Experiments on the future of AI-driven science — Google Labs

A collection of experiments exploring the future of AI-powered scientific discovery
Comprehensive literature search to find relevant papers, structure results in a data table, and create high-fidelity artifacts like reports, slide decks, infographics, and more.
Generate novel research ideas through a multi-agent system that simulates the scientific method to help identify knowledge gaps and propose testable research plans.
Discover models and algorithms with an agentic research engine that generates and scores code variations based on your optimization metrics.
Comprehensive literature search to find relevant papers, structure results in a data table, and create high-fidelity artifacts like reports, slide decks, infographics, and more.
Perform high-signal filtering and search to find relevant papers for your research.
Extract data, such as complex metrics and variables, from your papers to compare side-by-side.
Extract comprehensive literature search with citations that can link claims back to the highlighted text in the source.
Transform deep literature analysis into shareable team resources by using the studio panel to instantly produce slide decks, mind maps, audio overviews, infographics, and more.
Generate novel research ideas through a multi-agent system that simulates the scientific method to help identify knowledge gaps and propose testable research plans.
Chat with a specialized agent to help expand and refine your research challenge, preferences, and focus areas before initiating a run.
Employs a multi-step tournament process to generate a breadth of novel research directions while rigorously evaluating their potential.
Generated ideas are linked to a comprehensive knowledge base of verified scientific references used by the agent during the run.
Surfaces insights that can distinguish high-potential directions from non-viable ones by helping to identify key limitations and critical flaws.
Discover models and algorithms with an agentic research engine that generates and scores code variations based on your optimization metrics.
Work with the agent to help define your optimization task, evaluation requirements, and apply relevant scientific methods.
Generates and scores thousands of code variations in parallel to test modeling directions that would be extremely difficult to explore manually.
Inspect the evolutionary lineage of your code and use the agent to help understand the specific "jumps" in performance scores.
Delivers expert-level, vetted code that can be quickly integrated into your existing research workflow or used for further iteration.