About the project
This project will take a cross-disciplinary approach to the design and use of benchmarks for evaluating ethical and moral reasoning in Large Language Models (LLMs). Combining approaches from experimental philosophy and natural language processing, the project will address research questions like the following:
- How can we best design benchmarks for evaluating ethical reasoning in LLMs? What are the strengths and weaknesses of current approaches?
- How do the different stages in the LLM training pipeline or different sources of training data impact the knowledge learned about ethics?
- How robust and consistent are the moral judgments of a given LLM? This could be investigated with respect to, e.g., word choice or framing, and cross-lingual or cross-cultural effects.
This project forms part of a collaboration between researchers from several departments at the University of Oslo, including the Language Technology Group at the Department of Informatics, the Practical Philosophy Group at the Department of Philosophy, Classics, History of Art and Ideas, and the Department of Biostatistics at the Institute of Basic Medical Sciences.
Project team
PhD Fellow Hannah Clausen, University of Oslo, Department of Informatics
Professor Erik Velldal, University of Oslo, Department of Informatics
Professor Anna Smajdor, University of Oslo, Department of Philosophy, Classics, History of Art and Ideas
Associate Professor Andrey Kutuzov, University of Oslo, Department of Informatics