During Google I/O, we published our research on ERA (Empirical Research Assistance) in Nature. Check out this conversation between Yossi Matias (VP, Google & GM, Google Research) and Lizzie Dorfman (Principal Product Manager for Science AI) as they discuss how this agentic coding system is unlocking scientific breakthroughs that were previously thought impossible. ERA is one of the systems used to build Computational Discovery, a new experimental tool that is starting to roll out more broadly through Gemini for Science. As Yossi often says: AI is an amplifier of human ingenuity. 🚀 Resources to learn more👇 Blog link: https://lnkd.in/gG7htjyE Nature Paper link: https://lnkd.in/gr_uQPhP Google for Science: https://lnkd.in/gYcvzpcb
About us
From conducting fundamental research to influencing product development, our research teams have the opportunity to impact technology used by billions of people every day. We aspire to make discoveries that impact everyone, and sharing our research and tools to fuel progress in the field is fundamental to our approach.
- Website
-
https://research.google/
External link for Google Research
- Industry
- Technology, Information and Internet
- Company size
- 1,001-5,000 employees
Updates
-
An impressive milestone for our crisis resilience efforts. WeatherNext, developed with @GoogleDeepMind, delivers forecasts 8x faster and accurately predicted Hurricane Melissa's path 5 days in advance with 80% confidence. Check out the post from Yossi Matias to learn more. ↓
A new milestone in weather forecasting from Google I/O this week highlights how we are applying core AI research to address critical global challenges and help with crisis resilience. We developed WeatherNext—an AI weather model by teams in Google DeepMind and Google Research - that excels at predicting both track and intensity simultaneously. The model achieves this dual capability by training on decades of global weather patterns alongside specialized datasets of tropical cyclones. 🌀 Rather than providing a single best guess, WeatherNext can run ensembles of 50 different "what-if" scenarios, giving experts a broader range of possibilities to inform decision-making. During the 2025 hurricane season, this multi-year collaboration with the National Hurricane Center (NHC) faced a historic test with Hurricane Melissa, the strongest hurricane on record to make landfall in Jamaica. 🌀 WeatherNext helped the NHC anticipate rapid intensification, predicting a Category 5 strength landfall in Jamaica five days in advance with 80% confidence, which increased to near 100% three days in advance. This marked the first time a storm was successfully predicted to reach Category 5 intensity starting from Category 1 wind speeds, capturing a top-tier hurricane from modest beginnings. 🌀 Across the entire 2025 season, the NHC’s annual verification report found that WeatherNext was the top-performing individual model for track and intensity. This performance gave forecasters the confidence to provide the Meteorological Services of Jamaica with unprecedented lead time to safely mobilize resources and coordinate evacuations. By combining the speed and accuracy of AI with the irreplaceable experience of expert forecasters, our goal is to reduce the human and economic toll of natural disasters. This technology is already integrated into forecasts on Search in geographic areas covered by NOAA and are working to expand access to other regions. Moving forward, these research capabilities will be brought to other critically affected regions, including active partnerships and future collaborations with local agencies in the Philippines, Taiwan, Indonesia, Vietnam, Japan, Australia, and India. You can explore the data and tracking models through our experimental showcase at Weather Lab: https://lnkd.in/gEqf_7Cv The weather models are part of Google Earth AI, our collection of geospatial models and datasets for planetary intelligence. Read about the announcement: https://lnkd.in/gvx6hB_g Read about Google Earth AI: https://lnkd.in/d33UYtae Read about Google climate resilience efforts: https://lnkd.in/gp34Cyq8
-
-
Today, Yossi Matias shares 5 research highlights (of many) from Google I/O, showcasing how these advancements in personalization, intelligence, and agentic systems are shaping the future of AI and transforming technology's impact. Watch the video below to learn more. ↓
Hello from Google I/O! Yesterday and today we showcased how our advancements across personalization, intelligence, and agents are helping to solve complex problems and deliver real-world helpfulness—from daily life to scientific discovery. What a great energy at the show! And thanks to the Google social team for discussing with me 5 (out of so many) research highlights. Read more about everything we announced: https://lnkd.in/g7Jg9PrM
