Google ATLAS: scientists save ~6 hours a week on AI

Google's AI & Economy ATLAS is a long-term research project, and this update brings two things together: a new interactive, open-access experience for exploring its millions of data points, and new research from Google, Google DeepMind, and MIT FutureTech on how scientists use AI. The study draws on ATLAS data, an analysis of 2,600 specialized AI models, and a survey of over 600 U.S. and U.K. scientists organized using a new taxonomy from MIT FutureTech.

The findings suggest that scientists are adopting AI at a higher rate than many other occupations, but the way they use it differs by tool. LLMs like Gemini spread widely across fields, while specialized models are relatively more common in health and life sciences and in domain-specific data prediction, generation, and simulation tasks. The reported time savings of just below seven hours a week are significant, yet the research also points to time spent validating AI outputs, a growing backlog of hypotheses, and bottlenecks in physical experimentation and clinical validation.

That combination suggests the benefits of AI in science may hinge less on the tools themselves and more on whether research workflows are redesigned around them. For research institutions, funders, and lab managers, the question is likely to be how to absorb faster hypothesis generation without letting testing and validation become the limiting step. The article notes that many questions about the future of AI and the economy remain, and that ATLAS will continue with partners in academia and elsewhere.