AI Certification for Copyright Compliance

A new certification scheme called Fairly Trained has been launched for copyright-compliant AI, aiming to address the use of copyrighted content by generative AI companies. The initiative is led by Fairly Trained’s CEO, Ed Newton-Rex, who resigned from Stability AI over concerns about the use of copyrighted content. The certification aims to differentiate companies that use licensed training data for their AI models from those that don't. However, major text generation models, such as ChatGPT, currently do not qualify for certification due to their training on copyrighted works. While large language model (LLM) giants like OpenAI argue that training on copyrighted works is necessary, Newton-Rex believes that alternative approaches are possible and necessary to address the existential threat GenAI poses to creative industries and human creativity.