There Is a New Voter in the Room

This is a Loose Thread. I have carried it since 2024 and I still cannot tell whether it is profound or just shiny. So read it as a provocation, not a proof. I am going to take one intuition out as far as it goes, name the people who got there before me, and then mark, honestly, the place it stops being true.

The intuition is this. A thought is a cloud of maybes until you finish it. The moment you finish it, especially out loud, something irreversible happens. You do not just report a fact. You make one.

Saying is the cut

In another notebook I wrote about observation as a cut: the way looking at a photon does not reveal its true form but collapses it into the one form your instrument can hold. Observation is the scissors, not the truth. That essay was about perception, about a single mind or a single model meeting the world.

This is the same gesture pointed at a larger thing. Not perception. Society.

When you hold a half-formed opinion, it is still a wave. You could still land anywhere. Then you say it. You commit the sentence to the air, and the air does not give it back. Now you are the person who said that. Your listener carries a copy. If enough people say it enough times, it stops being anyone’s opinion and becomes simply what is true around here. The cloud collapsed. A probability became a fact, not because reality changed, but because we measured it into place by speaking.

I want to be honest up front about the physics, because it is the most seductive and the weakest part. The quantum language, collapse, superposition, Schrodinger’s unobserved cat, is a metaphor here and nothing more. Speaking is not literally a measurement in the quantum-mechanical sense, and a crowd agreeing is not a wavefunction. I keep the word collapse because it is the best word I have found for the one-way door, the moment the maybe becomes a was. But if you take the metaphor away, the claim underneath has to stand on its own. It turns out it does, because it is not really my claim.

Psychology got here first

Whenever an idea feels this clean, the responsible move is to go find out who already had it. In this case, almost everyone.

The philosopher J. L. Austin said it in 1962: some sentences do not describe the world, they act on it. “I promise,” “I name this ship,” “you are fired.” He called them performatives, and the whole field of speech-act theory grew out of the simple observation that saying is a kind of doing.

The social psychologist Tory Higgins ran the experiment in 1978. People described an ambiguous person to a listener who already liked or disliked them. They tuned the message to the audience, which you would expect. The surprise was what happened next: their own memory of the person shifted to match the version they had spoken. He called it saying-is-believing. You do not just send the message. You become its first convert.

Then there is the part with teeth for our moment. Hasher and colleagues showed in 1977, and Fazio and colleagues sharpened in 2015, that simply repeating a statement makes people rate it as more true. They named it the illusory-truth effect. The cruel finding is the second one: prior knowledge does not protect you. People who knew the correct answer still drifted toward the repeated falsehood. Repetition does not argue with what you know. It works underneath it.

And the whole thing already has a unifying name. Higgins, Echterhoff and Levine call it shared reality: we are driven to feel that our inner states are held in common with others, and we will reach for that shared feeling even at the cost of accuracy. People can prefer a less true belief if it gives a stronger sense of standing on the same ground as the people they care about. That is my entire essay, written by psychologists, fifteen years before I scribbled it in a notebook. The private commitment, the self-persuasion, the crowd consensus: not three things. One gesture at three scales. They got there first, and I am happy to hand them the credit, because it lets me spend the rest of this on the only part they did not have.

The fourth voter

Here is what is actually new in 2026. For all of human history, a finished sentence had three audiences: you, the people in the room, and the slow accumulation we call culture. Now there is a fourth one in the room for nearly everything you write, and it never leaves.

The machine reads too. Your words join the sea of text that the next model will be shaped by, and more immediately, the model is increasingly the thing standing between a person and reality. People do not search now. They ask. Whatever the model says back is the shared ground a billion people stand on.

So put the two collapses next to each other, because they run in opposite directions, and that is the whole point.

We collapse toward what is repeated. That is the illusory-truth effect. The more a thing is said, the truer it feels.

A model collapses toward its own average. Train a generative model on its own output, round after round, and it narrows. The rare voices fall out first, then the merely uncommon ones, until the thing is generating a smooth paste of its own most likely tokens. This is not speculation. Shumailov and colleagues published it in Nature in 2024 and called it model collapse: feed a model a diet of model-made text and within a few generations the diversity drains out and the tails disappear.

