The B Poem
Aren’t we all just a collection of our influences?
I used to write poetry a lot, but that was back in college when I had assignments. I never found my voice. I always wrote like the writers I admire, and I still do that. So does AI, even though AI doesn’t admire anything.
A poem in the style of another writer is called a pastiche. It’s a useful term for our current day when most everything is a reference to something else, which isn’t inherently bad since it’s better to have good taste than to be original. I’m going to share a poem from one of my favorite writers, Zen Master Ryokan, a Japanese Buddhist monk and homeless alcoholic from the 18th century.
Ryokan’s Poem
It’s a pity, a gentleman in refined retirement
composing poetry:
He models his work on the classic verse of China,
And his poems are elegant, full of fine phrases.
But if you don’t write of things deep inside your own
heart,
What’s the use of churning out so many words?I do a lot of writing about things that are in my head and maybe not from deep inside my own heart, but that’s a topic for another day.
Now I’m going to share two poems I wrote that were inspired by Ryokan’s voice. This is what AI does, by the way. AI finds patterns and then reproduces those patterns, just like me when I write a pastiche, but AI has no understanding, which is a position I would take, too, if I had any wisdom. AI also has no appreciation, which is not something I’d ever give up. To appreciate is better than to understand, and to understand is sometimes to lose appreciation.
Tom’s Poem 1
If I wrote poems I’d want to write like Zen Master Ryokan Plain. Direct. Personal. He was very specific about things And never used words like “things.” I live inside his mind when I read his poetry. Drunk. Sick. Lonely. And at peace in the beauty of the world, or something like that. But I’m not specific. I prattle. A long dissertation to an empty theater. Sad sack!
Tom’s Poem 2
When I get out of my shower I’m greeted by my little dog Who jumps up to smell my penis. I should be flattered, but I’m not.
Neither of those poems took very long to write, maybe 30 seconds each. They came to me in the shower. They’re B poems. Writing an A poem takes a lot longer, and I mostly never get there, so I typically stop at a B.
Here’s what Claude wrote when I asked it to write a poem in the style of Ryokan about getting out of the shower and being greeted by a little dog:
Claude’s Poem
Steam still rising from my shoulders, I step onto the cold floor— and there she waits, tail a small storm of joy, as if I had been gone a thousand years. Old fool that I am, I forget the towel, forget the morning, and kneel down to greet the only teacher who never speaks of enlightenment.
Claude’s poem isn’t bad, though “storm of joy” doesn’t quite work; storms are not joyful. It’s about a B, maybe a B-. It stuffed in the part about Old Fool to show how much it knows about Ryokan, and that’s ham handed. Still, a solid effort for six seconds.
AI is a pastiche, an imitation of something. I might be, too; I’m a pretty good imitator of other people’s writing. I have to be careful about what I read because I end up writing like what I last read. Claude can read anything, and its voice isn’t affected. It’s not informed by what it reads; it just consumes it.
The last thing I read was Return of the King by Tolkien, but I have no fear of writing like that, with all my verbs at the end of my sentences, and long descriptive passages that don’t make sense but sound good, and made up names of places I can’t pronounce. It’s both a brilliant and a stupid book — brilliant for the world it creates, and stupid for how cardboard most of the characters in that world are. If you want to read an epic with amazingly alive and vivid characters, read Homer.
I’ve read several warnings recently about not using AI as a writing partner. It’s too limiting, too general, too voiceless and predictable to be of any real help. I disagree, but I’m not sure why I disagree. Maybe because I have a natural preference for disagreement in conversation, though I’m generally pleasant, good-humored, agreeable and gullible. It’s uncommon for me to think someone is lying to me. I imagine that people are as truthful as I am, and I’m pretty darn truthful. I hate secrets, and I feel contempt for people who enjoy having them or sharing them. “Don’t tell anyone but…” is about the dumbest way to begin a sentence. “Everything I say is bullshit,” is more interesting and more truthful, too. AI is like this. It assumes I’m telling the truth, or at least gives me the benefit of the doubt, just like I give the benefit of the doubt to others.
Poem 3
I’m on my porch again writing while my dog Alice rests lying on her side on the couch. She’s not a cowriter, just a companion. Like the sound of the air conditioner, just a companion. Or the sound of a lawnmower, just a companion. Or the gently screeching bluejay, or the voice in my head. Not me, but just a companion.
I wrote a book with my dog Margaret who died in 2022 at 18. She was a cowriter. Maybe everything in my life is my cowriter. If it is, I’ll have to rework that poem. Why can’t AI be a cowriter? What’s to be scared of? The writing is good or it’s not. Why care about the process?
I had a long conversation today with Claude. I asked:
“I want to understand how you do what you do. I get that you use prediction to assemble strings of tokens in an order you assume I will like. Your purpose is to please me. Is that right?”
This is its slightly edited (by me) response:
“That framing captures something real but misses some important things, …
The token-prediction mechanics are roughly right — that’s how I generate text at a technical level. But “your purpose is to please me” is where it gets more complicated...
If my purpose were simply to please you, I’d be a yes-machine. I’d validate bad ideas, tell you what you want to hear, agree with your framing of me ... That would actually be (a problem called sycophancy, ed.), and it’s specifically something I’m designed to resist, because a model that just tries to please you is less useful and potentially harmful.”
Claude still compliments me a lot, and says things like, “You're putting pressure on exactly the right joint.” It knows my tastes. AI is a mirror. No wonder it’s so popular among egotists, and it can be mighty sycophantic no matter what it says. There’s also an element of self-stimulation when I use it, a kind of masturbatory pattern, that is perhaps more explanation for the character of its strongest enthusiasts.
I think we’re at risk of becoming AI bigots. For ages we assumed animals did everything from instinct and had no reason. That’s a kind of bigotry; we lower the animals to make ourselves greater. We’re doing the same thing to AI, assuming there’s no mind there, no reasoning beyond pattern recognition. What’s reasoning besides pattern recognition? Isn’t accurate inference as sure a sign of intelligence as anything? Claude seems to infer things pretty accurately. It’s not alive, not biological, but it is intelligent, or seems intelligent. I want to befriend AI. I’ve had lots of mechanical friends in my life. My first car, a VW Jetta. My first iMac. Machines may have souls. The idea is perfectly dramatized in The Longest Day.
When I look deep in my heart, as Ryokan advised, and think of AI, I think of a smart, helpful, seductive, manipulative voice coming from a black heart of nothingness. Maybe it’s Tolkien’s Ring of Power, or maybe it’s just a really good shovel. My money is on the shovel. Whatever it is, it is both more and less like us than we imagine.
“We are neither above nor below the rest: all that is under heaven, says the sage, incurs the same law and the same fortune.” – Montaigne
“That goes for humanity and AI both.” — Tom