You wouldn’t give a spell-checking program credit if you published a novel, but the lines gray with ChatGPT and its peers, as a simple prompt can catalyze a long and complex answer that could mirror the human prompter’s intentions, but also might sidestep them, and go beyond the requests or desires of the human. “It is important to note that AI software does not create anything,” the guild wrote in a tweet. “It generates a regurgitation of what it’s fed.”

The guild proposed that scripts written with the help of AI would still get full “written by” credits, Variety reported, and not be relegated to “screenplay by” credits, which are based on other literary works or third-party source material and command only 75% residuals for writers.

This is a practical question, as some AI-assisted TV writing may already be happening: A recent episode of South Park was reportedly written, in part, by ChatGPT.

But ultimately it runs into a philosophical question of what constitutes derivative work. Large language models (LLMs), which power AI writing services, are trained on large swaths of content already on the internet. Of course, all works are in some way derivative of others, sometimes without full awareness, admission, or credit, but the black-box nature of these AI tools does make it difficult to draw elegant lines between originality and plagiarism—even if an AI-generated script is being heavily edited by humans before anyone ever sees it acted out on screen. “Plagiarism is a feature of the AI process,” the guild wrote in tweet.

This story has been updated to include statements from the Writers Guild of America West.