"The River" Revised by AI

Comparing an old poem and its heinous AI revision.

The River

The woman told me the saddest thing I had ever heard. I told her I loved her because of what she had told me. Her expression soured. She warned me not to love her for her telling me that. She told me it was okay, and maybe even good, to love her—only not for that. I responded that I did not love her for that, exactly, and that she had misunderstood me. I admitted that why I loved her was related to what she had told me, yes, but only tangentially, and was that alright? She asked me to elaborate, so I told her that I loved her, not for the thing she had told me, but for the courage involved in telling someone something like it, something that sad, which seemed to me to be a great deal of courage—and I told her I also loved her, though far less than for the courage part, although plenty still, for the way in which she told it to me, which I explained had been, in all seriousness, eloquent and mesmerizing. She had a small build and at that point she laughed like a flower, wilting and blooming. Her nose was in the center. I decided to show her the river. I picked her up in my hands and carried her, crisscrossing back and down through the steep and elaborate cragwork of the slope of the riverbank. When my feet were finally in the water I looked at her and said, the river is deep, and fast, and it drowns many people, but I still love it. I still love the river, I told her. But I do not love it because it is deep, and fast, and drowns many people. I love it because it runs behind my house, and I have lived above it forever.

This poem is from my out-of-print first book, Beauty Was the Case That They Gave Me.

I asked ChatGPT if it was any good. It praised the poem endlessly for its metaphorical and descriptive detail—praise I thought was dubious. I think the metaphors are predictable and the description is only detailed if you’re not familiar with good descriptive writing.

It’s not that I think the poem is bad, it’s just that its strong suits aren’t its metaphorical and descriptive precision. Perhaps this poem demonstrates that predictability of metaphor, or clumsiness of description, can make a poem good, if, say, those things bring a particular voice to life. For me, the voice of the narrator is much more specific than the description and metaphor. That said, even the voice is not so uncommon that you couldn’t encounter a similar voice in a thousand other poems.

In any case, I asked ChatGPT if it could revise the poem to make it even better. It said it could easily do so and seemed eager to prove itself—which surprised me because it had just been praising the poem to no end. Here’s the AI version:

The River

She shared with me the saddest tale,
I told her my life would never fail.
She frowned, warning not to love her for this,
I reassured her that’s not what I’d miss.

Not for the tale, but her bravery to tell,
And the way her words cast a spell.
Her small frame laughed like a flower,
And I knew she had my heart’s power.

I took her to the river to show,
The place I loved that others fear to know.
Deep and fast, it drowns many lives,
But I love it for reasons that thrive.

It flows behind my house, my lifelong friend,
A constant presence that never bends.
Not for its darkness or dangerous flow,
But for the memories that it helps to grow.

So, I love her for her bravery and grace,
And the river for the beauty in its space.
Both are tied to me with a special bond,
And the love I have for them is ever so fond.

I suppose I’m most surprised at how little I hate it. Though maybe that’s more a reflection of how I have changed over the years—away from a more defensive mindset when it comes to aesthetic choices I might have previously reflexively dismissed, and toward a more open or perhaps indifferent mindset. The older I get, the greater tolerance I seem to have for amateurish art and poetry. Not sure why that is, but I’m grateful for the shift since most poetry and art available is amateurish. Maybe everything on the internet is so commoditized and aligned with corporate interests that “bad” writing is the last bastion of authenticity. It’s tragic then that I can already see the “bad” writing of a robot programmed by the greediest people on Earth as an exemplar of redemptive amateurism. I don’t want to keep thinking about that.

I admire the robot for how it seems to understand things that I myself didn’t understand about the poem, such as the fact that the reason why the speaker loves the river is for “the memories that it helps to grow.”

Whenever I’ve thought about the original, I wasn’t aware that it was memory, specifically, that makes something that is familiar resonate with heightened meaning; however, now that this AI-language model has made the theme of memory, and the things that anchor memories, more explicit, I do get some kind of new understanding—of the poem and of myself—from the artificial intelligence’s revision.

Maybe an AI translation of a human-generated poem is less like poetry and more like literary criticism. It can help one illuminate themes or vectors of meaning that one might otherwise have remained unaware of. And like literary criticism (when it is good), it isn’t hard to dwell on the parts that are insightful while discarding the rest.

I don’t always say this, but thanks for reading and subscribing.