Understanding in the Age of AI: Preserving Scientific Achievement in AI-Assisted Research Production

Over the past several years, the rapid integration of AI tools into scientific research has dramatically accelerated the pace of inquiry and, by extension, increased the volume of published work. This surge in research output appears to advance scientific progress. At the same time, however, it presents a serious puzzle. Historically, science has been driven not merely by data accumulation but by the pursuit of genuine scientific understanding—a pursuit characterised by features that explain why we take scientific understanding to be a genuine achievement: for instance, a skilled grasp of explanatory and coherence-making relations, conceptual clarity, and rigorous interpretation. As AI systems demonstrate increasingly sophisticated capabilities in processing and analysing vast datasets—typically in ways that are opaque not only to scientific researchers but even to the algorithms’ designers—the question arises: to what extent does AI-assisted research contribute to the depth of understanding that has paradigmatically defined genuine scientific progress, as opposed to something epistemically ‘cheaper’? This question calls for careful examination of both its epistemological and ethical dimensions. The epistemological challenge concerns how we evaluate and characterise the kind of epistemic gain that is achieved by papers published as a result of AI-aided inquiry. Pertinent sub-questions here are: What constitutes genuine scientific understanding, and how does it differ from mere data processing or pattern recognition? Can AI-assisted research foster the same depth of insight and achievements in grasping explanatory connections that has historically characterised scientific breakthroughs? Meanwhile, the ethical dimension focuses on researchers’ responsibilities when publishing AI-assisted work—particularly when their own understanding of the underlying phenomena or methodologies may be incomplete due to AI outsourcing. This project will systematically investigate these challenges through a rigorous analysis of how AI-aided scientific inquiry relates to traditional scientific understanding. It will develop a new and much-needed framework for evaluating the quality of understanding in AI-assisted research and establish guidelines for responsible publication practices. The research will be conducted at Cogito, the largest and broadest epistemology research centre worldwide (over 60+ members). The project will be mentored by Prof. Christoph Kelp, a world-leading expert in the epistemology of scientific understanding and inquiry, as well as in ‘extended’ epistemology – the pursuit of knowledge by means of technological scaffolding. This research will also be led with the support of two large-scale scientific organisations at the European level: the Initiative for Science in Europe (ISE) and the Young Academy of Europe (YAE). The project will unfold over 24 months through three core work packages—one epistemological, one ethical, and one applied which draws from the results of the first two—culminating in a monograph titled AI Inquiry, Scientific Understanding and Progress, alongside peer-reviewed articles, conference presentations, and knowledge-exchange activities, including scientific policy guidance in the form of an AI-publishing toolkit. This guidance will be disseminated in Year 3 through initiatives at the YAE and ISE. My unique research profile—which combines expertise in epistemology, ethics, and philosophy of technology—positions me ideally to conduct a systematic investigation of these challenges. More generally, this fellowship provides the perfect opportunity and career support for me to develop my initial findings into an ambitious, cohesive project with both theoretical depth through dedicated research time for the monograph and meaningful policy impact.