'AI Alignment' Encompasses Competing Technical Priorities
This study addresses the fragmentation in AI alignment research stemming from competing conceptual frameworks, which can lead interventions to have opposing effects under different alignment perspectives. By systematically analyzing three dominant alignment paradigms, the work reveals that their fundamental disagreements arise from divergent threat models and normative orientations. Through conceptual analysis, comparison of research programs, and clarification of policy–science distinctions, the paper articulates— for the first time—the internal pluralism and tensions within alignment discourse. It proposes five recommendations to improve research practices and makes a key contribution by developing a refined conceptual framework that distinguishes idealized alignment goals from empirical proxy metrics. This framework provides a clearer terminological foundation and methodological guidance for interdisciplinary communication and technical intervention in AI alignment.