The paper presents an active inference account of psychedelic-induced ego-dissolution, arguing that psychedelics lower precision on high-level priors (raising the Bayesian learning rate) which collapses the temporal thickness of the deep generative model and thereby dissolves the self-model. This framework links ego-dissolution to mechanisms underlying ordinary self-consciousness and its disturbances in conditions such as psychosis, autism, depression and dissociative disorders.
Psychedelic drugs such as psilocybin, LSD and DMT are known to induce powerful alterations in phenomenology. Perhaps of most philosophical and scientific interest is their capacity to disrupt and even “dissolve” one of the most primary features of normal experience: that of being a self. Such “peak” or “mystical” experiences are of increasing interest for their potentially transformative therapeutic value. While empirical research is underway, a theoretical conception of the mechanisms underpinning these experiences remains elusive. In the following paper, psychedelic-induced ego-dissolution is accounted for within an active inference framework, as a collapse in the “temporal thickness” of an agent’s deep temporal model, as a result of lowered precision on high-level priors. The argument here is composed of three moves: first, a view of the self-model is proposed as arising within a temporally deep generative model of an embodied organism navigating an affordance landscape in the service of allostasis. Next, a view of the action of psychedelics as lowering the precision of high-level priors within the generative model is unpacked in terms of a high Bayesian learning rate. Finally, the relaxation of high-level priors is argued to cause a “collapse” in the temporal thickness of the generative model, resulting in a collapse in the self-model and a loss of the ordinary sense of being a self. This account has implications for our understanding of ordinary self-consciousness and disruptions in self-consciousness present in psychosis, autism, depression, and dissociative disorders. The philosophical, theoretical and therapeutic implications of this account are touched upon.
Papers cited by this study that are also in Blossom
Bayne, T., Carter, O. · Neuroscience of Consciousness (2018)
Carhart-Harris, R. L. · Current Opinion in Psychiatry (2019)
Carhart-Harris, R. L., Friston, K. J. · Pharmacological Reviews (2019)
Carhart-Harris, R. L., Goodwin, G. M. · Neuropsychopharmacology (2017)
Carhart-Harris, R. L., Nutt, D. J. · Journal of Psychopharmacology (2017)
Psychedelic drugs such as psilocybin, LSD and DMT produce profound alterations in perception, emotion, time perception and self-consciousness, among which ego-dissolution (a felt loss of self or blurred self/world boundary) is especially noteworthy. Deane frames ego-dissolution as philosophically and therapeutically important yet mechanistically under-explained, noting prior suggestions that such experiences relate to relaxation of high-level beliefs within predictive processing frameworks and to measurable therapeutic outcomes when experienced as “peak” or mystical states. This paper sets out to provide a theoretical account of psychedelic-induced ego-dissolution within an active inference / predictive processing framework. Deane proposes three linked moves: (1) to characterise the self-model as arising from a temporally deep generative model that supports allostatic control; (2) to characterise the action of classical psychedelics as lowering the precision of high-level priors (operationalised as a very high Bayesian learning rate); and (3) to argue that reduced precision at high hierarchical levels collapses the temporal thickness of the generative model, producing the phenomenology of ego-dissolution. The account is intended to illuminate ordinary self-consciousness, pathological self-disruptions and therapeutic mechanisms.
This is a theoretical and conceptual paper rather than an empirical study. Deane develops an argument by integrating concepts from the Free Energy Principle, predictive processing, active inference and existing empirical and phenomenological literature on psychedelics. No original experimental data are reported; instead, the paper synthesises prior computational models, neurophysiological findings, perceptual paradigms and phenomenological reports to build a coherent explanatory framework. The methodological approach consists of (a) explicating key theoretical constructs (free energy, predictive processing, precision-weighting, active inference, temporal thickness and allostatic control), (b) mapping accounts of the self onto hierarchically deep generative models that encode distal and proximal goals, and (c) translating proposals about psychedelic action (notably the REBUS idea of relaxed high-level beliefs) into the language of precision-weighting and a high Bayesian learning rate. Deane also reviews converging empirical observations (e.g. neural plasticity under psychedelics, perceptual paradigms such as Kanizsa figures, binocular rivalry and oddball mismatch responses) to assess the plausibility of the high learning-rate hypothesis and to motivate the proposed link to ego-dissolution. Where relevant, alternate mechanistic accounts from the literature are considered (for example, proposals that psychedelics increase sensory noise rather than reduce high-level precision). The paper remains agnostic where causal direction is unclear and highlights the need for empirical work to adjudicate between competing mechanisms.
