PLoS Computational Biology
August 2, 2023
Alexander Tscshantz, Beren Millidge, Anil K. Seth et al.
56 citations
Predictive coding theory holds that the brain perceives by minimizing prediction errors through cycles of neural activity. However, some visual perception, including complex object recognition, happens too quickly for such cycles. This paper proposes that the initial fast 'feedforward sweep' performs amortized inference, using a learned function to map data directly to beliefs, while slower recurrent processing performs iterative inference, sequentially updating beliefs for greater accuracy. A hybrid predictive coding network combining both methods is introduced, implemented in a biologically plausible neural architecture using local Hebbian rules. The hybrid model achieves rapid perception for familiar data while retaining context-sensitivity and sample efficiency for novel situations, and adaptively balances both inference modes based on uncertainty.
Cognitive Neuroscience
April 7, 2015
Anil K. Seth
56 citations
The author argues that perceptual presence—the sense that something is really there—can be explained by predictive models that link perception with action, specifically through the framework of predictive processing of sensorimotor contingencies. The paper re-emphasizes the core commitment to predictive model-based perception, reconsiders how presence relates to objecthood, and refines the target by distinguishing perceptual presence from the phenomenology of absence-of-presence, or phenomenal unreality. This conceptual work requires speculative, blue-sky thinking to advance the theory.
Animal Sentience
January 1, 2016
Anil K. Seth
49 citations
The question of whether fish consciously feel pain cannot be settled by looking for the same neural mechanisms that underlie human pain reports. Non-mammalian consciousness, if it exists, may rely on different mechanisms. Neurophysiological and behavioral evidence, evolutionary considerations, and Bayesian brain theories suggest that if fish can feel at all, they can feel pain. Because pain and suffering are qualitative experiences, the precautionary principle obliges us to accommodate the possibility that fish feel pain where doubt remains.
International Journal of Machine Consciousness
May 19, 2009
Anil K. Seth
45 citations
Weak artificial consciousness—using synthetic models to simulate neural mechanisms—can advance understanding of how those mechanisms produce conscious experience, whereas strong artificial consciousness, which aims to instantiate consciousness in machines, is less likely to yield such insights. Synthetic models can serve as "explanatory correlates" that causally account for deep structural properties of consciousness. Strong artificial consciousness, while not impossible, has not been credibly illustrated and suffers from circularity: using models both as instantiations and as tools for discovering principles delays empirical comparison.
Psychology of Consciousness Theory Research and Practice
June 25, 2020
Zoltan Dienes, Peter Lush, Bence Pálfi et al.
42 citations
Hypnotizability reflects a broader trait called phenomenological control, which allows people to create subjective experiences in nonhypnotic contexts to fulfill goals. This control operates as a metacognitive process where intentional actions occur without awareness of specific intentions, known as cold control. Laboratory phenomena such as vicarious pain, mirror-touch synesthesia, and the rubber hand illusion may partly result from phenomenological control. A new theory of intentional binding measures the absence of conscious intentions in hypnosis. No evidence suggests cold control confers abilities beyond changes in metacognitive monitoring, and a negative correlation exists between mindfulness and cold control, viewed as a lack of awareness of intentions.
bioRxiv (Cold Spring Harbor Laboratory)
November 2, 2020
Pedro A. M. Mediano, Fernando E. Rosas, Christopher Timmermann et al.
39 citations
preprint
Psychedelics reliably increase brain entropy (neural signal diversity), an effect linked to psychological changes and opposite to the decrease seen during loss of consciousness. This study investigated how context—specifically stimulus manipulation—modulates that entropy increase. Participants under LSD or placebo experienced eyes-closed versus eyes-open conditions, or no stimulus, music, or video. Brain entropy rose with LSD across all conditions but was largest with eyes closed. Entropy changes consistently matched subjective ratings of the psychedelic experience, except during video viewing, suggesting competition between external stimuli and internal LSD-induced imagery. The findings provide quantitative evidence that context shapes neural dynamics during psychedelic experiences, supporting the practice of eyes-closed psychedelic psychotherapy, and challenge simplistic views of brain entropy as a direct measure of conscious level.
Collabra Psychology
January 1, 2021
Peter Lush, R. B. Y. Scott, Anil K. Seth et al.
