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.
External link for Google Research
Google Research reposted this
Around the world, scientists are using our technology to better understand and respond to infectious diseases. From mapping outbreaks to screening patients, researchers are making progress faster than ever. Learn more about how we’re supporting the science behind these breakthroughs: ai.google/societal-impact
It's a milestone to see the Personal Health Agent (PHA) research published in Nature and open-sourced for the community.
A new Nature paper by our teams on the Personal Health Agent (PHA) —the foundational multi-agent AI architecture behind the Google Health Coach. Conventional monolithic LLMs often struggle to simultaneously balance numerical time-series reasoning, clinical factuality, and open-ended coaching dialogue. The PHA overcomes these limitations through a modular architecture that orchestrates three specialized sub-agents over multimodal personal health data—including wearable time-series and blood biomarker data: ▶️ Data Science Agent: Analyzes longitudinal wearable and laboratory data to derive quantitative insights contextualized against population-level statistics. ▶️ Health Domain Expert Agent: Integrates personal data with grounded clinical health knowledge to provide evidence-based interpretations. ▶️ Health Coach Agent: Applies proven psychological strategies (such as motivational interviewing) to support multi-turn dialogue, goal setting, barrier identification, and progress tracking. 🧠 Central Orchestrator: Dynamically decomposes complex queries, routes tasks to specialists, and synthesizes unified responses through iterative reflection and persistent memory. Evaluated across 11 benchmark tasks with over 7,000 human annotations and 1,350+ hours of assessment from health experts and end users (grounded in the WEAR-ME study), this research represents a most comprehensive evaluation of a personal health agent. To help further advance safe, personalized, and trustworthy health AI, we are open-sourcing the PHA multi-agent framework for the broader research community. Read the full paper in Nature: https://lnkd.in/gSq8zkfp
We are incredibly proud to celebrate Dr. Marian Croak, Vice President of Human-Centered AI and Foundational ML at Google, who is being honored by the New York Hall of Science! From her pioneering work advancing VoIP technology to her current leadership ensuring AI is developed responsibly and equitably, Marian's impact on the tech landscape is immeasurable. She continues to be a passionate advocate for STEM initiatives and a mentor to the next generation of innovators. Join us in congratulating Marian on this well-deserved recognition! 🌟 🔗 Support the NYHS Annual Gala and learn more: nysci.org/gala
Running AI directly on wearables and IOT takes milliwatt-level efficiency, but building for the edge is still a challenge. Because the ecosystem is split across many of proprietary chips and custom toolchains, developers end up rewriting code for every device they target. Today, we are sharing three milestones for Coral NPU to make edge AI portable and open from model to silicon: First, we have completed our 32-bit open-source RISC-V architecture with a new Matrix Execution Unit, delivering up to 1 TOPS of compute and targeting 10 mW average power for inference so small devices can run continuous, multimodal models like Gemma. Second, we are releasing v1.0 of our open-source compiler, letting teams take models from JAX, PyTorch, and TensorFlow straight to Coral NPU hardware without writing custom assembly. Finally, alongside silicon readiness work with VeriSilicon and GlobalFoundries, Synaptics Incorporated is launching its new Astra SR200 chip series, the Coralboard SR, and global availability of the Coralboard SL. Read the blog and explore the code on GitHub: https://goo.gle/4zGSu17
Our new research shows how blind sweep ultrasound AI generalizes to new care settings, enabling non-sonographer operators with low-cost probes to deliver clinical-grade gestational age assessments, expanding life-saving maternal care globally. Read the blog: goo.gle/4siHGUG
How does Google Research decide which problems are worth solving? 💡 In our latest episode of The Google Research Podcast, host Sinéad Bovell sits down with Google Research’s Michael Brenner and Lizzie Dorfman to break down the core framework behind how Google Research selects its moonshots — and why long-term investments in computational neuroscience, genomics, and atmospheric science are paying off today. Here are a few key takeaways from the conversation: 1️⃣ The Three Core Questions Every project at Google Research is evaluated against a simple yet rigorous framework: Is it a big problem? Is it a problem technology can address? And is it something Google is uniquely positioned to solve? 2️⃣ Patience in Foundational Science Breakthroughs rarely happen overnight. Google Research invests early in long-horizon domains — like weather modeling, genomic sequencing, and neural mapping — years before their practical AI applications reach critical scale. 3️⃣ Intersecting AI with Earth & Health Sciences By applying advanced machine learning to climate patterns and complex biological systems, researchers are moving from pure data analysis to predictive modeling that impacts sustainability and global health. 🎬 Watch the cutdown below, or tune in to the full episode here: https://lnkd.in/gJK3yJy8 Explore more Google Research films → https://goo.gle/3NvR4kt Subscribe to the Google Research channel → https://lnkd.in/eeV2KKpB
Google Research reposted this
