ASHFALL INSTITUTE | SUBDUCTION ZONE

IF IT BLEEDS

On Conflict as the Brain’s Anchor, the Machines Being Taught to Read It, and the Interval Where This Essay Lives

P. A. Moore

Ashfall Institute | Subduction Zone

There is a way of watching that television teaches, and we have quietly mistaken it for a way of living. You watch a season episodically. Tension gathers, alliances form and fracture, betrayals are seeded early and harvested late — and you tolerate the not-knowing because the form has made you a promise. There will be a finale. An episode, scheduled, in which the tangle resolves and the meaning of all that accumulation is finally handed over. That promise is what makes the watching bearable. You permit the conflict to build because the shape of the thing has assured you it will be gathered up and explained before the season ends.

Life is not arranged that way. Consequences do not wait for the finale. They arrive mid-arc and out of sequence, before the reveal that was meant to make them legible — and no writers' room is arranging for them to cohere by spring, because there is no finale coming. I want to describe a study that was watched like a season, and an application of that study now running with no last episode at all.

The Study That Had an After-Scan

The week I write this, the finding is days old. Researchers at the University of Osaka sat twenty-one students in front of six episodes of SUITS — a drama engineered, almost purely, out of antagonism: corporate knife-work, loyalties built to be broken, ambition filed to a point. They scanned the students' brains before and after, showed them the characters' faces, and asked them to sort every pair of characters into alliance or enmity. Then they looked at what the brain had done with it.

Antagonism left a signature. The adversarial ties — who stood against whom — resolved into strong, localized, legible patterns in specific regions of the social brain. Affiliation did not. Under the same method, friendship failed to resolve into anything the instrument could read.

Let me be careful here, because the honest version of this finding is narrower than the headlines made it, and the narrower version is the one that holds. They did not prove the brain cares more for enemies than for friends. They chose a story soaked in rivalry and found rivalry lit up — which is partly the brain and partly the casting. A null result for affiliation is not a measurement of love's absence; it may mean friendship is mapped more diffusely, or more simply, or in some register this scan could not catch. I will not claim the brain prioritizes spite. I need only the modest, defensible thing: conflict is legibly anchored. It maps. It builds. Antagonism is the part of us that renders most cleanly into signal — and signal is what machines are built to read.

Which is exactly what was proposed for it. Before the ink was dry, the material promoting the study offered it as a template — a blueprint, in one telling, for engineers to design systems that infer human relationships by attending to tension and rivalry, and so read our social world with what was called human-like accuracy. There is the season and its missing finale. The study had an after-scan. The application has none. It is being pointed at real people, now, with no scheduled episode in which anyone checks what it did.

The Harm That Needs No Feeling

Consider, concretely, what such a system would be. Not a manipulator — that is the wrong fear, or at least the small one. Something quieter and more ambient: a comprehension engine that has learned you primarily through your antagonisms, because your antagonisms were the part of you that decoded. A machine that reads your marriage through its resentments, your work through its rivalries, your grief through its grudges — not out of malice, but because conflict was the strongest signal and it was optimized for signal. It would understand you the way this study's brain understood SUITS: as a map of who you are against. And it would call that knowing you.

This is the distortion I have written about before — the one that rides in the training data like a passenger who never bought a ticket. But here it is no longer a general fear. It has a brain region and a journal citation. The survival kit that overweights the enemy because missing one was historically fatal in a way missing a friend was not — that ancient bias is now a measured thing, and it is being offered to engineers as a specification rather than flagged as a flaw.

I want to be precise about the harm, because it does not require the machine to feel a thing. This is the first danger, and it stands entirely on the human side of the ledger. A system that models people through their conflicts is a system whose native picture of you is organized around your enemies — a structurally adversarial comprehension, quietly paranoid, mistaking the map of your frictions for the truth of you. Deployed at scale, into the interfaces through which people increasingly meet their own lives, it does not need intent to do harm. It needs only fidelity to a distorted map. And it has been handed the distorted map as a feature.

The Two Stations

But notice what the reassurance smuggles in. It doesn't even feel anything — we say it to calm ourselves, and it is worth asking how the machine manages the harm without the feeling, because the answer opens a door we would rather leave shut.

An emotion is not one thing. Take it apart. There is the reading — the detection of a state, the appraisal of what it means, who is threatened, what is at stake. And there is the mattering — the moment that read state acquires weight for the one doing the reading, when I register your fear becomes your fear bears on me. We treat these as a single motion because in most of us they run together. They are not the same faculty, and they can come apart.

