Schrödinger’s chat: can AI and scientists co-exist?

OpenAI recently dumped hundreds of solutions to unsolved mathematics problems into the public domain, and mathematicians are not happy about it. The computer scientist Scott Aaronson called it the “math apocalypse.” The sheer quantity of proofs—covering a huge array of problems no human had previously solved—showed not only that AI could do work beyond human capability, but that it could do so in a blink of an eye. And the reaction from mathematicians reveals something deeper about how AI is beginning to reshape the scientific enterprise itself.

The state of AI in science

AI is already embedded across many scientific domains. As Alex Hearn, The Economist’s AI writer, explains, the applications are dizzying. In biology, AI models can read DNA the way chatbots read text—predicting genomes, building genomes from scratch, and understanding how genetics works. Specialized models design proteins in academic labs and help develop better drugs. AI controls plasma in fusion experiments, potentially bringing that energy source closer. Weather forecasting has been dramatically improved. And then there is the recursive loop: AI scientists are using AI to make better AI, accelerating the very process that underpins fears of runaway self-improvement.