Twitch activates data collection to train Amazon’s generative AI by default and faces criticism from content creators

The live streaming platform Twitch has faced a wave of criticism from its users and content creators, after revealing the activation of an option that allows its parent company Amazon to use their content to train generative artificial intelligence models by default. This setting allows the American technology company to collect data from creators and audiences and exploit it to develop its algorithms, while providing an option to disable it. However, the announcement has sparked widespread outrage and questions about the platform's motives in making the collection available without prior consent.

In a live stream dedicated to discussing the new feature, Mike Minton, head of products at Twitch, defended the decision by pointing out that the platform respects users by giving them the option to opt out.But he frankly admitted to the default activation, saying: if it required prior consent, no one would agree, and that is the honest answer.He also indicated that the platform does not expect the public to welcome this decision.

Minton explained during the stream that the content of any channel that remains within the default option can be used to train models, but he also asserted that the collected data will not be resold to external companies. In response to questions about whether user data had already been pulled for training purposes, Minton acknowledged that he did not know if the data had been pulled, stating that he was unsure what Amazon had specifically implemented in terms of model training and what it had used or ignored so far.

According to Twitch's FAQ page about the use of generative artificial intelligence, not opting out means that content may be used to develop models capable of producing new texts, images, and videos, similar to famous models such as ChatGPT and Gemini. The platform clarified that audio recordings may be used to refine models that convert speech to text, which contributes to improving the automatic translation read on Twitch streams and Amazon videos.

The concerns expressed by creators have extended to the potential implications for video game developers, as Amazon's algorithm training relies on the many game clips that content creators broadcast daily. Many users have expressed their rejection of the exploitation of their creative content to feed artificial intelligence,where one creator wrote a comment describing the default activation of the option as unacceptable, while others affirmed that the platform has missed an opportunity to set a new standard that limits data collection in the tech sector.

For its part, Mary Kish, head of the community department at Twitch, provided an educational explanation to listeners on how to disable training through channel settings, by going to the broadcast control panel, then heading to the security and privacy tab, and scrolling down to turn off the artificial intelligence training switch. However, a number of users reported a technical glitch, where they notice that after disabling the option and leaving the page, it automatically returns to the enabled state when they reopen the settings.

Mary Kish acknowledged during the live stream that data collection for training smart models has become a common standard in the tech sector.These developments come amid Amazon's large expansion, which acquired Twitch in 2014 for nearly $1 billionand continues to invest heavily in artificial intelligence services to compete with major tech companies like Google and Meta.