Deane's central theoretical “results” are a set of interlinked propositions rather than new empirical measurements. First, the self-model is characterised as an allostatic control model implemented by hierarchically deep generative models: pre-reflective self-consciousness arises from the system's temporally thick expectations about the sensory consequences of action across multiple interlocking timescales. These deep models allow agents to anticipate distal outcomes, arbitrate between policies and assign salience to affordances in service of continued viability. Second, psychedelics are interpreted within predictive processing as agents that lower precision on high-level priors, which the paper operationalises as producing a very high Bayesian learning rate. Deane marshals indirect empirical support for this view: psychedelics increase neural plasticity; psilocybin reduces Kanizsa-triangle filling-in and associated evoked potentials (consistent with weakened influence of prior sensory history); psilocybin alters binocular rivalry dynamics towards fusion or reduced switching; and LSD blunts mismatch negativity responses in oddball paradigms. Phenomenological reports (heightened sensory vividness, dynamic distortions) are also invoked as consistent with diminished top-down constraint. Third, the paper argues that lowering high-level precision (high learning rate) causes a collapse in the temporal thickness of deep generative models. Practically, this means the system shifts minimisation of prediction error to much shorter timescales, failing to contextualise lower-level prediction errors with distal expectations. The immediate consequences include impaired sensory attenuation and corollary discharge (so self-generated sensations are less effectively cancelled), increases in low-level prediction error and potential misattribution of endogenous events to exogenous causes. Deane links these computational changes to the phenomenology of ego-dissolution: bodily boundaries and the sense of agency may blur early, with narrative or higher-level aspects of selfhood breaking down at higher doses or later in the experience. The paper contrasts the high learning-rate/low high-level precision account with alternative proposals (e.g. increased sensory noise at lower levels), noting that the high learning-rate framing is compatible with either mechanism because it emphasises the net change in the balance between top-down and bottom-up influence rather than uniquely specifying the locus of change. Deane also reports theoretical implications and brief empirical implications: the account could extend to cognitive affordances (mental action), explain why mental phenomena may feel “objective” under psychedelics, and suggest mechanisms by which peak experiences could produce lasting therapeutic shifts through retuning of entrenched high-level priors. Finally, therapeutic and adverse-outcome considerations are outlined. The model suggests psychedelics can act as a “reset” by relaxing overly precise high-level priors (for example, priors that underpin depressive low allostatic self-efficacy), potentially enabling perceptual revision and retuning of self-models. Conversely, incomplete ego-dissolution—when a high-precision prior on control persists—may generate intense fear or distress, consistent with reports of challenging experiences. The importance of context and set/setting is emphasised. The extracted text does not report new quantitative effect sizes or statistical results because the paper presents a conceptual synthesis.
Deane interprets the preceding argument as offering a parsimonious way to link phenomenology, computational theory and emerging empirical findings: ego-dissolution follows from a collapse in the temporal thickness of hierarchically deep self-models when psychedelics lower precision on high-level priors. This collapse explains why early effects can be bodily (loss of sensory attenuation and agency) while more abstract, narrative aspects of the self require stronger perturbation to be affected. The account situates psychedelic experiences within broader theories of perception, action and affect, framing the self as fundamentally affective and action-oriented because it serves allostatic regulation. The paper relates its proposal to prior models, notably the REBUS hypothesis (relaxed beliefs under psychedelics), and acknowledges alternative mechanistic accounts (such as increased sensory noise) while arguing the high Bayesian learning-rate formulation preserves theoretical benefits and remains agnostic about exact neural loci. Deane positions the account as having implications beyond psychedelics: similar computational disruptions of precision have been proposed to underlie hallucinations, delusions and perceptual styles observed in psychosis and autism, and the framework may illuminate self-disruptions in depression and dissociative conditions. Several limitations and uncertainties are acknowledged. The paper is preliminary and conceptual; it does not provide direct causal evidence linking specific neuropharmacological actions to the proposed precision changes. Disentangling whether relaxation of high-level priors drives reduced sensory gating or vice versa is noted as difficult; the author calls for empirical work to adjudicate causation. Deane also highlights the need to connect neurocomputational mechanisms to the temporal dynamics of experience using microphenomenological interviews and to map these dynamics onto neural correlates. Implications for therapy and research are discussed cautiously. The hypothesis offers a mechanistic rationale for why peak or mystical experiences may predict positive therapeutic outcomes (by relaxing entrenched high-level priors), and suggests possible extensions to treatment of conditions such as depression and chronic pain. At the same time, the account underscores the role of psychological set and context: maintaining or encouraging high-precision priors about control can produce challenging experiences. The author thus stresses controlled conditions and therapeutic support when employing psychedelics and calls for further empirical and computational studies to test and refine the model.
