One of the largest debates about large language models (LLMs, imprecisely called “AI”) and authors has to do with rights: who owns what? Who consented to what? Are creators being fairly compensated? Those questions matter, and they’re not going away anytime soon. But the rights argument doesn’t explain the sheer intensity of the opposition to LLMs among authors, especially since there are plausible solutions available via tweaks to content licensing, copyright law, and compensation structures.
The other big argument centers around quality, but even that isn’t quite enough to explain the absolute fury I’ve seen in some quarters. Poor-quality writing has never been in short supply, and it has never posed a threat to high-quality writing, even though it has always existed in greater volume. Before now, though, it was largely treated as an annoyance. Authors ignored or avoided it (including, at times, looking down on those who self-published, but that’s a separate conversation).
As a writer who uses an LLM for developmental editing*, I’ve pondered why there seems to be such vehemence in its opposition. Ethics, rights, and quality aren’t minor issues — they’re very real and very important — but the emotionality of the reactions points to something more.
I suspect I know what that “something more” is: writers and non-writers alike expect writing to be hard. Anything that makes it easier is automatically suspect.
For a long time, effort has functioned as a sort of proof for creators, including writers. It’s a defense against the common accusation that we’re “just daydreaming” or “not doing actual work.” It’s also something that’s much easier to measure than quality, a determination that can become subjective fairly quickly. Effort, then, was a measure not just of how much work went into a creation. It was also an indicator of credibility and legitimacy. In addition, effort reflects skill, care, and discipline, so it’s evidence that creating is work.
Most writers — and most readers — are quite accepting of the use of LLMs for mechanical support (i.e., spelling and grammar). Within author circles, the use of tools for assistance with outlining and plotting is also long-established. So is the use of critiquing for refining the prose and making it “pop.” The use of LLMs strictly for research also isn’t strongly opposed.
All of these are tools to make writing easier. The same is true for the use of generative LLMs in drafting, but that evokes shrieks of dismay. Those of us who use LLMs are attacked as “not creative,” “wanting shortcuts,” and even “unethical.” This isn’t an abstract concern; I’m disqualified from some significant writers’ groups as a result of my use of an LLM, and when the time comes for me to seek an agent, it will be an obstacle that requires explanation. (I’ve even felt the need to explain it in this post, via use of a footnote.)
So what’s the difference? How is objection to use of LLMs any different than objections to use of typewriters and computers to write — which, by the way, actually happened?
I think it’s because generative LLMs reduce effort during the drafting phase. Or, in other words, they make the actual writing less difficult, thus reducing the amount of effort needed. And that strikes at the very heart of creatives’ defensiveness about whether they do “hard work.”
If the amount of effort needed for good writing is proof of skill, care, and legitimacy, then effort matters. Tools that reduce friction in other ways, or around the edges of that process, aren’t threatening. But something that reduces effort, right at the center of the process, is going to generate a huge amount of opposition. If something doesn’t take effort, the reasoning goes, it’s probably not useful or good. “Easy” writing is also easily dismissed as sub-standard or “slop.”
But here’s the issue with that assumption: effort and quality aren’t always related. In fact, they’re more frequently unrelated to each other. One of my stepchildren used to struggle for hours with writing a basic five-paragraph essay. Another one could crank one out in about half an hour. But, aside from mechanics, there was no substantive difference in the sophistication of their reasoning or the quality of their writing. They simply learned and thought differently — one was an auditory/verbal learner, one was a kinetic/mechanical learner — and the latter can be difficult to express using the former.
In other words, effort isn’t a placeholder for quality or value. It never has been, but creatives have often used it as a proxy this way. (That’s not limited to creators, but that too is a different conversation.) That’s why something that reduces effort is seen as being so very problematic. On the surface, it takes away from the legitimacy of writing as a vocation or career. Keeping writing difficult also serves as a gatekeeper for authors themselves: if writing takes a lot of effort, fewer people will do it. But if writing becomes easier, more people will “break” into the field.
The question, then, isn’t whether LLMs create (or replace) “good” writers and “good” writing. The question is whether reduced effort means less credibility.
I don’t think it does. But I do think it makes it harder to quantify the process of writing. Instead of looking at effort, we’re pushed back to looking at what’s actually on the page: does it show original thinking? Is it clear and coherent? Does it provide insight? Is there a strong and unique voice? Those aren’t as easy to evaluate since they can’t be assigned numerical values. They also require some subjective judgment: original as opposed to what? Unique as opposed to what?
But the fact that these types of evaluations are subjective doesn’t diminish their value. They’re just harder to do, and they emphasize that aesthetics itself is also subjective. De-emphasizing effort in the writing process actually increases effort in the evaluative process. The work hasn’t disappeared. It has just moved, and that really does seriously threaten some of our baseline expectations.
* Primarily brainstorming, checking structure, refining prose, and research support. Every word I actually write is my own.