“Psychedelics can profoundly alter consciousness by reorganizing brain connectivity, producing acute experiences that shape lasting psychological change, Psychedelic dynamics are commonly described as desynchronized or entropically disordered, yet the brain organization underlying self-dissolving and boundary-dissolving experiences that participants often report, and how context shapes that organization, remain unresolved. To address this, we acquired the largest single-site psychedelic neuroimaging dataset to date.”
The central move is a reinterpretation, not a new phenomenon. Prior work, most prominently Siegel et al.’s 2024 “psilocybin desynchronizes the human brain”, characterized the psychedelic state as entropic, disordered, desynchronized. This paper accepts those time-averaged findings (they replicate the GFC redistribution, reduced modularity, increased signal diversity) but argues that when you look at the brain’s activity as a temporally resolved trajectory rather than a time-averaged summary, what looks like disorder is actually latent structure that locks onto whatever the person is doing (rest vs. meditation vs. music vs. movie). They call that “context alignment,” its felt correlate “embeddedness,” and they show the degree of alignment correlates with subjective intensity (MEQ30/ASC) and next-day mindset change.
It’s a genuinely interesting conceptual reframing, and the “time-averaging obscures dynamic structure” point is legitimate and well-taken.
Real strengths, First
Largest single-site acute dataset in the field, truly multimodal (fMRI + 64-channel EEG), naturalistic contexts rather than artificial tasks, a registered trial with an open data descriptor, and, importantly, honest reporting of nulls (the 8-week meditation arm did nothing; modularity did not predict the behavioral outcomes that their embedding measure did). The convergence of the embedding structure across four dimensionality-reduction methods (CEBRA, PCA, t-SNE, Isomap) is a meaningful control against “the fancy method invented the structure.” And using EEG convergence to rebut a purely vascular/BOLD-artifact explanation is a smart, appropriate defense.
Weaknesses
- Open-label, no placebo, and the two headline results are both self-report-adjacent. “No blinding was performed at any stage.” Everyone knew they’d received psilocybin. The two central claims, that neural context-alignment scales with subjective experience, and that it predicts next-day mindset change, are both correlations between a brain measure and a self-report measure in people who know they’re dosed and are in a heavily engineered “supportive” setting. Expectancy can inflate both sides of that correlation simultaneously. This is the field’s oldest problem and it isn’t controlled here at all. The authors even concede the mindset link is “probably mediated by subjective experience,” which quietly undercuts the claim that the brain marker adds independent information.
- Fixed condition order is confounded with pharmacokinetics. Conditions ran rest → meditation → music → movie in a fixed sequence (a different fixed sequence in EEG). Drug level, fatigue, and adaptation all drift monotonically across a session. A classifier that separates “conditions” can exploit temporal drift, so some of the “context” separability may be time-since-dose separability wearing a context costume. Their rebuttal, that rest (the first block) already clustered tightly, so it can’t be pure sequence, is partial. It rules out a pure accrual artifact but not the more general point that a time-varying pharmacological state co-varies with condition.
- The interpretive leap is large and arguably backwards. Their finding is that conditions become more neurally separable under psilocybin. They interpret more separability as more continuity-with-environment (“embeddedness”). But naively you’d expect boundary-dissolution to make states blur together, not become more distinct. The story they tell (the brain “locks to context”) is internally coherent, but it isn’t forced by the data, the same separability could be read as the brain following external stimuli more slavishly under reduced sensory gating. Which leads to:
- Stimulus-following vs. reorganization. Movie and music are strong external drivers. Under psilocybin with relaxed sensory filtering, external stimuli plausibly entrain cortex more strongly, automatically making stimulus-conditions look more distinct from the eyes-closed ones. Part of “context alignment” may be ordinary stimulus entrainment amplified, rather than a deep latent reorganization. The framing absorbs this (“embeddedness = coupling with the environment”), but that rhetorical move blurs the line between a profound claim and a fairly mundane one.
- Small-sample brain–behavior correlations. The machine-learning analyses run on n = 54 (complete four-condition data), and the individual-differences correlations sit around n = 53–54. The headline r = 0.64 (accuracy vs. MEQ30) is strong, but individual-differences brain-behavior correlations at this sample size are exactly the regime Marek et al. (2022) flagged as unstable and inflation-prone. Several of the subscale correlations (Fig. 4h) are explicitly uncorrected. I’d treat the specific effect sizes as provisional.
- “Embeddedness” is largely a relabeling. There is no new instrument measuring embeddedness, it’s inferred from existing scales (MEQ unity/blissful, ASC oceanic boundlessness). The construct is also defined somewhat circularly: embeddedness is the phenomenological correlate of context alignment, and context alignment is the neural signature that correlates with self-dissolving effects. Whether “embeddedness” carves nature at a new joint versus renaming ego dissolution / oceanic boundlessness is, at this point, more assertion than demonstration.
- Baseline is a separate day, and the modalities are at different drug times. The “no psilocybin” comparison is a different session, so drug effects are confounded with session order, scanner habituation, and day. And fMRI was ~80 min post-dose while EEG was ~150 min, so the celebrated cross-modal “convergence” is a convergence across different pharmacological time points, not a clean simultaneous replication.
- One methodological detail can be only partly verified from the main text: exactly how CEBRA’s contrastive objective was supervised. If condition labels informed the embedding and then an SVM classified condition on that embedding, some separability is baked in. The unsupervised-method replications (PCA/Isomap/t-SNE) are what rescue this concern, but it’s the thing I’d read the Methods and Supplement most carefully to confirm.
The usual fMRI caveats (no global-signal regression, defended in Methods; motion in a moving population, QC’d and partially checked) apply but are handled about as well as the field currently does.
Bottom line
This is a strong, data-rich contribution whose conceptual claim, that temporally resolved analysis reveals organization that time-averaging hides, is well-supported and worth taking seriously as a corrective to the “pure desynchronization” narrative. But the interpretive superstructure (“embeddedness,” continuity of self and world, a brain-derived marker of therapeutic change) rests on open-label correlations between two expectancy-sensitive measures in a small, elaborately staged sample with a fixed condition order. It’s a compelling hypothesis-generator dressed in the confidence of a demonstrated mechanism. The right next step is the obvious one the design couldn’t deliver: a blinded, active-placebo-controlled, counterbalanced-order replication before “embeddedness” gets treated as an established construct.
NB: This critique was assisted with the use of Claude.
Dr Bob, Brain Mechanic 