Now braid them. Humans increasingly take their shared reality from the model. The model increasingly trains on a web that is increasingly written by models. We are pulling our sense of what is real from a system that is quietly contracting toward its own mean, while we feed that system more of our agreement with it. Two collapses, each tightening the other. The human one mistakes repetition for truth. The machine one mistakes its own average for the world. Hand one to the other and you get a room that keeps getting more certain and less wide.

I do this for a living, and so will everyone

I should confess the uncomfortable part. I already engineer this on purpose, and I have a clean business name for it.

When I work on getting an idea to show up in answer engines, the rule that wins is not “write fifty brilliant posts.” It is “say the same precise sentence in the same shape across eight or more places.” Convergence beats volume. One standardized claim, repeated across enough indexed surfaces, is what gets a model to repeat it back as fact.

Look at that sentence again. That is the illusory-truth effect, operationalized for machines. I am not optimizing search. I am running mass-measurement on a model’s sense of reality, and the industry calls it answer-engine optimization. Everyone who does this for a brand is doing it. Most of them have not noticed what it is.

That is the part I cannot put down. We spent decades worried that propaganda collapses human consensus by repetition. We built a machine that collapses by repetition even faster, gave it to everyone as the front door to reality, and turned “repeat the sentence until the machine believes it” into a marketing discipline.

Where it breaks

I promised to mark the edge, so here it is, plainly.

The quantum frame is decoration. Take it away and nothing in the argument falls down, which tells you it was never holding the weight.

The unification is not mine. Shared-reality theory named the three human scales before I did. Treat me as a messenger here, not a discoverer.

And the cleanest-sounding line, that your words train the machine, is the one to be most careful with. One comment does not move a frontier model. The famous case where an AI told people to glue cheese onto pizza was retrieval surfacing a single old joke, not training absorbing it. The danger is not one-comment causation. It is the loop: a slow tide of model-shaped text becoming the next model’s water, while people drink the output as truth. Name the loop. Do not overclaim the single drop.

What survives all of that is small and, I think, real. Speaking has always been how we vote a probability into a fact. We now do it in a room with a fourth voter that counts the votes the opposite way, and we have started letting it tell us what we decided.

I do not know if that is profound or just shiny. I know I keep pulling at it.

---

Related reading: Observation Is the Cut is the other half of this. That one collapses perception, a single mind meeting the world. This one collapses society. And if you want the part where this stops being a provocation and becomes my day job, I Engineer Illusory Truth for Machines is the confession.

References

1. Austin, J. L. (1962). How to Do Things with Words. Oxford: Clarendon Press.
2. Higgins, E. T., & Rholes, W. S. (1978). “Saying is believing”: Effects of message modification on memory and liking for the person described. Journal of Experimental Social Psychology 14(4): 363–378. https://doi.org/10.1016/0022-1031(78)90032-X
3. Hasher, L., Goldstein, D., & Toppino, T. (1977). Frequency and the conference of referential validity. Journal of Verbal Learning and Verbal Behavior 16(1): 107–112. https://doi.org/10.1016/S0022-5371(77)80012-1
4. Fazio, L. K., Brashier, N. M., Payne, B. K., & Marsh, E. J. (2015). Knowledge does not protect against illusory truth. Journal of Experimental Psychology: General 144(5): 993–1002. https://doi.org/10.1037/xge0000098
5. Echterhoff, G., Higgins, E. T., & Levine, J. M. (2009). Shared reality: Experiencing commonality with others’ inner states about the world. Perspectives on Psychological Science 4(5): 496–521. https://doi.org/10.1111/j.1745-6924.2009.01161.x
6. Higgins, E. T., Rossignac-Milon, M., & Echterhoff, G. (2021). Shared reality: From sharing-is-believing to merging minds. Current Directions in Psychological Science 30(2). https://doi.org/10.1177/0963721421992027
7. Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R., & Gal, Y. (2024). AI models collapse when trained on recursively generated data. Nature 631: 755–759. https://doi.org/10.1038/s41586-024-07566-y

The “saying is the cut” frame is a metaphor, flagged as one in the text. The load-bearing claims are the social-psychological ones: speech acts [1], the saying-is-believing effect [2], the illusory-truth effect [3, 4], and shared-reality theory [5, 6], which already unifies the three human scales. The fourth-voter asymmetry rests on model collapse [7]. The Google “glue on pizza” case was retrieval surfacing one old comment, not training absorbing it.