The Pulse
Google Opens AI Atlas as Scientists Report 7 Hours Saved
Google has opened an interactive version of its AI & Economy ATLAS, pairing global adoption data with a new study of AI use in scientific research. The study finds that scientists save nearly seven hours a week with AI, while facing growing
AI.info Team ·
Nearly half of surveyed scientists say they use artificial intelligence every day, and those who use it report saving almost seven hours a week. Google is making that finding part of a newly opened public data project that maps how people use AI across occupations, countries and household activities.
Google’s AI & Economy ATLAS now includes an interactive, open-access experience built from millions of de-identified AI interactions. The company also published a new study, produced with Google DeepMind and MIT FutureTech, examining how AI is entering scientific workflows and where productivity gains run into limits.
15 million interactions become a public map
The ATLAS project draws on about 15 million anonymized interactions across the Gemini app, Google AI Mode and the Gemini API. Google says the data spans more than 150 countries, 140 languages, 800 occupations and 4,000 tasks.
The new interface allows users to examine AI use by occupation, region and activity. It includes comparisons between jobs such as electricians and purchasing managers, as well as patterns in household use and country-level adoption. Google says the data shows that India’s arts, design and media occupations account for 19% of work-related AI use, or 1.6 times the global average. In the United States, computer and mathematical occupations account for 30% of work-related AI use, twice the share found elsewhere.
Regional differences also appear in manual and technical work. Google reports that equipment diagnostics and troubleshooting make up 7% of work-related AI use in Brazil and Germany, compared with 4% in Japan. Brazil and the United Arab Emirates also show higher adoption than their gross domestic product per capita would predict.
Scientists use general and specialized models together
The accompanying AI in Science: Early Insights paper combines three sources: Gemini interaction data, an inventory of more than 2,600 specialized scientific AI models, and a survey of 637 active researchers in the United States and United Kingdom.
The research finds that scientific occupations use AI more heavily than most other occupations. In the United States, life, physical and social science roles are about 2.7 times more likely to use AI than the employment baseline. Nearly half of the surveyed researchers say AI forms part of their daily workflow.
Researchers use general-purpose language models and specialized systems for different jobs. Gemini-style language models appear across coding, statistical analysis, literature review, troubleshooting and manuscript preparation. Specialized models are more common in health and life sciences, where they support tasks such as predicting disease outcomes, designing molecular constructs and running simulations.
Google’s researchers describe the two model categories as complements rather than substitutes. The paper’s inventory covers 2,690 notable scientific models linked to publications and official code repositories, while the wider model search identified more than 5,500 systems before the study narrowed its analysis.
Seven hours saved, but experiments still set the pace
About three-quarters of surveyed scientists report saving time with AI. The average saving is just below seven hours per week, and researchers say they mostly put that time back into research. Roughly eight in ten report higher laboratory output over the past three years, while 89% expect further increases.
The gains do not remove the physical and organizational limits of scientific work. More than four in ten scientists say their primary constraint has shifted downstream into laboratory execution, clinical validation or field data collection. Forty-one percent report a growing backlog of untested hypotheses.
Checking AI output also absorbs a substantial share of the time saved. Among scientists who report time savings, 89% spend more than one-tenth of that time verifying AI results, and 46% spend more than one-quarter. The survey also finds a risk preference split: 49% say AI pushes them toward safer, more incremental projects, while 28% say it helps them pursue riskier questions.
A useful dataset with clear limits
The study is an early measurement rather than a full account of scientific AI use. The Gemini analysis isolates about 360,000 interactions judged likely to belong to scientific workflows from the broader 15 million-interaction ATLAS sample. Google says that classification cannot identify users directly and may miss routine scientific tasks that are difficult to distinguish from general technical work.
The survey itself is not representative of all scientists. It covers researchers in the United States and United Kingdom and was conducted between July 27 and August 11, 2026, through a third-party online panel. The specialized-model inventory is also described by the authors as a work in progress rather than a complete registry.
Google’s new ATLAS interface makes the adoption data easier to inspect, but the science study points to a harder question than usage rates: whether faster analysis and writing can be matched by enough lab capacity, clinical testing and independent verification to turn more hypotheses into tested results.