Anthropic CEO Dario Amodei strongly defends his company's use of copyrighted materials for AI training, claiming it qualifies as fair use under existing law.

Photo Credit: Dwarkesh Podcast (Dario Amodei)

In a New York Times interview with Ezra Klein, Amodei explained that their AI training process is "much more like the process of how a human learns from experiences" rather than simply copying and reproducing content. He emphasized that their models shouldn't output verbatim copyrighted content.

This stance comes amid an ongoing lawsuit filed by major music publishers Concord, Universal, and ABKCO against Anthropic in 2023. The publishers allege copyright infringement, claiming Anthropic's Claude chatbot unlawfully uses scraped song lyrics in its training data.

Anthropic's defense rests on three key arguments:

1. The use of lyrics is "transformative" and adds different character to the original works
2. Song lyrics comprise a "miniscule fraction" of the training data
3. The scale of licensing required would be practically impossible, involving "trillions of snippets across genres"

The company also contends that the publishers themselves triggered Claude to produce the infringing content they're complaining about, shifting responsibility for the alleged infringement.

This case represents a crucial legal battle that could set precedents for how copyright law applies to AI training data and fair use in the age of artificial intelligence.

Related Articles

Previous Articles