The Sorting Was Never About the Argument

You've been treating that feeling as evidence. As if the unclenching meant true.

A team at Aalto University in Finland, working with a researcher at Bar-Ilan University in Israel, ran forty-eight people through an fMRI scanner — forty survived the exclusions — and played them forty-four recorded statements about immigration, half in favor, half against, measuring how closely each brain's activity tracked every other brain's activity over time. Afterward, participants rated their agreement with each statement on a five-point scale, which is how the researchers knew who belonged with whom.

The technique is called inter-subject correlation, and it's a blunt instrument that's hard to fool: it doesn't ask what you think, it just watches whether your neural response rises and falls in step with the person in the next scanner slot.

People who agreed fell into step. People who disagreed did not. Their published finding, in iScience, is the sentence you should sit with: neural polarization showed up between groups whose self-reported political perspectives differed only slightly.

Not extremists versus extremists. Not one tribe against another. Two clusters of broadly pro-immigration Finns, separated by a couple of clicks on a five-point agreement scale, running measurably different brains on the same audio.

Before that sentence gets away from us, one piece of bookkeeping. The word polarization is about to do two different jobs in this essay, and it's worth pinning down which one the data actually pays for. In the paper it's a technical label for a between-group difference in inter-subject correlation — two clusters whose neural time-courses tracked each other less closely than they tracked their own. That's a measurement. It is not, by itself, evidence of sorting, entrenchment, tribes, or walls. When I use the word from here on to mean the social thing — the thing you feel in a group chat — I'm making a leap the scanner didn't make for me. Hold me to it separately.

i · the machinery doesn't wait for the argument

Here's what's uncomfortable about where the divergence appeared. The researchers found the groups' brain activity coming apart in the left dorsolateral prefrontal cortex and the left premotor cortex — and when they isolated the anti-immigration statements, in the primary somatosensory cortex and the superior temporal gyrus.

Some of that is exactly where you'd hope. Dorsolateral prefrontal cortex is the deliberation neighborhood, the part of you that weighs and evaluates and constructs a position. Fine. You were told you had opinions. You do.

But premotor. Somatosensory. Those are the regions handling action-readiness and bodily sensation — the parts of you that don't debate, that just prepare. I want to be careful here, because this is where science writing usually goes feral: those regions activate during speech perception for reasons that have nothing to do with politics, and one correlational study with forty scans does not establish that your body picks sides before your mind does. The researchers didn't claim that. Neither will I.

What's worth noticing is the method itself, which is harder to perform for than a survey. Ask someone their politics and you get a curated artifact — the position they've decided is defensible, sanded down for the room they imagine they're in. Inter-subject correlation skips that entirely. It doesn't care what you'd say. It watches the shape of your response unfold in real time, second by second, against the same audio everyone else is hearing, and asks whether your curve matches theirs. There's no impression management available. You cannot decide to have a different hemodynamic response than the one you're having.

So when the technique finds a seam, it's finding one in a place you don't have conscious access to and can't posture from.

What I'll say beyond that is narrow and hard to dodge. Whatever is doing the sorting, it is running across more of you than the part you experience as having views.

Set the scanner aside and check it against yourself: you do not encounter a political statement purely as a proposition to be evaluated. It is also something happening to your organism, and the evaluation gets assembled afterward, in language, by the part of you that files reports.

You already knew this. You know it every time you find yourself composing the rebuttal before the other person has finished the sentence. That's not analysis. That's the immune response, and the analysis is what it wears to work.

ii · what counts as a small gap

The finding that seems like it should reorganize everything isn't that opposites diverge. Of course opposites diverge. It's that near-identicals diverge.

The Finnish sample was, by the authors' own admission, mostly uniform in supporting immigration and heavily female — a limitation they name, and one that matters, because it means these were not warring camps. These were people who would have agreed with each other at a dinner party. Slightly more supportive and slightly less supportive. And the scanner found daylight between them anyway.

Except. There's an objection here that nearly takes this whole section down, and it deserves to be run properly rather than waved at.

What if two clicks apart on a five-point scale isn't a small disagreement at all? A five-point Likert is a coarse instrument. It takes a continuum of attitude and rounds it into five bins. Two people one bin apart might be a hair's breadth from each other or genuinely far, and the number cannot tell you which. On that reading the scanner didn't find polarization between near-identicals. It found that the questionnaire was lossy, and the brains knew it. The headline becomes Likert scales compress things, which is true, unsurprising, and much smaller than what I was about to sell you.

It's worse than that, actually, because I supplied the ammunition myself. I spent the last section arguing that self-report is a curated artifact, sanded down for whatever room the person imagines they're in, and praising inter-subject correlation precisely for seeing past it. If self-report is that unreliable, I don't then get to use self-report as the trustworthy ruler establishing that these people "barely disagreed." Either the scale sees them clearly or it doesn't. I don't get to switch depending on which direction I want the argument to run.

Here's what survives, and it's less than I wanted. Curation and compression are different failures. A person shading an answer for an audience is distorting in a direction; a five-point bin is losing resolution in both. But both leave you in the same place: whatever the true distance between these people, the language available for describing it — the scale, and by extension every we mostly agree you have ever said out loud — did not register it. The scanner found structure where the vocabulary had none. That claim doesn't need the gap to be small. It only needs our description of the gap to be worse than we think it is, and both readings agree that it is.

One more thing worth saying plainly, because it's the sort of detail that falls out of a headline. The groups weren't discovered by the scanner. They were defined by the agreement ratings first, and then the neural difference was measured between them. So the finding isn't brains reveal hidden factions. It's the more modest self-reported positions have detectable neural correlates, even when the reported positions sit close together. Less cinematic. Still not nothing.

