The Use of Generative AI for Science of Science and Higher Education Studies. A scientific workshop at the Università della Svizzera italiana
Institute of Communication and Public Policy
The development of generative AI language models, such as ChatGPT and Microsoft Copilot, represents one of the most significant technological change of the recent years. The development of the models has sparked wide interest in the research community, but also in various application fields of research results. The versatility of such models makes them suitable for a wide range of scientific research and its application. The literature highlights both risks and challenges of the use of generative AI including increased efficiency and accuracy, but also security concerns, capability limitations, and ethical and data privacy issues
The workshop, organized by Professor Benedetto Lepori and Pinar Eldemir – Institute of Communication and Public Policy – and supported by the Swiss National Science Foundation, has gathered about fifty international experts on the use of Generative AI for the study of science and higher education. The presentations have illustrated the potential and limitations of GenAI for applications such as comparing universities, identifying and labeling research topics, and retrieving data on academic careers. The participants agreed on the fact that GenAI has the potential to foster research in this area, but also that proper reporting of usage is essential to ensure robustness and replicability of results.
The keynote speech by Dr. Kayvan Koshua of the University of Wolverhampton, moderated by the pro-rector for innovation Prof. Luca Gambardella, addressed one of the most contested application of GenAI, i.e. usage for evaluating the quality of research. Empirical data showed that GenAI evaluations are to some extent correlated with peer assessment, but contain also important sources of biases. Hence, GenAI can at be used as an additional tool to support expert assessment rather than replacing it.
It is planned to published selected results in a special issue of a leading scientometric journal.
