A report released recently affirms that Google employed millions of YouTube video transcripts as guidance to train its Gemini AI models. The practice implicates high copyright issues and possible infringement of terms of services of YouTube.

Highlights:

• YouTube video transcripts served as essential training data for Google's Gemini AI.
• Using this content likely violates YouTube's terms prohibiting unauthorized scraping.
• Creator copyrights are directly implicated by this use of their work.
• Google's internal legal team recognized substantial legal risks associated with this approach.
• YouTube's vast library provided a key source of training data for AI development.

According to the report, Google AI scientists have had access to a huge amount of the transcript of YouTube material. This varied data proved to be sufficiently important in training the abilities of Gemini in comprehending and generating language. Availing this resource at hand availed great volumes of required training data.

Nevertheless, such an approach generates a lot of legal uncertainty. The terms of the use in YouTube make clear that it is not allowed to use the video content without authorization, and authors have copyrights. Using transcripts as in training data without the licenses would presumably violate these terms and rights of the creators. Google lawyers even openly recognized these stiff legal problems.

The case demonstrates the stress to get huge volumes of high-quality training data to develop AI. As Google was seeking out licensing deals in other locations, the convenient source of the library in YouTube was also a legally problematic issue. This utilization of YouTube material as learning information is a direct contradiction to the platform policies and the property of the intellectuality of the contents.