When AI Helps Too Well: Transformative Inquiry and the Risk of Second-Order Drift in Ethics Research

Abstract AI tools lend themselves to ethics research because they may help improve its speed or rigour. But the better they become at assisting ethical inquiry, the greater the risk that ethics research drifts towards the kinds of reasoning AI assists best. This paper argues that the impact of large language model (LLM) assistance on ethics research may be structurally asymmetric. LLM use may be especially helpful for research tasks operating within relatively stable evaluative frameworks, such as argument mapping, conceptual clarification, or consistency checking, and may yield smaller productivity gains in forms of inquiry aimed at revising the evaluative framework itself. I call this latter transformative inquiry : inquiry whose success depends on enabling the recognition of morally salient features not fully accessible within some current evaluative framework. The paper argues that this asymmetry risks contributing to a specific second-order drift in the field of ethics, namely a gradual reorientation away from the forms of inquiry through which new moral concerns become visible. This risk is especially acute in frontier applied ethics, where existing moral categories are especially likely to be under pressure and transformative demand is correspondingly high.