Half of scientists now reach for AI every day, Google's global data shows

Key points

• Nearly half of surveyed scientists in the US and UK now use some form of AI every day, according to new research from Google, Google DeepMind and MIT FutureTech.
• Scientists report saving just under seven hours a week thanks to AI tools, but a backlog of untested hypotheses is building up.
• Arts, design and media jobs account for 19% of work-related AI use in India, 1.6 times the global average.
• Computer and maths jobs account for 30% of work-related AI use in the United States, double the share seen elsewhere.
• Brazil and the United Arab Emirates are adopting AI faster than their income levels would predict.

Google has opened a much bigger window into who actually uses AI on the job. The data comes from ATLAS, its long-running project tracking AI in the global economy, now paired with an interactive tool anyone can explore.

AI isn't spreading evenly. It's clustering in specific jobs, specific countries, and most strikingly, in science.

Who is using AI most at work?

Scientists, by a clear margin. A study from Google, Google DeepMind in collaboration with MIT FutureTech, drawing on ATLAS data, found nearly half of the 600-plus researchers surveyed in the US and UK now use some form of AI every single working day. That's a higher daily rate than most other professions ATLAS tracks.

The study covered 2,600 specialised AI models and used a new classification from MIT FutureTech that maps the actual tasks scientists perform. Two kinds of AI are doing the work, and they don't overlap much. Large language models, the technology behind chatbots like Google's Gemini, get used across nearly every field for reading and summarising. Specialised models, built for one narrow job like predicting a protein shape, dominate in health, life sciences and domain-specific simulation.

Is AI actually speeding science up?

Partly. Scientists told researchers they save just under seven hours a week using AI tools, a serious chunk of a working week freed for actual research.

But those saved hours aren't yet turning into a flood of new discoveries. Scientists spend real time checking whether AI outputs are correct, and a queue of untested ideas is piling up behind the parts of research AI can't handle: physical experiments, clinical validation, the slow grind of getting samples processed. We've tracked AI in science across eight stories since July, and the pattern that keeps surfacing is exactly this one.

Here's my read after covering this beat: the bottleneck has moved, not disappeared. AI drafts the hypothesis in an afternoon; the ethics board and the laboratory bench still take months. Productivity gains will show up on scientists' calendars long before they show up in journals, and labs that don't redesign their whole workflow will capture only a fraction of what's on offer.

How does AI use differ around the world?

ATLAS breaks work-related AI use down by country and occupation, and the regional splits are sharp.

In wealthier OECD countries, computer, maths, business and finance roles lead. Lower-income countries tell a different story: the top users are office admin staff, creative workers and teachers, which says something about what AI actually does for people when technical infrastructure isn't the starting point.

What happens next?

Google is treating ATLAS as a long-term project and says it'll keep working with academic partners on new research. The interactive tool is open access, so anyone from a policy analyst to a curious electrician can look up their own occupation and country.

For everyday readers, the practical picture is smaller than the headlines suggest. Creative, technical and scientific workers probably have colleagues using these tools daily already. Hands-on physical jobs are seeing AI too, mostly as a diagnostic assistant rather than a replacement, and more slowly than the science headlines imply.