State-of-the-art conversational AI exhibits a level of sophistication that promises to have profound impacts on many aspects of daily life, including how people seek information, create content, and find emotional support. It has also shown a propensity for bias, offensive language, and false information. Consequently, understanding and moderating safety risks posed by interacting with AI chatbots is a critical technical and social challenge. Safety annotation is an intrinsically subjective task, where many factors—often intersecting—determine why people may express different opinions on whether a conversation is safe. We apply Bayesian multilevel models to surface factors that best predict rater behavior to a dataset of 101,286 annotations of conversations between humans and an AI chatbot, stratified by rater gender, age, race/ethnicity, and education level. We show that intersectional effects involving these factors play significant roles in validating safety in conversational AI data. For example, race/ethnicity and gender show strong intersectional effects, particularly among South Asian and East Asian women. We also find that conversational degree of harm impacts raters of all race/ethnicity groups, but that Indigenous and South Asian raters are particularly sensitive. Finally, we discover that the effect of education is uniquely intersectional for Indigenous raters. Our results underscore the utility of multilevel frameworks for uncovering underrepresented social perspectives.

How people interpret content is deeply influenced by their socio-cultural backgrounds and lived experiences. This is especially crucial when evaluating AI systems for safety, where accounting for such diversity in interpretations and potential impacts on human users will make them both more successful and inclusive. While recent work has demonstrated the importance of diversity in human ratings that underlie AI pipelines, effective and efficient ways to incorporate diverse perspectives in human data annotation pipelines is still largely elusive. In this paper, we discuss the primary challenges faced in incorporating diversity into model evaluations, and propose a practical diversity-aware annotation approach. Using an existing dataset with highly parallel safety annotations, we take as a test case a policy that prioritizes recall of safety issues, and demonstrate that our diversity-aware approach can efficiently obtain a higher recall of safety issues flagged by minoritized rater groups without hurting overall precision.

We draw from the framework of relationality as a pathway for modeling social relations to address gaps in text classification, generally, and offensive language classification, specifically. We use minoritized language, such as queer speech, to motivate a need for understanding and modeling social relations–both among individuals and among their social communities. We then point to socio-ethical style as a research area for inferring and measuring social relations as well as propose additional questions to structure future research on operationalizing social context.

Majority voting and averaging are common approaches used to resolve annotator disagreements and derive single ground truth labels from multiple annotations. However, annotators may systematically disagree with one another, often reflecting their individual biases and values, especially in the case of subjective tasks such as detecting affect, aggression, and hate speech. Annotator disagreements may capture important nuances in such tasks that are often ignored while aggregating annotations to a single ground truth. In order to address this, we investigate the efficacy of multi-annotator models. In particular, our multi-task based approach treats predicting each annotators’ judgements as separate subtasks, while sharing a common learned representation of the task. We show that this approach yields same or better performance than aggregating labels in the data prior to training across seven different binary classification tasks. Our approach also provides a way to estimate uncertainty in predictions, which we demonstrate better correlate with annotation disagreements than traditional methods. Being able to model uncertainty is especially useful in deployment scenarios where knowing when not to make a prediction is important.

Tasks such as toxicity detection, hate speech detection, and online harassment detection have been developed for identifying interactions involving offensive speech. In this work we articulate the need for a relational understanding of offensiveness to help distinguish denotative offensive speech from offensive speech serving as a mechanism through which marginalized communities resist oppressive social norms. Using examples from the queer community, we argue that evaluations of offensive speech must focus on the impacts of language use. We call this the cynic perspective– or a characteristic of language with roots in Cynic philosophy that pertains to employing offensive speech as a practice of resistance. We also explore the degree to which NLP systems may encounter limits to modeling relational context.

A common practice in building NLP datasets, especially using crowd-sourced annotations, involves obtaining multiple annotator judgements on the same data instances, which are then flattened to produce a single “ground truth” label or score, through majority voting, averaging, or adjudication. While these approaches may be appropriate in certain annotation tasks, such aggregations overlook the socially constructed nature of human perceptions that annotations for relatively more subjective tasks are meant to capture. In particular, systematic disagreements between annotators owing to their socio-cultural backgrounds and/or lived experiences are often obfuscated through such aggregations. In this paper, we empirically demonstrate that label aggregation may introduce representational biases of individual and group perspectives. Based on this finding, we propose a set of recommendations for increased utility and transparency of datasets for downstream use cases.