AI ethics after generative AI: from consequences to effects

Abstract

Large language model (LLM)-based applications are becoming increasingly integrated into everyday practices of communication, learning, and creativity. Their widespread adoption has intensified debates in AI ethics concerning how their societal significance should be understood and evaluated. Existing approaches to AI ethics have developed important concepts and governance frameworks for evaluating the consequences of AI systems, particularly in relation to their design, deployment, and identifiable impacts. These approaches have proven indispensable for analyzing harms, assigning responsibility, and guiding governance. However, the widespread incorporation of LLMs into everyday practices also raises questions about more gradual and cumulative transformations that emerge through repeated human–LLM interaction. Drawing on John Dewey’s pragmatist account of inquiry, habit formation, and moral reconstruction, we distinguish between consequences, identifiable outcomes that can be evaluated within existing normative and regulatory frameworks, and effects, the cumulative transformations of the habits and conditions through which inquiry, judgment, and action are organized over time. We argue that interactions with LLM-based applications give rise to both consequences and effects, and that these should be understood as complementary rather than competing analytical perspectives. Building on this distinction, we use Dewey’s conception of habits to examine how repeated engagement with LLMs may contribute to transformations in three domains: epistemic inquiry, affective-relational self-understanding, and cultural meaning-making. Rather than proposing an alternative to existing AI ethics, we argue that attending to effects complements current approaches by directing ethical inquiry towards the cumulative transformations that emerge through everyday human–LLM interaction.