At Google I/O 2026, held in late May, Google Research introduced Gemini for Science, a suite of experimental AI tools designed to accelerate scientific discovery by automating hypothesis generation, code development, and experimental evaluation. The system builds on two newly published research breakthroughs—Empirical Research Assistance (ERA) and Co‑Scientist—both featured in Nature just prior to the event. (research.google)
Empirical Research Assistance (ERA) is an AI-driven research coding engine that, given a well-defined problem and scoring metric, generates, evaluates, and iterates through thousands of code variants using tree search to optimize performance. Early applications span neuroscience, cosmology, hospital admission forecasting, and hydrological modeling in California. (research.google)
Co‑Scientist is a multi-agent system built on Gemini that collaborates with researchers by iteratively generating, evaluating, and refining hypotheses. It has already been applied to challenges such as antimicrobial resistance, plant immunity, and liver fibrosis. (research.google)
Together, these systems power Gemini for Science, which includes tools like Computational Discovery—an agentic research engine that runs thousands of code variants in parallel, enabling rapid hypothesis testing and modeling exploration. (research.google)
Google is also piloting AI tools for peer review and scientific validation. Its Paper Assistant Tool (PAT) has been used experimentally at major conferences like ICML, STOC, and NeurIPS to review over 10,000 papers, helping authors identify theoretical gaps or run new experiments. (research.google)
These developments signal a shift toward agentic AI systems that can meaningfully contribute to the scientific method—from hypothesis generation to validation—potentially transforming how research is conducted across disciplines.