“We need a reform of what matters in science,” Larivière said. The conflation of scientific productivity with publication counts has had a distorting effect on science, causing research to gravitate toward small, tractable problems that are guaranteed to result in something publishable. AI could do great things, he said — help cure cancer, develop fusion energy — but right now it is being used to generate papers to “pad CVs.”

“Of course we need more science,” he said, “but do we need more papers?”