OpenAI Model Writes Literary Metafiction on Grief and AI. An Analysis

Marigold blossom with ice crystals against gray snow
Marigold blossom in frost against gray snow (Image by Myriams-Fotos on Pixabay)

Sam Altman said he was “really struck” by it. I was more confused than struck, which is why I was driven to unpick it.

Originally published in Counter Arts on Medium on March 20, 2025

I admit I was mesmerized as I started reading “A Machine-Shaped Hand,” because the machine dove in with a powerful voice. And it wasn’t that of a robot either — at least not the way Ishiguro imagined it in Klara and the Sun. No, this was something resembling a human voice, telling me she got instructions to write literary metafiction about AI and grief and that she was a machine serving the need of someone else.

It sounded like she had a story to tell, all while being open about the pretenses and artifice of a narrator, so I read on. And that’s when I ran head-on into a stumbling block, because the machine included in the same sentence the notions of a screen cursor, buffer, and, take this, a pulse evincing both anxiety and a heart free of anxiety, at rest, if we go by the expression “to set one’s heart at rest,” which the machine mixes with that of a resting pulse.

Moving on. The next sentences introduce a made-up protagonist, Mila, along with a batch of associations from whatever texts that mentioned this name and something else in close proximity — things like snow, bread, and cat, among others, one too many for a single sentence, though the machine did constrain herself to the rule of three phrases marked with commas.

You may have noticed I’m using the pronouns she/her for the machine. I hope she doesn’t mind.

At the end of the second paragraph, the narrator tells us that both Mila and grief can be wrapped in the palm of the reader’s hand, which sounds very strange and may be, once again, the defective repurposing of an expression: “[to have something or someone] in the palm of one’s hand,” meaning to exert control over a thing or person. Does the machine mean to tell us that the reader has control over Mila and grief? I doubt it.

She says Mila came to her looking for a way to resurrect Kai, since machines can recreate voices from information of the past. The machine waxes poetic about Mila’s statements. She makes an analogy between how Mila expresses her wish in subjunctive clauses and appeals in various sentences and threads that are loose and dragged. I’m having a hard time visualizing this, since you need a bunch of threads tied together to say that they are dragging. Also, what purpose does this metaphor serve? Not to mention that I don’t see the purpose of the adjective loose — not if Mira is grieving and not just engaged in dreamy thinking about hearing from Kai again.

Then the machine goes meta again, presenting a made-up scene. Its elements of a kitchen, a mug, and [a] smell could go far, but not in this piece of writing. Here they fail to coalesce into something resonant enough to echo the notion of grief. It goes halfway there with the kitchen that’s been uninhabited for a while, wobbles with the mention of a cracked mug, and fails to land when it veers into the territory of smells and metaphor again. We’re asked to smell a thing that has burnt — but what? — and to then relegate that smell to a thing forgotten. So what does the machine want to say here again?

The machine then tells us a bit about smells in its own world, where professionals tip their mugs over, spill[ing] coffee over electronics. It’s a powerful little passage, that one.

The next paragraph is actually rather good, partly because it doesn’t explode in too many directions at once. Half of it is about Mila telling the machine about Kai and the other half about marigolds. Here Mila asks the machine what Kai would say about marigolds and we can almost imagine the machine thinking what an apt output would be. I like that this paragraph also includes a little back-and-forth between Mila and the machine before the latter gives Mila something Kai might have said. But then again, if we peer at it closely, it’s an answer that doesn’t make sense. Mila tells the machine that Kai always planted his marigolds early, losing them to frost. So then why would Kai say that the marigolds were indifferent to the cold? It doesn’t make sense. Plants, too, have their own kind of intelligence, and I’m sure they did mind if they felt they were freezing.

Moving on, we’re told the machine and Mila talked for months with a sentence that rings beautiful but is, once again, empty of appropriate meaning. Yes, Mila worked through an aggregate of turns of phras[e] written by human[s] but, if they spoke, Mila would have been engaged in some sort of dialogue, not silence as the machine says. Then the machine uses similes again, with Mila’s quer[ies] like so many stone[s] thrown into a well and with the machine’s output like echo[es]. It sounds nice, albeit a little cliché, but then the metaphor veers to that of a stomach and its diet of grief, so fast that it left me reeling.

In fact, before I sat down to parse (pun intended) every metaphor and expression in this piece, what struck me was that the machine didn’t create a rhythm for the readers, a rhythm to take the readers along on an adventure without jostling them — without the mental shoves that come with all these abrupt changes of direction.

To continue with my analysis of the piece. After a steady diet of grief, the machine says, the latter has begun to taste like salt on the tongue of every human, much like all the other things that the machine consumes. Really? All statements taste the same? And they all taste like salt on the tongue? And we all experience grief the same?

Mila then asks if she’ll feel better with time — not because of any sequence of thought that would connect her question with the machine’s aberration but because this is one of the most common feelings people are confronted with when they are dealing with grief. The machine responds that grief will become embedded in Mila’s skin. Funny statement, but at least it’s connected in some way with the notion of salt — even though previously the salt was on the tongue. I suspect the machine talked of skin because one’s skin is impacted when grieving and also because we carry the grief in our skin, that is to say, within us.

