Statement on AI Use

I don’t use AI for creative writing.

There. That’s it. That’s my statement.

Well, maybe there’s a bit more to say.

I’m not generally opposed to AI. I’ve found the tools useful in a range of applications. I’m also just plain fascinated by them, and I’ve read a bit of auto-generated fiction, trying to get a handle on the hallmarks of the style. I’m open to the possibility that authors will find exciting new uses for these inventions.

Still, I’m fundamentally an old fogey of a writer, with a preference for old-fashioned ways of generating prose. I built my repertoire of habits and hangups in an earlier era, when writing meant sitting down and choosing words yourself. I like that tradition. I want to keep it alive. And I’m interested in associating with other people who also value it.

If we’re going to build communities where different kinds of writing are championed—fully mechanical, fully human, and everything between—we need ways of distinguishing among different types of craft, much like musicians who specialize in either digital tools or conventional instruments. I don’t think it’s wrong for AI-boosted writers to get together and gab about their interests. But I also don’t think it’s wrong for readers to express a preference for human-made work, or to inquire into a writer’s process. So, when I say I do my writing myself, here’s how you can know I’m for real.

First, I have a twenty-year track record of publishing fiction in both professional and independent venues, going back long before the age of AI. You can find a partial list of those stories on this site. I hope you’ll check ‘em out!

Second, I encourage readers to use the AI detector Pangram on my prose. Just be mindful of the tool’s limitations. It works poorly with short clips and excerpts, it tends to return somewhat iffy percentages, and it will occasionally give wrong results. The best route is to test multiple long samples, preferably of 500 words or more. If they mostly come back human-generated, there’s almost surely a real brain behind the words.

Third, I’m interested in the details of the writing process and willing and eager to discuss them at length. Versions? Revisions? Logs? Notes? Embarrassingly amateurish first drafts? I can tell you—and show you—more than you’d ever want to know.

Finally, I believe in full disclosure. In that spirit, here’s a detailed breakdown of where I currently do and don’t use LLMs.

Yes

• For vibe-coding: making games for home use, building simple tools to handle clerical tasks, creating websites like this one, working on projects with my son, etc. -For troubleshooting: explaining error messages, summarizing technical documentation, providing basic computer help, etc.
• For enhanced search: getting book and article recommendations (most of which have been terrible, so far, but that’s another matter), locating books I read years ago and can almost but not quite remember, building lists of associated keywords to improve standard searches, etc.
• For text-to-speech: I use AI narrators where human readers aren’t available. For general reference: explaining mathematical concepts, paraphrasing dense texts, providing help with foreign languages, etc.
• For experimentation: testing the limits of current models, training myself to recognize generated writing, tinkering with detection tools, building AI fluency, etc.

No

• For writing: I don’t use AI to generate prose, and I tend think most such uses should be watermarked, flagged, or freely disclosed.
• For editing: I don’t ask AIs to offer feedback on my prose, but I have no objection to others doing so, and I don’t think this kind of usage needs to be disclosed.
• For brainstorming: I don’t bounce story ideas off AIs, but that’s mostly just a personal preference. I’m currently unsure what to think of this practice.
• For spellchecks: I don’t rely on AI-powered spellcheck features, but I don’t have any objection to others doing so, and I don’t think this kind of usage needs to be disclosed.
• For grammar checks: I’m personally leery of AI grammar checkers (I may produce solecisms, but they’re my solecisms, damn it)—but honestly, this seems like a perfectly legitimate way to use the tools.

It’s Complicated

Research: Oh, man. Research is a tough one. I’ve experimented with a slew of different approaches, and I’m still not sure how to proceed. The problem isn’t really that AIs spit out errors. Those are fairly easy to check. Nor is it that they quote other writers’ work—all you have to do to get around that hazard is avoid cut-pasting chatbot answers into your manuscript. The problem is that they editorialize relentlessly, insinuating themselves into the creative process. If you ask an AI to recommend some books, it’ll cobble together a whole research plan. If you ask it for names of near-future technologies, it’ll start speculating about sci-fi applications. If you ask it for background material on the Everglades, it’ll ask if you’re planning a trip, then throw together an itinerary. OK, thanks, I guess—but that’s my job, buddy. Sure, you can train the bots not to be so voluble. One way or another, though, their insistent helpfulness, their weird mix of pushiness and submissiveness, becomes both seductive and intrusive. I still haven’t found a good solution to this problem.