Twitch has ignited controversy with its new generative AI training system, which utilizes streamer content for Amazon's AI models. The system defaults to an opt-in setting, meaning all streamers' content is used for training unless they actively choose to opt out. This decision has drawn significant criticism and has led to an explosion of activity on Twitch's user feedback forums.
While rumors of an AI training program had circulated for weeks, Twitch confirmed the system through a support announcement on X, detailing an opt-out setting. The default opt-in nature of the feature has been met with widespread disapproval, with many users on X either explaining how to disable the setting or expressing their frustration.
A Twitch User Voice thread, initially posted in May, calling for AI features to be optional and opt-out by default, has garnered nearly 14,000 votes and grown to 228 pages. Mike Minton, Twitch's chief product officer, acknowledged the controversy during a livestream, stating that the system is opt-in by default because "If it was opt-in, nobody would opt in." This admission has been seen as a candid, if unflattering, confirmation of the platform's awareness of creator sentiment.
Opting out of the generative AI training is relatively straightforward. Streamers can navigate to the "Security and Privacy" tab in their Twitch settings and locate the "Training for Generative AI" toggle. However, critics argue that Twitch has not made sufficient efforts to ensure widespread awareness of this option. The announcement of the feature through a post on the Twitch Support channel, rather than a formal announcement on the Twitch Blog, has been cited as a low-visibility approach.
Further compounding the confusion, a response from Twitch to a post by streamer Zach Bussey stated, "Twitch is not training GenAI models on streamer content." However, Twitch's own FAQ on account settings clarifies that "Generative AI content models are AI tools that can help create or synthesize content outputs like text or images when given instructions or prompts because they are also trained on content. When you allow training, that means that your content may be used for future Gen AI model improvements." The FAQ provides an example where audio content might refine models for speech-to-text, improving captions not only on Twitch but across Amazon.
Twitch's explanation suggests a distinction between Twitch training models and Amazon training models, a difference that appears negligible given Twitch's ownership by Amazon. The platform has also noted that opting out of generative AI training does not exclude users from all AI or machine learning applications, as data can still be used for features like auto-generated captions, though this data is not retained for training new content-generating models.
This latest development adds to Twitch's history of contentious decisions. Industry professionals are concerned about the broader implications. Mike Futter, co-founder and analyst at game consultancy F-Squared, highlighted potential consequences for game developers, questioning whether Twitch considers itself entitled to feed developers' game content into AI training if a creator streams that game and is opted in. Futter stated that many studios will likely consult their lawyers, calling the move a "clear and present danger to our businesses and our creative work" and "extremely hostile" during a period when the industry requires support. He added that for studios staunchly anti-AI, this poses a significant threat to their creative output and that a critical mass of studios and creators will need to voice their opposition.
Given Twitch's history of reversing course due to backlash, there is a possibility the feature could be altered or removed. However, Amazon's strong commitment to generative AI suggests the feature may persist. The situation remains a significant point of contention for Twitch, with the outcome uncertain.