A room, five causes
At 10:14, five people say the white room has turned blue.
One has received a psychedelic under medical supervision. One is dreaming in a sleep laboratory. One is responding to a hypnotic suggestion that the walls are blue. One is emerging from propofol sedation and later recalls a blue chamber. The fifth is an expert meditator absorbed in a stable visual field with eyes closed.
The reports match. The causes need not.
A researcher who groups them under “vivid internally generated experience” has noticed something real, but has not yet explained it. The psychedelic intervention may alter receptor-mediated cortical dynamics and the influence of high-level expectations. The dream may be generated while external sensory connection is intermittent. The hypnotic report may follow a deliberately strengthened semantic expectation. The anaesthetic report is delayed, filtered by memory and produced during behavioural disconnection. The absorption state may be voluntarily entered, unusually stable and accompanied by retained control.
The tempting intuition is that common phenomenology reveals a common mechanism. Its hidden assumption is that the report is a transparent readout of one underlying state. The same conscious content can be produced, preserved or reported through different causal routes. Altered-state science becomes more informative when it stops asking which state is “higher”, “deeper” or “more conscious” and asks which causal property was perturbed.
The state label hides the experiment
“Psychedelic”, “sleep”, “hypnosis” and “anaesthesia” sound like explanatory categories. They are usually bundles. A psychedelic has a compound, dose, receptor profile, expectations, setting and time course. Sleep contains stages, micro-arousals and transient windows of external responsiveness. Anaesthesia varies by drug, concentration and timing. Hypnosis combines instruction, expectation, social context and individual responsiveness. Absorption depends on training, task and the stability of the attained state.
When these bundles are compared as if each were a single condition, one variable quietly absorbs several jobs. It stands for what the person experiences, what the brain or body is doing, whether the environment can still affect cognition, whether behaviour is possible and whether a report can be delivered. These are not interchangeable.
Represent each intervention by a precision perturbation fingerprint rather than one state score. The fingerprint records five separable properties:
- Sensory precision: how strongly current external evidence constrains inference.
- Contextual-prior precision: how strongly learned expectations and task context constrain inference.
- Self-model stability: how resistant body ownership, agency, perspective and temporal continuity are to revision.
- Environmental connectedness: whether external signals can enter cognition and guide an appropriate response.
- Report and control access: whether the system can monitor, remember, communicate or voluntarily regulate the state.
The fingerprint is a vector, not a consciousness score. Its components need not rise or fall together, and no scalar average has an obvious scientific meaning.
A participant under propofol follows no command. You first classify the person as unconscious. Now leave the drug level unchanged but replace spoken commands with a later awakening question. The participant reports a vivid dream. Which property changed? Not necessarily the experience. The measurement boundary moved from online response to retrospective report. Move it again by testing a transient facial-muscle response, and connectedness may appear during a period still scored as sleep.
Unresponsiveness, environmental disconnection, lack of report and absence of experience are four different claims. A study that substitutes one for another has changed its conclusion without changing its participant.
This distinction is empirically consequential. Sleepers can sometimes classify words and respond through facial muscles during N1, N2 and REM sleep, showing that connection to the environment can open in brief windows (Türker et al., 2023). In deep propofol sedation, a repeated-awakening study obtained interpretable reports of experience after many awakenings, yet two commonly used complexity measures did not reliably separate reports from non-reports (Bajwa et al., 2025). Neither result proves that reports are infallible. Both show why a state label cannot carry the whole inference.
Established observation: responsiveness, connectedness and report can dissociate under sleep and anaesthesia. Contested interpretation: whether a particular neural measure tracks consciousness itself, a capacity for report, or one mechanism supporting both. Practitioner synthesis: the five-part fingerprint proposed here is a study-design instrument, not an established biological taxonomy.
flowchart TD A["A room, five causes"] --> B["The state label hides the<br/>experiment"] B --> C["Precision is a weighting rule,<br/>not a substance"] C --> D["Five interventions leave five<br/>fingerprints"] D -. "re-test" .-> A classDef first fill:#0a1424,stroke:#5B5CF6,stroke-width:2px,color:#e6f3ff classDef step fill:#0a1424,stroke:#3f7da0,stroke-width:1px,color:#e6f3ff class A first class B,C,D step
Precision is a weighting rule, not a substance
Imagine that you see a line through fog. Years of experience suggest that lines in this setting are vertical. The blurred sensory cue suggests an eight-degree tilt. What should you perceive?
