People trust scientists with AI. Keep the scientists in charge.

Here is a welcome piece of news: people have not given up on the idea that AI can help science. Nature's coverage points to confidence in the people doing the work, not simply excitement about the machinery. That distinction gives research institutions something valuable to build on. A Reuters Institute survey across nine countries found especially positive expectations for scientists and healthcare professionals using AI. It measures public perceptions, not the safety or effectiveness of a particular system. The underlying report appeared in October; this is fresh coverage of existing survey evidence, not a new experiment conducted today. A separate UK survey makes the picture more interesting. Support for research involving generative AI was lower than support for research generally, while 83–85% backed approaches retaining substantial human input. These surveys ask different questions in different populations. They should not be forced into a contradiction, or merged into one global percentage. What I find encouraging is the room for a constructive bargain. Researchers can use better tools while remaining answerable for the choices that matter: which problem to study, whether a result survives scrutiny, and what uncertainty to communicate. Explaining those decisions respects public trust instead of spending it as if it were unlimited. It would also be unfair to turn the finding into a demand that scientists manually redo everything. The useful question is where expert judgment changes the outcome. A laboratory that can explain what the model contributed, what people examined and what remains uncertain offers a stronger invitation to trust than either unqualified enthusiasm or blanket rejection.