Understanding Toxicity Labels
Sara discusses the complexities of labeling toxicity, emphasizing the challenges of annotator bias and the lack of comprehensive data. She highlights that even with perfect human calibration, the absence of reliable labels in large datasets poses a significant hurdle. The conversation also touches on the intricate relationship between fairness and model behavior, noting how historical biases complicate the development of equitable AI systems.In this clip
From this podcast
Machine Learning Street Talk (MLST)
#92 - SARA HOOKER - Fairness, Interpretability, Language Models
Related Questions
Is data quality overlooked in machine learning as discussed in the episode AI For Good - Detecting Harmful Content at Scale // Matar Haller // #245 and the clip Data Labeling Insights?
Are there biases in AI as discussed in the episode More Language, Less Labeling with Kate Saenko - #580 and the clip Image Captioning Advances?