Copyright and AI

AI vs Content Creators

Indian news outlets have joined the courtroom battles of content creator corporations against AI companies. This has been an emerging battlefield in the world of copyright. Not only OpenAI, but several generative AI companies like Google, Anthropic, and Meta AI are currently involved in over two dozen lawsuits in the US alone, with content creators ranging from publishers to authors to news outlets joining the cases. Question is, are generative AI technologies infringing on copyright? Let’s take a look.

AI models like ChatGPT are trained through data mining of large amounts of online content. In this training, the AI model analyzes the content, determines, and then learns to replicate, patterns. However, much of this training content is copyrighted. So when AI scrapes, downloads, or processes copyrighted content, it may violate copyright law.

Is it Fair Use?

That said, copyright law also includes the doctrine of fair use, which allows limited copying without permission. Fair use is determined using the following four-part test:

1. The purpose of use (e.g., is it for profit or nonprofit?).

2. The nature of the original work.

3. The amount of content used and whether it changes the original in a significant way (making it “transformative”).

4. The impact on the original work’s market (e.g., whether it competes with or replaces it).

AI companies currently claim that their use indeed qualifies as fair use, and therefore, they can train their models without paying or seeking permission. The reason given for that is that AI training focuses only on non-expressive use. That is, AI doesn’t actually copy the creative parts of works, but instead, extracts facts and patterns, which aren’t copyrightable. For example, if AI analyzes photos of dogs, it’s using the images only to be able to identify dogs and differentiate between dogs and other things, not using the actual photos for anything else. In fact, courts have supported this idea in earlier cases, such as plagiarism detection tools and searchable databases like Google Books.

However, content companies’ argument is that generative AI is different, because these models can create outputs similar to their training data, sometimes competing directly with the original works. They can even replicate specific elements, like fictional characters, with high accuracy. Because of these abilities, generative AI’s use of copyrighted content may not be entirely fair use.

Conclusion

These cases are still in courts, with none of them having been finalized. Some say that the development of AI boosts innovation in diverse fields like science, linguistics, and internet. Also, licensing of AI training data may be impractical because of the massive datasets involved. Critics, however, argue that using copyrighted works without permission—especially for commercial purposes—is unfair. Artists and authors are concerned about losing control, compensation, and licensing opportunities.

At NorthWind, this is a hot topic of discussions, and it’s going to be super interesting what turn this evolution of copyright takes. Stay tuned!

Photo by Tara Winstead/Pexels.com