-
Our latest research on Co-Scientist is out today in Nature! Built with Gemini, this multi-agent system powers the new Hypothesis Generation tool within Gemini for Science, helping researchers navigate the rigorous cycle of ideation, critique, and refinement. Read more from Yossi Matias and explore the full announcement: https://lnkd.in/gCZ5qxSV
Today, our research on Co-Scientist was published in Nature. It introduces a new multi-agent AI system built with Gemini that iteratively generates, debates, and evolves novel hypotheses for complex scientific problems. Hypothesis Generation introduced today at #GoogleIO as part of the Gemini for Science experimental tools is built with Co-Scientist. Scientific discovery is rarely a straight line; it is a rigorous cycle of ideation, critique, and refinement. The core research question behind Co-Scientist was: How can an AI system empower researchers in this rigorous structured thinking for scientific discovery? Rather than thinking linearly, Co-Scientist relies on a collaborative coalition of specialized agents built with Gemini, orchestrated by an adaptive supervisor agent acting as a freeform planner. It operates across three distinct phases: 👉 Generating Ideas: A specialized Generation Agent proposes novel hypotheses grounded in literature, while a Proximity Agent maps and clusters these ideas to ensure the system thoroughly explores diverse areas of the research space. 👉 Debating Ideas: To enforce scientific rigor, a Reflection Agent acts as a virtual peer reviewer to critically evaluate hypotheses. A Ranking Agent then orchestrates a "tournament of ideas," utilizing pairwise comparisons and simulated scientific debates to prioritize the most robust concepts. 👉 Evolving Ideas: An Evolution Agent continually refines and combines the top-ranked hypotheses. Finally, a Meta-Review Agent synthesizes insights from the entire tournament to optimize system performance and generate a final research proposal for human review. To ensure these generated hypotheses are robust and testable, the majority of the system's computation is dedicated to verification. The system runs an Elo-based tournament to continuously rank ideas while injecting fresh knowledge. It cross-checks claims against scientific literature and integrates web search alongside specialized databases such as ChEMBL and UniProt. We are already seeing this move into real-world validation. At Stanford University School of Medicine, Co-Scientist helped identify drug-repurposing candidates for liver fibrosis—one of which successfully blocked 91% of a scarring-linked response in lab tests. AI handles the complexity of data synthesis and hypothesis ranking, the "human spark"—intuition, judgment, and asking the right questions—becomes even more critical. Science remains a team sport, and we believe the future is one where human creativity is amplified by accessible AI partners. AI as an amplifier of human ingenuity Read the paper in Nature: https://lnkd.in/gcu9wjM8 The announcement blog: https://lnkd.in/gN_W73yf
-
-
Today at #GoogleIO, we announced Gemini for Science, a new collection of tools and experiments designed to expand the scale and precision of scientific exploration. Learn more from Yossi Matias and read the blog: https://lnkd.in/gYcvzpcb
I am delighted that today at #GoogleIO, we are introducing ✨Gemini for Science✨ — a collection of science tools and experiments designed to expand the scale and precision of scientific exploration. Today’s scientific paradox is that our collective knowledge is growing so fast that it’s becoming harder for individual scientists to see the full picture. Scientific breakthroughs often rely upon making creative connections between data, but the time required to do this manually can take weeks or even months. AI can help eliminate this bottleneck and serve as a force multiplier for scientific work by handling complex tasks. Researchers can now focus on identifying and tackling the most impactful scientific problems. AI as an amplifier of human ingenuity. Key pillars of this new scientific workbench: ✨ Hypothesis Generation (built with Co-Scientist): Ideation is the heartbeat of science. Our Co-scientist system is a multi-agent AI system that iteratively generates, debates, and evolves novel hypotheses for complex scientific problems. We are making Co-Scientist available to individual researchers through Hypothesis Generation. The research