Generative AI tools like ChatGPT and GitHub Copilot are now writing research software, and no one has yet worked out what that means for the integrity of science. This fellowship tackles a gap: research software engineers (RSEs) write the code that processes data, runs simulations, and produces results in fields from climate modelling to drug discovery. But as scientists increasingly rely on AI-generated code, no framework exists to judge whether that code is robust, transparent, or reproducible. The project will first clarify how RSEs actually contribute to scientific knowledge—what counts as a meaningful software contribution—then build a practical assessment framework for AI-generated research code, and finally produce evidence-based policy guidance for funders and institutions. If successful, this work could reshape how research institutions manage software quality, preserve professional coding skills, and maintain academic authorship standards. It will also produce training workshops for RSE practitioners. The immediate impact is on the infrastructure of science itself—the code that quietly underpins most modern research—rather than on everyday life directly. Without such guidance, the growing use of generative AI risks eroding the reproducibility that scientific results depend on.
View original technical description
Software is an increasingly important part of modern science. With it, scientists are able to process data, simulate complex systems, and visualise their results. This fellowship will explore how generative AI tools, such as ChatGPT and GitHub Copilot, are reshaping the production of reseaerch software, and provides guidance on how software engineering scientists can use them in a way which promotes academic integrity and ensures robust scientific research. The rise of generative AI presents both an opportunity and a challenge for Research Software Engineering (RSE). On one hand, these tools could lower barriers to entry, making science more inclusive and improving productivity. On the other hand, they raise urgent questions about academic authorship, loss of professional skills, and the trustworthiness of AI-generated output This fellowship investigates the emerging role of generative AI in science from three key angles: Assessing the role RSE plays in knowledge creation: To evaluate the impact of AI on RSE, we first need to understand how RSE contributes to science. What counts as a meaningful software contribution? How do RSE practices shape the reliability of scientific results? This part of the fellowship draws on insights from philosophy and science studies to offer a clearer picture of how scientific knowledge is made—and the specific roles RSE plays in that process. 2. Evaluating AI-generated research software: The second focus is more practical: testing what generative AI tools can actually do. Can they write software that is robust, transparent, and reproducible? Or do they introduce hidden risks? This work will involve building a framework for assessing the quality of AI-generated research code and identifying the characteristics that make it distinct from generic software. 3. Developing guidance on the introduction of generative-AI into RSE practice: the final stages of this fellowship will explore how we turn these insights into actionable research policy. I will develop training workshops for RSE practicioners, enabling them to use generative AI in a way which retains important professional skills and preserves academic integrity; and will produce evidence based policy guidance for research institutions and funders. Together, these studies will provide a deeper understanding of how generative AI is changing the landscape of research software, and how we can ensure that that change improves science as a whole. Given the rapid growth of generative AI and the increasing centrality of RSE to modern science, this is a critical moment. If we fail to act, we risk undermining the foundations of reproducible and trustworthy research. This fellowship will ensure that decisions about the use of generative AI in science are grounded in evidence, and guided by a commitment to integrity, inclusion, and excellence.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know