Ed Newton-Rex Highlights Creator Harm and Regulatory Need for 'Derived Data' in AI Training

Ed Newton-Rex highlighted the issue of derived data in the AI training process, arguing that social attention and regulation are urgently needed. Previously, over 100 lawsuits focused on illegally collected data, such as pirated or web-scraped content; however, the method where AI first rewrites creative works and then uses them for training is difficult to detect and causes the same harm to creators.

How Derived Data Works and Its Risks

Derived data refers to using AI to rewrite sentences and other content from original creative works for training, rather than having the AI model learn directly from the originals. This approach makes detection difficult because the model is less likely to output the original text verbatim, but it effectively amounts to using creative works to develop competing products without permission. As AI model performance improves, generating such derived data becomes easier and more frequent, making it increasingly difficult for creators to recognize and respond to the misuse of their work.

Circumventing Opt-Outs and the Need for Regulation

This issue can also be exploited as a means to circumvent opt-out measures by AI companies. Even if a platform promises not to train on specific data, companies can avoid legal liability if they do not exclude derived data created by rewriting that data with AI from the training set. Newton-Rex emphasized that the only solution to this problem is mandatory disclosure of training data, urging legal regulations that require AI companies to transparently disclose their training data.