Deane concludes that framing psychedelic-induced ego-dissolution in terms of predictive processing and active inference provides a coherent theoretical account: classical psychedelics relax high-level beliefs, which can be expressed as a high Bayesian learning rate, leading to a collapse in the temporal thickness of the self-model and the phenomenology of ego-dissolution. This framework links phenomenology, computational theory and tentative empirical findings, and offers a rationale for how ego-dissolution might produce enduring therapeutic change by retuning entrenched self-models and opening the affordance landscape to new modes of engagement.
Psychedelic ("mind-manifesting") drugs are known to occasion radically altered states of consciousness, including profound changes in sensory perception, emo-tion, cognition, time perception, and self-consciousness. One of the most interesting of all of these effects is the experience of egodissolution. Although the experience is notoriously difficult to articulate and even considered ineffable, psychedelic researcher Stanislas Grof, who considers egodissolution the "main objective" of psychedelic therapy, describes it as "an ecstatic state, characterized by the loss of boundaries between the subject and the objective world, with ensuing feelings of unity with other people, nature, the entire Universe, and God").Ego-dissolution is of considerable philosophical and theoretical value for understanding selfhood and the nature of consciousness). 2 It is also considered to be central to the therapeutic potential of psychedelics; see alsoin this special issue). Despite this, very little is known about the mechanisms underpinning psychedelic-induced ego-dissolution. "Predictive processing" theories of brain functionhave recently taken precedence in cognitive science, affording a novel theoretical framework to approach cognitive phenomena. In this paper I propose a novel account of ego-dissolution within an active inference framework. To this end, I initially furnish an account of self-modelling within active inference, where pre-reflective self-consciousness emerges in organisms as a consequence of "temporal thickness", the need to model the consequences of potential actions over time. I then give an account of the action of psychedelics within a predictive processing framework, unpacking the view that psychedelics "relax" high-level priorsin terms of a high Bayesian learning rate. Finally, I argue that low precision at high-levels of the inferential hierarchy results in a collapse of the temporal thickness of the generative model and the corresponding self-model, leading to the phenomenon known as ego-dissolution (see also.
The Free Energy Principle (FEP) has the most ambitious explanatory scope of all "predictive processing" style theoretical frameworks. It combines, subsumes and links to several brain theories, including the Bayesian brain hypothesis, predictive coding, efficient codingand reinforcement learning. The mathematics of the theory are complex and beyond the scope of this paper (for a review see. According to the FEP, simply in virtue of existing, all organisms tend to minimise the entropy or dispersion of their states. This much is intuitive: the conditions that are viable for an organism are fairly narrow -deviation from homeostatic bounds, such as having a body temperature of 50 degrees centigrade, is incompatible with continued existence. Organisms that fail to stay within their "species-specific window of viability") simply cease to exist. Life, on this account, resists the tendency towards disorder imposed by the second law of thermodynamics, and this principle applies at all levels -"from their gross morphology to fine details of cortical microcircuitry as well as at timescales from the neuronal to the phylogenetic". Organisms, then, must resist entropy, the long-term average of (information-theoretic) surprise. Because this quantity is beyond direct epistemic access to an organism, according to the FEP, organisms minimise a proxy variable or upper bound -dubbed (variational) free energy. Free energy (under some simplifying assumptions) is equivalent to precision-weighted prediction error in predictive processing.
