This article was written entirely by a human

Recently, at my day job, I was tasked with writing a blog post for the company website, about Google Analytics and whether you should be using it to track your website visitors. This may sound like a pretty dry assignment, and I guess it was, but I always enjoy putting together blog posts at work. It feels like a life hack, getting paid a tech job wage to write. Even if it's not subject matter I'd otherwise choose to write about, it's always an interesting challenge trying to engage an audience — especially on a topic they might not otherwise choose to read about.

Also recently, I took a week off work to attend a creative writing workshop with a group of nine other fiction writers. As part of that, I needed to read each of those writers' workshop submissions and offer my thoughts on improvements they could make. This isn't as easy as it sounds; the challenge with workshop critiques, much like when I was writing that blog post for my day job, is that you often find yourself reading stories you'd never choose to write yourself, and in styles you might never dare to try. When it comes to giving constructive feedback on work like that, you really have to struggle sometimes to grasp what a piece is trying to do, and to understand it on its own terms. It can be a fairly robust exercise in empathy.

What really struck me about it this time, though, having not been in a workshop since getting my MFA in 2010 (👴🏻), was that "exercise" is exactly the right word for it — because it really felt like working a muscle. It took me some time to warm up again, after so long away, and the first critique I finished was definitely stiffer and a little less effective than the later ones. By the last, though, I felt limber again, supple, and when I was finished I had that same, satisfying feeling of having worked out. The feeling that, hey, look what I'm capable of when I'm really trying.

More to the point, like when you do a leg day after months of working only arms, I could feel a slightly different set of brain cells firing, compared to when I write and revise my own work. After all, I always know what I'm going for, which is rarely the case when editing other people. So it felt really useful, taking the time to go back to that very different skill — and trying to identify what someone else had been going for, based only on what was on the page. It felt like I was really thinking about writing more deeply than I had in a long time.

Anyway: back to that blog post for work. After I sent it to my boss, she came back to ask for revisions. It seemed, she said, that I was really trying to make two slightly different points for two slightly different audiences, and it would be better if I just split the thing into two separate articles. I had to begrudgingly agree; I'd actually had the same feeling myself as I was writing it, and had been hoping I'd got away with smushing them together.

I also knew it was going to be a real slog untangling them again. But we had just come off a company meeting enjoining us to find ways to use AI to improve our work, so I tried something I would never do with my own creative writing: I fed the original post into Claude (with my boss's blessing), told it about the two audiences I was trying to talk to, and asked it to split the thing up for me. It was curiosity more than anything; I wanted to see just how much I should be worried about the robots taking my job.

It will perhaps not surprise you and almost certainly relieve you to hear that what it came up with was awful. I never in a million years would have considered using it, even though the alternative was working extra hours to do it properly myself.

On the other hand, while the writing itself was terrible, the AI did a pretty decent job of separating out and reorganising the distinct points that were relevant to each audience. So while I did still go back to rewrite the articles myself, I did so with something like an outline for each, which undeniably sped things up. I'd written the original post from scratch, and I ended up writing the two new posts from scratch as well, so a human had still done the important work of having an opinion and expressing it in writing — but the AI had removed a somewhat tedious part in the middle of the process.

This, I suppose, is the promise of AI: automating the tedious tasks so humans can focus on the creative ones. And as far as using AI in content creation goes, I would say generating an outline from a draft you've already written yourself is at the innocuous end of the scale. Clearly, though, even that is debatable. Recently, award-winning author Olga Tokarczuk said that she used AI in similarly innocuous ways, by asking it to do historical research for her, and suggesting ideas for how to develop a story — and the controversy was swift and withering. Even the headline of that linked news story above implies wrongdoing: "author admits using AI," as if a crime has been committed.