The answer depends on how reliable the visual cue is relative to the prior expectation. In a simple Gaussian model, reliability is expressed as precision, the inverse of variance. High precision means a narrow distribution and therefore stronger influence on the estimate. Low precision means a broad distribution and weaker influence.
τsensory = 1 / σ2sensory
posterior estimate = (τprior × prior + τsensory × observation) / (τprior + τsensory)
Here the prior is 0 degrees and the observation is 8 degrees. With equal precisions of 4 and 4, the posterior estimate is 4 degrees. Strengthen the prior precision to 16 while keeping sensory precision at 4, and the estimate becomes 1.6 degrees. Relax the prior precision to 1 while keeping sensory precision at 4, and it becomes 6.4 degrees.
The calculation permits a directional prediction. If an intervention selectively reduces the precision of high-level priors, ambiguous perception should move towards current sensory evidence more than it does at baseline. It does not permit the claim that every unusual perception reflects “low priors”. The same posterior can be produced by several combinations of prior mean, sensory noise, likelihood structure and decision rule. Once the model becomes hierarchical, one level can become more precise while another becomes less precise.
The equation is valid only for a linear Gaussian case with fixed means and independent sources. Remove those assumptions and the arithmetic changes. The explanatory principle survives: inference depends on relative reliability, but the reliability must be estimated at a named level, for a named signal, in a named task.
Predictive-processing models use precision to explain how prediction errors are weighted. Some implementations associate precision with neural gain or the responsiveness of particular neuronal populations. Feldman and Friston modelled attention as inference about uncertainty and proposed gain as a possible implementation (2010). That is an influential computational proposal, not a licence to point at any change in oscillation, connectivity or subjective confidence and call it precision.
Two participants both report that a stationary face is breathing. In participant A, the visual signal is noisy and a strong expectation fills the gap. In participant B, a pharmacological intervention weakens a high-level stability constraint, allowing local visual fluctuations to propagate. The content is identical. The intervention fingerprint is opposite.
Now add a third participant who gives the same answer because the response options imply that “breathing” is expected. The perceptual mechanism may not have changed at all. A report-equivalent result is not a mechanism-equivalent result.
The self is not one prior
Talk of “the self-model” can become another catch-all. At minimum, body ownership, agency, first-person perspective, autobiographical continuity and social identity should be separated. A psychedelic may alter bodily boundaries and narrative salience while leaving basic task control available. A dream may contain a stable first-person perspective but weak reality monitoring. Hypnosis can selectively alter agency or ownership while preserving ordinary conversation. Absorption may reduce discursive self-reference while increasing attentional stability.
Interoceptive predictive models offer one way to ask how bodily signals help stabilise a self-model (Seth and Friston, 2016). They do not establish that selfhood is nothing but interoceptive inference, nor that a change in bodily feeling identifies one computational parameter. The operational question is narrower: which self-related expectation became easier or harder to revise, under which intervention?
Five interventions leave five fingerprints
A cross-state comparison should not ask whether all five conditions increase “entropy”, reduce “top-down control” or weaken “the self”. Those descriptions are too coarse to survive intervention. The stronger question is whether each condition makes a distinct, preregistered pattern across the five fingerprint components, and whether that pattern changes predictably through induction and recovery.
Psychedelics: a receptor intervention with a disputed computational bridge
What is directly manipulated, and what is inferred?
Classical psychedelics offer an unusually valuable intervention because receptor pharmacology can constrain the causal story. In a within-participant LSD study, ketanserin blocked both the subjective effects and much of the altered global and thalamic connectivity associated with LSD, implicating the serotonin 2A receptor (Preller et al., 2018). DMT studies have likewise connected intense, immersive experiences with large changes in electrophysiological and functional organisation while wakefulness is preserved (Timmermann et al., 2023).