When researchers gather from around the world for our yearly Research conference to share knowledge, spark bold new ideas, and open our doors to share our top work with the world, the energy is unmatched. But what happens in the sessions is only half the story. Behind every smooth demo and thoughtful session is this team — navigating endless details, solving last-minute hurdles, and creating a space where great science can thrive. Our Research conferences over the years are not just built by the people on stage. It’s also built by the incredible event organizers, program managers, and cross-functional teammates who work tirelessly behind the scenes to bring it all to life. To everyone who put their heart into making last week happen: thank you - I'm so lucky to work alongside you all. Tag someone who made the Research conference better this year 🧡
Will AI soon "solve science"? Or is scientific discovery an endless frontier? 🔬✨ In Episode 2 of The Google Research Podcast, host and AI Futurist Sinéad Bovell sits down with Google Research’s Michael Brenner and Lizzie Dorfman to break down the realistic, transformative role of AI in scientific discovery. While viral headlines often frame AI as a tool that will single-handedly solve complex fields, the real power of AI lies in its ability to amplify human curiosity and accelerate how we ask and test questions. Here are key takeaways from their discussion: 1️⃣ AI as an Accelerator, Not a Replacement. AI isn't replacing the scientific method, it’s compressing execution timelines. By sifting through vast datasets, generating complex hypotheses, and running high-throughput simulations, AI tools allow researchers to move from idea to breakthrough faster than ever before. 2️⃣ Science Isn't a Finite Puzzle Science isn't something to be "solved." Every question AI helps us answer inevitably reveals a new set of unanswered questions, expanding the boundaries of human knowledge rather than closing them. 3️⃣ The Irreplaceable Human Element No matter how advanced AI models become, intentional curiosity, critical evaluation, and intuitive judgment remain uniquely human. Humans frame the problems worth solving and direct the trajectory of discovery. 💡 Discover how AI is reshaping the scientific frontier and where human creativity fits into the future of research. 👉 Watch the full episode now: https://lnkd.in/gJK3yJy8
Google Research reposted this
This week at our Research@ flagship event, we announced new updates for our experimental Geospatial Reasoning agent and Earth AI capabilities. Following promising tests with partners, we are expanding early access to our experimental agent to more partners. In a new technical report, we demonstrate how combining domain-specific foundation models, spatial tools, and upfront planning enables a Geospatial Reasoning agent to carry out complex spatial analysis and planetary prediction that standalone LLMs can’t. Highlights from this work: --> The agent connects Gemini to the capabilities of Earth AI’s foundation models (WeatherNext, Flood Forecasting, PDFM, AlphaEarth) using modular skills, dedicated spatial workspaces, and private data support (CSV, GeoJSON). --> Beyond answering complex spatial questions, the agent powers our Planetary Prediction Engine—combining satellite embeddings and population dynamics to train validated models for disease spread, crop forecasting, or environmental risk. --> Real-World Impact with Partners: Milken Institute: Identifying the highest-impact community resilience hubs for solar energy build out—layering rooftop yield with transit access, heat vulnerability, and local demographics data to compress 6-week siting feasibility studies into 15-min reports to unlock $179 million in clean energy capital. WHO AFRO & DRC’s INRB: During the recent Ebola outbreak response in the DRC, teams mapped remote mining corridors by cross-referencing satellite forest loss, mining locations, and mobility data—pinpointing 48 exposed settlements and 45,500+ at-risk people in minutes rather than weeks, and modeling transmission risks to plan mobile laboratory deployment. People in Need: Analyzed historical cyclone paths to identify high-exposure villages in minutes rather than days of manual GIS work, unlocking vital time for anticipatory action. GiveDirectly: Evaluated population exposure across wind-speed zones during Cyclone Maila and benchmarked tornado damage in Mississippi within a single workspace. WPP: Connected population dynamics with proprietary data to optimize geospatial marketing while preserving privacy. --> Evaluating Geospatial Reasoning: Planetary datasets are massive and multi-layered. Across four benchmarks—including two new benchmarks we are releasing to researchers, GeoExpertQueries and GeoKTools—our evaluations quantify the essential role of specialized Earth AI tools and upfront planning over out-of-the-box LLMs, enabling the agent to execute multi-step analytical workflows reliably. Congratulations to teams across Google, and thank you to our partners for collaborating with us on the ground. Read the blog by Shravya Shetty and Luke Barrington: goo.gle/4s13F25 Early access interest form: https://lnkd.in/gxt9EfqB
Accelerating scientific discovery starts with empowering the researchers already leading it. Last week we announced the 17 organizations selected for the $36M Google.org Impact Challenge: AI for Science. These academic and nonprofit teams from around the world are tackling complex challenges in health and life sciences and in climate resilience and environmental science. Over the next six months, they will receive pro bono engineering support from Google experts and $3M in Google Cloud credits as part of a dedicated Google.org Accelerator program, helping them turn complex research into robust and scalable solutions. Their teams will be creating models, agents, and datasets that peers in the fields can also use to drive further innovation and discovery. See how they are shaping the future of science: https://lnkd.in/gQuTKMMK