On the predictive processing view, the brain has stored prior beliefs (in the form of probability distributions) about the causes of sensory inputs in the world. Prior beliefs are hierarchically organised, where higher-levels encode predictions about representations at lower levels. Prediction errors, arising from the discrepancy between the low-level predictions and incoming sensory signals, are passed up the hierarchy, where higher-level predictions are updated to minimise further prediction errors. Perception, then, both exteroceptive and interoceptive, is the product of (approximate) Bayesian inference, whereby the influence of prior beliefs and sensory evidence are weighted according to "expected precision", e.g. confidence in the given context, to generate a posterior. Inference in these schemes is thought to occur across a hierarchy of inferred causes, where higher levels encode regularities that occur at larger spatial and temporal scales. In perceptual inference, sensory prediction errors can be minimised by tweaking the parameters of the generative model -that is, generating predictions to quash the influx of prediction error. Prediction error can also be minimised though action by changing the incoming sensory data to fit a prediction -for instance, I can move my eyes to bring my coffee cup into view, to fulfil the prediction of a coffee cup. Actions can be thought of as the fulfilment of proprioceptive (or oculomotor) predictions -an intended movement occurs as a result of predicting the proprioceptive consequences. There are detailed accounts of the neural implementation of these schemes available.
A key feature of predictive processing schemes is the contextual flexibility afforded by precision-weighting. Precision regulates the interaction between top-down and bottom-up signals, through the synaptic gain on neuronal populations signalling prediction error, in order to approximate optimal inference over time. Precision can be thought of as tracking both the reliability and relevance of the incoming sensory information, where weighting by reliability is analogous to assigning greater weight to more reliable information when updating a belief. Prediction error signals with high precision (inverse variance) have greater influence in updating the top-down predictions. Precision itself has to be inferred, both by the empirical variance in the sensory data itself, and according to prior expectations about precision. The optimisation of precision weighting, through updating of the precision expectations (precision-related priors), is frequently equated to attention within predictive processing. Importantly for the current treatment, precision is thought to mediate both sensory attenuation -the top-down filtering out of afferent information, and affordances, where affordances refers to the latent possibilities for action given the capabilities of the agent.
In mandating that existence necessitates maintaining oneself within a limited repertoire of states via control-oriented predictive regulation (instrumental active inference), the FEP aligns itself with precursors of this view, cybernetic theories that build on control, feedback and predictive modelling (e.g., the "good regulator theorem"). Note that while a purely Helmholtzian view of the brain might cast it in terms of inferring hidden causes in the world, casting the predictive machinery in terms of being for ensuring continued existence means that the generative model is not constrained to veridicality. Rather than faithfully reconstructing the world, perception is "ultimately geared towards driving actions that preserve [the] physiological integrity of the organism. In other words, we do not perceive the world (and self) as it is, but as it is useful to do so".
Homeostasis refers to the tendency of living systems to keep an "internal balance" despite changes in the surrounding environment. This has long been described in terms of control theoretic and cybernetic mechanisms, and more recently this homeostatic control is thought to involve interoceptive signals that report current physiological states (e.g., heart rate, or blood-bound glucose levels). One way to restore bodily conditions to favourable states is to engage autonomic reflexes -for example, a hyperthermic animal can perspire to cool down. Of course, autonomic regulation alone is not sufficient to ensure continued existence -to avoid hunger or thirst the animal must engage actions, such as seeking out food and water. Collectively, these actions are termed allostasis, the process via which the brain regulates the needs of the body. Crucially, to stay viable on longer timescales, this action must be anticipatory -avoiding dyshomeostatic conditions before they arise.