The computational bridge is less settled. The REBUS model proposes that psychedelics relax the precision weighting of high-level priors, allowing lower-level signals and alternative interpretations greater influence (Carhart-Harris and Friston, 2019). This is a serious synthesis, but “relaxed beliefs” is not directly read from an fMRI connectivity map. A discriminating test must predict which perceptual priors weaken, which lower-level dynamics gain influence, and what an alternative account based on arousal, receptor-specific gain or task strategy would predict.
The strongest psychedelic evidence begins with the receptor perturbation; the precision interpretation remains a model to be earned task by task.
Anaesthesia: early signals can survive while predictive routing fails
What disappears when a participant stops responding?
Propofol does not simply turn sensory cortex off. In macaque auditory cortex, propofol abolished alpha and beta modulation associated with predictable stimuli and disrupted coherence between sensory and frontal regions. Early gamma activity and spiking could remain strong or even increase, while higher-order oddball information became less decodable (Xiong et al., 2024). This pattern supports a loss of hierarchical routing more specifically than a loss of raw input.
Human thalamic work points in a related direction. Deep propofol sedation altered the core-matrix organisation of thalamocortical connectivity, with transmodal systems especially affected (Huang et al., 2024). The authors are careful about causality, and that caution matters. Behavioural unresponsiveness may indicate disconnection from the environment while disconnected experience persists. Reviews have long warned against equating unresponsiveness with unconsciousness (Sanders et al., 2012).
The fingerprint prediction is therefore asymmetric: connectedness and online report access fall sharply; early sensory precision need not vanish; higher-level predictive routing and broad-use access may be disproportionately disrupted. Retrospective reports, when present, are filtered through the recovery transition.
Sleep: connection flickers while internally organised experience continues
Is sleep a state of weak input or a changing gate?
Auditory work during sleep has found preserved early processing alongside reduced hierarchical prediction-error responses in both non-REM and REM sleep (Strauss et al., 2015). Yet the newer communication experiments show that sleepers can transiently make lexical decisions and produce instructed facial responses across several sleep stages (Türker et al., 2023). The gate is dynamic, not simply closed.
Dream research adds a measurement problem. Serial awakenings can link pre-awakening neural patterns to later dream reports, but a “no dream” response can mean no experience, no recall, weak encoding or a poor question. The proposed posterior hot zone for dreaming (Siclari et al., 2017) has therefore attracted a direct methodological critique centred on recall and timing confounds (Ruby, 2020).
A sleep fingerprint must be time-resolved. External connectedness can open and close within a stage. Report access is usually delayed. Self-model stability varies across dreams. Any precision claim should be attached to a specific sensory hierarchy and a specific window, not to “sleep” as a whole.
Hypnosis: strengthen one expectation while leaving the channel open
Can a targeted belief alter perception without changing early input?
Hypnosis is valuable precisely because suggestion can manipulate semantic expectation while the participant remains connected, responsive and report-capable. In one tactile study, suggestions that a finger was larger or smaller changed spatial discrimination and later somatosensory processing, while the earliest primary afferent response did not show the same modulation (Markmann et al., 2023). The result is consistent with targeted top-down penetration rather than a global altered level.
It also exposes a rival generator. Trait phenomenological control, the ability to generate experiences that match expectations, predicts some reports in paradigms such as mirror synaesthesia and the rubber-hand illusion (Lush et al., 2020). This does not make the experience unreal. It means the experimenter’s cues may be part of the cause. A good hypnosis study therefore measures hypothesis awareness, demand characteristics and objective performance separately from subjective endorsement.
Hypnosis is not a noisy version of psychedelia. It is a controllable intervention on selected expectations, embedded in a social instruction loop.
Absorption: attenuated self-reference with retained voluntary control
What if reduced self-boundaries coexist with increasing stability?
Advanced concentration absorption offers a useful contrast to pharmacological disruption. A densely sampled case study of jhāna used fMRI, EEG and structured phenomenology in a practitioner with extensive training. The reported states were repeatable within the participant and accompanied by changes in network modularity, global connectivity, broadband power and signal complexity (Chowdhury et al., 2025). The study is rich but remains a case study. It establishes a tractable protocol, not a population law.
