Tag: authorship

A note on process. September 2026.

I built a 44-card oracle deck with a machine. I wrote an essay asking whose hand is really on the instrument when an AI helps finish a lyric. I’ve used AI to draft artwork, restructure this whole website, even to sketch the sentence you’re reading right now before I rewrote it in my own words. AbeArtsSweden is, in a lot of visible ways, a collaboration between a person and a set of tools that didn’t exist a few years ago.

And then there’s Testimony.

Seven episodes in, heading toward an eighth and last, that series has never touched a language model. Not for structure, not for phrasing, not for the parts that are hardest to get right. It’s the one corner of this site I’ve kept deliberately, stubbornly analog — and lately I’ve been trying to work out why, because “it just feels wrong” isn’t a real answer, even to myself.

Craft versus witness

Here’s the closest I’ve gotten. Most of what I make with AI is craft — music, visual art, an oracle card’s little myth, a think-piece about authorship itself. Craft has room for iteration, for a tool that offers you eight versions of a line and lets you pick the one that lands. An AI in that role is closer to an instrument than a co-author: a synthesizer adds range to a voice without pretending it had the experience the voice is singing about.

Testimony isn’t craft. It’s witness — a record of getting a lymphoma diagnosis in Sweden, of hospital corridors and taxi rides and a body doing things I didn’t ask it to. There’s nothing to iterate on. There’s only what happened, and my job is just not to smooth it into something better-shaped than the truth was.

What fluency costs

This is the part of the AI conversation I think gets skipped in favor of arguing about jobs and copyright, both real arguments, but not this one: a language model is trained to resolve things. Give it a fragment and it wants to complete the sentence, tie off the paragraph, arrive somewhere satisfying. That instinct is exactly what makes it useful for a lyric or a card description — and exactly what would ruin a memoir. Memory doesn’t resolve. It repeats, contradicts itself, gets a date wrong and a feeling exactly right. The awkward, unfinished texture of an actual account is the whole evidence that it’s true. Hand that to something optimized for fluency and you don’t get a better version of the story. You get a more readable version of a different, false one.

So the line isn’t “AI good, AI bad.” It’s narrower than that: AI can stand in for craft, and craft can be shared, remixed, iterated on in the open — which is most of what this site argues, in the Reflections essays and the Whistle of the Signal piece on music authorship. But witness has to stay singular, or it stops being witness. The tool doesn’t know which one it’s holding unless the person using it decides first.

Two answers on the same site

I don’t think this makes me a purist. It makes me someone who had to draw the boundary himself, on purpose, rather than let a default setting draw it for me. Whistle of the Signal and Testimony sit on the same domain, a few clicks apart, and they’re really two answers to the same question — what belongs to a machine, and what belongs to a person alone. I’d rather keep answering that question deliberately, episode by episode, than let it get answered for me by whatever the tool happens to be good at that week.

If you want to see both sides of that line: the Whistle of the Signal essay is where I think out loud about AI and authorship in music. Testimony is where I don’t think out loud about anything — I just tell you what happened.