mlsanityData to 14 Oct 2026

BiasX: "Thinking Slow" in Toxic Content Moderation with Explanations of Implied Social Biases

arXiv 2305.13589 ↗First posted 23 May 2023cs.CLcited by 7

Not on the map. Why?

Abstract (arXiv)

Toxicity annotators and content moderators often default to mental shortcuts when making decisions. This can lead to subtle toxicity being missed, and seemingly toxic but harmless content being over-detected. We introduce BiasX, a framework that enhances content moderation setups with free-text explanations of statements' implied social biases, and explore its effectiveness through a large-scale crowdsourced user study. We show that indeed, participants substantially benefit from explanations for correctly identifying subtly (non-)toxic content. The quality of explanations is critical: imperfect machine-generated explanations (+2.4% on hard toxic examples) help less compared to expert-written human explanations (+7.2%). Our results showcase the promise of using free-text explanations to encourage more thoughtful toxicity moderation.

Key facts. Cited by 7 papers in the map (data through 14 Oct 2026). Builds on 21 of its 44 references that are in the map. First posted on arXiv 23 May 2023.

Builds on

21 of 44 references.

References · 21

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  3. 2020DeBERTa: Decoding-enhanced BERT with Disentangled Attentioncited by 2,181
  4. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,255
  5. 2017Explanation in Artificial Intelligence: Insights from the Social Sciencescited by 1,206
  6. 2020StereoSet: Measuring stereotypical bias in pretrained language modelscited by 924
  7. 2018e-SNLI: Natural Language Inference with Natural Language Explanationscited by 468
  8. 2020Does the Whole Exceed its Parts? The Effect of AI Explanations on Complementary Team Performancecited by 399
  9. 2019Social Bias Frames: Reasoning about Social and Power Implications of Languagecited by 380
  10. 2021Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detectioncited by 264
  11. 2019Build it Break it Fix it for Dialogue Safety: Robustness from Adversarial Human Attackcited by 176
  12. 2021MultiVerS: Improving scientific claim verification with weak supervision and full-document contextcited by 143
9 more
  1. 2022Explanations Can Reduce Overreliance on AI Systems During Decision-Makingcited by 111
  2. 2017Measuring the Reliability of Hate Speech Annotations: The Case of the European Refugee Crisiscited by 96
  3. 2020"Why is 'Chicago' deceptive?" Towards Building Model-Driven Tutorials for Humanscited by 86
  4. 2022Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderationcited by 67
  5. 2021Misinfo Reaction Frames: Reasoning about Readers' Reactions to News Headlinescited by 32
  6. 2023Selective Explanations: Leveraging Human Input to Align Explainable AIcited by 29
  7. 2022Listening to Affected Communities to Define Extreme Speech: Dataset and Experimentscited by 22
  8. 2022A Trade-off-centered Framework of Content Moderationcited by 11
  9. 2022Cascading Biases: Investigating the Effect of Heuristic Annotation Strategies on Data and Modelscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 2; 2025: 3; 2026: 2.

7 papers.

  1. 2024AEGIS: Online Adaptive AI Content Safety Moderation with Ensemble of LLM Expertscited by 136
  2. 2025Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrailscited by 132
  3. 2025Lost in Moderation: How Commercial Content Moderation APIs Over- and Under-Moderate Group-Targeted Hate Speech and Linguistic Variationscited by 13
  4. 2024Tox-BART: Leveraging Toxicity Attributes for Explanation Generation of Implicit Hate Speechcited by 11
  5. 2025Real-World Gaps in AI Governance Researchcited by 5
  6. 2026ChartAttack: Testing the Vulnerability of LLMs to Malicious Prompting in Chart Generationcited by 0
  7. 2026NaiLIA: Multimodal Nail Design Retrieval Based on Dense Intent Descriptions and Palette Queriescited by 0