At Google I/O 2026, researchers unveiled a suite of AI tools that work as a scientist's collaborative partner—systems designed not to replace human ingenuity, but to amplify it across the most pressing questions facing our world. The announcement marks a turning point: AI moving beyond chat interfaces into the laboratory itself, where it can help researchers generate hypotheses, write expert-level code, and synthesize findings across millions of published papers.
The centerpiece of this initiative is Gemini for Science, a collection of experimental tools built through collaboration between Google Cloud, Google DeepMind, and Google Labs. Two foundational systems behind it—Empirical Research Assistance (ERA) and Co-Scientist—were published simultaneously in Nature, signaling the scientific rigor undergirding these advances. ERA is a research coding system that helps scientists write expert-level empirical software, while Co-Scientist operates as a multi-agent system based on Gemini, functioning as an AI research partner that can collaborate with humans on complex problems.
The real-world impact is already visible. ERA has accelerated discoveries spanning neuroscience to cosmology, with recent applications predicting hospital admissions for respiratory illnesses and forecasting seasonal runoff across California's river basins. Co-Scientist has been used to tackle antimicrobial resistance, plant immunity, and liver fibrosis—some of the most intractable challenges in modern medicine. These are not theoretical demonstrations; they are active research efforts producing publishable results.
Within Gemini for Science, a new tool called Computational Discovery acts as an agentic research engine that generates and scores thousands of code variations in parallel, enabling scientists to test multiple hypotheses and modeling approaches that would otherwise take months to explore manually. A companion tool, Hypothesis Generation, uses Co-Scientist to help researchers define a problem and then runs a multi-agent "idea tournament" that generates, debates, and evaluates potential hypotheses—each backed by clickable citations to ensure scientific rigor. Literature Insights, built with NotebookLM, tackles another growing pain: synthesizing findings across the millions of papers published annually, which has become nearly impossible to do by hand.
Google is also pioneering support for the scientific publishing process itself. At leading conferences including ICML, STOC, and NeurIPS, a tool called the Paper Assistant Tool (PAT) has been piloted in an experimental capacity, reviewing over 10,000 papers to help authors identify theoretical gaps or suggest entirely new experiments to strengthen their work. This peer-review augmentation is expanding gradually, with the company working closely with the scientific community to ensure these tools are used responsibly.
The broader vision at play is simple but profound: AI as an amplifier of human ingenuity. These tools are designed to handle the computational grunt work—the hypothesis generation, the literature synthesis, the code variations—freeing researchers to focus on the creative and strategic thinking that only humans can do. Google is gradually opening access to these tools through labs.google/science, partnering with the global scientific community to ensure that this technology accelerates discovery at scale, not just in a few well-resourced labs.