That's because many publishers have staked out positions on AI that forbid all but its most mechanical uses. According to Bloomsbury's AI policy, for example — which I quote only because Bloomsbury have made their policy so transparent and available online — using AI for proofreading is okay, but other uses, including outlining, are prohibited, as it "raises concerns around authorship [and] originality." I confess, too, that when I plugged that blog post into Claude at work, I could see where AI policies like that were coming from. I felt slimy doing it, as if I were cheating somehow.

Then I went to my writing workshop, and the waters seemed to muddy somewhat. Because the whole premise of a workshop is to have other readers dissect your writing for you, and help you see the things you've overlooked yourself. The feedback I got in workshop helped me understand what I needed to do next, and gave me something like an outline for continuing — just like Claude had with my blog post at work.

Of course, it would be ridiculous to convene an entire workshop just to revise a 500-word corporate blog post that's mostly there for SEO bots to digest anyway. Indeed, when you look at job descriptions for this kind of content creation, they almost all emphasise that you should be able to file copy without needing someone else to edit it. So in this case it seemed like AI wasn't removing humans from any real-life process they'd otherwise have been involved in. Rather, it was providing a new opportunity for a helpful feedback loop, one that this human would never have had otherwise. It suddenly seemed a little harder to fault that.

In fact, as Tokarczuk put it, "collaboration with artificial intelligence" can clearly give writing advice to lots of people who might never get it otherwise. In that light, the fiats contra AI begin to seem less like protection for creative expression, and more like propping up another structural barrier to equality, restricting authorship to only the people who can spare the time and money for human feedback and instruction. While it would be great, of course, if every aspiring writer could engage an experienced editor for advice, or take a week off work to attend a writing workshop, like I did — or heck, to take three years off to do an MFA — requiring that level of privilege to be an author is not, I don't think, a great editorial policy. If nothing else, it can lead to work just as homogenous and repetitive as prose generated using AI.

I'm also deeply uneasy about evaluating AI using standards that we've never previously applied to creative work. (A few years ago I made a similar point about protesting AI on the grounds of its energy usage.) It's legitimate to say that using an AI to create an outline "raises concerns about authorship." But as my workshop showed, revising from an early draft based on external feedback is also fundamentally the process of writing. You write something; you get other people to tear it apart; you write it again. And while editors and agents and writing workshops often get mentioned in the acknowledgements of a novel, there's no formal policy requiring you to disclose your "use" of any of them — even though some workshop critiques and editorial notes end up being particularly formative for later drafts. Should that raise concerns about authorship too?

The more you think about it, in fact, the more painfully evident it becomes that literary authorship has never been confined to a single person. If we start leaning too hard on that belief in order to justify anti-AI policies, even sensible and legitimate ones, the concept of authorship might start to collapse in unexpected ways that are just as undesirable.

One last observation, before I wrap up: I've revised this blog post dozens of times now. I've softened and rewritten contentious sentences over and over again, I've cut sections entirely, and I've tried to examine the issue from every point of view I can imagine. And this, by the way, is exactly the sort of deep thinking and creative work that AI restrictions are trying to protect.

But I've also come close to just scrapping the entire thing several times, and I'm still terrified of putting it out there in public — because I am an author, and I worry that if a prospective publisher ever reads this they'll instantly disqualify all my work, just for having fed a single blog post into an AI at my day job, or for having the temerity to suggest that might have been okay.

To be clear, other than that one experiment at work, I've never used AI to generate content, and I don't plan to ever use AI in my personal, creative work — and the fact that I feel compelled to say so in such uncertain terms may well say more about my anxieties than about the realities of publishing industry. But equally, if that's where the AI debate is right now, then it's plainly stifling creative work at the same time as trying to protect it. When the default position from the industry's gatekeepers is that any use of AI is assumed to be wrong unless proven otherwise — when even a Nobel laureate can get dragged over the coals for "admitting" that she used AI — that has a chilling effect on authors even talking about it, or thinking about how they could use it creatively, and that's bad too. Because it's important we have these discussions, and parse out what, if anything, is okay. That seems like the only way AI won't devour us all.