Practical Philosophy Seminar: Hannah Clausen

Recent advances in language modeling have contributed to a growing emphasis on machine ethics, including researching and assessing moral judgment by large language models (LLMs). This talk will provide a critical overview of the state of research with a special focus on existing data resources and approaches. We will explore available data specifically designed for LLMs as well as repurposed data designed for humans and how these are applied to LLMs. While the former are marked by an over-reliance on two particular datasets, reinforcing existing shortcomings, the latter arguably contain a critical element of anthropomorphism. To advance the field, we recommend adopting a shared and consistent terminology and creating carefully curated datasets based on interdisciplinary work. To ensure a clear delineation from normative approaches, we propose focusing on the assessment of moral consistency and certainty of LLMs as an effective and descriptive approach to evaluate the performance on moral judgment.