Using A.I. in Writing
Ah, artificial intelligence, the generative and large language model platforms that have become so ubiquitous. Depending on who you ask, A.I. will either give us three-day work weeks, make our daily lives easier, take all of our jobs, destroy humanity, or something else in that very wide span of possibilities. Given the risks and opportunities, the subject of A.I. is controversial, especially since we’re already seeing positive and negative applications and results. But what about using A.I. in writing, or other creative pursuits?
Well, the simplest answer to that question is a solid no.
To say that the use of A.I. in creative pursuits is unwelcome would be an understatement. But there’s clearly more nuance than that, which is why there’s a whole bunch of text below this. So, let’s start out with expanding upon that “no” before looking at possibilities for “yes,” and why those might be okay.
The State of A.I. in Creative Works
Right now, the attitude towards the use of A.I. in creative works is intensely negative, and it’s understandable. Instead of a human using their artistic and creative talents to create a novel work of literature, entertainment, or art, someone prompts an A.I. platform to rapidly generate a work based on synthesizing existing works. Such outputs can easily come across as lifeless…and a small number of people could use this technique to flood the market, with each person generating more content than tens or hundreds of human artists or writers. And while most A.I.-generated works are trash, there are two issues here: first, the sheer volume of artificially-generated content can drown out original works and second, A.I. could start producing content that competes effectively against human-made works, simply by ruthlessly optimizing towards the most marketable content.
This is why many creative professions are hurrying to shore up protections for their fields. Because if A.I. becomes dominant in any creative field, it would not only dry up talent pools and rob us of creative individuals (not to mention those individuals’ livelihoods), it would also subject us to lower-quality, derivative works as a standard.
At the same time, the reactions to A.I. use have been intense. In some cases, they’ve been zero-tolerance. For instance, the critically-acclaimed video game Clair Obscur: Expedition 33 (full transparency: I have not played that game) had two major awards withdrawn because the studio used A.I.-generated placeholder textures early in development. And as I’ve started to look into literary agents and publishers, many include questions about A.I. use, including versions of the question that basically ask you to verify that you haven’t used it. I’ve been able to truthfully say that I haven’t used A.I. in my novel writing, but I imagine that many people might turn to A.I. for some background work…but that still counts as A.I. use. At this point, even using search engines often engages an A.I. agent.
So, if you want to be on the safe side, simply do not use A.I. for anything related to your writing.
Need to do background research? Use books, films, articles, and other non-A.I. content. Developing characters? Don’t turn to A.I. for help. You can use some of the tips on this site for that! Writing itself? It might be tempting to pop your text into ChatGPT or Claude or something to have it review it for you, but avoid that: use some editing tips and/or find a friend or beta reader to review it.
If you use A.I. for any part of your process, you may need to disclose that, and it could severely limit which literary agents and/or publishers will be willing to take you on. Use it too much, and you might end up having to compete with swarms of A.I.-generated rubbish.
But what if things change? Well, before we get to that, let’s cover some more general topics:
The Promise and the Downsides
When we think about A.I., we often jump to the big promises and the big risks. But those headline-grabbing topics probably don’t feel immediate when you’re asking ChatGPT for a book recommendation or Claude for the history of soup (or something; I don’t know what you use A.I. for). So let’s bring it down to the human scale: what are the pros and cons when you open up your favorite A.I. and get its help?
On the positive side, we get efficiency and detail:
You’re accessing what, in many cases, is a smart search tool, able to find information faster and more thoroughly than ever before.
A.I. can provide you with sources (and you should ask for those), and you can explore more.
Conversing with A.I. allows you to build upon information you’ve already gained, and to dig into areas of interest.
For some A.I. uses, you can reduce or eliminate costs and time associated with retrieving information or generating content.
You can train an A.I. to remember your interests, preferred sources, topics related to what you’re working on, and even your voice and style.
An A.I. can potentially be a partner in the development of your works: we’ll go into that soon, but you’ll need to weigh it against the downsides.
