Awake, Asleep, Comatose: What Measurement Actually Shows, and What It Doesn't

Every previous article in this series has treated human consciousness as the one fixed point everything else gets checked against. This piece is where I actually check it. If consciousness is always existing, non-emergent, waiting to be reflected rather than generated, that claim should have some visible signature, or lack of one, across the full range of states a human brain actually passes through: waking, dreaming, deep sleep, coma, the vegetative state, and the states in between that turn out to be stranger than the ordinary vocabulary suggests. I want to go through what's actually been measured, state by state, method by method, before asking what any of it settles.

The objective method: perturb the brain and measure the echo

I covered the Perturbational Complexity Index in the AI article, but I need its actual numbers here, because they're doing real work in this piece. Casali and colleagues established, across a large benchmark of healthy subjects, that PCI reliably separated conscious from unconscious states using a specific empirical cutoff: unconsciousness never produced a PCI above roughly 0.31, and alert wakefulness never produced one below roughly 0.44, leaving a clear gap between the two¹. A later refinement, PCIst, found the same separation with even starker numbers, an average of about 48 during wakefulness against about 14 during NREM sleep or anesthesia, a threefold difference².

Applied across actual states, the pattern is clean at first and then gets genuinely interesting. Wakefulness: consistently high. NREM slow-wave sleep and general anesthesia: consistently low, below threshold, matching the near-universal report of "I experienced nothing" on waking. REM sleep is the case that should unsettle any purely behavioral definition of consciousness: PCI stays high, comparable to waking, even though the body is behaviorally disconnected, paralyzed by muscle atonia, unresponsive to the outside world entirely³. The brain, measured this way, looks conscious. The behavior says nothing is happening. The subjective report, on waking, says otherwise, vivid dreaming, and the objective measure agrees with the report, not the behavior. Ketamine anesthesia does something similar: some patients under it remain behaviorally unresponsive while producing high PCI and later reporting vivid, dream-like experience, a dissociative state rather than a truly unconscious one⁴.

The clinical states: coma, vegetative state, minimally conscious state, and locked-in

These four categories exist specifically because behavior alone turned out to be an unreliable guide, and the numbers behind that unreliability are larger than I expected going into this.

Coma involves no wakefulness at all, eyes closed, no sleep-wake cycling, and PCI here is reliably low. The vegetative state, now more precisely called unresponsive wakefulness syndrome, is stranger: patients cycle between eyes-open and eyes-closed states, appear awake by every behavioral criterion, yet show no reproducible sign of awareness on standard bedside assessment. Most patients in this category do show low PCI, consistent with the diagnosis. But a real, non-trivial subgroup does not, they show PCI values above the same 0.31 threshold that separates conscious from unconscious everywhere else it's been tested⁵. The minimally conscious state sits between the two, patients show inconsistent, fluctuating behavioral evidence of awareness, and PCI values cluster in an intermediate zone accordingly. Locked-in syndrome is the cleanest case of all: full awareness, confirmed by whatever residual communication channel the patient retains, typically eye movement, paired with virtually no other behavioral output, and correspondingly high PCI. Locked-in patients are proof, on their own, that behavior and consciousness can come apart completely in either direction, full awareness with almost no behavior, or apparent wakeful behavior with, in some cases, none.

The clinical stakes of getting this wrong are not abstract. Multiple studies estimate that roughly 40 percent of patients diagnosed as vegetative are misdiagnosed once assessed by specialist teams using more careful behavioral protocols, and some covertly aware patients escape even that level of scrutiny for years⁶.

Cognitive motor dissociation: the discovery that changed the field

In 2006, Adrian Owen and colleagues put a woman diagnosed as vegetative into an fMRI scanner and asked her to imagine playing tennis. Her brain activated the same motor-planning regions a healthy volunteer's would, reliably, on command, a clear sign of intentional, willed cognition in a patient every bedside behavioral test had classified as having none⁷. This was initially treated by parts of the field as a fluke. It wasn't. A 2019 study using EEG rather than fMRI, cheaper, more portable, usable at the bedside, found that 15 percent of clinically unresponsive brain-injured patients showed clear EEG evidence of brain activation in response to spoken commands, despite showing no observable behavioral response at all⁸. A 2016 meta-analysis put the figure for chronically unresponsive patients at 14 percent⁹. A larger 2024 study, across six international centers and 353 patients, found the rate climbing to 25 percent among patients with no observable response to commands¹⁰. This condition now has a formal name, cognitive motor dissociation, colloquially "covert consciousness," and its detection changes real clinical decisions, prognosis estimates, and, in the most serious cases, decisions about withdrawing life support¹¹. This is, as far as I can tell, the single strongest existing proof that a real, activation-based signature can catch consciousness that behavior alone misses entirely, exactly the kind of result the whole PCI and Butlin-indicator research program in earlier articles was reaching for.

