Amazon Leverages Twitch Data for Generative AI Training
AI-generated image · Bay Street Wire
Parent company Amazon is utilizing streamer content by default to fuel AI models, prompting backlash over opt-out mechanics.
Amazon is now using content from the streaming platform Twitch to train its generative AI content models, according to reporting from TechCrunch and The Verge. The data being harvested includes stream recordings, VODs, clips, channel pictures, text, and stream chats.
Twitch has implemented the change as an opt-out system, meaning creators are enrolled by default. During a live stream with nearly 3,000 users, Twitch Chief Product Officer Mike Minton addressed the decision to avoid an opt-in model, stating, "If this was opt-in, nobody would opt in. That’s honestly the answer."
Streamers have expressed concern that their voices and likenesses are being used without explicit consent. When questioned by users regarding whether content had already been utilized for training, Minton stated he did not know what Amazon had previously used.
Twitch Head of Community Mary Kish noted that the company is not unique in this practice, citing Meta's use of public Facebook and Instagram data for AI training. Kish described the inclusion of the opt-out toggle as a reaction to the community's opposition to generative AI.
According to The Verge, opting out prevents content from being used in "future training" of models designed to synthesize video, audio, images, or text. However, it does not disable "AI-supported" tools such as AutoMod, viewer discovery recommendations, or real-time sponsorship assistance. Additionally, Twitch specifies that if a user participates in a chat on another creator's stream, the training status of that chat is governed by the stream owner's preferences.