Bullshit

"One of the most salient features of our culture is that there is so much bullshit. Everyone knows this." — Harry Frankfurt, On Bullshit (1986)

Frankfurt's small book makes one large distinction, and it has stayed with me for twenty years.

The liar, he says, is in a relationship with the truth. An intimate one, even. To lie well you must first know what is true, then carefully construct its inverse, then keep track of both. Lying is expensive. It requires a map of reality accurate enough to navigate away from.

The bullshitter has no such burden. He is not turned away from the truth — he is simply facing a different direction entirely. He is optimizing for effect, for plausibility, for the impression he makes in the room. Whether his words happen to correspond to anything is not a consideration he has to process at all. It's not that he gets the answer wrong. It's that there is no answer being sought.

This is why Frankfurt calls bullshit "a greater enemy of the truth than lies are." The liar reinforces the authority of truth even as he violates it. The bullshitter dissolves it.

I have been thinking about this a lot lately, for the obvious reason. But also for a less obvious one: I am not sure bullshit is the enemy.

Manure

The word is worth taking literally for a moment.

Bullshit — actual, steaming, agricultural bullshit — is the single most valuable substance in a garden. It is what turns dead dirt into living soil. Every civilization that figured out how to feed itself figured out how to move dung around. Before synthetic nitrogen, the wealth of a farm was measured in the size of its manure heap.

And yet: too much of it kills. Fresh, unmixed, undiluted manure will burn the roots of whatever you plant in it. The nitrogen is too hot, too fast. The plant that should have been nourished is instead scorched. The problem was never the substance. The problem was the dose, the velocity, and the absence of soil to hold it.

I think this is precisely right about the other kind of bullshit too.

The gift of not caring

Here's the uncomfortable part. Our capacity for bullshit — for producing language untethered from any state of affairs — is not a defect in the machine. It is the machine.

Every animal communicates about what is. Bees dance about actual flowers. Vervet monkeys have distinct calls for actual eagles and actual snakes. The signal is welded to the world. What humans did, somewhere in the fog of the last hundred thousand years, was break the weld. We learned to say things about flowers that aren't there, eagles that never were, and worlds that could be.

Once you can say what isn't, you can say what might be. Hypotheses. Fiction. Contracts. Money. Gods. Mathematics. Plans. Every single thing we have built started as a sentence that was not, at the time it was uttered, true.

The scientist proposing a hypothesis and the con man proposing a scheme are performing the identical cognitive act. The difference comes entirely afterwards — in whether the utterance is ever dragged back and made to face reality. Bullshit is the raw material. Grounding is the process. The value isn't in the manure or in the soil; it's in the ratio between them.

Social life runs on this too. "How are you?" "Fine." "We should catch up sometime." None of these are truth-claims. They are gestures, lubricant, warmth. A person who insisted on strict propositional accuracy in all social exchange would be unbearable, and possibly diagnosable. A culture needs a certain amount of shared, comfortable, agreed-upon nonsense to hold itself together. We call it politeness, or ritual, or in the good cases, poetry.

So the question was never how do we eliminate bullshit. The question is: what is our ratio, and what happens to the excess?

The industrialization

Which brings us, finally, to the machines.

We have been calling it "hallucination," and this is a category error that costs us clarity. Hallucination implies a perceptual apparatus that has malfunctioned — a system that was trying to see reality and saw wrong. That's not what's happening.

A language model has no perceptual relationship with the world at all. It is optimizing for the plausibility of the next token given everything before it. It is, in the purest and most technical sense available, Frankfurt's bullshitter: a system that produces utterances without any reference to their truth, because truth is simply not a variable in its loss function. When it gets things right — which is most of the time, and impressively so — it is right the way a broken clock is right, except the clock is very well-calibrated and broken only occasionally.

This is not an insult. It is a specification. And once you see it as a specification rather than a bug, the engineering problem becomes clear: we have invented the Haber-Bosch process for bullshit. We can now synthesize unlimited quantities of it at near-zero marginal cost, in any style, on any subject, at any hour.

Synthetic nitrogen fed four billion people. It also created dead zones in the Gulf of Mexico the size of small countries, because we produced fertilizer faster than the soil could absorb it and the runoff went somewhere.

The runoff always goes somewhere.

Our grounding mechanisms — peer review, journalism, courts, reputation, the slow accumulation of trust between people who know each other — were all built for a world where producing plausible language was expensive. They are absorption systems calibrated to a manure supply that no longer exists. They are, right now, being overwhelmed. Not by lies. By something much harder to fight, because there is no liar to catch, no intention to expose, no hidden truth being concealed. Just volume.

The only test that works

I have stopped trying to detect bullshit by examining the content. The content is too good now. Instead I ask three questions, and they are all about the structure of the claim, not its substance:

What does this cost the speaker? A claim that costs nothing to make is not a claim, it's a noise. Skin in the game is the oldest grounding mechanism we have and it still works.

Could this be wrong? Not "is it wrong" — could it be. If there is no state of the world that would falsify it, if the thing is compatible with every possible outcome, then it isn't connected to the world at all. It's decoration.

What changes if it's false? If nothing changes — if the plan proceeds, the meeting ends, the deck gets approved either way — then the statement was never load-bearing and you can safely ignore it. Most corporate language, and increasingly most AI output, fails here.

None of these questions ask whether something is true. They ask whether it is the kind of thing that could be true — whether it has been dragged back and made to face reality, or whether it is floating free.

The garden doesn't need less manure. It needs more soil, and a longer composting season, and someone paying attention to what's actually growing.

I don't know how to build those things at the speed the runoff is arriving. But I am fairly sure that's the job now.