AI Training Is Copying. Is It Infringement? | Keith Kupferschmid, CEO Copyright Alliance
AI is forcing one of the biggest copyright debates in decades. So what does copyright law actually say about training AI on creative work? This episode is paid for by the Future of Life Institute. In this episode of Technically Creative, Orlando Wood sits down with Keith Kupferschmid, CEO of the Copyright Alliance, to unpack the increasingly complicated relationship between artificial intelligence, copyright, fair use, creators and the companies building generative AI. Keith represents a coalition that includes individual creators as well as organizations across film, television, music, publishing, photography, journalism and other copyright industries. He argues that generative AI presents a fundamentally different challenge because AI models can be trained on copyrighted works and then produce material that competes in the same markets as the people who created the original work. But what does copyright law actually say about that? Keith explains an important distinction that often gets lost in the AI copyright debate: the legal question surrounding the copying of copyrighted material during AI training is separate from the question of whether an AI-generated output is substantially similar to an original work. From there, Orlando and Keith dig into fair use, licensing, market harm, training data, pirated material and the growing number of AI copyright lawsuits working their way through US courts. They discuss why different types of creative work may produce very different legal outcomes. A language model trained on books but used to write résumés presents a different fair-use question from an AI music system trained on music to generate more music. Keith explains why he believes cases involving music, journalism and some visual imagery may provide particularly strong tests of the fair-use argument. The conversation also looks beyond the lawsuits to the question that matters for the future: if AI isn't going away, how do we build a system that works for creators and technology companies? Keith proposes three essential ingredients: licensing, transparency and safeguards. And while Keith is forceful about protecting copyright, this isn't a conversation about rejecting AI. He believes creators can use artificial intelligence productively and responsibly while remaining in control of their creative work. As he puts it: “Don't put AI in the control room.” We explore: How copyright law applies to generative AI and AI training data The difference between input infringement and output infringement Whether training AI on copyrighted material can qualify as fair use Why permission, compensation and licensing matter to creators The major AI copyright lawsuits moving through US courts Why different courts are reaching different conclusions about AI and fair use How market harm affects the fair-use analysis Why AI-generated music presents a particularly interesting copyright case Copyright issues affecting writers, musicians, filmmakers, photographers and software developers Why Keith believes licensing, transparency and safeguards could provide a path forward How creators can use AI without surrendering creative control Why AI copyright law may take years to become clearer For artists, filmmakers, writers, musicians, photographers, publishers, creative companies and anyone trying to understand the fight over AI training data and copyright, this is a practical explanation of what the law says today, where the uncertainty remains, and what a workable future could look like. Technically Creative is a technology podcast for people who don't necessarily think of themselves as tech people, exploring how AI and technology are changing film, television, advertising, publishing, media and the creative industries. This episode is sponsored by the Future of Life Institute.
- Transcript