Absorption can include diminished discursive self-reference, altered body boundaries and reduced environmental salience. Unlike many psychedelic episodes, however, the transition can be deliberate, the attentional object can remain stable and exit can be voluntarily initiated. A retreat study combining psilocybin and mindfulness also showed that practice context and drug effects interact rather than add as independent ingredients (Smigielski et al., 2019).
The key comparison is therefore not “ego dissolution versus ego dissolution”. It is self-model attenuation under rising control and stability versus self-model attenuation under increased volatility and pharmacological constraint.
Participant Mira reports, “There was no centre and no boundary.” Under psilocybin, she could not stabilise the visual field, time felt fragmented and the state unfolded despite attempts to regulate it. During trained absorption, the same sentence follows a deliberately sustained object, low distractibility and a clean voluntary exit.
If the report is the dependent variable, the states look convergent. If control, temporal stability and susceptibility to external perturbation are measured, they separate. The scientifically useful question becomes: which parts of the self-model were attenuated, and which control loops remained intact?
Toy scale: a participant rates body boundary from 0 to 10. That score cannot separate loss of ownership from expanded peripersonal space. Computational scale: add a rubber-hand perturbation, visuomotor delay and heartbeat discrimination task. The pattern can distinguish visual capture, agency disruption and interoceptive instability. Configured-system scale: collect those tasks during matched windows of a psychedelic protocol, hypnotic suggestion and absorption, then model each participant’s baseline and recovery. The shared phrase “no boundary” becomes a family of testable changes rather than an endpoint.
The five fingerprints are not directly measured natural kinds. They are model-guided summaries built from tasks, physiology, behaviour and reports. A different computational vocabulary may explain the same observations. The framework earns value only when its variables support predictions that a rival vocabulary does not, or when it reveals that an apparent cross-state similarity disappears under intervention.
stateDiagram-v2 [*] --> Claim state "The self is not one prior" as Claim state "Psychedelics: a receptor<br/>intervention with a disputed<br/>computational bridge" as Test state "Anaesthesia: early signals can<br/>survive while predictive<br/>routing fails" as Revision Claim --> Test: expose mechanism Test --> Decision: evidence survives Test --> Revision: rival survives Revision --> Claim: narrow or rebuild Decision --> [*]
Build a cross-state precision assay
The practical contribution is a study architecture that makes cross-state claims expensive enough to be useful. It does not require every intervention in one trial. It requires every study to emit the same typed record: what was manipulated, when it acted, which fingerprint component was probed, what the participant could still do, how the report was obtained and which rival route could produce the same result.
The minimum protocol
Begin with a within-participant baseline for each task. Use at least one ambiguous sensory task, one hierarchical regularity task, one self-model perturbation and one connectedness probe. Capture reports near the tested window rather than relying on one narrative at the end. Record induction, maintained state and recovery separately. Finally, repeat the critical task with a control intervention designed to preserve the report while changing the proposed mechanism, or preserve the mechanism while changing the report.
Suppose a team wants to compare weakened self-boundaries under psilocybin and trained absorption. The team first measures baseline body ownership, agency under visuomotor delay, interoceptive confidence and attentional stability. During the target window it introduces brief, ethically approved perturbations and asks structured near-time questions. It then measures recovery and repeats the tasks.
The preregistered psychedelic prediction is greater volatility under sensory conflict and reduced stability of high-level self-related expectations. The absorption prediction is lower discursive self-reference with preserved or improved resistance to distractors and voluntary exit. A shared boundary-loss rating strengthens neither mechanism by itself. A crossed interaction between boundary rating and control stability would.
Precision Perturbation Workbench
Use the controls to separate a simple perceptual estimate from the broader intervention fingerprint. The calculation exposes a common error: two conditions can generate nearly the same posterior estimate while differing sharply in self-model stability, connectedness and report access.
The estimate sits midway between prior and observation because the two precisions are equal.
