The Prosthetic Voice

On AI, authenticity and the new artificial voice.

An idea has started to percolate repeatedly that we shouldn’t even open an AI tool if we have a creative thought. Which is wild, because we’ve all been Googling at will for thirty years and nobody wrings their hands about that.

I think the distinction depends on the kind of writing. When I’m working in a more journalistic, essayistic mode, I love having an editor like Claude. I write the draft, then ask it to line-edit and catch what I’ve missed. If I’m editing my own work before I post, I always miss things — and see them only after it’s live, knowing the version that went out had typos and formatting glitches I’d glossed over. Getting those cleared out before posting is genuinely useful. I honestly wish I had a paid editor and worked for a famous magazine, but it’s not happening and those days are long gone (kiss Benchley for me).

When I’m stuck for an ending, Claude is good at the buttonhole. To help me close the loop. The suggestions I don’t want, I say no and delete. I’m in charge.

What I feel less at peace with is AI-generated writing, whole cloth, ideas on a platetr, whever they came from out of the primordial online muck and plaigiarism. I saw something today on Substack (having written off Facebook long ago) that was written with AI, top to bottom. Substack still feels to me like it should be different. Readers know what that “AI voice” sounds like now. It’s strident. That cadence is becoming the default register for so much writing. I miss the style of a human voice.

Literary style matters.

And yet sometimes a story is worth being told even if the prose isn’t beautiful, and even if someone needed AI to get it onto the page. Still, using AI to write the story seems to disrespect the story.

But what about the person who simply can’t write well enough to tell the story? The Substack poster who wanted to write about their dying father, the agony of living with dementia, and a kind Uber driver — probably composite, probably shaped in the retelling, but emotionally true — and just can’t write it on their own? Is that story worth telling if AI barfs it out? I think it is.

Or this, as long as we’re travelling through Metaphoria: no one was going to deny Stephen Hawking a voice because ALS took his own voice. He had new and incredible insights to share. And he said them in the fake voice. But what he said was still every bit as valid as his human voice.

Or Jean-Dominique Bauby — the French journalist who suffered a stroke that left him completely locked in, able to move only his left eyelid, who dictated an entire memoir that way, one blink at a time. The Diving Bell and the Butterfly exists because someone found him a mechanism.

Will we apply a different standard to people who have genuinely important things to say but simply aren’t equipped to say them on the page? Working fiction writers could sit down and have a short story drafted in two hours. Most people can’t. What do we do — say, you can’t use a tool that helps you tell your story?

What about older relatives and wanting to get their stories down? I would a thousand times prefer to have their AI-assisted written memories than nothing at all. I would recognize their human self in the words.

Ghost writers have existed forever. That’s not a new ethical problem.

Even when I am reading human-generated writing, I judge it on its merits. A bad story told beautifully is still a bad story. A good story told badly is still a good story. Classic literature is a good story told skillfully.

At present I don’t think that contemporary AI can approach a human voice. There was a fantastic essay by Will Self¹ — recently on X, reported and reposted in part by my new hero Becky Tuch. The words are Will’s and he lays out why AI can’t recreate a wholly human voice.² His argument: AI works at the sentence level. Individual sentences can be structured correctly, even elegantly. But current AI can’t assemble a piece the way a human mind does, where sentences cohere into something larger than their parts. The sentences are modular. They don’t add up.

And tone. The default register of LLMs — that striving, relentlessly affirmative stance — it’s just too much. Like an evangelical grandmother. I don’t need every point hammered home at volume. We’re losing emotional depth and spectrum. A flattening out, if you will.

I want to convene a panel on these AI questions. The gradations of AI use. What constitutes an AI voice. The cases where AI is genuinely justified — and where it becomes a prosthetic for someone who simply needs one. Let’s not conflate these conversations.

I’m also noticing that people who claim they never use AI are using AI to write and edit. Absolutely this is happening. I can sniff it a mile away. The question becomes — why is it so important to say they wouldn’t, while they are? Why the hiding? Did we collectively decide that using AI makes someone inauthentic, and then quietly keep using it anyway? Is it shameful to use an editor? To seek to post the cleanest copy possible? The gap between public position and private practice is worth unpacking on its own. I don’t care if you write with AI. But if I think the writing is boring or jacked up, I will judge it on its merits and move on.

We’re not going back to quill and parchment. We’re not pretending the internet doesn’t exist.

I don’t have clean answers. But these are my questions.

Thoughts? I’d genuinely like to know. This thread is open!

1. If you don’t know Will Self, inform your(self): https://en.wikipedia.org/wiki/Will_Self.

2. Self’s piece was originally posted on X, but I deleted my account around the time of the hostile takeover. Grateful to Tuch for her summarizing and posting. Go and subscribe to her if you like to read new writing as much as I do!