Scientists Embrace AI But Humans Still Outperform Best Bots

Scientists are falling in love with artificial intelligence, and the numbers prove it.

A new report from Stanford University shows that mentions of AI in natural sciences papers grew nearly 30 times from 2010 to 2025. Today, between 6% and 9% of publications in any given field discuss AI in some way.

More than 80,000 papers, preprints, and publications mentioned AI in 2025 alone. That's 26% more than the previous year.

Earth sciences led the pack with 9% of papers mentioning AI. Physical sciences had the highest raw numbers, with 33,000 publications.

"Scientists have really embraced this AI era," says computer scientist Yolanda Gil, who led this year's Artificial Intelligence Index Report. She notes that researchers are now so dependent on these tools that removing them would cause a riot.

The past year brought exciting new AI platforms built specifically for science. AION-1 became the first foundation model for astronomy, trained on more than 200 million celestial objects to help classify galaxies and estimate their properties.

The Bright Side

Despite concerns about AI's rapid adoption, there's good news for anyone worried about being replaced by robots.

When researchers tested the best AI agents against human specialists with PhDs on complex, multistep tasks, the humans won decisively. AI agents scored roughly half as well as their human counterparts.

"Agents are wonderful, but we are still far from a place where we understand how to use them effectively," Gil explains. These AI assistants still struggle with the kind of complex scientific workflows that researchers handle daily.

The explosion in AI tools hasn't yet proven to boost scientific productivity in measurable ways. Studies remain limited, and debate continues about whether this growth is meaningful or simply trendy.

Some scientists worry the shift is happening too fast. Computer scientist Arvind Narayanan from Princeton University suggests that scientific norms haven't had time to adjust, which may have hurt research quality.

But researchers keep adopting these tools anyway. While hard evidence of improved productivity is scarce, scientists clearly find value in AI assistance for their work.

The technology has also reached beyond research labs. The world's first operational AI-powered weather forecasts went live in early 2025, bringing cutting-edge science directly to everyday life.

The future of science will likely blend human expertise with AI assistance, playing to each side's strengths.