As lawsuits mount against AI companies over copyright infringement – most recently with Dow Jones and NY Post suing Perplexity AI for “massive illegal copying,” Microsoft’s CEO Satya Nadella calls for more flexible copyright laws.
Nadella asks: “If I read a set of textbooks and I create new knowledge, is that fair use?” This analogy fundamentally mischaracterizes how LLMs work. When humans read textbooks, we don’t make copies of them – we learn from them. AI systems, in contrast, must first make copies of millions of copyrighted works, store them in various forms, and can often reproduce them verbatim.
But there’s an even more crucial distinction: When I read a textbook and build on its knowledge, I don’t create a system that replaces the need for others to read that textbook. LLMs, however, are explicitly designed to bypass the need for future engagement with the original works. Look no further than Perplexity’s own explanation as to why use its service “Skip the Links” – a direct admission that their goal is to eliminate the need to consult original sources, sources they copied without authorization or payment and now claim as “fair use.”
As highlighted in “The Heart of the Matter,” we must carefully navigate the intersection of these domains to foster advancement that respects both progress and rights. To quote directly: “To foster a future that is both pro-copyright and pro-AI, it is essential to carefully navigate the intersection of these two critical domains, harnessing the power of both copyright and technology as engines of innovation.”
Proper licensing is an important part of the solution of a comprehensive system to ensure both AI technology and creative works can thrive together. Without such a balanced approach, we risk undermining the very foundation of human creativity that fuels innovation.