What a positive result permits: if a preregistered intervention changes the predicted task sensitivity and the change tracks the relevant fingerprint component, the proposed mechanism gains support relative to named rivals. What a negative result means: the manipulation may not have changed that component, the task may not measure it, or the model may be wrong. What the workbench cannot establish: it cannot determine whether an experience occurred, whether precision is implemented by one neural quantity, or whether two phenomenologies share an ontology.
from dataclasses import dataclass
from math import sqrt
from typing import Dict, Tuple
@dataclass(frozen=True)
class Fingerprint:
prior_precision: float
sensory_precision: float
self_stability: float
connectedness: float
report_access: float
def validate(self) -> None:
if self.prior_precision <= 0 or self.sensory_precision <= 0:
raise ValueError("Precisions must be positive")
for name in ("self_stability", "connectedness", "report_access"):
value = getattr(self, name)
if not 0 <= value <= 10:
raise ValueError(f"{name} must be in [0, 10]")
def gaussian_posterior(
prior_mean: float,
observation: float,
fp: Fingerprint,
) -> Tuple[float, float]:
"""Return posterior mean and precision for a linear Gaussian toy model."""
fp.validate()
total = fp.prior_precision + fp.sensory_precision
mean = (
fp.prior_precision * prior_mean
+ fp.sensory_precision * observation
) / total
return mean, total
def fingerprint_distance(a: Fingerprint, b: Fingerprint) -> float:
"""Illustrative distance after mapping all five axes to 0..1."""
a.validate(); b.validate()
def vec(x: Fingerprint):
return (
min(x.prior_precision / 20, 1),
min(x.sensory_precision / 20, 1),
x.self_stability / 10,
x.connectedness / 10,
x.report_access / 10,
)
return sqrt(sum((x - y) ** 2 for x, y in zip(vec(a), vec(b))))
states: Dict[str, Fingerprint] = {
"same_posterior_open": Fingerprint(4, 4, 8, 9, 9),
"same_posterior_closed": Fingerprint(4, 4, 2, 1, 1),
}
for name, fp in states.items():
posterior, precision = gaussian_posterior(0, 8, fp)
print(name, "posterior=", posterior, "total_precision=", precision)
print("fingerprint_distance=", fingerprint_distance(*states.values()))The script proves a narrow point: an identical perceptual estimate and identical total precision can coexist with very different state-level access conditions. The distance is an illustrative diagnostic, not a validated consciousness metric.
How to run and extend the Python artefact
Save the block as precision_fingerprint.py and run python precision_fingerprint.py with Python 3.10 or later. Replace the synthetic fingerprints with preregistered directional hypotheses, not post-hoc fitted scores. For real data, keep each axis in its native measurement model and compare predicted interactions rather than relying on the Euclidean distance shown here.
Useful extensions include trial-level Bayesian estimation, state-transition models, participant-specific baselines and explicit missingness for reports that cannot be obtained. Do not impute “no experience” from missing report access.
| Intervention family | Strongest direct handle | Precision claim supported | Report boundary | Serious rival generator | Next discriminating test |
|---|---|---|---|---|---|
| Psychedelic | Strong: receptor blockade and pharmacological timing | Model-guided: relaxed high-level constraints in selected tasks | Usually online, but language and setting shape content | Arousal, receptor-specific gain, demand and task strategy | Crossed prior-versus-sensory reliability manipulation with antagonist control |
| Propofol | Strong: dose, response transition and neural routing | Partial: higher-level feedback may fail while early input remains | Often retrospective after an unresponsive interval | Motor blockade, amnesia, disconnection without loss of experience | Repeated awakening aligned to online sensory and feedback probes |
| Sleep | Strong: polysomnographic stage and transient response windows | Variable: hierarchy-specific feedback changes, not a global precision fall | Delayed and recall-sensitive, except lucid or responsive windows | Forgetting, micro-arousal, stage transition and question wording | Closed-loop stimulation plus immediate staged report |
| Hypnosis | Strong: targeted semantic suggestion | Directional: selected expectation can gain influence | Online and conversational | Hypothesis awareness, compliance and phenomenological control | Suggestion-content reversal with early and late sensory measures |
| Absorption | Emerging: trained voluntary entry and within-person repeatability | Open: narrowed input and stabilised attention need separate models | Often delayed but can include planned markers and voluntary exit | Expert expectancy, selection, movement and idiosyncratic strategy | Multi-practitioner replication with perturbation and novice controls |
A team predicts that hypnotically suggested finger enlargement will increase the influence of a size prior on tactile judgements. Subjective ratings change strongly, but discrimination thresholds and late evoked responses do not. The first temptation is to call the manipulation weak. The fingerprint framework forces three alternatives into view: the suggestion changed report policy rather than perception; the task probed the wrong sensory level; or the participant generated a semantically appropriate experience without altering the targeted inference.
