Study says AI boosts output but narrows the range of science topics

In short: A large study suggests that while AI helps individual scientists publish more, it may also push the whole research community to study a narrower set of topics.

What's going on

Researchers led by James Evans at the University of Chicago looked at 41.3 million research papers in a January 2026 study published in Nature. They compared scientists who use AI tools with those who do not. The study found AI users published 3.02 times more papers and received 4.85 times more citations.

The study also found career advantages for AI users. They became research leaders about 1.4 years earlier than non-users. In other words, AI looked like a productivity boost for individuals.

But the same study reports a downside when many scientists adopt AI at once. The overall range of scientific topics being studied fell by 4.63%, and engagement between scientists dropped by 22%. The authors argue this happens because researchers move toward “data-rich” problems, meaning areas where lots of clean, organized information already exists for AI to learn from, like a well stocked library.

Evans also warns that AI can favor “optimization over exploration.” That means it can encourage many teams to keep improving results on popular, measurable tasks, rather than taking risks on new ideas. The study describes “lonely crowds,” where a topic gets lots of attention, but researchers interact less, and work overlaps.

What to watch

The researchers suggest policy changes that reward work in “data-poor” areas and encourage AI tools that help gather new data, not just analyze existing datasets. If funders and institutions do not adjust incentives, science could drift further toward questions that are easiest to measure, not necessarily the most important.

Source: Financial Times