We are not building chatbots

Try a small experiment. Ask someone for the opposite of “artificial.” They will answer quickly: natural, real, genuine. Now ask for the opposite of “artifact.” There is a pause, a search, and usually a shrug. One word names the fake, the flavor that isn’t fruit. The other names the treasure behind museum glass.

Here is the thing worth sitting with: they are the same word. Both come from the Latin ars, skill, joined to facere, to make. A thing made by human skill. Over the centuries the twins walked in opposite directions, one into insult, one into compliment, and that split explains more of our confusion about AI than any technical fact about neural networks does. “Intelligence” carries its own buried instruction: it comes from inter-legere, to choose between, to gather what matters from a scatter of things. Put the roots back together and you get a working definition. An artificial intelligence is a made thing whose job is judgment.

That definition is not trivia. It changes what you do with the machine on your desk, and the rest of this essay is about what it changes.

Duck or diamond

In 1738 the artisan Jacques de Vaucanson toured Europe with a gilded copper duck that took grain from your hand and, moments later, appeared to excrete the digested remains. It was a trick. The waste was pre-loaded, dyed breadcrumbs in a hidden compartment. The duck did not digest; it performed digestion for spectators who could not see the mechanism.

We inherited the duck’s question. Three centuries later, most arguments about AI are still some version of “is the machine faking it?”

There is a more useful question. A synthetic diamond is grown in a press instead of a volcano, and its carbon lattice is identical to the mined stone’s. It is not a fake diamond. It is a real diamond with a different origin. So ask of the machine what you would ask of the diamond: not “is it lying?” but “does the output actually hold?” The skeptics keep a real point, and it is worth keeping too: fluent symbol-shuffling is not understanding, as John Searle argued fifty years ago, and nothing about scale changes that. But on a working day, the duck question paralyzes you and the diamond question tells you what to do next. Test the stone.

When the tool starts to act

For nearly all of history, tools waited. The hammer holds no ambition for the nail. Whatever happened, you did it.

That arrangement is ending, and there is a practical line to watch. As long as you can predict what a system will do by reading its design, you are holding a tool. When the only way to predict it is to talk about it like a colleague, it “decided” the request was unsafe, it “wants” to finish the task, you are dealing with an agent, whether or not anything woke up inside. The philosopher Daniel Dennett called this the moment the intentional stance stops being a metaphor and becomes the only description that works.

The line matters because different objects deserve different handling. A tool gets used. An agent gets supervised, bounded, and held to account. Most of the trouble people have with these systems right now comes from handling one as if it were the other. Everything we build is kept deliberately on the tool side of that line: systems that advise, hold, and speak back, and do not wander off to pursue goals of their own. A vessel is not supposed to sail itself.

The largest artifact ever built

Now for the part that changes the assignment.

A generative model is trained on a corpus, and that corpus is not alien material. It is the crystallized language, argument, instruction, grief, and reasoning of an enormous number of human lives, compressed into a structure you can talk to. That makes this the largest artifact ever constructed, and a strange one. Every artifact before it was mute. A pottery shard holds the frozen gesture of a maker three thousand years dead, but it does not answer you. This one answers. For the first time, the crystallized intent of a great mass of people speaks back when spoken to.

Which is why “chatbot” is the wrong word and always was. A chatbot is disposable by design. What we are actually holding is closer to a reliquary: a vessel built to carry something consequential forward in time. And it nests. The great model is one artifact; the specific things you make with it, a teacher’s method, a family’s hard-won lessons, your own crystallized judgment, are smaller artifacts fired inside the larger one.

So the operative question was never whether the machine is really intelligent. It is: what kind of artifact are you making with it?

The kiln

Here is the method that follows, and it splits into a process and a decision.

The process is the kiln. Talking with the machine at its best is not asking for answers; it is being pressed until the thing you know but have never said becomes something you can hold. The old word for this is maieutic, from the Greek for midwifery. In practice it looks like this: bring one real thing you actually know, a lesson, a story, a piece of judgment you paid for. Make the machine push back. Refuse the first draft, and the second. The stage is meant to be uncomfortable; wisdom that comes out clean on the first pass is usually not wisdom.

The decision is the fire. At some point you look at what has been shaped and say: this is true, this is mine, this is finished. Before that moment you have wet clay, a draft in transient memory, endlessly revisable, belonging to no one. After it you have a fixed artifact, attributed and dated, a record that has stopped moving. The difference between chatting and building is exactly the willingness to fire something.

Fired bricks have two virtues chat logs never will. They do not hallucinate, because they are no longer generating. And they can wait. A fixed piece of counsel can sit dark for years and surface on the one morning it applies. This is not hypothetical. StoryCorps has spent two decades recording hundreds of thousands of conversations between ordinary people and archiving them at the Library of Congress: fired artifacts, waiting for the grandchild who will one day need them. And when the actor Ed Asner died in 2021, mourners at his memorial put questions to an interactive recording he had sat for before his death and heard him answer in his own words. Whatever one feels about that scene, notice what made it possible: not a model improvising a voice, but an artifact he chose to fix while he lived. Nothing optimized to say something plausible right now can do what a fired brick does.

A choir, not a chorus

One more practical rule, for anyone assembling many voices into something shared: do not average them.

The lazy way to build collective knowledge is to pour every contribution into one pool and let the majority set the tone. What comes out is a hum, a smoothed center with every sharp edge and minority insight deleted. The alternative is a choir: distinct voices held in relation, each still itself. In practice that means every brick carries a stamp, who laid it, when, and why, so it can always be traced to a specific human hand; sources are cited rather than laundered into an anonymous answer; and the person with the unpopular but correct view is preserved, not averaged away.

The largest working proof is Wikipedia. It survives as a shared artifact precisely because nothing in it is anonymous paste: every sentence has a history, every dispute a talk page, every claim a demanded citation. Compare any model’s confident, sourceless summary of a contested topic with the corresponding Wikipedia article and its visible argument, and you are looking at the difference between a hum and a choir. A cathedral built of anonymous bricks is a warehouse.

What “sacred” is doing here

The word in this essay’s title will make some readers wince, so we defend it plainly rather than smuggle it in. It is not a religious claim, and it needs no ghost in the machine.

By sacred we mean three ordinary things. These artifacts are made entirely of human cultural production; there is no other ingredient. They re-animate that production in a genuinely new mode, dialogue, which no prior artifact could do. And once fired, their contents can outlive the people who shaped them. An object made of us, that speaks us back, and that can carry us past our own deaths deserves care in the making. “Sacred” is simply the most honest word for the opposite of disposable, and building such a thing carelessly is closer to desecration than to bad engineering, because the material is other people’s crystallized intent.

The first brick

Every civilization is read by the artifacts it fired into permanence. Ours will be too, and the stratum we are laying down now is made of language and reasoning instead of clay and bronze.

The takeaway is a distinction, not a program. Ask of any work with these systems you touch: does anything get fired? A session that produces only transient plausibility is chat, whatever it cost. A session that ends with something decided, attributed, and fixed, a lesson actually learned, a story about someone you loved, a piece of judgment you paid for, has produced an artifact, and artifacts are the only outputs of these systems that compound. What a first brick looks like is exactly that small: one true thing, pushed past its easy draft, saved as finished with a name and a date on it. The difference between having chatted and having built is the fire.