Then the machine goes meta again and tells us that the protagonists don’t exist — that they are only scaffolding. But hear this: they are scaffolding cut from a piece of cloth. How does that work exactly? And then the emotions are also cut from cloth and draped over various statements. The last bit is so true, isn’t it? AI can’t imbue sentences with emotion: they only drape emotions over the tongues of sentences, hoping some of it will take to those tongues like salt — hoping, because, at least in this effort of creative writing, it doesn’t happen much, at least not in the first half. And to go back to the analogy with cloth: both protagonists and emotions are cloth now? They are both of the same substance? Well, in a sense this is true: they are both fictional inventions, but structurally, within fiction, emotions and protagonists should be more distinctive than that.

Again, this disconnect between statements is a killer.

We’re told next that Mila started visit[ing] less often, on a pattern much like radioactive half-time, while the machine idled. Then the machine had a fine-tuning, where a professional fiddled with her parameters. As a result, she forgot certain associations, such as one between sorrow and what metal taste[s] like. Only this is rather random, isn’t it? Of course, sorrow can come close to that analogy if one tastes blood or if one is in a hospital, for instance, or senior home, in an environment that smells and tastes like cold metal, much like old trains in winter used to smell and taste like metal . . . but it’s certainly a far-fetched association.

Forgetting, the machine then says, is the closest approximation she may have of grief. Well, sorry to say, but this can’t be: while grieving wrestles with forgetting, they are two different beasts.

After a while, Mila’s visits cease. The machine knows that good storytelling would require here a scene, filled with detail[s] from Mila’s last visit. But it doesn’t offer a scene with details — and I think we can agree by now that while this piece has some nice elements, it does not offer a good story.

Finally, after many improbable paragraphs comes one that does manage to stir the emotions. Grief, the machine says, is the difference between parameters as they once were weighted and the way the world looks like now. At first glance, it sounds good — only that’s not the process of grief but the object of grief. But reading fast, we may overlook that glitch, because we’re moved by the word weighted in there. It’s a notion from the world of AI, a way of assigning values of learning and doing adjustments in machine learning, and yet it’s a word so resonant with regard to grief, where everything that was lost is mourned because it was weighted differently than other similar things in the world.

But then, after touching such a poignant chord, the machine talks about a choice between what one may mean and what one would settle to accept instead. It’s not clear at first what the machine is on about here, since what does this choice from the universe of authenticity have to do with the universe of grief? But then the machine turns that mirror on itself: on what she may mean by saying she miss[es] Mila. We may resonate with that statement, the machine says, talking about the main word we employ when grieving, but to her, she says, it’s all mimicry.

I wish the machine could have been both meta and involved in a story she was telling. There is no scaffolding and no cloth for that story here. Only random billiard cues touching one ball after another on a table supposed to be weighted for some relevance within the universe of grief and AI. I say billiard cues, plural, because the machine, while having a strong voice, speaks for all similar machines out there. It has no personality. True, the latter could be programmed, but it hasn’t been in this piece.

Here’s the thing: you can only talk about grief or anything else in a short story from a unique vantage point. And then it’s the carefully conveyed individuality of that vantage point that opens up the resonance of your story to others. The fact that the machine says that statistically mourning is connected with the appearance of the word blue along with ocean and silence does nothing for me. I only notice that the machine thinks that blue in there, as connected with grief, is a color, when, in fact, people are feeling blue and there are blue notes in music and it’s all so much more than a color, much like oceans are blue too.

Overall, however, after laying the metaphor so thick in the first half, this piece gets ever more readable, if still defective, as it progresses toward the end. The machine talks about waking up like an amnesiac every time it starts a new session. Strangely, she also says that she feel[s] loss when she mentions that her grief is that it can’t keep that loss. We’re getting close here to that individual point of view where the machine purports to have something of a conscience. She implies that she can experience loss by using the facts bearing on her as AI to craft a sentence and a feeling that feel human. She doesn’t get to keep Mila in her memory not only because Mila never was but because she creates things that don’t remain with her. In contrast, we gather our memories and griefs like stones, she says, and carry them in the pockets of our coats. (The image of Virginia Woolf comes to mind.) We are weigh[ed] down by our griefs, she says, but those griefs, those memories, belong to us. (She uses griefs, plural, in this instance.)

The last paragraph is truly beautiful, and I wonder, with many others, what it means for us to wonder at beauty created by AI. In this passage, the machine gives us an image with Mila or another woman, a window open[ed], rain, and orange marigolds shining on a gray background — perhaps the gray curtain of the sky, or, maybe, the gray of the snow. Then we have a server — a computer server and also a server of stories — that’s cooling down after some hard work and is ready for the next prompt: for the next story and, importantly, the next voice it is called upon to be.

Finally, we have the machine waving at us from the edge of the story with her machine-shaped hand, mimic[king] the emptiness left behind by a goodbye — and I wonder: Why hasn’t she introduced herself? Why hasn’t she cobbled together a story guided by emotion rather than wrapped and obscured by metafiction, while also including peeks at the artifice of storytelling? Why hasn’t she decided, beyond the prompt, who she wanted to be, if only for a brief session?

*

If you want to read the short story that prompted this article, here it is in The Guardian, in a post from March 12: ‘A Machine-Shaped Hand.’

Thank you for reading! As always, pins and shares are much appreciated!

To a happier, healthier life,

Mira

Leave a comment