The models of Generative They need huge amounts of data to train and be able to develop their creation capabilities. How the developers get this data It is not always legitimateand we have already seen many face legal problems, as is the case of goal or openai.
The last scandal in this regard stars it Google. It turns out that the technological giant is serving as its wide library of YouTubecomposed of 20,000 million videos, To train Gemini, I see3 and other AI models.
This fact has been revealed by CNBC after having obtained confirmation by Google, which affirms that only uses a subset of these videos and does so legally, complying agreements with creators and media.
«We have always used the YouTube content to improve our productsand this has not changed with the arrival of AI. We also recognize the need for security measures, so we have invested in solid protections that allow creators to protect their image and likeness in the AI erasomething that we are committed to continue doing, ”explained a YouTube spokesman.
What do youtube and google policies say
However, the news has caused skepticism and concern About to what extent the creators, media and companies are aware of these agreements and how Google uses their content. Likewise, there is also no configuration that allows the technology to be used for the videos published on their platform to train their AI models.
In the terms of the YouTube service, the platform is responsible for making it clear that «when you upload content to the service, You give YouTube a world licensenot exclusive, free and royalty free, transferable and with the right of sublicence To use said content (including to reproduce, distribute it, modify it, transform it, show it, communicate it to the public and represent it) In order to operate, promote and improve the service».
It is in the part of “improving the service” where the trap could be found if Google considers that training its IAS corresponds to this condition. In addition, Google's privacy policy indicates that the legal basis to be treated could include the Use user or user information «publicly available online or other public sources To train new automatic learning models and develop foundational technologies on which various Google products are based, such as Google translator, Gemini applications and Cloud's abilities ».
It should be noted that Google's privacy policy «It applies to all services offered by Google LLC and its affiliates, including YouTubeAndroid and the services provided on third -party websites, such as advertising services ».
This reality reopens the debate about the use of the content generated by people to polish models that can become a direct competition of these creators and creators. In addition, the little control over how companies handle intellectual property, accentuates the unfair of this system.
Luke ArrigoniExecutive Director of Loti, a company focused on the protection of the digital identity of the creators, explained to CNBC that: «It is plausible that they are taking data from many creators who have dedicated a lot of time, energy and reflection to these videos. He is helping the model I see 3 to create a synthetic version, poor imitation, of these creators. That is not necessarily just for them ».
Photo: Canva
