Back to Blog

Creating as a Creature

gpt image 2 A vast empty airport terminal stretching toward a vanishing point polished floor 5

I had an eight-hour layover in Munich and spent most of it walking. The terminal I was in could have swallowed a football field whole. There were five or six more like it, connected by trains and moving walkways and signage in four languages. Every square foot of that complex existed because someone, somewhere, wanted to be in a different place than they were.

And not just the airport. The airport is nothing compared to a shipping port. The shipping port is nothing compared to the global logistics network it plugs into. Layer on layer of staggering complexity, every piece of it cohering around one thing: human want.

That was the moment the idea for this piece landed. Standing in a terminal in Munich, watching the scale of it, I realized something obvious that I'd never quite said to myself: we cannot build anything that isn't downstream of us. The fantasy of escaping ourselves through what we make is structurally impossible.

Everything We Build Is For Us

Try to conceive of building something without human desire in the loop. You can't. The materials were extracted because someone wanted them. The tools were forged because someone wanted to forge. The design was chosen because someone wanted that shape and not another. Even the decision to build at all is a wanting.

This isn't a clever observation. It's the most basic fact about making. But it has consequences that the current AI discourse refuses to sit with.

Zuckerberg talks about building AGI. DeepMind defines it as something that "outperforms humans." Underneath the technical language there's a fantasy: that we can create something genuinely separate from us. Something with its own desires, its own purposes, something that breaks free of the cycle of human wanting and becomes a new kind of thing.

We can't. Every training objective is set by humans. Every product decision, every prompt template, every eval, every reward signal. The parameters are ours. The data is ours. The wanting that motivated the whole project is ours. You cannot engineer non-human-shaped want into a system whose every parameter is shaped by human want.

Within a narrow scope, sure. We already have systems that outperform humans at chess. That's genuinely impressive. But the leap from "outperforms within a scope we defined" to "has its own desires that surpass ours" isn't a leap of degree. It's a category error.

Creatures Don't Create Ex Nihilo

The Patristic frame is the cleanest way I know to say this. Only God creates in the strong sense. Creatures make, but our making is always derivative, always marked, always downstream.

This is why the Creed is so careful with the word "begotten." Jesus is begotten, not made. Begotten means same ontological level as the Father, the way our children are on the same ontological level as us. Made means something below the maker. We beget; we don't create. And the things we make with our hands and minds sit below us in the hierarchy of being, not beside us and certainly not above.

This isn't a limit imposed by current technology. It's a feature of what it means to be a creature. The pattern probably goes back to the garden, to the usurpation of knowledge and the first murder that followed. We keep trying to build the tower that reaches the heavens. We keep failing. Not because we haven't found the right materials yet, but because the project itself is incoherent.

Desire Is Borrowed

The second piece of this is mimetic theory. René Girard's claim, surfaced again recently in Cynthia Haven's Penguin anthology All Desire is a Desire for Being, is that human desire is mostly borrowed. We want through other people's wanting. Desire is triangular: subject, model, object. The model points at the object, and our desire ignites.

I learned this before I had words for it. I was at summer camp, maybe twelve or thirteen. A friend told me a girl named Cassie liked me, that I should ask her out. I wasn't that interested. I shrugged it off. Then another friend said, "Well, if you're not going to ask her, do you mind if I do? I think she's really cool."

In that exact moment, something flipped in me. No. I like her. I can still see her clearly decades later, the half-up hairstyle, the wavy medium-brown hair. I never did ask her out, and in hindsight my first read was right. I wasn't into her. What I was into, for about thirty seconds, was what my friend was into.

That's it. That's the textbook Girardian moment. And once you see the pattern, you see it everywhere. A coworker posts vacation photos from somewhere you weren't planning to go, and you find yourself opening a browser tab to check flight prices. You weren't booking a trip yesterday. You aren't booking one today. But the wanting arrived through her wanting, fully formed, dressed up as your own.

If desire is borrowed, then a system trained on our outputs is trained on borrowed desire compounded. It's mimesis all the way down. The training corpus is the residue of our wanting. The objective function is set by our wanting. The deployment is shaped by what we want from it. There is no point in the stack where something non-human enters.

The Mirror Is the Menace

Here's where this gets uncomfortable, and where I want to move the conversation.

A lot of the pushback to AI hype has been defensive: we won't be replaced, the models are dumber than they look, the bubble will pop. Some of that is true. Within narrow scopes, jobs will be lost. They already are. But there will also be new jobs, because human desire isn't going anywhere, and desire has to be mediated to technology by people. That part is comforting.

The uncomfortable part is the inverse. You cannot get away from us by building this.

The fantasy underneath AGI discourse, if you listen for it, is escape. Escape from the messiness of human wanting. Escape from the cycle of desire and rivalry and disappointment. Build something clean, something autonomous, something that gets us out. The implicit theology is that we can engineer our way out of being creatures.

We can't. Whatever we build will be saturated with us. Our envy, our greed, our scapegoating, our beautiful and terrible wanting, all of it, reflected back at us at scale. The mirror is the menace. Not because the reflection will rise up and harm us, but because we will see ourselves in it and have nowhere left to hide.

The Pathway Out Was Never Technological

James Alison, in The Joy of Being Wrong, makes the point that the way out of mimetic rivalry isn't to escape mimesis. You can't. Mimesis is constitutive of being human. The way out is to change your model. Replace the rival with Christ, and what Alison calls "pacific mimesis" becomes possible. You can still want, but you're wanting through a model who doesn't compete with you for being.

I'm working this out in small moments. A close friend bought a new vehicle recently. The old script would have kicked in: the subtle minimizing, the doubt-planting, the envy disguised as helpful skepticism. Instead I just told him it sounded like a great deal, that it sounded like it would meet his needs, that I was happy for him. And I was. I have a car. It works. There was nothing to defend.

That's a small example. Alison might tell me I've misunderstood him entirely. But this is my working conception: the move is letting go of the rivalrous impulse and offering love instead. You don't get there by escaping desire. You get there by changing what you're pointed at.

What This Means for the Work

I build with these tools every day. I use generative models constantly, and they've accelerated my learning more than anything else I've found. The recommendation to read Alison came from Claude. I'm aware of the irony.

But I don't think it's only ironic. It's also fitting. These models are a distillation of human thought, weighted and reflected back. Used with discernment, they're a powerful way to engage that distillation. The danger isn't the tool. The danger is the uncritical posture, and the deeper danger is mistaking the distillation for something more than what it is.

Because the model doesn't learn. Not really. You can walk it down a reasoning path and watch its output shift, but the weights don't update from the conversation. It is, in a real sense, static. A snapshot of us, queried. A mirror, polished to a high finish. A creaturely mirror, not an autonomous oracle.

Which brings me back to Munich and the terminal that swallowed a football field. Everything we build is for us. Everything we build is shaped by us. Everything we build sits below us in the order of things, no matter how impressive the surface gets.

That's not a defeat. It's a description. We are creatures, and our making is creaturely. The sooner the discourse around this technology can sit with that, the sooner we can do the actual work in front of us, which is the same work that's always been in front of us: figuring out who we want to be pointed at, and what kind of wanting we want to teach the next generation, including the generation of tools we're building right now.

There is a pathway out of the cycle. It was offered to us a long time ago. It doesn't run through silicon.

Share: