Twitch has confirmed that it has been using livestreamed content to train generative AI models for its parent company, Amazon. The admission came with the announcement of an opt-out setting, which is buried deep within the platform's security settings and was enabled by default. This move has ignited significant backlash from streamers and viewers who are concerned about their content and personal data being used without explicit consent.
During a stream to explain the policy, Twitch's chief product officer Mike Minton and head of community Mary Kish acknowledged the use of streamed video for AI training. However, Minton admitted to not knowing if personal user data had been included in the scraping process or the full extent of Amazon's AI model training activities. The platform did not prompt users with notifications about this data usage upon login, and the opt-out toggle required manual navigation through account settings.
While Twitch also offers other "AI-supported" tools, the primary concern revolves around the use of generative AI, which streamers fear could lead to copyright issues. Currently, the relationship between streamers and game publishers relies on implied or explicit permission for content creation, often for mutual marketing benefit. Amazon's approach appears to disregard this symbiotic relationship, with the potential to scrape game footage without equivalent benefit to rights holders. Concerns have also been raised regarding agreements with unionized voice actors whose performances may be protected from AI training.
In a candid statement during the explanatory stream, Minton acknowledged the likely outcome of a user-facing choice: "If it's opt-in, nobody would opt-in. That's the honest answer." This statement underscores the platform's understanding that widespread consent for such data usage would be unlikely.