The FEP regards homeostasis and allostasis as the central aspects of organic life, thus the autopoietic principles at the basis of the FEP act as a kind of "first prior". In other words, "[t]he brain is in the game of predicting the world, but only as a means to the end of embodied self-preservation". In so doing, the free energy principle collapses expected utility (instrumental value) and information gain (epistemic value) under a single quantity. On this approach, action planning is itself a form of inference, where preferences and goals are framed in terms of prior beliefs, such that these priors are fulfilled by action. Casting value and utility purely as inferential problems may at first appear unintuitive -if an agent finds itself in consistently adverse circumstances, then such adverse circumstances should, at first pass, seem to have high probability. However, "[t]he critical step in this logic is the assumption that evolution has equipped us with the belief that low utility states are low probability, due to the fact that if our ancestors spent a lot of time in those states they would be less likely to reproduce"). The so-called "first prior", that of maintaining existence via homeostatic and allostatic regulatory behaviour, ensures that organisms seek to actively maintain internal and external conditions conducive to their own persistence. Active inference refers to the process by which agents actively sample states of the world so as to reduce uncertainty and realise prior preferences, rendering the action selection process itself an inference problem. This arbitration occurs according to priors pertaining to expected free energy over a given course of action, or policy. Expected free energy is the free energy an agent predicts itself to average in opting to pursue a particular course of action. Intuitively, some courses of action are more likely than others to lead to "expected" or desirable outcomes. A policy that has lower expected free energy is going to have a higher prior probability than a policy with higher expected free energy, because agents equipped with prior beliefs about their continued existence will pursue policies that reduce expected free energy. Crucially, agents engaging in active inference do not merely restrict themselves to the states they expect; rather they anticipate in order to minimise uncertainty about potential future outcomes. This prospective form of control relies on the contextualization provided by higher levels in the inferential hierarchy, which anticipate the downstream consequences of actions and select policies accordingly. Contextualisation here depends on the relative precision at various hierarchical levels, where "precision dynamics subsume the role of arbitration". This approach bears similarities to other control-theoretic approaches, such as the affordance competition hypothesis, where an affordance is a potential for action that avails itself to an organism in its action-oriented perception of environmental features.On this view, perceived affordances jostle for precedence and are arbitrated on the basis of the desirability of their predicted outcomes.
This section outlines an account of how pre-reflective self-consciousness -an implicit sense of being a subject present in all experience -is structured within an active inference framework. Here, the self-model is underpinned by the same inferential Bayesian schemes that are increasingly being used to describe perception and action. This predictive-modelling approach to selfhood has roots in Thomas Metzinger's work on conscious and unconscious self-models, and the "self-model theory of subjectivity"where "[a] self-model, an inner image of the organism as a whole [is] built into the worldmodel, and this is how the consciously experienced first-person perspective develop[s]". The account presented here follows the increasing focus on the embodied nature of selfhood, where "being" or "having" a body is thought to be one of the most basic aspects of the experience of being a self. A growing number of researchers seek to ground selfhood and emotion in interoceptive processes, particularly in their functional relation to allostatic regulation. A key reason for this is that interoceptive inference is apt to put greater emphasis on control over discovery, due to "a priori hyper-precision of visceral channels", in which interoceptive signals are assigned very high precision in virtue of communicating information about key physiological variables. Grounding the self-model in control-oriented active inferenceinflects perception of the affordance landscape in terms of bodily states, an idea which is nicely expressed by Montague and King-Casas: A sated and comfortable lioness looking at two antelopes sees two unthreatening creatures against the normal backdrop of the temperate savanna. The same lioness, when hungry, sees only one thing -the most immediate prey. In another circumstance, in which the lioness may be inordinately hot, the distant, shaded tree becomes the prominent visual object in the field of view.This forms the basis for the view that will be unpacked in more detail in what follows, that the self-model can be understood as an "allostatic control model", arising from the system's sense of control of the temporally deep consequences of actions for allostasis. On this view, pre-reflective self-consciousness is underpinned by the inference about endogenous control of the sensory consequences of actions within deep goal hierarchies, where goals and preferences are framed in terms of prior beliefs, such that goals are fulfilled by actions. Recall, action allows an organism to change the sensory input in order to conform to its generative model, as opposed to perceptual inference that involves revising model parameters to conform to the sensory input. In order to act, then, the system implicitly infers its own ability to bring about the intended sensory consequences -it is in this sense that "implicit in a model of sampling is a representation or sense of agency", which is closely related to what has been called the "primacy of the 'I can' ". Crucially, organisms with deep temporal models have "temporal thickness" -expectations regarding the sensory consequences of actions on multiple interlocking timescales. The following sections will unpack this conception of the self-model in terms of hierarchically deep allostatic control, starting with the notion of temporal thickness, and then moving to how motivated control hierarchies "attune" organisms to action opportunities on multiple timescales, for both proximal goals, for instance, pain motivating an organism to act so as to fulfil a "healthy body condition" prior; and distal goals, for example emotions motivating a change of circumstances pertaining to longer timescales such as moving to a different city. The discussion will then move to how deep self-models allow organisms to arbitrate between different policies and trade off outcomes on different timescales.
