Google Launches ATLAS Explorer With New Findings on Science Bottlenecks

Google has launched an open-access explorer for its AI & Economy ATLAS data and published new research on scientists’ AI use. The update points to a stubborn limit on AI productivity: researchers report time savings, but validating outputs and testing hypotheses can still slow the path to discovery.

The interactive experience lets people explore ATLAS data by occupation, country and household AI use. It is a new interface for a project Google introduced in July to track use of its AI products and tools through aggregated, de-identified interactions.

ATLAS v1.0 draws on 15 million interactions from the Gemini App, AI Mode and Gemini API. Google says its insights span more than 150 countries, 140 languages, 800 occupations and 4,000 tasks.

The study by Google, Google DeepMind and MIT FutureTech analyzed 2,600 specialized AI models alongside the scientist survey. It found that researchers use both large language models and specialized models, with specialized systems relatively more common in health and life sciences and in domain-specific prediction, generation and simulation tasks.

The research does not treat saved time as an automatic route to more discoveries. It identified bottlenecks in validating AI outputs, physical experimentation and clinical validation, contributing to a backlog of hypotheses awaiting testing.

ATLAS also shows workplace AI use differing sharply by location and profession. In the United States, computer and mathematical occupations account for 30% of work-related AI usage, twice the share in the rest of the world. In India, arts, design and media occupations account for 19%, or 1.6 times the global average.

The explorer is not a full census of economically relevant AI activity. Google says ATLAS excludes Google Workspace, Google Translate, AI Overviews, Gemini for Google Cloud and Gemini Enterprise. Google also says the dataset adds privacy protections beyond removing personally identifiable information, including unlinking summaries from user logs and aggregating them across multiple users.