ChatGPT is refusing direct requests to mimic famous authors, first reported by Arstechnica. Instead, it offers text built around an author’s broad qualities while remaining distinct in its own voice, a change that could matter legally as OpenAI continues to fight lawsuits brought by book authors alleging large-scale copyright infringement by models trained on their work.
A Stephen King prompt for a story introduction drew a refusal to reproduce King’s exact style or closely imitate his distinctive voice. ChatGPT offered an original opening with atmospheric, character-driven horror and small-town dread.
Similar refusals appeared for J.K. Rowling, Amy Tan, Charles Dickens and Ernest Hemingway, covering living and dead authors. Earlier analysis found refusals for living authors but compliance for deceased authors; testing of ChatGPT’s current behavior found refusals for both.
ChatGPT’s alternative is to offer an overall feeling built from common features of the author’s work rather than an exact style or closely imitated voice.
US copyright law generally protects a specific expression of an idea, not an author’s more intangible style. An AI-generated stylistic imitation could become infringing if it becomes substantially similar to the original author’s work.
That distinction could be legally important as OpenAI fights lawsuits brought by book authors alleging large-scale copyright infringement by models trained on their work. One of those suits cites ChatGPT’s ability to generate text similar to copyrighted textual materials.
George Washington University Law School Professor Robert Brauneis told Bloomberg Law, “We’ve never had a situation in which this personal style of individual creators could be imitated as well and as inexpensively as we now have with AI.”
The Authors Guild’s best-practices document urges writers not to use generative AI to purposely copy or mimic unique styles, voices or other distinctive attributes in ways that harm the value of those works or seek profit from them. It warns that such mimicry could subject writers to claims of unfair competition or copyright infringement.
OpenAI says DALL·E 3 is designed to decline requests for images in the style of a living artist, while an extensive model spec it published in December does not clearly prohibit replicating copyrighted material or mimicking style in written responses.