The subjective method, and where it disagrees with the objective one

Here's the part of this research that surprised me most, and it complicates the clean PCI story above. Researchers have directly tested what people experience during NREM sleep using the serial awakening paradigm, waking subjects repeatedly through the night and asking, immediately, whether they had been experiencing anything. The classic finding, refined across studies from the 1990s through 2024, sorts responses into three categories: full dream reports, "white dreams," a clear sense of having been experiencing something with no recallable content, and true reports of nothing at all¹². Recent reviews of studies from 2000 to 2024 found that NREM awakenings produce dream or white-dream reports roughly 60 percent of the time, versus about 83 percent for REM, with true "nothing" reports being the minority outcome even in supposedly dreamless sleep¹³. Philosopher Evan Thompson, whose work on witnessing sleep I cited in the field-question article, has argued directly from this evidence that dreamless sleep cannot accurately be described as a uniform state of unconsciousness at all¹⁴.

This creates a real tension I don't want to paper over. PCI says NREM sleep drops below the threshold for consciousness in the clear majority of cases. Serial awakening says a majority of NREM awakenings still yield some report of experience, full dream or white dream. These aren't measuring exactly the same moment, PCI is typically assessed at a specific instant, while dream reports reflect whatever happened across the preceding sleep period, so the two methods aren't strictly contradictory. But they are, at minimum, telling two different stories about how empty deep sleep actually is, and I think it's more honest to hold that disagreement openly than to quietly pick whichever number fits the point I want to make.

What all of this actually settles about generation versus reflection

Here's my honest answer, and I don't think it's the one either side of this debate wants to hear: none of it settles the question, and I don't think it's a matter of needing better instruments. It's structural.

Every result above, PCI dropping in NREM sleep, rising in REM, dropping further in coma, sometimes surprisingly present in patients diagnosed as vegetative, is fully compatible with two entirely different explanations, and I can't find a way to make the data prefer one. The emergent reading says exactly what it looks like it says: consciousness is being generated by the brain in proportion to how much integrated, differentiated activity the brain is currently producing, and it genuinely dips toward zero when that activity collapses. The reflection reading says something different but equally consistent with every number above: consciousness, as a field or ground, doesn't fluctuate at all, what fluctuates is the instrument's capacity to catch and express it, the way a radio's reception quality has nothing to do with whether the station is broadcasting. On this reading, a low PCI during deep anesthesia means the receiver is temporarily broken, not that the signal stopped.

This is exactly the same problem I ran into with the field question two articles ago, now showing up with real clinical data instead of philosophical abstraction. PCI works by having the brain perturb itself and measuring the complexity of its own response. That's a measurement entirely internal to the instrument. It cannot, even in principle, distinguish "there is nothing here to reflect" from "there is something here, but the instrument's own machinery for catching it is currently offline," because both produce the identical signature, a flat, undifferentiated echo. The eye-cannot-see-itself problem from the Drig-Drishya Viveka argument applies with full force here: every method in this article, PCI, metabolic imaging, EEG command-following, is a measurement made by and of the instrument. None of them, by construction, can get behind the instrument to check whether something being reflected is there or not when the instrument goes quiet.

Is this a chicken-and-egg problem?

I don't think "chicken and egg" is quite the right description, that phrase usually points at a temporal circularity, which came first. What's actually happening here is closer to what philosophers of science call underdetermination of theory by data: two genuinely different explanations account for exactly the same observations, equally well, and no experiment currently proposed, including the entire battery covered in this piece, breaks the tie between them. It's not that we haven't measured carefully enough yet. It's that every measurement available is, by its very design, a measurement of the instrument's own activity, which is precisely the one thing both theories already agree fluctuates. The disagreement was never about whether the instrument's activity fluctuates. It's about what that fluctuation means, and that's a layer of interpretation no instrument reading could arbitrate.

I want to end without pretending this makes the whole research program pointless, because it doesn't. Cognitive motor dissociation detection is already changing real medical decisions for real patients, independent of how the deeper metaphysical question eventually resolves. That's the same lesson from the animals and plants article: the practical, functional question and the deep metaphysical one can be worked on largely separately, and real progress on the first doesn't have to wait on the second. What I don't think I can honestly claim, after actually doing this research rather than assuming it would resolve things, is that any of it moves the generation-versus-reflection question forward at all. It sharpens exactly where the wall is. It doesn't get behind it.