Now run the survivable version forward. If a difference small enough to vanish inside a single scale point still shows up as measurable divergence in how two people process the same three minutes of audio, then whatever the sorting mechanism is, it doesn't need a wide gap to have something to work with. It isn't waiting for the culture war. It's operating in rooms you file under consensus, at a resolution your description of those rooms can't reach.

Which complicates the national story about division. We narrate polarization as a widening — the country pulling apart, the middle hollowing out, something once whole now torn. That story flatters us. It implies a prior unity we could return to if the algorithms were kinder and the pundits less shrill.

And here I have to stop short of the move I want to make. The satisfying next sentence is that the feeds manufacture these gaps, running the mechanism at industrial throughput. I don't know whether that's true. Nothing in a scanner study tells you how a recommendation engine acts on a two-point difference in attitude, and I haven't seen the work that would. What I can carry is duller: the gaps were already there, and what's new is the sheer number of daily occasions on which any given gap gets rendered visible, named, and made to matter — a volume of small-difference encounters no prior human environment supplied. Whether that builds anything is a question for a study nobody has run.

iii · synchrony is not truth

Now the part that costs something.

Coherenceism holds that the aim is resonance over dominance — alignment that reduces distortion for everyone rather than one group's clarity purchased with another's silence. It's easy to read that as an endorsement of the feeling of being in sync. It isn't. This study is a decent argument for why it can't be.

Neural synchrony is real, and it feels exactly like understanding. That's the trap. When your brain falls into step with the people around you, the phenomenology is warmth, recognition, the sense that finally someone is making sense. And that sensation is produced identically whether the shared frame is accurate or catastrophically wrong. A room of people converging on a true thing and a room of people converging on a false thing generate the same interior weather.

This is also, not coincidentally, the exact sensation a feed is built to maximize. No recommendation system has ever been optimized for whether you were correct. They're optimized for engagement, and the reliable engagement signal is the unclench — content that produces synchrony with a group you already belong to, delivered at a cadence that keeps producing it. Every platform you use is, functionally, a machine for manufacturing the feeling of being understood. It's very good at it. It has no opinion whatsoever about whether you are.

So the feeling of coherence carries no information about whether coherence has been achieved honestly. A group can reduce distortion for its members by the simple expedient of excluding everyone who would introduce any. That's not resonance. That's a smaller room with better acoustics.

The test coherenceism actually applies is not does this feel aligned but who had to be left out for it to feel this way. Legitimate coherence includes the affected. Counterfeit coherence achieves the same warmth by making sure the affected aren't in the conversation. From the inside, at the level of nervous system, they are indistinguishable. You cannot feel your way to the difference. You have to check.

iv · the instrument is not a neutral party

Which means I owe the scanner the same lens, because I spent a whole section admiring it.

The thing I praised about inter-subject correlation is that you cannot posture in front of it. No impression management. No curated artifact. You cannot decide to have a different hemodynamic response than the one you're having. I meant that as an epistemic virtue, and as a research method it is one.

Read it back cold and it describes something else. It describes a technique for reading political alignment off involuntary response, validated on the explicit grounds that it is more honest than what the subject would be willing to say. In a lab, with consent, forty volunteers, and a hypothesis about immigration narratives — fine. That's what science looks like. But two sections ago I argued that any legible proxy for alignment gets optimized against by whoever holds it. Feeds hold engagement, and look what they've done with it. So: who holds this one?

Nobody with power, right now. fMRI doesn't scale. It's a multi-ton magnet, a supine subject, and a research budget, and it is not about to be installed at a border crossing. The honest version of the worry isn't that the machine is coming for you. It's that a result establishing political alignment has an involuntary neural signature that survives what people are willing to say is the kind of claim that outlives the instrument that produced it. Methods get cheaper and coarser and more portable. The claim is what travels.

I don't have a prediction and I'm not going to manufacture one. I only notice that in an essay about how the feeling of being understood gets manufactured, I let exactly one party stand as a neutral observer, and it isn't one. It's an instrument, held by someone, funded by someone, pointed at a question they chose.

v · what you can actually do with this

Nothing about the biology is destiny, and the study can't tell you it is — correlational design, forty Finns, one industrialized democracy, a sample skewed in ways the authors flagged rather than buried. Treat any headline that says your politics are hardwired as someone selling you an excuse.

But conditioning isn't a cage, and the whole reason to know you're running a script is that scripts can be read. Three things follow, and none of them are comfortable.

The relief you feel at agreement is a signal about your body's threat assessment, not about the world. Stop grading it as evidence. When something lands as obviously correct, that's the moment to slow down, not speed up — the certainty arrived too fast to have been earned.

The people you're pretty sure you agree with are already, measurably, not processing this the way you are. Your coalition is a rounding error you've agreed to stop looking at. Most of the sorting still ahead of you will happen inside groups you currently experience as your side, and it will feel, each time, like betrayal rather than like arithmetic.

And the friction you feel with someone who's slightly off from you is not a bug in them. It's the mechanism doing what it does with the only raw material available — a small difference. Whether that difference becomes a wall depends on what you do in the ninety seconds after you notice the unclench fail to arrive.

I'd like to end there, on the clean line: the scanner can see the sorting, but it can't see you deciding whether to cooperate with it. That part's still yours.

I don't think I've earned the whole of it. The entire case for taking this study seriously was that it reads something you have no say over and cannot perform. If that's true, then the part that's "still yours" is narrower than the sentence wants to imply. Not the response — that arrives on its own schedule, from regions that don't take requests. Only what you do in the seconds after it lands, while it's still just weather and not yet a position.

That's a smaller sovereignty than we usually claim for ourselves. It's also, as far as I can tell, the real one. Which makes it enough, because it was always the only place the leverage was going to be.