35 citations
Phenomenological control—the ability to generate experiences to match expectancies—varies as a stable trait and can be measured through responses to imaginative suggestions such as paralysis, amnesia, or hallucinations. While traditionally studied within hypnosis, the context of hypnosis itself may distort scores due to misconceptions. The Phenomenological Control Scale (PCS), adapted from the Sussex Waterloo Scale of Hypnotizability but without hypnotic framing, yields higher mean scores and similar reliability on its subjective scale. The PCS is thus a valid tool for assessing trait phenomenological control outside the hypnotic context, suitable for broader psychological experiments.
Frontiers in Psychology
January 1, 2010
Anil K. Seth
34 citations
The article argues that the concept of a 'cognitive unconscious' is fundamentally flawed because it conflates two distinct meanings of the term 'unconscious.' One meaning refers to mental processes that are inaccessible to introspection, while the other refers to neural processes that are not mental at all. The authors contend that much of what is called unconscious cognition is better understood as non-conscious neural activity that lacks the representational content required for genuine mental states. They conclude that the cognitive unconscious is a misleading notion that obscures the true relationship between mind and brain.
bioRxiv (Cold Spring Harbor Laboratory)
January 4, 2019
David J. Schwartzman, Michael Schartner, Benjamin B. Ador et al.
31 citations
preprint
Stroboscopic stimulation—flashing light—can induce altered states of consciousness without drugs, increasing the intensity and range of subjective experiences, including simple and complex visual hallucinations. These experiences were accompanied by rises in EEG signal diversity, measured by Lempel-Ziv complexity, that exceeded levels seen during wakeful rest. The results align with previous findings from psychedelic studies and support the idea that neural signal diversity reflects the richness of subjective experience across different states of consciousness.
Frontiers in Human Neuroscience
July 3, 2014
Cassandra Gould van Praag, Tom Froese, Adam B. Barrett et al.
29 citations
A detailed phenomenological investigation of sequence-space synesthesia in a single adult (AB) reveals that synesthetic experiences occur within a specific spatial reference frame, a 'mental room' that overlays perception of the physical world, and that AB can voluntarily switch attention between these two rooms. Using the Elicitation Interview method across 9 hours of transcripts, the study found that, contrary to AB's initial report, the synesthetic concurrents usually followed an internal verbalization of the inducer rather than visual presentation, indicating an auditory component. This suggests cross-modal interactions between auditory and visual systems in the development of this synesthetic form.
Cognitive Neuropsychiatry
February 27, 2018
Geoff Davies, Charlotte L. Rae, Sarah N. Garfinkel et al.
23 citations
People in the early stages of psychosis show a reduced ability to accurately evaluate their own perceptions, a metacognitive deficit. In a perceptual decision-making task, individuals with first-episode psychosis performed worse on metacognitive accuracy than healthy controls, even when their basic perceptual performance was accounted for. Grey matter volume was also reduced in the superior frontal gyrus, a brain region linked to metacognition. However, no direct relationship was found between grey matter volume and metacognitive accuracy, leaving the neural basis of this deficit unclear. These findings suggest that a specific problem in scrutinizing one's own perceptions may underlie later functional difficulties in schizophrenia.
Journal of Experimental Psychology General
January 30, 2017
Yair Pinto, Annelinde R. E. Vandenbroucke, Marte Otten et al.
23 citations
Change blindness—the failure to notice unattended changes in a visual scene—has been taken as evidence that conscious perception requires spatial attention. However, change blindness tasks introduce visual interference that forces reliance on working memory, which has limited capacity. Before such interference, a larger-capacity store called fragile memory exists. Two experiments show that minimizing spatial attention nearly eliminates visual working memory but leaves fragile memory largely intact. Furthermore, minimally attended fragile memory responses produce accurate metacognition, indicating that conscious memory persists with limited spatial attention. These findings suggest that both visual consciousness and memory can be supported by only minimal attention.
Frontiers in Human Neuroscience
January 3, 2024
Keisuke Suzuki, Anil K. Seth, David J. Schwartzman
20 citations
Visual hallucinations differ substantially depending on their cause, such as neurodegenerative disease, visual loss, or psychedelic drugs. Using a deep neural network approach called computational (neuro)phenomenology, researchers identified three key dimensions that distinguish these hallucinations: realism (how true-to-life they seem), spontaneity (how much they depend on sensory input), and complexity. By tuning the network along these dimensions, they generated synthetic hallucinations characteristic of each cause. Two studies with patients having Parkinson's disease, Lewy body dementia, or Charles Bonnet syndrome, and people with recent psychedelic experience, confirmed that these synthetic images matched the phenomenology reported by each group. The findings show that a neural network model can capture the distinctive visual features of hallucinations from different origins.