On the negative side, we get into some of the known issues with A.I., and the ethical questions that arise:
A.I. can hallucinate: it can and will tell you things that are completely untrue, which is why it’s important to ask it for sources on how it came to its conclusions.
Some A.I. can also be rather sycophantic, going along with whatever you’re saying without providing meaningful feedback or criticism.
And now we start into the bigger ethical questions: for some A.I. uses, you could be directly or indirectly taking business away from humans. For instance, if you need to generate some artwork for your story, turning to A.I. can be cheap and fast, but you’ve denied a real artist the opportunity to create something new (and as an aside, people will be able to tell that you used A.I.; I just saw an advertisement that clearly used A.I., and it made me more than a little disgusted that a large company would choose to cheap out that way).
A.I. processing requires enormous amounts of power and water, and every interaction with A.I. adds just a bit to those resources used. Data center impacts are already causing electricity rates to rise for residential users (and for other businesses as well, at least in some cases), and some areas are calling upon residents to conserve water and power, essentially giving priority to data centers. Now, spending hours searching for information on the internet might actually use more energy than using A.I. for the same task, but that’s hard to tell.
So from an ethical perspective, are there ways to use A.I. for creative works? Ways that you can improve your efficiency without causing harm? Or at least without causing notably more harm than we all do with our use of resources? I think so…but I’ll continue to emphasize the disclaimer that any A.I. use in creative works is viewed dimly, so proceed with utmost caution.
Ethically Using A.I. in Creative Works
Can A.I. be used ethically in creative works? Let’s find out; for this, I think it’ll be more useful to examine types of writing/creative labor, rather than the writing process or the PETS system in my introductory post. So we’ll look at research, character-building, art, the actual act of writing, editing, and getting your work into the wild. For each of these categories, we’ll look at what they encompass, how A.I. can be used at a basic level, how A.I. can give you grief, what A.I. can potentially do, and what pieces seem to be ethical in the creative space.
Research
Background research is an essential part of writing. Writing research can be quite different from academic research, and can be as simple as revisiting your own memories…or as complex as learning about life in a completely different time and place. And because research isn’t writing, it can be very tempting for many writers to get it over with quickly. And that sure sounds like a use case for A.I.
Basic A.I.: at the simplest level, A.I. can be a research search engine, collecting and summarizing information that you ask it about. A.I. can essentially sift through information for you, finding the pieces most pertinent to the subject you’re examining. And from there, you can converse with the A.I. agent, having it refine and/or expand its findings to better match exactly what you’re looking for. In this usage, A.I. is effectively a highly efficient search engine.
A.I. Downsides: when relying on A.I. to retrieve and contextualize information for you, there are two big traps to watch for: hallucination and sources. A.I. platforms can and will just make stuff up. This is the infamous hallucination problem, where an A.I. will tell you what you want to hear (without any evidence) or just confidently spill oodles of bullshit. And that brings us to the second point: sources. It’s always best to prompt A.I. to provide sources, allowing you to examine those sources yourself. This is where skills with A.I. start to come into play, and if you’re going to use A.I. for research, you’ll want to practice: prompt the A.I. into explaining where it’s sourcing its information and how it’s synthesizing it.
Maximum A.I.: with time, effort, and tweaking your conversations with A.I., you can potentially have it generate entire reports on the subjects your researching. You could even train it enough to provide you with quick references, bulleted lists, and other reference content that relates directly to your work. This goes beyond just a search engine, and would have you adopting your A.I. agent as a research partner. And that’s a key idea here: the A.I. is your partner in research, not your researcher. To achieve real results, you must spend the time to work with the A.I. to teach it exactly what you need and to set expectations for well-supported information.
Ethics: the nice thing about research in this topic is that it’s something that you’d otherwise be doing exclusively by yourself (or nearly so). And because you’re synthesizing content out of previous work or simple facts, you wouldn’t be transferring A.I.-generated content to your audience.