform, underscores a commitment to democratizing advanced research tools. The rigorous testing across diverse benchmarks and the successful application to open scientific questions, as evidenced by eight manuscripts, lend substantial credibility to ERA’s capabilities. This initiative positions AI not merely as an assistive technology but as a core component in the engine of scientific progress.

The potential for ERA to expand the capabilities of existing experts while also democratizing access to expert-level computational modeling is particularly noteworthy. It suggests a future where the bottleneck of complex coding and iterative experimentation is substantially alleviated, allowing researchers worldwide to pursue more ambitious and intricate scientific inquiries. The demonstrated success in areas like epidemiological forecasting, where ERA’s models consistently outperform others, highlights the tangible, real-world benefits this technology can deliver, reinforcing the argument for AI’s profound role in addressing pressing global challenges.

FUTURE IMPLICATIONS

  • Near-term (3–6 months): Expect increased adoption of ERA within academic and industrial research labs as scientists explore its capabilities through the Gemini for Science platform and trusted tester programs. The release of additional research papers leveraging ERA is highly probable.
  • Medium-term (1–2 years): ERA’s success in specific benchmarks could lead to its integration into more specialized scientific software and platforms, becoming a standard tool for computational modeling across various disciplines. Training programs and educational curricula may begin incorporating AI-assisted coding methodologies.
  • Long-term (3–5 years): ERA and similar AI tools could fundamentally alter the scientific discovery pipeline, significantly reducing research cycles and enabling the exploration of previously intractable problems. This could lead to breakthroughs in medicine, environmental science, and materials research at an accelerated pace.

ACTIONABLE INSIGHTS

  • Explore the ERA introductory blog post and detailed applications blog post to understand its current functionalities and use cases.
  • Review the official ERA code and experiments to gain practical insights into its implementation and benchmark performance.
  • Investigate the Computational Discovery prototype via Google Labs’ trusted tester program to experience AI-assisted research firsthand.
  • Consider how ERA’s capabilities in scientific coding and optimization could be applied to current research challenges within your organization.
  • Stay informed on new scientific manuscripts applying ERA, particularly those relevant to epidemiological forecasting and environmental modeling.

What is Empirical Research Assistance (ERA)?

ERA is an AI tool developed by Google Research that uses Gemini to write and optimize scientific code. It is designed to assist scientists in iteratively testing and refining computational experiments, speeding up the scientific discovery process.

Who developed ERA and when was it published?

ERA was developed by Lizzie Dorfman, Product Manager, and Michael Brenner, Research Scientist, at Google Research. Its design and performance were described in a paper titled “AI system designed to help scientists write expert-level empirical software,” published today in the journal Nature.

What scientific problems has ERA been applied to?

Google scientists and collaborators have used ERA to investigate eight scientific problems, including epidemiological forecasting for flu, COVID-19, and RSV hospital admissions, and modeling seasonal runoff in California’s snow-fed river basins.

How does ERA achieve expert-level performance?

ERA employs a tree search approach, considering thousands of options to optimize its output code against a given goal. It has been tested on benchmark problems across various disciplines, consistently achieving expert-level results.

What is Computational Discovery and how is ERA related to it?

Computational Discovery is a new experimental tool now available through a trusted tester program in Google Labs, and more broadly through Gemini for Science. ERA is one of the systems used to build this prototype, showcasing its foundational role in Google’s AI-driven scientific initiatives.

KEY TAKEAWAYS

  • Google Research has launched Empirical Research Assistance (ERA), an AI tool that leverages Gemini for expert-level scientific coding.
  • ERA significantly streamlines the time-consuming process of testing and refining computational experiments, as detailed in a Nature publication.
  • The tool has demonstrated expert-level performance across diverse benchmarks, including genomics, public health, and neuroscience.
  • ERA is a foundational component of Google’s new Computational Discovery prototype, accessible via Google Labs and Gemini for Science.
  • Initial applications of ERA include highly accurate epidemiological forecasting for respiratory viruses and environmental modeling for California’s water resources.