The most comprehensive annual report on the state of AI is out, and it tells a nuanced story: AI is transforming science at unprecedented speed — but human expertise remains irreplaceable for complex tasks.
The Stanford HAI AI Index 2026, released this week, tracks the explosive growth of AI in the natural sciences — and the gap between hype and reality.
The Numbers: A 30x Explosion
Since 2010, the number of scientific publications mentioning AI has grown by a factor of almost 30. In 2025 alone, more than 80,000 papers in the natural sciences referenced AI — a 26% increase over 2024.
The Reality Check
Despite the hype around autonomous AI agents, the best AI systems still score roughly half as well as human PhD specialists on complex, multistep scientific workflows.
As Yolanda Gil, the report lead researcher at USC, put it: "Agents are wonderful, but we are still far from a place where we understand how to use them effectively."
The Productivity Paradox
There isn't yet strong evidence that AI is improving scientists' productivity in measurable ways. But scientists "can't live without it."
Science Foundation Models
This year saw the emergence of science foundation models — AI systems trained on massive domain-specific datasets. The first astronomy foundation model, AION-1, was trained on over 200 million celestial objects.
What This Means for Enterprise AI
- AI is a tool, not a replacement
- Adoption does not equal productivity
- Domain-specific AI outperforms general AI
At Apex Aion AI, these findings validate our approach: we build domain-specific AI solutions that augment human expertise rather than replace it.
Sources: Stanford HAI AI Index 2026 | Nature