Using AI to identify moral categories in ancient taboo lists

The past two decades have seen an emergence of sophisticated theories for understanding the multiplicity of concerns that characterize human morality. While these frameworks aspire to identify universal moral foundations, it is questionable whether they are adequately representing the full range of moral concerns present in non-Western societies. The present research sought to evaluate the potential of a bottom-up (data-driven) methodology for identifying moral categories based on textual evidence from traditional societies. In particular, our point of departure was the existence of ‘taboo lists’ – inventories of behaviors incurring divine punishment – that have been found in ancient (Mesopotamia, Egypt, Israel) and contemporary ethnographic sources. The extrapolation of general categories based on these lists was carried out by artificial intelligence (AI) models. The performance of this new inventory of moral categories was evaluated by human respondents and additional AI models in comparison with existing frameworks. The results of these experiments demonstrated a human-like competence of AI models to both generate and use moral categories. More importantly, the success of the novel inventory confirmed the viability of a bottom-up approach for the study of human morality.