on Co-Scientist is published today in Nature. ✨ Computational Discovery (built with AlphaEvolve and ERA): Scientific progress is often limited by how many computational experiments we can run. This agentic engine can generate and score thousands of code variations in parallel, accelerating modeling in fields like epidemiology and climate science. The research on ERA (Empirical Research Assistance) is published today in Nature. ✨ Literature Insights (built with NotebookLM): Understanding vast amounts of literature is a core research bottleneck. This tool synthesizes findings across papers, identifying research gaps and uncovering new opportunities through searchable side-by-side analysis. We are collaborating with over 100 institutions—including Stanford University School of Medicine, Imperial College London, and the Crick Institute—to validate these systems. To ensure the integrity of AI-generated insights, we have established a trusted tester community of researchers across a wide range of scientific fields to stress-test our tools against real-world challenges. We keep exploring more ways to help with the scientific method, including tools that help review papers such as our Paper Assistant Tool (PAT) and ScholarPeer. Congratulations to the amazing Google teams and partners involved in this work! Our Gemini for Science announcement - posted with my colleague Pushmeet Kohli : https://lnkd.in/gdtYkqfh Co-Scientist announcement: https://lnkd.in/gN_W73yf ERA announcement: goo.gle/3Rjc0A2 Co-Scientist Nature paper: https://lnkd.in/gcu9wjM8 ERA Nature paper: https://lnkd.in/gbuZSYjR
-
-
Our research on Empirical Research Assistance (ERA) was published today in the journal Nature. ERA uses Gemini to achieve expert-level performance for scientific coding and helped build the new Computational Discovery prototype in Google Labs. Learn more: goo.gle/4ueFSvY
-
-
Google Research reposted this
I am incredibly proud to share that AI Quests has been named the Best Global K–12 Learning Solution at the inaugural EDTech Innovation Hub Awards 2026! 🏆 This recognition is so meaningful because it validates the work we’ve been doing at Google Research in collaboration with the Stanford Accelerator for Learning. Together, we created AI Quests, a game-based learning experience inspired by Google Research projects. By turning research into play, we help students navigate the "healthy struggle" of learning while building true AI literacy. A massive thank you to the whole team, our brilliant partners at Stanford and the hundreds of Googler volunteers who have helped bring these quests to life in classrooms globally. We are just getting started. The future of learning is about building agency, confidence, and curiosity—and I’ve never been more optimistic about the path we’re charting together.
-
-
We’re teaming up with CGIAR to accelerate the development of climate-resilient crops for farmers in the Global South. Through the Google.org AI Collaborative, we’re supporting the creation of a new AI model that’ll help breeding programs automatically analyze crop traits via drone and smartphone imagery — transforming how we fight food insecurity. The Goal: Impact all 150 field stations across 89 countries to: • Modernize breeding: Moving from manual checks to AI-driven efficiency. • Unify data: Standardize agricultural data across global research centers. • Scale fast: Deliver climate-ready crops to the world’s most vulnerable populations in record time! Together, we’re not just helping to breed better crops; we’re building a better food-secure future. 🌍✨. Read the full announcement: https://goo.gle/4un3bTU Learn more about the AI Collaborative: https://goo.gle/49hzTOn
-
-
Google Research reposted this
How can we help scientists get better data out of their instruments? Google Research is using Google DeepMind's AlphaEvolve to identify new ways to train DeepConsensus, a model that corrects errors in DNA sequencing data. Its newest version is launching on PacBio instruments, increasing both output and the amount of high-quality, highly accurate data produced. Stronger data can lead to clearer insights, helping researchers better understand genetic variation and its role in patient health and disease. Explore the updates and open-source improvements on the DeepConsensus GitHub. https://goo.gle/4d01ogy
-
Scientific breakthroughs reach their full potential only when they empower others to replicate & expand upon findings. We are dedicated to responsible, collaborative research that empowers a global community to innovate across disciplines. Learn about our partnerships & impact → goo.gle/4t7fAKA
-