Understanding Annotator Safety Policy with Interpretability
Safety policies define what constitutes safe and unsafe AI outputs, guiding data annotation and model development. However, annotation disagreement is pervasive and can stem from multiple sources such as operational failures (annotators misunderstand or misexecute the task), policy ambiguity (policy wording leaves room for interpretation), or value pluralism (different annotators hold different perspectives on safety). Distinguishing these sources
By Alex Oesterling, Donghao Ren, Yannick Assogba, Dominik Moritz, Sunnie S. Y. Kim, Leon Gatys, Fred Ho