Generative AI models like GPT-4, DALL-E, and Midjourney are trained on massive datasets scraped from the internet. This dataset includes copyrighted books, articles, artwork, and code. The creators of this content received no compensation. The legal and ethical implications are still being resolved.

Copyright lawsuits are emerging. New York Times sued OpenAI and Microsoft for training on copyrighted content without permission. Authors sued for similar reasons. Lawsuits in the US are ongoing; outcomes will establish precedent. EU regulations are stricter—requiring explicit permission for training data in many cases.

The 'fair use' question is central. Fair use (in US law) permits limited use of copyrighted material for purposes like commentary or criticism. Does training an AI model on copyrighted content constitute fair use? Arguments on both sides: (1) training is transformative and doesn't replace the original, (2) the dataset is the economic asset enabling the AI's value.

Attribution becomes difficult with generative AI. When an AI generates an image, which artists contributed to the training data? Can you even determine this? Some researchers argue generated content should clearly disclose that AI was involved and list trained artists. Others argue this is impractical.

Copyright offices are developing guidance. The US Copyright Office clarified that purely AI-generated works lack copyright protection, but works with sufficient human creativity do. This creates ambiguity—how much human input is required?

Open licensing models are emerging. Some creators willingly license their work for AI training. Stability AI trained Stable Diffusion partly on openly-licensed LAION dataset. This model respects creator wishes while enabling AI development.

Fair compensation models are being explored. Imagine if generative AI companies were required to compensate creators for using their work. This could be direct payments, licensing fees, or pooled arrangements. Implementation challenges are significant—identifying all contributors and calculating fair compensation is complex.

The path forward likely involves legal clarity (court decisions and regulations), industry standards (attribution, licensing), and technology (watermarking to identify training data). Creators are rightfully concerned about their work enabling commercial AI systems without compensation. Resolving this tension fairly while enabling AI innovation is one of the more important unsolved questions in technology.