Wired founding executive editor Kevin Kelly has backed OpenAI and other defendants in a US copyright battle, arguing that using copyrighted books to train AI models qualifies as fair use and helps advance the creativity copyright law is intended to encourage.
In a proposed friend-of-the-court brief dated October 16, submitted as an exhibit in the consolidated OpenAI copyright litigation in the Southern District of New York, Kelly urged the court to rule in the defendants’ favour on the training issue.
Kelly, himself an author of fiction and nonfiction, argues that training large language models creates a new technological tool rather than a substitute for the books used to develop it. His central contention is that copyright protects authors’ original expression but does not give them exclusive control over the ideas, information and knowledge contained in their works.
“Copyright is not a shield from disruptive technology,” the brief says.
The filing draws a parallel between AI training and the way people learn from reading and experience. It describes models as analysing vast quantities of text to identify patterns and relationships that can subsequently help generate new material.
Kelly argues that writers have always drawn on earlier works without necessarily infringing them. The brief points to Shakespeare’s Romeo and Juliet, which built on earlier storytelling, and Jane Smiley’s A Thousand Acres, a modern retelling of King Lear. Its argument is that drawing on shared ideas and themes differs from copying protected expression.
The brief also asserts that models do not store works in their original expressive form. That assertion forms part of Kelly’s case for fair use; the proposed brief does not establish it as a judicial finding.
A key commercial issue in the filing is whether competition from AI-assisted books constitutes harm that copyright law should recognise. Kelly characterises the plaintiffs’ position as seeking protection against new, non-infringing works that compete with their books. He argues that such competition differs from supplying readers with copies that substitute for the originals.
To illustrate AI’s role as an author’s assistant, the brief cites self-published writer Aaron Black, who reportedly published four books in ten months while writing the prose himself. According to the account cited in the filing, Black used AI for research, manuscript continuity and visual references, while retaining creative control.
Kelly’s broader argument is that these tools can give individual writers support previously associated with teams of researchers, editors and assistants.
Beyond publishing, the brief invokes AI applications in medical research, mathematics and criminal investigations to argue that restricting training could impede wider scientific and creative advances. It asks the court to weigh those benefits in assessing fair use.
Kelly acknowledges that AI raises serious questions about its capabilities and effects on society. But his position is that copyright should not become a mechanism for resolving every concern about the technology and that allowing the training at issue better serves copyright’s underlying purpose.
