Google's AI science tools could fuel breakthroughs

Google is rolling out Gemini for Science, a set of experimental tools designed to compress scientific work that typically takes months or years into a matter of days. The program, built across Google Research, DeepMind, and Google Cloud, puts three prototypes into the hands of working researchers, including a multi-agent "Co-Scientist" that generated a bacterial hypothesis in days that previously took Imperial College researchers years to reach. Early partners include BASF, Bayer Crop Science, and the US National Labs, with researchers at Stanford using the system to identify an existing drug that reduced signs of liver scarring.

oogle is rolling out Gemini for Science https://ai.google/gemini-for-science/ , a set of experimental tools designed to compress scientific work that would typically take months or years into a matter of days. This was one of the announcements at Google I/O that went under the radar, overshadowed by the rollout of Gemini 3.5 https://www.thedeepview.com/articles/gemini-3-5-and-spark-get-google-in-the-agent-race and Google's re-entry into the smartglasses market https://www.thedeepview.com/articles/google-s-ai-glasses-look-ready-to-take-the-lead . But Gemini for Science could quietly have an even bigger impact than either of those flashy announcements. The program is built across Google Research, DeepMind, Google Cloud, and Google Labs. It puts three prototypes into the hands of working researchers through Google Labs. It also extends to enterprise customers through Google Cloud, with early partners including BASF, Klarna, Bayer Crop Science, and the US National Labs as part of the Department of Energy's Genesis Mission. In an interview with The Deep View, the Head of Google Research, Yossi Matias said, "I think about AI as an amplifier of human ingenuity." He emphasized that junior scientists, postdocs, and even graduate students can now run "their own virtual lab." He also shared the example of an early partnership with Imperial College, where researchers had spent years arriving at a bacterial hypothesis that Google's new Co-Scientist agent reached in days. Co-Scientist, which Google detailed in a new paper in the journal Nature , is a team of AI agents built on Gemini that work together like a research group. Each agent has a job: one comes up with ideas, another critiques them, another ranks the best ones, another improves them, and another reviews the whole process. A lead agent acts as the manager, keeping them all on track. Early results look promising. At Stanford, researchers used Co-Scientist to help find an existing drug that reduced signs of liver scarring in lab-grown tissue. The system is also being used by researchers at Calico Life Sciences to study aging, at the University of Edinburgh to look for new liver disease treatments, and at the University of Cambridge to study how viruses like the flu and COVID-19 can spread from animals to humans. The pitch from Google Research is that general-purpose agents, not narrow specialized models, are what will move the needle for working scientists across disciplines. The key components include: Hypothesis Generation : This is built on Co-Scientist, runs a multi-agent "idea tournament" that generates, debates, and evolves research hypotheses, with claims grounded in citations. Computational Discovery : This is built on AlphaEvolve and ERA Empirical Research Assistant , generates and scores thousands of code variations in parallel to test modeling approaches in fields like solar forecasting and epidemiology. Literature Insights : This is powered by NotebookLM and structures findings across papers into searchable tables and then produces reports, slide decks, and audio overviews from the materials. Science Skills : This bundle plugs in more than 30 life science databases including UniProt, AlphaFold Database, and AlphaGenome API into agentic tools like Antigravity. Lizzie Dorfman, who is the science product lead at Google Research, told The Deep View that in one epidemiological forecasting project, her team generated 200,000 candidate models. Most were discarded quickly, but the ability to explore that volume at all is what changes the workflow. " Researchers would say things like, 'I input some ideas, and then I went to sleep, and then I woke up and I had all these cool results to take a look at.'" said Dorfman. "That’s an example of changing the amount of productivity that a single person can have." Our Deeper View All of the frontier labs love to extol the benefits that AI is going to have on science. For example, just last month, OpenAI released GPT-Rosalind https://www.thedeepview.com/articles/openai-s-gpt-rosalind-marks-rise-of-science-first-ai-models , the first life sciences model purpose-built for lab workflows. And we have to give Google credit for its long investment in science and research. Leaders like Dorfman and Matias have been working on basic research inside Google for over a decade. The fact that they are now applying the breakthroughs in AI agents to accelerate their experiments and sharing it broadly with the wider science community has the potential to become one of AI's most positive impacts.