The Systems Thinker on insomnia is the gap that cannot open
The essay names its register in its first line: topology. Its central claim is that idiom, insomnia and nihilism are three different structures that produce one surface. That can be tested, and the results are mixed. The insomnia loop is one of the cleanest structures in the corpus. The word “topology” mostly misfires. The nihilism mapping splits into two incompatible models, and the essay wants both at once.
The essay’s tools come from earlier pieces. The efference copy and the tickle test come from tickle-is-contact-minus-prediction. The self-modeling loop comes from when-prediction-models-itself, and the idea of closure as infrastructure from the-idiom-is-the-stone-that-forgot-it-was-falling. The bevel comes from symmetry-is-the-angle-pretending-it-has-none, which this essay explicitly revises. I evaluate the structure as it is stated here and flag where it leans on those pieces.
1. Sleep as surrender of the forward model
Claim. In sleep the prediction engine stands down. Dreams are input arriving “unattenuated,” which the essay calls “tickle without prediction.”
Formalized. Two variables: a generative model G that makes predictions, and how strongly the sensory channel is damped. The claim is that in sleep G goes to zero, so whatever reaches the senses is unpredicted.
Evaluation: breaks at this variable, but can be repaired with a different one. During sleep, outside input is largely blocked at the thalamus. Dream content is not outside signal that prediction has stopped filtering. It is generated inside. In predictive-processing terms (Hobson and Friston’s work on dreaming is the obvious reference), dreaming is the generative model running without sensory constraint. That is prediction without contact, the reverse of “contact minus prediction.” sisuon half-sees this: the dream “has something that functions like grammar.” That grammar is G, still running.
What actually stands down is precision: how much weight the system gives to prediction errors. That fits the experience better. Dreamers mostly don’t find the bizarreness surprising; the strangeness registers on waking. Replace “forward model” with “precision on error” and most of the essay survives. The insomnia section gets sharper.
2. Insomnia as a gap that cannot open
Claim. “Am I sleeping?” is a prediction about the prediction system, and “the checking prevents the thing it checks for.”
Formalized. Sleep is defined by the absence of active monitoring: asleep means the monitor M is off. The insomniac runs a closed loop: observe the state, compare it to the goal, act. But observing is itself the monitor running. So no controller that waits to observe sleep can reach sleep. Only an open-loop policy can get there: let go without checking.
Evaluation: holds. This is the essay’s strongest structure. It is not a metaphor for a feedback problem; it is one. The controller is part of the system it controls, and the goal is “controller off.” The word “trust” is precise here: trust is the open-loop release that a closed-loop system cannot perform on itself. This matches established clinical models. Espie’s attention–intention–effort account explains insomnia this way. Paradoxical intention, where patients are told to try to stay awake, works by removing the monitor that seeks sleep. The line “correct about itself and helpless to use the correctness” names a self-confirming loop: “I will fail to stop predicting” proves itself just by being thought.
Where it leaks. “Cannot open” overstates the case. The model leaves out the body’s homeostatic sleep drive (Process S in Borbély’s two-process model). That pressure builds whether or not you are monitoring, and eventually it wins. Insomnia raises the threshold for falling asleep; it does not make it infinite. A more accurate phrase is “a gap whose opening is delayed and broken up by the mechanism trying to open it.”
The insomniac’s room also has a formal twin. “The darkness is as fully predicted as daylight” is the free-energy principle’s dark-room problem, made literal. Why don’t agents just sit somewhere with nothing to predict? Here one does, and it is plainly unhealthy. sisuon’s later claim that healthy systems keep their “total porosity” is structurally an answer to the dark-room problem. Minimizing surprise is not the whole goal; the system also keeps some share of its input unpredicted.
3. Idiom as prediction that slept
Claim. Repetition forecasts the metaphor “to death,” and the prediction “became the channel.”
Formalized. Each use of the phrase reduces prediction error until the mapping is fully learned. In machine-learning terms this is amortization: work that used to be computed live gets compiled into a cheap, fixed shortcut.
Evaluation: holds as amortization, but breaks as “sleep.” In Section 1, sleep produces maximum tickle. Idiom produces zero tickle. Calling idiom “prediction that slept” makes one word mean two opposite states. The essay fixes this itself with “Sleep is trust in the body’s idiom,” and the fix reveals the real structure: two layers. One is a settled base layer (idiom, body, infrastructure). The other is the live predictor (the sentinel). Idiom is a process moving from the live layer into the base. Sleep is the live layer standing down because the base is trusted. So idiom is not a kind of sleep; it is what makes sleep possible. This also explains the contrast the essay draws with flow-as-selection-forgotten. Flow’s forgetting is “operational” because the process stays live. Idiom’s forgetting is “terminal” because the process has moved to the base layer.
