Latest / AI Ethics with Fexingo: Bias, Safety, and Responsible Artificial Intelligence
How AI Models Learn Bias From YouTube Comments
Lucas and Luna explore how YouTube's comment sections are training AI models to absorb and amplify toxic language, political polarization, and racial bias. They focus on a 2025 study from the University of Washington that analyzed 10 million YouTube comments and found that AI models trained on them picked up hate speech, conspiracy theories, and gendered insults at rates far higher than models trained on curated text datasets. The hosts discuss why YouTube comments are uniquely dangerous training data — unmoderated, algorithmically boosted, and filled with bot-driven content — and what it…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.