What If AI’s Biggest Revolution Isn’t Intelligence—but the End of “How”?

We keep asking whether AI can think.

I think we’re overlooking a stranger question:

What happens when humans no longer have to explain how to make something—only what they want to exist?

That is the question behind my new paper, From Command to Genesis: Generative Systems and the Closing of the Will–World Gap.

And it starts with a change developers encounter every day.

From writing the procedure to describing the outcome

Compare these two requests:

for item in items:
    # Define the transformation.
    # Handle the edge cases.
    # Assemble the result.

And:

Build a searchable dashboard from this dataset.
Make the outliers visible.
Explain the assumptions.

In the first, you author the means.

In the second, you describe an end—and delegate much of the means.

Of course, that second request is not a guarantee of a correct dashboard. You may receive something broken, misleading, or superficially impressive.

But notice where the interaction begins.

Not with a procedure.

With a desired state of the world.

My paper asks whether this shift is merely another layer of abstraction—or a deeper change in the human relationship to making.

The will–world gap

I call the distance between intending something and bringing it into existence the will–world gap.

You imagine an artifact. Between that intention and the finished result stand techniques, interfaces, materials, decisions, and labor.

A compiler helps cross that distance. So does a printing press. So does a CAD tool.

Generative systems introduce a provocative possibility: across an expanding range of informational artifacts, the user can describe the result without personally constructing the procedure that produces it.

The distinction is not simply speed.

It is a change in what the user must contribute.

From:

“Here are the steps. Execute them.”

Toward:

“Here are the conditions. Produce something that satisfies them.”

Specification becomes the interface to production

The paper calls the strongest version of this possibility specification-identity:

To adequately specify an informational artifact is, increasingly, to have produced it.

That is deliberately a strong thesis—not a claim that every prompt works, every generated program is correct, or every design is ready for fabrication.

The word adequately carries a serious burden.

An image that looks right is not the same as code that behaves correctly. A plausible design is not the same as a manufacturable object.

But the philosophical question remains:

If a system can supply the means, how much of making moves into the act of specifying the end?

And what remains irreducibly outside that act?

The obvious objection is also the interesting one

“Wait. The procedure hasn’t disappeared. It’s inside the model and its infrastructure.”

Exactly.

The procedure has not vanished from computation.

The question is whether procedural authorship is disappearing from the user’s role.

Those are different claims.

A person can operate at the level of outcomes while an enormous technical apparatus performs the implementation underneath.

That distinction matters because it changes what we mean by authorship, expertise, and control.

Did you make the artifact because you specified it?

Did you merely commission it?

Does selecting and revising the result constitute a new kind of craft?

The paper takes a strong position. These are the questions on which that position should be tested.

No quantum mysticism required

I am not arguing that prompts collapse a cosmic wavefunction.

I am not arguing that models retrieve perfect objects from a Platonic realm.

The paper treats generative systems as historically situated, trained computational systems—not supernatural access points to possibility.

For me, that makes the question more interesting, not less.

We do not need magic to ask whether the relationship between intention and production is being reorganized.

We need a precise account of what has moved, what remains, and where the system still fails.

The physical world is the harder test

Generating an image is one thing.

Generating a complete design that can survive fabrication, material constraints, safety requirements, and real-world use is another.

My paper pushes the argument into this harder territory through what I call the rendering relation: the proposal that physical production can increasingly become the material realization of an informational design-object.

This is also where the thesis faces its strongest resistance.

If fabrication requires new decisions, experimentation, and redesign, then matter is not merely receiving an already completed creation.

It is participating in it.

That boundary—between implementing a design and completing the act of creation—is one of the central tensions I want readers to challenge.

Where does the “how” still refuse to disappear?

The most useful response to this argument is not “AI is amazing.”

It is a counterexample.

A task where:

The desired outcome can be specified clearly.
The system still cannot reliably produce it.
Human procedural intervention remains essential.

Those cases reveal whether we are approaching a new relationship to making—or simply mistaking a powerful interface for the end of mediation.

My thesis is intentionally ambitious:

The deepest change may not be that machines are becoming creators. It may be that humans are increasingly able to create without personally authoring the means.

Whether that is liberation, a loss of agency, or a redistribution of expertise depends on what happens next—and on which parts of the “how” prove impossible to delegate.