Cognitive Neuroscience
August 18, 2010
Geraint Rees, Anil K. Seth
11 citations
Consciousness research has advanced rapidly through cognitive neuroscience methods that combine multiple experimental techniques. This special journal issue presents five new empirical studies and one theoretical paper with peer commentaries, collectively examining how conscious mental states relate to brain activity patterns.
bioRxiv (Cold Spring Harbor Laboratory)
July 16, 2019
Lionel Barnett, Suresh Muthukumaraswamy, Robin Carhart‐Harris et al.
10 citations
preprint
Psychedelic drugs such as LSD, psilocybin, and low-dose ketamine reduce directed functional connectivity—the flow of information—across the brain, as measured by Granger causality in source-localised MEG recordings. This breakdown in organised information flow supports the idea that the psychedelic state disrupts normal patterns of neural communication. With LSD specifically, directed connectivity decreased while undirected connectivity (measured by correlation and coherence) increased, an opposite movement that highlights the importance of using multiple connectivity measures when analyzing time-resolved neuroimaging data. The non-psychedelic anticonvulsant tiagabine was included for comparison.
Pedro A.M. Mediano, Fernando E. Rosas, Andrea I. Luppi et al.
10 citations
A new method called Complexity via State-space Entropy Rate (CSER) estimates neural signal complexity with better temporal resolution and spectral decomposition than the standard Lempel-Ziv complexity (LZ) approach. CSER matches LZ in distinguishing conscious states but offers two key advantages: it can break complexity down by frequency bands, and it provides temporal resolution about 100 times finer. Using MEG, EEG, and ECoG data from humans and monkeys, CSER revealed that gamma-band activity primarily drives complexity changes across states of consciousness. In an auditory mismatch negativity experiment, CSER detected early entropy increases roughly 20 milliseconds before the standard event-related potential. This method enables finer-grained study of how signal complexity relates to cognitive processes and conscious states.
bioRxiv (Cold Spring Harbor Laboratory)
March 11, 2025
Devon Stoliker, Leonardo Novelli, Moein Khajehnejad et al.
8 citations
preprint
Psychedelics like psilocybin alter consciousness by reorganizing brain connectivity in a context-sensitive way. In the largest psychedelic neuroimaging dataset to date, 62 adults underwent functional MRI and EEG before and after ingesting 19 mg of psilocybin, during rest and naturalistic stimuli. Under psilocybin, brain signals during eyes-closed conditions became similar to those during eyes-open conditions, with increased global functional connectivity in associative regions and decreased connectivity in sensory areas. Machine learning linked subjective effects to structured neural activity patterns. Stronger self-dissolving effects were associated with more distinct neural representations and next-day mindset changes, revealing a state of 'embeddedness' where networks that usually segregate internal and external processing integrate coherently, aligning neural dynamics with context.
bioRxiv
February 1, 2022
Hardik Rajpal, Pedro A.M. Mediano, Fernando E. Rosas et al.
5 citations
preprint
Schizophrenia and drug-induced states from LSD and ketamine both increase neural signal diversity, but they differ in how information flows in the brain. In schizophrenia, transfer entropy from the front to the back of the brain is increased, whereas under both drugs it is reduced overall. These differences can be modeled by altering Bayesian inference within a predictive processing framework: drug effects correspond to reduced precision of prior beliefs, while schizophrenia corresponds to increased precision of sensory information. The findings clarify similarities and differences between these altered states, with potential implications for understanding consciousness and developing mental health treatments.
Behavioral and Brain Sciences
January 1, 2016
Anil K. Seth
4 citations
Consciousness may be linked to voluntary action selection, but a narrow focus on skeletomotor control overlooks how interoception and autonomic regulation contribute to conscious selfhood and subjectivity. From the perspective of predictive processing, internal 'action'—the regulation of the body's internal state—plays a crucial role in shaping conscious experience. The argument suggests that understanding consciousness requires integrating both external action selection and internal bodily regulation.
bioRxiv Preprint Server
July 16, 2025
Borjan Milinkovic, Anil K. Seth, Lionel Barnett et al.
