Treat AI as a Debate Partner, Not an Answer Machine, Oxford Researcher Says

Generative artificial intelligence has become deeply embedded in academic environments, with usage rates among students reaching unprecedented levels. According to a March survey conducted in the United Kingdom, 94% of undergraduates utilize these technologies to complete assessed coursework. The shift is similarly evident in professional research; data indicates that over half of scientific papers published in 2025 displayed linguistic markers characteristic of large language models (LLMs), a sharp increase from approximately 10% in 2023.

This rapid integration has triggered concern among educators and researchers who fear the proliferation of AI could degrade critical thinking skills and compromise the integrity of the scientific record. However, Marc Schmitt, a researcher at the University of Oxford, UK, argues that this alarmist perspective misidentifies the core issue. Schmitt, whose work examines how AI influences decision-making in financial markets, contends that the technology itself is not the threat.

Instead, Schmitt suggests that universities should focus on teaching students how to effectively challenge AI systems. He draws a parallel between AI models and intellectual sparring partners, noting that access to a capable counterpart who can rigorously test reasoning and evidence is historically rare. Typically, such rigorous debate requires a skilled teacher or a sharp academic colleague, resources that are not always available. LLMs, by contrast, offer widespread, on-demand access to an entity capable of engaging in real-time argumentation.

The prevailing narrative often portrays AI as a danger to human cognition, with fears that students will cease thinking independently and researchers will outsource scientific judgment to algorithms. Schmitt acknowledges these risks but argues they stem from passive usage—treating the machine as an oracle that provides finished answers—rather than from active engagement. He emphasizes that scientific ideas rarely emerge fully formed; instead, they develop through iterative conversation and the clear articulation of concepts.

By utilizing AI as a tool for active intellectual exchange, researchers and students can simulate the collaborative processes essential to scientific discovery. This approach allows for continuous dialogue with AI models, complete with searchable transcripts, thereby distributing intellectual firepower more broadly across the academic community.