New Era of Discovery: Google Research at I/O 2026
Google/DeepMind
DeepMind
Synaptics
At Google I/O 2026, the company unveiled cutting-edge technologies, including the Gemini for Science platform with tools like Empirical Research Assistance (ERA) and Co-Scientist, as well as advances in healthcare, hurricane and flood forecasting. Google's research projects, such as AMIE and WeatherNext, have shown significant results in clinical practice and natural disaster prediction.
Key achievements of Google Research were presented at Google I/O 2026. In the science domain, the Gemini for Science platform was launched, including Empirical Research Assistance (ERA)—a system for scientists to write empirical software—and Co-Scientist, a multi-agent system based on Gemini that functions as a collaborative AI partner. Both systems were published in the journal Nature. ERA has already accelerated discoveries in neuroscience and cosmology, while Co-Scientist is applied to combat antimicrobial resistance, plant immunity, and liver fibrosis. Also part of Gemini for Science are Computational Discovery, an engine for parallel hypothesis testing; Hypothesis Generation, a system for generating and evaluating hypotheses based on an "idea tournament"; and Literature Insights, a tool for synthesizing scientific literature based on NotebookLM. In healthcare, the Google Health app and Google Health Coach were announced, built on years of research. Symptom AI, a symptom analysis tool, was presented, which in a blinded comparison was preferred by clinicians twice as often. The AMIE system demonstrated new capabilities in multimodal diagnosis (medical histories, lab data, complex images), published in Nature Medicine. The MedGemma and MedASR models were also released for developers. Hurricane forecasting: the WeatherNext model accurately predicted the intensification of Hurricane Melissa five days before it approached Jamaica, helping save lives. For urban flood forecasting, the Groundsource method was presented, which used Gemini to transform 20 years of unstructured news reports into a dataset of 2.6 million records, enabling the training of models for Flood Hub, which covers 2 billion people in 150 countries. The Coral NPU accelerator and Synaptics Coralboard were launched for energy-efficient AI at the edge, featuring the Gemma 3 270M model.
Source: Google Research —
original
