Reproduce “The Files are in the Computer: On Copyright, Memorization, and Generative AI” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
A. Feder Cooper, James Grimmelmann
A. Feder Cooper and James Grimmelmann precisely define 'memorization' in generative AI as the potential to reconstruct exact training data from a model, contending that such models constitute 'copies' under copyright law. Their work synthesizes machine learning insights to inform legal discourse on AI training data and intellectual property.
Key insights
- Memorization is quantifiably encoded within a generative AI model's parameters, making it an inherent property rather than just an output phenomenon.
- A generative AI model that has memorized training data fulfills the functional definition of a 'copy' under U.S. copyright law.
- The stochastic nature of AI outputs does not negate the existence of memorized content within the model; non-zero probability of regurgitation indicates memorization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.