To successfully navigate the world over longer timescales, and select policies that result in survival -and not dispersion or non-existence -organisms must possess models of the future; in other words, they require deep temporal models. The generative models that endow organisms with the capability of inferring the consequences of future actions must have the property of temporal thickness, which allows the organism to anticipate the downstream consequences of potential actions, conferring the ability to select policies or action scripts that are favourable to the organism's continued existence. The minimi-sation of surprise through active inference on the FEP involves acting so as to reduce uncertainty, and to do this the system must model itself across time and counterfactuals -that is, it must model what kind of agent it is at varying degrees of temporal depth. Self-modelling, then, emerges as a natural consequence of prospective action selection, where the principal function of a counterfactually rich self-model is to facilitate navigation of the affordance landscape and action selection across multiple interlocking timescales -for example expectations of what an agent can do on shorter timescales inform expectations of what the agent can do over longer timescales. The functional role of having a rich self-model, then, is that it enables the organism to predict outcomes across diverse policies, and endows the organism with "what if?" capabilities, which puts this picture into contact with mental time travel and offline simulation.
Conceiving of the self-model through an active inference framework, a hierarchically deep self-model guides policy selection over various timescales in service of minimising expected free energy. In what follows, pain perception, viewed as arising through the violation of the prior of "healthy body condition", will be used to illustrate how inferences about the self "attune" an organism to adaptive action opportunities. One key advantage that the active inference account of self-modelling has over strictly Bayesian approaches is that it is goal-directed. Classical models of pain perception as the consequence of physiological dysfunction are challenged by the efficacy of placebo treatments in relieving pain, and cases in which pain is experienced without physiological disruption, as is often the case in chronic pain. Instead, there is evidence to suggest that affectively charged percepts, such as pain, are best understood as resting on the same inferential mechanisms as are assumed to underpin perception and action under a predictive processing framework. Bayesian models of pain perceptionindicate that prior beliefs about the generation of painful percepts are integrated with current sensory data to infer the posterior or hidden worldly cause (the painful percept). Crucially, these pain percepts incorporate the "weight" or precision of past experiences when computing the current painful percept. On their own, however, these models of pain perception are silent on the functional role of pain as a motivator to an embodied organism. Optimal inference about pain to the allostatically concerned organism is heavily dependent on the context, as anyone who has felt the pain of an injury only after danger is averted can attest to. In this way, pain perception is allostatically "tuned": "organisms can tune their own pain perception according to both their prior beliefs and the specific biological goals they believe are attainable in that context". A Bayesian framework of pain perception, therefore, needs to represent the agency and aims of the organism. This is precisely what is afforded by conceiving of the self-model within an active inference framework-as this provides the necessary context to study the self-model, across multiple hierarchical levels. Like physical pain, and sharing the neural underpinnings of physical pain, social pain is similarly understood in inferential terms, and does not scale with "damage" per se (for instance, social rejection), as evidenced by the wide range of sensitivity people have to the same physical manipulation. Accordingly, there is evidence to suggest that appropriately "tuning" emotional responses in social contexts allows for agents to approximate Bayesian inference in policy selection given bounded cognitive capacity and rationality. For example, on a "stag hunt" game,agents with "prosocial" preferences can outperform agents of similar cognitive sophistication that lack social biases.
