Latest / AI Ethics with Fexingo: Bias, Safety, and Responsible Artificial Intelligence
How AI Models Learn Bias from Video Game Voice Chat
In this episode of AI Ethics with Fexingo, Lucas and Luna explore a surprising new source of bias in AI training data: voice chat from online video games. They discuss how millions of hours of toxic, gendered, and racially charged speech from gaming platforms are being scraped to train speech recognition and sentiment analysis models. The hosts examine a real 2025 study that found models trained on game voice data performed 22 percent worse on female and non-native English speakers. They also unpack the broader implications for AI safety when training data includes deliberate adversarial…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.