A new eLife paper by Colin Bredenberg, Fabrice Normandin, Blake Richards, and Guillaume Lajoie takes a swing at one of neuroscience's strangest questions: why do classical psychedelics produce visuals that are not random TV static, but not exactly reality either? Their answer is the oneirogen hypothesis - basically, psychedelics may push the brain into a more dream-like mode while you're still awake and upright and, ideally, not trying to text your ex (Bredenberg et al., 2026).

The trick is that they do not model hallucinations as pure chaos. They use a neural network trained with the classic Wake-Sleep algorithm, a setup with two modes. In the wake phase, the model learns from actual sensory input. In the sleep phase, it generates its own internal activity, like the brain's overnight cleanup crew deciding to improvise a little. The researchers then simulate psychedelic effects by nudging the system partway toward that sleep-like state, increasing the influence of top-down signals that reflect internal expectations rather than raw incoming data.

That matters because perception is never just your eyeballs doing honest labor. Your brain is constantly guessing what is out there, then correcting itself. Under normal conditions, this works great. Under psychedelics, this paper argues, the guessing machine gets a little too much executive power. Imagine a manager barging into the sensory department yelling, "I already know what the memo says," while the interns with the actual memo are still flipping pages.

Hallucinations, but Make Them Mechanistic

What is cool here is not just the vibe of the theory. The model reproduces several things people actually report and researchers actually measure. It generates hallucination-like outputs, increases stimulus-conditioned variability, and predicts a boost in synaptic plasticity. In plain English: perception gets wobblier, internally generated content gets louder, and the learning machinery may become more flexible.

That lands neatly in a broader literature suggesting psychedelics alter large-scale brain communication rather than simply "turning things up." Recent reviews describe shifts in functional connectivity, sensory processing, and subjective experience under compounds like psilocybin and LSD (Yu et al., 2024; Erritzoe et al., 2024). Another study on psychedelic visual imagery found connectivity changes consistent with stronger internally driven imagery and weaker ordinary sensory constraint, which sounds very much like the brain replacing "what's there?" with "I have ideas" (Mason et al., 2025).

There is also a nice conceptual overlap with predictive coding, the idea that the brain is always building a running model of the world and updating it when reality objects. Psychedelics may loosen those predictions or change how much the system trusts them. This new paper adds a sharper mechanistic angle: maybe the relevant shift is specifically toward a replay-heavy, dream-like regime, not just generic noisy weirdness.

Why This Is More Than a Very Educated Trip Report

The real value of the paper is that it gives researchers something testable. Not "consciousness expands into cosmic jellybeans," which is hard to put in a grant proposal, but specific predictions about neural activity, variability, and plasticity. That is a much better deal.

It also tackles a real gap in the field. Psychedelic neuroscience has plenty of big theories, including debates around the entropic brain idea and how altered states should be classified (Miller et al., 2023). What it has had less of is a concrete model that links pharmacology, circuit effects, perception, and learning in one package without immediately wandering off into interpretive modern dance.

That said, nobody should oversell this. A computational model is not proof that the human cortex literally runs Wake-Sleep like it is a 1995 Helmholtz machine with feelings. The eLife assessment itself notes that the model gives a useful theory of hallucinations, but it is not yet clear that it specifically captures 5-HT2A psychedelic mechanisms and not hallucination-like dynamics more broadly. That is the right level of caution. In psychedelic research, the field has already had a few reminders that exciting stories and clinical reality are not the same thing. On August 9, 2024, the FDA declined to approve MDMA-assisted therapy for PTSD, underscoring how quickly enthusiasm runs into demands for cleaner evidence.

Still, this paper is genuinely interesting because it treats hallucinations not as mystical smoke, but as computation with the guardrails loosened. Your waking brain and your dreaming brain may not be opposites. They may be neighbors with a thin fence, and psychedelics may just open the gate.

References

1. Bredenberg C, Normandin F, Richards B, Lajoie G. Modeling the hallucinatory effects of classical psychedelics in terms of replay-dependent plasticity mechanisms. eLife. 2026;14:RP105968. DOI: 10.7554/eLife.105968.3. PubMed: 42011872

2. Hatzipantelis CJ, Olson DE. The Effects of Psychedelics on Neuronal Physiology. Annual Review of Physiology. 2024;86:27-47. DOI: 10.1146/annurev-physiol-042022-020923

3. Erritzoe D, et al. Exploring mechanisms of psychedelic action using neuroimaging. Nature Mental Health. 2024;2:141-153. DOI: 10.1038/s44220-023-00172-3

4. Yu Z, Burback L, Winkler O, et al. Alterations in brain network connectivity and subjective experience induced by psychedelics: a scoping review. Frontiers in Psychiatry. 2024;15:1386321. DOI: 10.3389/fpsyt.2024.1386321

5. Mason NL, et al. Neural mechanisms of psychedelic visual imagery. Molecular Psychiatry. 2025;30:1259-1266. DOI: 10.1038/s41380-024-02632-3

6. Tipado Z, Kuypers KPC, Sorger B, Ramaekers JG. Visual hallucinations originating in the retinofugal pathway under clinical and psychedelic conditions. European Neuropsychopharmacology. 2024;85:10-20. DOI: 10.1016/j.euroneuro.2024.04.011

7. Miller J, Kiverstein J, Rózsa N, et al. Psychedelics, entropic brain theory, and the taxonomy of conscious states: a summary of debates and perspectives. Neuroscience of Consciousness. 2023;2023(1):niad001. DOI: 10.1093/nc/niad001

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.