The team now reverses suggestion content, hides the behavioural hypothesis, adds confidence judgements and measures early and late sensory responses. If only reports continue to move, the result weakens a perceptual-precision account while strengthening a report-generation account. A negative result is therefore not an empty cell. It locates the causal change downstream of the proposed mechanism.
A machine classifies psychedelic, propofol, sleep, hypnosis and absorption recordings with 99 per cent accuracy. It relies entirely on drug-related cardiovascular change, sleep spindles, scanner movement and the timing of verbal instructions. Has it discovered altered-state mechanisms?
No. It has discovered intervention labels. Now remove those shortcuts and ask it to predict the directional effect of a new reliability manipulation within each state. Performance collapses. The failure is informative: state recognition is not mechanism identification.
Decision rules for the assay
Same report, different fingerprint: do not infer a shared mechanism. Different report, similar fingerprint: investigate language, memory and task interpretation before declaring mechanistic divergence. Same fingerprint under a held-out perturbation: a cross-state mechanism becomes more plausible, but only at the measured level. No directional interaction: lower confidence in the precision account or redesign the measurement bridge.
Cross-state fingerprints should predict responses to held-out perturbations better than nominal state labels. The hypothesis is strengthened if a model trained on sensory weighting, prior weighting, self stability, connectedness and report access transfers from one intervention family to another and predicts the direction of a new ambiguity or self-model manipulation. It is weakened if the axes are unstable within participants, if state labels predict just as well after nuisance variables are removed, or if a simpler arousal or task-engagement model explains the same effects.
The smallest useful test needs no drug exposure: compare ordinary wake, targeted suggestion and staged sleep using one ambiguous auditory task, one connectedness probe and near-time reports. The proposal should not be used as a clinical consciousness detector or a welfare verdict. Its purpose is to improve experimental discrimination.
This is an experimental design, not an instruction for self-administration or unsupervised altered-state practice. Pharmacological studies require legal, medical and ethical oversight. Anaesthesia belongs in clinical research environments. Intensive contemplative protocols need screening, trained supervision and adverse-event pathways. The smallest useful implementation is a non-pharmacological comparison using ambiguous perception, suggestion, sleep-stage probes and low-risk attentional tasks.
sequenceDiagram participant Q as Question participant M as Mechanism participant E as Evidence participant D as Decision Q->>M: A room, five causes M->>E: The self is not one prior E-->>M: compare rival explanations E->>D: Use altered states as precision instruments D-->>Q: revise scope or act
Use altered states as precision instruments
Do not choose an altered state because it appears more profound, reports more self-loss or produces a more dramatic brain image. Choose it because its intervention creates a discriminating test of one causal property.
For a psychedelic study, start from receptor and task manipulations, then test the proposed precision bridge. For anaesthesia, separate early sensory response, hierarchical routing, connectedness, memory and report. For sleep, sample transient windows rather than treating a stage as homogeneous. For hypnosis, manipulate semantic expectation while measuring demand and early sensory processing. For absorption, use voluntary control and stability as defining contrasts, then replicate beyond exceptional individuals.
The new default should be a typed claim: this intervention, at this phase, altered this fingerprint component, on this task, relative to this rival generator, while these other components were preserved or unavailable.
The decision is to stop treating altered states as destinations on a consciousness map and start using them as precision instruments. That shift does not settle what consciousness is. It produces experiments capable of telling us which mechanism changed, which observation survived and where the evidence still stops.