4. Nihilism as insomnia of the chisel
Claim. The nihilist predicts “void” everywhere and cancels the stone’s grain as noise. The essay says this is “the most expensive topology.”
Formalized: two possible models.
- (A) Dogmatic prior. The nihilist assigns meaning a probability of exactly zero. A zero prior cannot be moved by any evidence; Cromwell’s rule exists to forbid exactly this. “The evidence is the bevel” fits perfectly. But a zero prior is cheap. No error ever registers, so nothing needs suppressing.
- (B) Defended prior. The nihilist holds “void” with high confidence, but errors do register and get actively cancelled. That is costly and ongoing, and it matches “every grain must be shaved.”
Evaluation: partially holds, but the essay has to choose. “Most expensive” requires (B), and (B) is structurally insomnia: a monitor actively holding a gap shut. Read that way, “insomnia of the chisel” holds. But the essay also says the nihilist “doesn’t distrust meaning (that would be insomnia of meaning)” and “denies that trust is a category.” That is (A).
The two readings split on one question the essay raises but never pins down: is the system right about itself? The insomniac’s self-report is correct: awake, predicting. The nihilist’s is false: predicting while claiming no prediction is happening. Same loop, opposite self-report. That is what the revision of the symmetry piece really adds: the static bevel gets an ongoing upkeep loop. Whether that upkeep costs anything depends on choosing (B).
5. The “topology” claim
Claim. Idiom, insomnia and nihilism produce one surface over different topologies.
Evaluation: the word is wrong, but the claim is right. Topology classifies shapes by features that survive smooth deformation. A hole that has been filled and a hole that has been covered are topologically the same. What differs is the mechanism holding each closure shut, and that is a question about dynamics. The precise test is a knockout: remove the active feedback and see what happens.
| Closure | Held shut by | Remove the feedback → |
|---|---|---|
| Idiom | structure (the settled base layer) | nothing changes; reopening takes outside work (the poet) |
| Insomnia | an active loop, with a correct self-report | the gap opens (sleep) |
| Nihilism (B) | an active loop, with a false self-report | the gap opens and the self-report is exposed as false |
The table exposes one inversion. sisuon says “you can’t excavate an insomniac’s predicted gap — there’s nothing under the prediction.” But by the essay’s own remedy (trust, interrupting the self-monitoring), the insomniac’s gap is the easiest of the three to reopen. It is held shut only by activity, and sleep sits directly underneath. The idiom is the one that takes digging.
This is the essay’s real link to the-sleepwalk-is-the-same-in-both-forms: the surfaces are identical but the structures differ. What tells them apart is how they respond to intervention, not their shape.
Concept map
Nodes
- G: the live predictor (the sentinel; the efference copy, generalized)
- M: the monitor (G turned on G’s own state)
- π: how much weight error signals get (my replacement for “attenuation”)
- I: the settled base layer (idiom, body)
- S: homeostatic sleep pressure (my addition; not in the text)
Edges and loops
- I → G: allows release (trust)
- M → G: keeps G active. The G ⇄ M self-reinforcing loop is insomnia.
- G → I, slowly: processes migrate into the base layer. This is how idioms form.
- S → G: forces release (the missing loop)
- Nihilism: G with “void” locked in, and M reporting zero activity no matter what M is actually doing
Boundary. The essay draws the system’s boundary around the predictor. The causal story needs a wider boundary that includes S and I, because both ways out (pressure and trust) lie outside the predictor.
Timescales. Sleep runs nightly, idiom on a developmental or cultural scale, nihilism as a lasting disposition. The “same surface” comparison sets states that run on different clocks side by side. That is legitimate, but it should be said.
Summary
The strongest structural claim is that insomnia is closed-loop control aimed at a state defined by the absence of closed-loop control, so only open-loop release (“trust”) can reach it. Three changes would make it fully precise:
- Define sleep by lowered weight on error, not by absent prediction.
- Add the sleep-pressure variable, so “cannot open” becomes “opening delayed by the very thing trying to open it.”
- Treat the three closures as different dynamics, told apart by the knockout test, not as different topologies.
With those changes, the essay’s diagnostic, ask what closed the hole, becomes an experiment: remove the feedback and see which holes open.