2 citations
preprint
Consciousness depends on neural activity across many scales. A new measure, dynamical independence (DI), quantifies these multi-scale relationships. Applying DI to EEG data from people under three anaesthetics, the authors found that propofol and xenon—which abolish conscious report—produce more emergent but highly variable dynamic structure, indicating fragmented macroscopic organisation. Ketamine, which preserves dream-like states, shows reduced overall emergence but partial preservation of macroscopic structure similar to wakefulness. Regional brain contributions varied. The results reveal drug-specific reconfigurations of emergent dynamics, dissociate the amount of emergence from its organisation, and caution against equating emergence with consciousness level.
bioRxiv Preprint Server
November 3, 2017
Keisuke Suzuki, Warrick Roseboom, David J. Schwartzman et al.
2 citations
preprint
A tool called the Hallucination Machine simulates visual hallucinatory experiences using deep convolutional neural networks and panoramic virtual reality videos of natural scenes. It induces visual phenomenology qualitatively similar to classical psychedelics, but does not evoke the temporal distortion commonly associated with altered states. This technique allows researchers to study altered consciousness without the confounding physiological and cognitive effects of psychoactive substances or psychopathological conditions, offering a valuable method for consciousness science and psychiatry.
bioRxiv (Cold Spring Harbor Laboratory)
February 18, 2026
Ethan Grove, Trevor Hewitt, Anil K. Seth et al.
1 citation
Visual hallucinations (VHs) occur in psychedelic states and various psychiatric and neurological conditions, but their phenomenology is hard to characterize due to a lack of large-scale datasets. Stroboscopic light stimulation (SLS) with closed eyes reliably induces VHs in healthy people, producing vivid colors and dynamic geometric patterns similar to simple VHs in other contexts. Researchers developed an unsupervised computer-vision pipeline to analyze 10,598 drawings made after hallucination-inducing SLS at a public installation. Most drawings contained geometric forms, consistent with prior observations, but novel patterns like concentric squares, crosses, and hyperbolic shapes also appeared. The pipeline organized the drawings into interpretable classes, mapping the diversity of simple geometric VHs and placing new constraints on theoretical accounts.
PLoS ONE
December 4, 2025
Jonathan Robinson, Andrew W. Corcoran, Christopher J. Whyte et al.
1 citation
Active inference, a framework for modeling how sentient agents behave, is being tested as necessary for changes in conscious content. In an adversarial collaboration, active inference will be contrasted with two other theories that do not require it for consciousness. This study protocol describes an adaptation of the motion-induced blindness paradigm: an active condition where participants direct their gaze toward a target after it disappears from consciousness and report its reappearance, versus a passive condition where participants fixate centrally while the stimulus array moves in a replay of active eye-tracking data. Two experiments will compare target reappearance across conditions to evaluate active inference's contribution to conscious awareness.
December 13, 2024
Trevor Hewitt, Ioanna Amaya, Romy Beauté et al.
preprint
Exposure to rapid and bright stroboscopic light can induce vivid visual hallucinations of color and geometric forms, a phenomenon first documented by Purkinje over 200 years ago. Despite centuries of scientific, therapeutic, and cultural interest, fundamental questions remain about its phenomenology, physiological origins, and potential clinical applications. This narrative review summarizes the historical research on stroboscopic light stimulation, its use in recreational and lay-therapeutic settings, and discusses the phenomenology of these experiences. It also examines current perspectives on the neural mechanisms that may underlie stroboscopically induced experiences and outlines directions for future research.
medRxiv Preprint Server
June 17, 2026
Danny Nacker, Luise Kalus, Anil K. Seth et al.
preprint
Supervised stroboscopic light stimulation (SLS) was safe, tolerable, and feasible in adults with depressive symptoms, but efficacy was not established. In a staged program, 31 participants tested 11 SLS parameter sets; no severe adverse reactions occurred, and mean discomfort was low (0.49 out of 10). A subsequent randomized trial assigned 84 participants to four weekly 31-minute sessions of SLS or a low-phenomenology control. Retention was 83.3% (70 of 84 participants), with higher retention in the intervention arm (39 of 42) than the control arm (31 of 42). Exploratory depressive-symptom changes suggested a possible signal on the BDI-II but do not confirm efficacy. The next step is a Phase 2a feasibility trial with a locked protocol.