AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Research@ 2026: 10 ways we're amplifying human ingenuity

Collected Oct 14, 2026

At its flagship Research@ event in Mountain View, Google Research presented ten updates that span geospatial intelligence, health AI, scientific tooling, numerical foundation models and edge hardware. Yossi Matias, Vice President at Google and GM of Google Research, framed the work around AI as an amplifier of human ingenuity and scientific inquiry.

The largest cluster of announcements concerns Google Earth AI, which the company describes as enabling planetary intelligence. Google is expanding early access and inviting partners to explore three new access paths: a deeper geospatial analysis feature inside Google Earth, risk assessments and scenario forecasting through a planetary prediction engine, and Geospatial Reasoning agent skills exposed to custom workflows via an experimental agent-to-agent API. That API gives developers programmatic access to capabilities including satellite imagery parsing, routing and demographic embeddings. The stack also adds Google Dataset Search, which queries more than 100 million public datasets, and TimesFM for planetary nowcasting and forecasting. These capabilities rest on over 70 geospatial tools, including foundation models such as the Remote Sensing Foundation Model (RSFM) and Population Dynamics Foundation Model (PDFM), and are validated by empirical benchmarks in an Earth AI Technical Report. Alongside the technical work, Google.org is committing $10 million to organizations focused on crisis resilience and public health, including $2 million to The Milken Institute for infrastructure bottlenecks around affordable housing, clean energy, broadband and safe water systems. Google says the geospatial foundation models are already being used to help predict disease outbreaks.

On the scientific workflow side, Google built a multi-agent orchestrator that lets scientists reproduce previous methods and analyses, then apply them in new settings. As one example, researchers applied a methodology for inverse design problems in fluid mechanics to urinary catheter design and found catheter designs that, in simulation, were better at preventing bacterial infections. This complements Empirical Research Assistance, which generates and tests thousands of code variations in parallel, and Co-Scientist, built with Google DeepMind, which generates novel hypotheses and has helped partners including Stanford University researchers uncover repurposed medicines to fight liver fibrosis.

Deep Review is an agentic system for rigorous verification across complex documents, checking logical consistency and grounding facts while leaving creative direction to the author. An early version was used by Google researchers and external authors to review over 16,000 research publications submitted to top-tier computer science conferences. Google is now expanding access to the general public, targeting technical documents ranging from legal agreements to financial reports.

On hardware, Google points to the first Coralboard from Synaptics, launched earlier this year using Google's Coral NPU accelerator core, which lets AI and ML engineers rapidly prototype devices. Next week Google will reveal a co-designed, next-gen, low-power Edge AI chip built by Synaptics using Google's open source IP. Google positions it for design engineers building smart glasses and watches, connected cameras, streamers, and industrial vision and voice systems.

For numerical data, Google highlights TimesFM and TabFM, foundation models engineered to reason over time-series streams and structured tabular data. Both deliver predictions out of the box with zero training, rank at the top of leading public benchmarks in their fields, and are available to all Cloud BigQuery users. Google announced an upcoming integration of these models into Gemini for business, aimed at forecasting trends such as product sales or hospital bed capacity from complex operational tables and millions of rows, with sensitive data staying inside an organization's perimeter.

In health, MedGemma was published last week in Nature Medicine and has been downloaded more than 10 million times by more than 65,000 unique innovators. Developers have adapted it for faster triage in overcrowded clinics, eye disease screening and cervical cancer screening. Google says its latest research adds tool use so MedGemma can draw on health records and other clinical applications, and that it is exploring modalities including mammography, ultrasound and electrocardiograms, with early signs of gains in answering clinical questions and interpreting radiology scans, skin images and pathology slides.

The AMIE program explores conversational AI for clinical reasoning. Research published this year in Nature advanced AMIE beyond initial diagnosis toward long-term disease management, and work in Nature Medicine showed reasoning across multimodal data such as medical images and lab results. A new study in The Lancet reports that at Beth Israel Deaconess Medical Center, 98 patients safely consulted AMIE ahead of urgent care visits, and clinicians said the AI summaries helped them prepare in 75% of cases. Google is advancing AMIE toward real-time video consultations and has begun a nationwide randomized study of AI in real-world virtual care settings with Included Health. AMIE builds on the earlier Med-PaLM line of work.

On disaster prediction, Google's global hydrologic model, published in Nature, provides reliable river forecasts up to seven days in advance for the most significant riverine floods, covering more than 2 billion people in over 150 countries. Flash floods were a harder problem. Groundsource is an AI-powered methodology that turns publicly available crisis information into high-quality historical datasets. Applied with Gemini to flood-related news reports, it produced a structured dataset of over 2.6 million flood events, which helped overcome the lack of ground-truth data and train a model that predicts the likelihood of flash flooding in urban areas up to 24 hours ahead. The dataset is now part of the Google Earth AI collection and is openly available.

In education, Google's generative UI research fine-tunes interface generation with pedagogical guardrails so teachers can create custom guided STEM simulations. A sample library features over 30 AI-generated learning interactives vetted by teachers for middle and high school physics, chemistry, biology and math. Schools can sign up to the Google for Education Pilot Program to give feedback.

Finally, Play with Putty lets teams build and prototype together in a shared environment, with multi-player editing and direct interaction with any part of an interface generated from natural language prompts. It was first designed for Google researchers and teams, including work on Google Earth AI visualizations, and the waitlist is now open on Google Labs.

Why it matters: For developers, the agent-to-agent geospatial API, BigQuery availability of TimesFM and TabFM, and the open MedGemma foundation are the most immediately actionable items, since they expose Google research as callable infrastructure rather than papers. Product teams in health, insurance, logistics and public sector work should watch the Earth AI early access and the Gemini for business integration, which together point toward forecasting and geospatial reasoning becoming default platform features. Hardware teams get a new low-power option for wearables and camera products, though the chip's real-world efficiency numbers will need independent validation. The recurring pattern here is packaging research into partner programs and pilot sign-ups, so access will likely be gated and staged rather than instantly general.

Read at Google Research

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

General Science