The article discusses the complex issue of copyright law in the context of AI models, specifically those developed by tech giants like OpenAI and Google, which use transcriptions of YouTube videos as training data. The author suggests that the focus should not be on the intricacies of copyright law, but rather on the political economy of copyrighted content in the AI world. The article proposes that the problem can be broken down into three parts: the content used for training, the outputs allowed, and who profits from those outputs. The author suggests that AI developers should respect signals indicating copyright intentions, produce outputs that respect the source and nature of copyright, and pay for the output, not the training. The article concludes by suggesting that a system similar to YouTube's Content ID could be a potential solution.
Key takeaways:
The New York Times has claimed that tech giants OpenAI and Google are in a copyright gray area by transcribing YouTube videos and using the text as training data for their AI models, despite terms of service and copyright laws that prohibit such actions.
There is a debate over who gets to profit from generative AI, with publishers arguing that AI-generated content competes with the creators whose work the AI was trained on, while AI model developers are seeking a business model that will repay their investments.
There are three main aspects to the copyright issue: what content is used for training, what outputs are allowed, and who profits from those outputs. AI model developers should respect signals indicating copyright intentions, produce outputs that respect copyright, and pay for the output, not the training.
The article suggests that the problem can be solved by creating a new architecture for the AI ecosystem and a new business model, which could lead to a new golden age for both AI model providers and copyright-based businesses.
