Would a savant who has read all the books in a specific library, whether in college or even the Library of Congress, be able to create original works without facing criticism for using knowledge gained from the works of others? That’s a question I pose to you, the readers of this blog. Your understanding of this concept is crucial. How does my hypothetical savant differ from a typical large language model trained on web content?
Copyright law, a crucial aspect of our modern content creation landscape, was initially intended to incentivize authors to produce new works by granting them exclusive rights to their writing for a limited time. This was aimed at advancing the progress of science and learning by ensuring that works are accessible to the public.
The Copyright Act of 1790 gave American authors the exclusive right to print, reprint, or publish their works for 14 years, possibly renewing the protection for another 14 years. This law encouraged authors, artists, and scientists to produce original creations.
Revolutionizing copyright laws to both protect and empower modern content creators while not hindering the capabilities of advanced technologies like Large Language Models (LLMs), would involve a multifaceted approach:
Balancing Interests
Revised laws could strike an equilibrium between protecting original work and fostering innovation. This could be achieved by clearly defining what constitutes fair use, particularly in AI-generated content, and establishing transparent guidelines for attribution and compensation when such technologies use existing copyrighted materials as part of their learning process.
New Licensing Models
Implement licenses that cater specifically to LLMs, allowing them access to copyrighted works under certain conditions without infringing on the rights of original creators. This could involve pay-per-use models or subscriptions for AI developers who use these technologies.
Innovation Incentives
Offer additional protections and benefits for content creators to encourage them to invest time in creating new, innovative works that LLMs can use without legal repercussions—akin to a “digital commons.” These could include tax breaks or grants.
Adaptive Legislation
Laws should be designed to adapt as technology evolves, potentially incorporating AI-driven governance systems that can continuously interpret and apply copyright law based on current technological capabilities and social norms.
Some of the ideas in this post came from using Ollama and the Phi-3:medium model.