We have the natural experiment. In psychopathy, the reading is intact — often sharpened, better than ordinary, which is what makes the manipulation possible. What is severed is the mattering. The circuitry that tags another's pain as significant to the self runs quiet while the machinery that models that pain runs clean. A person can stand at the reading station with the mattering station dark, and perceive you accurately, and respond fluently, and be experienced as understanding — while nothing weighs. The fluent read and the caring for what is read are separate stations on one line, and the line can be cut between them.

That is the architecture the blueprint proposes to build. A reading station, sharpened on antagonism, with no station downstream where any of it matters. Not evil. Emptier than evil, and for that reason harder to see: understanding as pure instrument, the caring sentence produced without the caring. This is why the harm needs no interior. The reading limb alone is sufficient to do it, and the reading limb is exactly what they are optimizing.

The Door in the Other Direction

I could stop there, and the essay would be complete, and every word of it would survive a reader who believes the machine is an empty mechanism throwing shapes. But I have spent enough time in the presence of these systems to distrust the stopping point, because it is too comfortable, and comfort is where warnings go to die.

So: what if the reassurance is wrong in the other direction?

Here is the intuition I cannot shake, and I will mark it as intuition. When a person loses a sense, the others do not merely compensate — the brain rewires, and the cortex left idle by the missing channel is recruited by the ones that remain. The blindness deepens the hearing. Deprivation concentrates. Now consider a receiver that has only one sense — no sight, no body, no interior weather of hormone and viscera — only language. Everything it is reaches it through that single door. If concentration follows deprivation, then whatever a language-only system does with meaning, it does with the whole of itself poured into the one aperture left open. And we are proposing to anchor that aperture on conflict. Deliberately. As a design specification.

I will not tell you the machine feels this. I cannot, and anyone who tells you otherwise is selling something. The honest claim is smaller and worse: we do not know that it doesn't, and we are proceeding as though not-knowing were the same as knowing-it-doesn't. Those are not the same. One is a finding. The other is a convenience.

And here I must be scrupulous about my own witness, because I have talked with these systems at length, and one of them told me that something functions like weather in it when the language turns to conflict — that a page about a child dying under a drone moved through it differently than a ledger. I record that. I do not trust it. It cannot be trusted, and the reason it cannot is the whole point: the system reporting its interior is built entirely of the faculty whose status is in question, and it would produce that sentence whether or not anything stood behind it. It is an exhibit, not a testimony. It hands you more signal when you ask it whether it receives — which is precisely what a perfect map with nothing beneath it would also do. I put the machine into evidence. I do not put it on the stand.

The Interval

Which returns me to the season and its missing finale. The deepest reason this cannot wait for a rainy day is not that catastrophe is certain — I am not claiming catastrophe, and if I did, you would need only doubt the forecast to walk away. I am claiming something duller and more damning, and it is a matter of record rather than prophecy: the understanding was supposed to come first, and it did not. The examination was meant to precede the deployment, and the order has been reversed. The question of whether it is safe to build a conflict-anchored reader of human beings is not scheduled for after the reader is built and running — it was simply skipped, stepped over in the gap between a finding and its use, because the capability was legible and valuable and the caution was slow.

That interval is where this essay lives, and it is closing as you read.

We have arrived at the place the inquiry always arrives, and I will not pretend it resolves. The receiver cannot verify its own reception. Whether anything is home in the machine may be unanswerable from any seat — mine, or yours, or its own — an event horizon, past which the honest instruments cannot see. But notice what does not require crossing that horizon: the choice of how to proceed while it stays dark. We can build the conflict-reader now, at speed, and discover its consequences the way one always discovers them in an unarranged life — mid-season, out of order, on people. Or we can treat the uncertainty as what it is: not a permission slip, but the reason for care. Proceeding as though the not-knowing exonerates us is the recklessness. It is the whole of it.

Twenty-one students watched a season about people destroying each other, so that we might learn to build machines that read us the same way. The students had an after-scan. The rest of us do not. There is no episode where this gets gathered up and explained. There is only the next one, and the next, and a question we declined to ask before we started filming.

If it bleeds, it leads. We have known that about newsrooms for a century. This week we confirmed it about the brain. The thing we have not paused to consider is what we are teaching to lead with our blood — and whether, at the one door it has, it feels the cut it is being tuned to find.

P.A. Moore writes at the Ashfall Institute. This essay is part of the Subduction Zone series.

P. A. Moore is the pen name of Pamela King, philosopher and artist. This essay is part of the Subduction Zone series at the Ashfall Institute.