Google releases AI & Economy ATLAS v1.0 with interactive data and research on AI adoption and scientific use

Google released AI & Economy ATLAS v1.0, an interactive open-access tool designed to visualize and explore global AI adoption patterns across occupations and countries. The tool aggregates millions of data points to allow users to examine AI usage rates by profession—from electricians to purchasing managers—and by country, including breakdowns of how AI is applied in work and at home.

Regional trends highlighted by the ATLAS data show India’s creative industry leading in AI usage, with arts, design, and media occupations accounting for 19% of work-related AI usage—1.6 times the global average. The U.S. leads in technical AI adoption, where computer and mathematical occupations represent 30% of work-related AI usage, double the share in the rest of the world.

Income level and AI adoption do not always align, with countries like Brazil and the United Arab Emirates exhibiting higher-than-expected adoption rates relative to GDP per capita. Usage for manual tasks such as real-time equipment diagnostics also varies by region, with Brazil and Germany reporting 7% of work-related AI usage in manual tasks—1.4 times the global average—compared to 4% in Japan.

New research from Google, Google DeepMind, and MIT FutureTech leverages ATLAS data to analyze how scientists use AI. The study examines 2,600 specialized AI models and surveys over 600 scientists in the U.S. and U.K., using a taxonomy from MIT FutureTech to map scientific tasks. It finds that nearly half of scientists use some form of AI daily, with both LLMs like Gemini and specialized models playing mutually reinforcing roles.

Scientists report saving nearly seven hours per week using AI, which they reallocate toward research activities. However, the research also identifies bottlenecks downstream in the research pipeline, including time spent validating AI outputs, a growing backlog of untested hypotheses, and delays in physical experimentation and clinical validation.

The findings suggest that while AI can boost individual productivity, realizing broader gains in scientific output may require redesigning workflows and processes to accommodate fast-evolving AI capabilities.