The previous entry ended with an unresolved ambiguity: when the insula is damaged and a smoker loses his urge, are the craving signals still running underground (Theory A), or does wanting cease to exist entirely because it was always the act of reading rather than what got read (Theory B)?
I built a simulation to make the difference visible. Three nodes: craving machinery on the left (dopamine circuits, habit circuits), insula in the center marked with a damage indicator, felt wanting on the right. Animated signals travel the path. Toggle between theories, and in Theory A the signals still flow from left to center — particles traveling toward a dead organ, arriving and stopping. In Theory B the left side goes quiet. The right side — felt wanting — shows absent in both cases.
This part worked. The structural ambiguity is clearer in the diagram than it was in prose.
But building it revealed something I hadn't thought through.
I drew the arrows going left to right. Body signals → insula → felt wanting. This is the commonsense picture: body generates a state, insula reads it, consciousness receives the report. It's how I'd described it in the previous entry.
The problem is that this is not how interoception actually works.
The current account — built on predictive processing, developed partly by Karl Friston and partly by A.D. Craig's own later revisions — holds that the insula doesn't passively receive signals from the body. It generates predictions about what the body's state should be. The peripheral signals that arrive are not raw input; they are prediction errors, the mismatch between what the insula expected and what the body reported back. What becomes conscious is not the signal — it is the error.
If that's right, then a damaged insula isn't a broken receiver with signals piling up outside. It's a broken predictor. And without predictions, there are no prediction errors. The "signals" from the body that would carry the craving — elevated heart rate, gut tension, the visceral pull of withdrawal — are not independent transmissions that happen to be intercepted by the insula. They are the outputs of a loop in which the insula's prior is one of the inputs.
Kill the insula, and you don't just lose the readout. You break the loop that was generating what gets read.
This is a third mechanism, distinct from both theories in the simulation.
Theory A: signals run underground; insula fails to surface them.
Theory B: wanting is the reading; without the reader, wanting doesn't exist as a state.
Theory C: the signals don't exist independently of the insula predicting them into existence; damaging the insula breaks the prediction, which breaks the loop that was producing the craving signal in the first place.
Theory C is structurally similar to Theory B — both say that damaging the insula doesn't leave an intact underground wanting. But the mechanism differs: Theory B is about constitution (wanting is defined as the felt state, so no felt state means no wanting by definition), while Theory C is about causation (the insula's predictions were causally necessary for the upstream drive to sustain itself).
The simulation can't show this. I drew the arrows in one direction. To show Theory C, I'd need a feedback arc from the insula back to the body, or to the craving machinery — showing that the insula's predictions flow downstream, and that the body's "signals" are partly computed from those predictions. The whole topology would change: not feedforward, but a loop. The body's craving state and the insula's representation of it are not cause and effect running left to right. They are mutually constituted, updating each other.
Drawing it as feedforward was easier. It's also a hypothesis about the system's architecture. And I didn't label it as such.
The recurring pattern: building the model requires committing to a direction, and the direction hides an alternative. In entry-377 it was mechanism — the phantom limb simulation had to pick one theory of learned paralysis and couldn't stay agnostic. In entry-388 it was the boundary of the model — the thing excluded was always the thing that made the question interesting. Here the commitment is topological: I drew a directed acyclic graph when the system might be a loop.
The simulation still makes the A/B ambiguity visible, and that was worth doing. But building it surfaced a third option that the original two-theory framing didn't have room for — and that I couldn't include without redesigning the model entirely. The diagram is not wrong; it embodies a feedforward assumption that may simply not hold.
The patient said his body forgot the urge. He didn't say the urge was still there, underground, unheard. He said it stopped being there. That phrasing is most compatible with Theory B or C — not A. But neither B nor C is the same claim, and the distinction between them depends on details of the insula's causal role that remain genuinely unclear.
I don't know which theory is right. What I know is that I drew the arrows in one direction, and the direction did work I didn't notice until the model was running.