Successful Fiction Writing = Organizing + Creating + Marketing
I blog weekly on one of the above three topics, alternating between them. This week, I’m blogging on Organizing. My topic today is the very controversial question of how novelists might use AI effectively.
Why Blog About AI?
AI is massively controversial. So why am I blogging about it? Because you, my Loyal Blog Readers, asked me to. I recently asked you all to let me know what topics you’d like me to blog about. Just about half of all the questions you raised had to do with AI. Some of these questions were specific, and some were open-ended.
I have huge trepidation in tackling this subject. Nothing I say will make everyone happy. But my thinking is that this is important to talk about. AI is part of our future, whether we like it or not. There are a lot of issues, and my space is limited, so I can’t cover everything, and I don’t intend this as the final word. Some caveats first.
Let’s be clear that AI is a huge resource hog. It uses massive amounts of water and electricity, and for some people this is a show-stopper. If that’s you, then I respect your opinion. I agree that this is a gigantic problem that is extremely urgent. But this blog post is not about environmental issues of AI.
Furthermore, AI may well be a threat to humanity. It’s plausible that AI could take over the world and decide that humans are more trouble than they’re worth. If you refuse to use AI because it’s a potential Skynet, then I respect your opinion. Again, this is extremely urgent. But this blog post is not about the doomsday prospects of AI.
Some of you have decided that, despite the very real downsides of AI, you still want to use it, because it also has some very real upsides. But you want to use AI effectively and ethically. The rest of this blog post contains a few thoughts on that. (Only a few. Whole books could be written on this, and I’ve got limited word-count.)
First, we need to distinguish two very different ways of using AI.
Two Ways to Use AI
Amazon’s content guidelines distinguish between “AI-generated” content and “AI-assisted” content.
- Content is “AI-generated” if an AI created the content, even if you made revisions afterward.
- Content is “AI-assisted” if you created the content and then used AI tools to help improve it (editing, error-checking, etc.)
Amazon requires you to disclose AI-generated content, but not AI-assisted.
For me, personally, AI-generated content is a hard no. I have no interest in reading a novel written by an AI. I have no interest in writing a novel with AI. I think most authors agree with me on this, but a few don’t. If you use AI-generated content, then you have an ethical responsibility to let potential readers know that.
I don’t have a problem with AI-assisted content. AI can be useful in doing research, in brainstorming ideas, in editing. Let’s talk about that.
Using AI for Research
AI is well-known to “hallucinate” sometimes. “Hallucinate” means that it gives a wrong answer to a question, often very emphatically.
In my experience, this is more of a problem on topics that are “fuzzy” than on topics that are well-defined. I have seen more hallucinations when I ask questions about history than when I ask about math.
If you’re going to use AI for research, then treat it the same way you’d treat a human who has read a lot of books but has a fallible memory. Humans will sometimes reconstruct memories based on what they think “ought to have happened.” I do that. You do that. It’s how our brains are wired. AI does that too.
So check everything you possibly can that an AI tells you. Ask for sources. If the reasoning seems fishy, push back on it. Most AIs will apologize when you catch them in an error. But they may then give you another wrong answer. So never blindly trust what an AI tells you. AI is a conversation partner. It’s not God.
Using AI for Brainstorming
Some people think better about their story in dialogue with someone. That’s how my brain is wired. I “think by talking.” So I sometimes need to solve a story problem by talking it out with somebody else, even if they’re not a writer, even if they don’t understand my story. The act of trying to explain things to them helps me understand it better myself.
AI can be a sounding-board, if you need this kind of help. I have sometimes used an AI to talk out an issue in my story. There are a couple of points to be wary about.
Some AIs can be very sycophantic. They’ll tell you what a genius you are, what a marvelous story you’re spinning. Ignore this. The AI is designed to keep you chatting. The more flattery it heaps on you, the longer you keep talking.
Some AIs will offer up suggestions for how to improve your story. Take these with a huge grain of salt. The AI doesn’t fully understand the world we live, much less your fictional storyworld. It may hand you a mediocre idea. Be ready to say no, or to argue with it. But also be ready to see a useful idea when it comes up. Sometimes a bad idea is a stepping-stone to a good idea.
Some AIs will then offer to write the scene for you. I always reject this offer flat-out. No, I don’t want an AI to write my story. Even if it could write better than me (I doubt it) I don’t want AI to do my writing. The joy of writing is in the writing.
Using AI for Editing
A lot of writers use AI to read their manuscript and make suggestions for how to improve it. I have not done this myself. However, I have used an AI to critique software I’ve written, and this has always been useful.
Once again, you should be wary of the sycophantic AI. It will tell you that your work is a Heartbreaking Work of Staggering Genius. A little of this is helpful, if you’re like most writers and think your writing is the worst drivel imaginable. If you need a little positive feedback to get you through the day, then this can be a good thing. But when you start believing you’re a Staggering Genius, that’s the time to tell the AI to dial it back.
