New Twitch Controversy Proves One Big Thing About AI

We’ve seen a number of crazy things surrounding AI and corporate efforts over the past few years. Previously, we’ve seen YouTube edit people’s videos with AI when they didn’t ask for it. This isn’t a problem limited to YouTube, either.

X (Elon Musks revamped Twitter) has also turned to user data to train its AI, and Musk has even told SpaceX employees in the past that he would use the company’s data to train his latest Grok AI models. And while some of these companies have been shady about how they launched their AI training systems, others have taken a much worse approach; like Twitch.

The latest controversy surrounding AI and a company is Twitch’s decision to begin training Amazon’s AI models on content coming from Twitch streamers themselves. However, instead of being open about this feature, informing streamers about it, and giving them the opportunity to enroll in the training program, the company decided to disable it and quietly began training on streamers’ content. And the reason why Twitch did it this way? Well, according to a statement made by the company’s Chief Product Officer, Mike Minton, the reason is quite simple: “If this were an opt-in option, no one would accept it.” » (via Kotaku)

Concerns about AI and its overall impact on the economy have grown in recent months, despite U.S. efforts to make data centers more attractive through legislation. Pair this with other news about how content is used to train AI – like Anthropic’s recent debacle around buying rare books and then destroying them to train Claude – and you have a recipe for people to hate anything to do with AI. And Twitch’s decision to train Amazon’s AI models on user content without asking is just more fuel on that fire.

The company further addressed the concerns in a livestream talking about the patch notes that introduced the toggle, and that’s where Minton made the above statement. Of course, this reaction, while frankly honest, has also sparked rhetoric online, with users noting that Twitch knows that no one wants AI training on their content, but the company doesn’t care, while others have noted that Twitch’s handling of the announcement seems illegal, as the company chose not to even email affected users, citing the fact that not everyone reads their emails as one of the main reasons.

The concerns don’t stop there. When asked how it would handle streaming content containing users who have opted out of the feature, Minton was unable to answer. Instead, he said it was “a good question,” and added that there were no plans to allow others to see who had or had not chosen to use their content for training.

Of course, training companies on your data is nothing new, and even before AI as we know it today became big, companies were always looking for ways to use your data to train their algorithms and systems in some way. However, with AI, the problem has become even bigger, especially since generative AI tools are not used to improve the systems you already use. Quite often your content is used to improve the model itself so that it can generate content on its own.

In the case of Twitch, the company’s FAQ page states that it will use channel content such as streams, VODs, discussions from streams and even images and text from your channel to train future Amazon models, including those designed to generate text, images, audio and video. This means your content could directly be used to train Amazon’s next generative video model, which might not be suitable for some people.

If you are one of those people who do not want your content to be used to train AI, you can always opt out by following these steps. First, click your account icon in the upper right corner of the page, then select Settings from the list. From there, go to Security & Privacy, then scroll down until you notice the Generative AI Training option at the bottom. Disable it to turn it off, and you should be fine. Twitch notes that turning off this feature does not opt ​​you out of AI-based features and that your content can still be used to improve these features.