GoKawiilGenerative artificial intelligence is now woven into academic life. In the United Kingdom, a March survey found that 94% of undergraduates use the technology to complete assessed coursework (see go.nature.com/45pced3). The scientific literature shows a similar shift: more than half of papers published in 2025 show linguistic traces of large language models (LLMs), compared with about one in ten in 20231.
This trend has alarmed educators and researchers, who worry that AI models could erode critical thinking and flood the scientific record with low-quality work. Yet as a regular user of AI tools and a researcher at the University of Oxford, UK, who studies how the technology reshapes decision-making in financial markets, I think the alarmists are getting it wrong.
Gen Z scepticism towards AI is a wake-up call — universities must take it seriously
Good intellectual sparring partners are scarce. To have someone who will challenge your reasoning, hold you to your evidence and push back on your arguments in real time, you need a strong teacher, a sharp colleague or the right seminar audience. Most people never get the opportunity. What LLMs offer, for the first time, is broad access to a capable counterpart that will reason with you on demand. This redistributes intellectual firepower.
Paradoxically, the dominant story runs the other way, casting AI models as a threat to human thought. The worry is that students will stop thinking and researchers will increasingly outsource scientific judgement to a LLM. Those concerns are real, but they are aimed at the wrong object. The danger is not AI itself. It is passive use: treating the machine as a source of finished answers, when its value comes from active engagement.
Most researchers know that scientific ideas do not arrive fully formed. They emerge through conversation and repeated attempts to articulate them clearly. Scientists can now have these intellectual exchanges with AI models at any time and in most places, with a searchable transcript of the discussion available on demand.