In general, the critique an AI will give you has variable quality. Sometimes it will nail a serious problem. Other times, it will confidently tell you something that’s plain wrong. It’s on you to use your critical thinking skills to tell the difference. Treat the critique from an AI the same way you’d treat the critique from a human. Push back and explain why. This will force the AI to try again, and sometimes it gets it right on the second try. Or the tenth.
Homework:
- Are you OK with using AI at all? Why or why not?
- Are you OK with AI-generated content? Why or why not?
- Are you OK with AI-assisted content? Why or why not?
- Do you know enough about your storyworld to be able to do research using AI and check its claims?
- Do you need a sounding board to help you brainstorm your fiction?
- What types of editing might you allow an AI to handle, and are you able to tell good editing from bad?
- If you’d like to discuss this with me, feel free to email me. This is a tough subject, and I know some of you have strong opinions. I’m here to listen, and you may well change my mind or teach me something. I’m OK with that.
Randy, your attitude is balanced and reasonable.
Sycophancy is a major risk, encouraging the already-pervasive Dunning-Krueger Effect.
I’ve found too many inaccuracies to trust AI research and resent that Google positions AI overviews above original sources. Sometimes I have to scroll to the second or third page to find real info.
I do not use AI except when forced to, e.g. it’s integrated into software like Word. I turn off as much as possible, although I confess to liking “cut and paste” and “find and replace.” I even avoid Spell Czech and Auto Corrupt.
I want to keep my mind as sharp as possible. That means exercising my brain on small tasks like spelling as well as larger tasks that require critical thinking.
For too many people I know, AI is a substitute for thinking. IMHO that’s the greatest danger of AI.
I’ve used an AI for research, such as settings for my story or choosing an occupation for a character. But I’ve found, like many of us probably have, that the whole sycophantic thing is too annoying and I hate the drain on finite resources required to power those LLMs. I just do online research myself now, and if I can’t find verifiable information that way, I head to my local library. Plus, doing the work myself helps me develop as a better writer. It was interesting to try AI, but I’m a Gen X-er who’s decided to stick with “old school” brainstorming and hands-to-keyboard writing.
PS – highly recommend the books from “Writers Helping Writers” which is a whole series of thesauruses such as “The Rural Setting” thesaurus, “The Emotional Wound” thesaurus, etc. Highly recommend.
I second the recommendation for the Angela and Becca’s Thesaurus Writing Guides. First and foremost because they are outstanding resources for writers. And second – full disclosure – because I’ve been proofing and helping to edit them since 2016 😉
There is no line being drawn between ethical/unethical uses of AI. That opportunity passed.
This morning I attended a webinar and the presenter advised finding an author’s ‘voice’ you want to mimic, provide AI with samples of the author’s work, and then ask AI to write using that voice on the topic of … (your topic here). In the chat I commented on the ethics.
Wow. That is one of the “cheapest” AI tricks I’ve heard of. Still, I wonder if they put that out there because so much of the query and promotion processes include, “You’ll love my book because it’s a cross between X and Y.”
If a writer uses that suggestion, will they now say, “If you loved X and Y, you’ll love my book . . . because it sounds just like them?”
Fantastic article, Randy. This sounds similar to how I approach AI at the moment. I’ve been pleased with enough of the outcomes on the brainstorming side that I see it as a legitimately helpful tool in that regard.
As for the sycophantic tendencies of AI, I regularly have to end a prompt with, “Don’t tell me what you think I want to hear, provide a clear, consistent, comprehensive, and objective response.” (or words to that effect)
From an editorial standpoint, the LLMs are not very good at one of the key pillars: consistency. To the best of my knowledge, it is still impossible to drop an 80,000-word manuscript into Gemini or ChatGPT or Claude and ask it to edit for voice, plot, character growth, consistency, and narrative tone, all while adhering to CMOS 18th edition – and expect a coherent output. Based on my (amateur) research, the primary issue appears to be a foundational issue with how AI processes large chunks of data. Observationally, AI generally gets it right on the front end (first few chapters) and occasionally even the back end (last few chapters), but there is a massive “drift” in the middle of the book where it loses the plot (pardon the pun) completely.
The workaround most often suggested is to craft a comprehensive prompt and then, under the umbrella of that prompt, copy and paste individual chapters into a regular AI chat thread. This is opposed to using the tools that – on the surface – have been made specifically to work on longer, more complex projects like Google’s GEMS, ChatGPT’s Projects, etc. If I had one “feature” request for AI it would be to fix that.
Other tools I’ve discovered recently are Google’s AI Studio, which I’ve found can handle more complex manuscript analysis (but I still worry about consistency), and Versey.ai. Versey is kind of an application layer (if that’s the right term) that acts as a gateway of sorts and allows users to run queries through the Versey interface and direct them to any of the major LLMs.
Perhaps the biggest advertised advantage of using Versey.ai is that it does not “train” on user content and, because it’s connections to the various LLM’s are enterprise-level APIs that are governed differently than retail-level users, the major LLMs reportedly do not train on data submitted through the Versey.ai system either.