Tech companies have downloaded more than 15 million videos from over two million YouTube channels to train their artificial intelligence models. Alex Reisner reports for The Atlantic that this practice often occurs without the creators’ permission and violates the platform’s terms of service. The investigation identified at least 13 data sets containing the scraped content, including nearly one million how-to videos.
According to the report, major tech companies like Microsoft, Meta, and Amazon have used these data sets. The goal is to develop AI tools capable of generating new videos from simple text prompts. This technology directly competes with human creators, whose work provides the raw material for their potential replacements. The woodworker Jon Peters, whose videos are part of the training data, expressed his concern about whether he should continue creating content.
The legality of using copyrighted videos for AI training is a contentious issue currently being debated in court. Tech companies often argue their actions fall under the “fair use” doctrine. Meanwhile, some AI-generated content is already competing with human-made videos on YouTube. If the platforms become flooded with synthetic media, it could devalue original work and make it difficult for human creators to compete.