Blue Garret AI Policy
Authors, editors, and other creatives have strong feelings about the emergence of Large Language Models (LLMs) and artificial intelligence (AI) tools and their use for writing and editing. My approach has been to listen, learn, and experiment. I work with authors who are already using these tools in their work, authors who are curious about them or experimenting with them, and authors who are adamantly opposed to them. My job as an editor is to understand the stance of the authors I work with and adjust my own practices accordingly.
My core beliefs about AI tools
Intellectual property and copyright should be respected; companies should ask for consent before including an author’s work in a large language model, and provide reasonable compensation for work that is included. We know that the current LLMs have been trained on material that was not provided with consent or compensation.
It’s important to recognize that the creation and use of these tools may be causing environmental harm. Similarly, we know that the tools recreate all of the many biases already present in their training data. We must continue to flag these issues and call on companies creating these tools to correct them.
Authors and editors working together should have clear and honest discussions about how they use AI tools. These conversations are necessary to establish trust and make sure their priorities are aligned. I include a clause in my contracts with authors specifying that I will never upload their manuscript to an AI tool without their explicit consent.
How I use AI tools
As I state in my contracts, I do not use AI tools for any core editing tasks, including developmental editing, analyzing or evaluating manuscripts, line editing, copyediting, writing manuscript comments, or writing editorial letters. My job as an editor is to be a stand-in for the (human) reader who will eventually be on the receiving end of the manuscript. My human instincts are an important element of my editing skills, especially when an author has used AI tools in creating their work.
However, I have found AI tools to be useful for specific tasks:
Writing Microsoft Word macros in Visual Basic that allow me to quickly perform specific behind-the-scenes tasks on a document that make me faster and more accurate: for example, highlighting all italicized words in a manuscript or identifying dialogue that might be missing a quotation mark.
Untangling a clunky sentence with a grammar-focused solution. Sometimes I’ll encounter a sentence that I know could be fixed by switching from a transitive to an intransitive verb, but I can’t immediately land on the correct verb. Pasting the single sentence into Claude and asking for suggestions for verb choices will get me to the best solution right away. For example, here’s an awkward version with a transitive verb: “The manuscript undergoes transformation through each editing stage.” And here’s a streamlined version with an intransitive verb: “The manuscript evolves through each editing stage.”
Similarly, Claude makes for an excellent thesaurus when a better word is hovering just outside the reach of my brain – I know what it is but can’t grab it. Asking Claude for a few suggestions for replacing a specific word in a sentence always shows me exactly the word I was reaching for.
AI tools often save me time when fact-checking, getting me to correct results I can verify with fewer steps.
I don’t routinely handle reference lists, but these tools can quickly and easily put bibliographies and notes in the correct style, according to Chicago Manual of Style or another source, and even fill in missing information. (This must be verified, but verifying is much faster than searching.)
AI tools are also very useful for summarizing, checking for consistency, and writing marketing copy and suggesting metadata. These tasks require uploading a full manuscript to an AI tool. I always ask for explicit permission from an author before uploading their work. Right now, I do not do this routinely, and I never require authors to allow me to use these tools to complete an edit. Anthropic’s Claude is my AI tool of choice because their privacy policy is clearly written and explicitly states that uploaded work will not be used to train future models.
Specific use cases for uploading an entire manuscript:
Creating detailed chapter summaries, character lists, and timelines, for a single manuscript or an entire series. Right now, I create these materials myself as part of copyediting, without relying on AI tools, except in special cases – for example, when working on a complex series when I haven’t edited the previous volumes.
Checking for consistency, within a single manuscript or across a series. AI tools are very good at helping you answer tricky questions: For example: “A character is celebrating a birthday; based on earlier mentions of their age, what age should they be turning in this chapter?” Or: “Please summarize, with chapter references, every place in the manuscript I’ve referred to this specific magical object.”
Writing a rough draft for a book description, synopsis, or plot summary for a query letter. AI tools are particularly good at marketing copy. Often the voice isn’t quite right, but the draft will show you which elements you can include to effectively showcase your plot and characters.
The future of creativity in the age of AI
I believe that humans will always want to consume art that is made by other humans. An AI tool can write a novel with a single click right now. I don’t see any evidence yet that novels written this way are popular with readers. As all published authors know, writing a book is, in some ways, easier than finding readers.
Right now, I see the value of AI tools in helping me be a better, faster editor. I plan to continue to use these tools in the limited ways listed above and experiment with new ones. I also will continue to listen to the various opinions circulating among writers and other artists and creatives about the value and ethics of using these tools, and I will always consult with authors I work with about their own beliefs and preferences.
Written December 23, 2024