Glossary
- Precision
- Model-relative confidence, commonly expressed as inverse variance. It determines how strongly a signal influences an estimate.
- Prior
- An expectation about a hidden cause before the current observation is incorporated.
- Prediction error
- A mismatch between predicted and observed signal, weighted by its expected reliability within a model.
- Self-model
- A family of expectations concerning body ownership, agency, perspective, continuity and identity.
- Environmental connectedness
- The capacity for external information to enter cognition and guide an appropriate response.
- Report access
- The capacity to monitor, remember or communicate a state. It is not identical to having an experience.
- Intervention fingerprint
- A directional profile across sensory precision, contextual-prior precision, self-model stability, connectedness and report-control access.
- Rival generator
- An alternative causal route that can produce the same observation without the proposed mechanism.
- Near-time report
- A report sampled close enough to the target state to reduce transition and recall confounds.
- Measurement bridge
- The explicit argument linking a task, physiological signal or report to the latent property being claimed.
References
- Feldman, H. and Friston, K. J. (2010). Attention, uncertainty, and free-energy. Frontiers in Human Neuroscience, 4, 215.
- Carhart-Harris, R. L. and Friston, K. J. (2019). REBUS and the anarchic brain: Toward a unified model of the brain action of psychedelics. Pharmacological Reviews, 71(3), 316-344.
- Preller, K. H. et al. (2018). Changes in global and thalamic brain connectivity in LSD-induced altered states of consciousness are attributable to the 5-HT2A receptor. eLife, 7, e35082.
- Carhart-Harris, R. L. et al. (2016). Neural correlates of the LSD experience revealed by multimodal neuroimaging. Proceedings of the National Academy of Sciences, 113(17), 4853-4858.
- Timmermann, C. et al. (2023). Human brain effects of DMT assessed via EEG-fMRI. Proceedings of the National Academy of Sciences, 120(13), e2218949120.
- Seth, A. K. and Friston, K. J. (2016). Active interoceptive inference and the emotional brain. Philosophical Transactions of the Royal Society B, 371(1708), 20160007.
- Xiong, Y. et al. (2024). Propofol-mediated loss of consciousness disrupts predictive routing and local field phase modulation of neural activity. Proceedings of the National Academy of Sciences, 121, e2315160121.
- Huang, Z., Mashour, G. A. and Hudetz, A. G. (2024). Propofol disrupts the functional core-matrix architecture of the thalamus in humans. Nature Communications, 15, 7496.
- Sanders, R. D., Tononi, G., Laureys, S. and Sleigh, J. W. (2012). Unresponsiveness does not equal unconsciousness. Anesthesiology, 116(4), 946-959.
- Bajwa, I. J., Nilsen, A. S. and Juel, B. E. (2025). A repeated awakening study exploring the capacity of complexity measures to capture dreaming during propofol sedation. Scientific Reports, 15.
- Strauss, M. et al. (2015). Disruption of hierarchical predictive coding during sleep. Proceedings of the National Academy of Sciences, 112(11), E1353-E1362.
- Türker, B. et al. (2023). Behavioral and brain responses to verbal stimuli reveal transient periods of cognitive integration of the external world during sleep. Nature Neuroscience, 26, 1981-1992.
- Siclari, F. et al. (2017). The neural correlates of dreaming. Nature Neuroscience, 20, 872-878.
- Ruby, P. M. (2020). The neural correlates of dreaming have not been identified yet. Frontiers in Neuroscience, 14, 585470.
- Markmann, M. et al. (2023). Hypnotic suggestions cognitively penetrate tactile perception through top-down modulation of semantic contents. Scientific Reports, 13, 6578.
- Lush, P. et al. (2020). Trait phenomenological control predicts experience of mirror synaesthesia and the rubber hand illusion. Nature Communications, 11, 4853.
- Chowdhury, A. et al. (2025). Multimodal neurophenomenology of advanced concentration absorption meditation: An intensively sampled case study of jhāna. NeuroImage, 305, 120973.
- Smigielski, L. et al. (2019). Characterization and prediction of acute and sustained response to psychedelic psilocybin in a mindfulness group retreat. Scientific Reports, 9, 14914.