Toxicity in AI-Generated Video Streams and Chats: How do Viewers React to Harmful Messages and Stereotypes?
Summary
This thesis examines the relationship between toxicity in streams and chats of AI-generated video streams. Specifically, YouTube streams consisting of AI-generated episodes of Family Guy and Spongebob Squarepants were analysed.
Utilizing the Perspective API, a free tool developed to detect toxicity in online spaces, both streams and chats of these YouTube channels were analyzed in terms of toxicity. Whenever there was a noticeable peak in toxicity, a qualitative analysis then took a closer look at the 1-minute intervals right around those peaks. A thematic analysis of these peaks provided insights into the causes and nature of toxicity in the communities around these YouTube channels and their streams.