Google launches 'Gemini for Science' with 3 experimental AI research tools

Google has launched Gemini for Science, a suite of experimental AI research tools announced at Google I/O. The set includes three new experiments in Google Labs: Literature Insights, built on NotebookLM, searches scientific papers and structures results into data tables, reports, slide decks, infographics, and video overviews. Hypothesis Generation uses Co-Scientist to help researchers define challenges, generate ideas, and evaluate hypotheses through a multi-agent 'idea tournament' with clickable citations and flaw detection. Computational Discovery leverages AlphaEvolve and Empirical Research Assistance (ERA) to generate and score code variations in parallel, targeting areas like solar forecasting and epidemiology. Google said the tools aim to make AI agents 'force multipliers for human ingenuity' in scientific discovery.

Alongside the experiments, Google unveiled Science Skills in Google Antigravity, a bundle that aggregates insights from over 30 life science databases and tools including UniProt, AlphaFold Database, AlphaGenome API, and InterPro. In early testing, the bundle reduced analyses that typically take hours to just minutes, leading to insights into a rare genetic disease caused by AK2 gene mutations. Google is collaborating with more than 100 academic institutions including Stanford Medicine, Imperial College London, and The Francis Crick Institute to validate the systems, with papers on ERA and Co-Scientist published in Nature. Enterprise-grade versions are already in private preview, with BASF named as a user. Access to Gemini for Science experiments begins gradually from May 2025 via Google Labs.