Conceptualising emotions in terms of a contextualisation of bodily states has historical roots dating back to the James-Lange theory of emotionand two-factor theory of emotion. Lisa Feldman-Barrett has developed this approach specifically within the active inference framework as the "theory of constructed emotion". According to the theory of constructed emotion, emotions are constructed in the same manner as percepts, where priors are recruited according to context to make a "best guess" at the hidden causes of (interoceptive) sensory signals. On Barrett's view, emotions arise through a context-sensitive inferential categorisation of interoceptive states. For this reason, emotions on this view are "constructions" -there are no neural or physiological signatures that reliably discriminate any emotional state. Rather, physiological reactions in the body occur in order to prepare it for action, and these are categorised as emotions only contextually through the predictive models recruited to explain away the incoming afferent interoceptive signals. For example, heart rate increases or decreases depending only on an anticipated action -e.g., fight or flight -and given an emotional ascription only contextually -e.g., the same bodily state could be categorised as fear in one context and anger in another. Interoceptive inference is experienced as emotion in service of producing allostatic action. In viewing the self-model in terms of hierarchical allostatic control, interoceptive inference on the hidden causes of bodily states pertaining to longer timescales tunes perception to the world and affordances differently, such that more abstract emo-tions might track regularities over longer time scales, informing policy selection thereon, and allowing for more abstract and distal outcomes to be motivationally salient.
Viewing the self-model in terms of allostatic control renders selfhood fundamentally affective and action-oriented, such that different aspects of the self in a given context -precision on goals and preferences at different levels of the hierarchymotivate behaviour and arbitrate between policies. On this view, the self-model inflects perception of possible actions in the world and mediates salience to facilitate the selection of policies with minimal expected free energy. Critical to this picture is the notion that these various models are associated with varying degrees of temporal depth. Deep generative models capture increasingly distal relations between actions and outcomes within hierarchical active inference, allowing for the coordination of behaviour across different hierarchical levels, enabling goals to become prioritised relative to current context. The result is an inferential framework of hierarchically nested contextual complexity, in which lower levels track basic (and sometimes evolutionarily hard-wired) motivations or affordances, while higher levels track motivations and plans over deeper timescales. In this way, higher-level contextualization of lower sensorimotor functions optimises expected actions in terms of both long-range consequences of actions and anticipated future affordances. Goals at different levels of abstraction may, of course, conflict -for instance, resolving proximal interoceptive prediction error by eating chocolate cake might conflict with the longer-term goal of sticking to a diet. Alternatively, temporary deviation from homeostatic set points at lower levels may be elicited to maintain higher level set points -such as a temporary change in blood pressure and adrenaline levels to engage fight-flight behaviour, with the goal of reaching safety and maintaining physiological integrity on a longer timescale. On the view of self-modelling in terms of allostatic control described, dimensions of the self at higher-levels constrain the self at lower-levels in that the self-model "actively shapes itself over time to align with those higher level regularities", for example long-term goals can be decomposed into intermediate short term-goals. This section has explored how the self-model arises as a consequence of a system engaged in temporally deep active inference, as prior probabilities over particular policies depend on knowledge about what and where the system finds itself, and what actions are available to it. Through active inference, agents can use their self-model to inform their goal and policy selection in order to arrive at high probability outcomes. This could entail assigning low probability to the self occupying states that are aversive, either physically or socially -with different hierarchical levels of the self-model contributing to different goal states. In this way, the hierarchical self-model determines salience -where "salience is literally defined by whatever has the most (or least) impact on visceral and autonomic homeostasis", at increasingly deep spatiotemporal scales and levels of abstraction.
One of the most striking and philosophically interesting effects of psychedelics is the radical disruptions of self-consciousness they can occasion, including apparently "selfless states". These states, instances of "Drug-Induced Ego-Dissolution" (DIED) are characterised by an experienced loss of self and/or loss of self/world boundary. DIED occurs most reliably under high doses of "classical" psychedelic drugs (5-HT 2A receptor agonists), such as dimethyltryptamine (DMT), lysergic acid diethylamide (LSD), and psilocybin. Ego-dissolution appears to be induced more reliably under psychedelics than meditation, in a dose-dependent manner, and prompted most reliably by high-doses. Recent theoretical work has explored the phenomenological and neurophysiological similarities and differences of ego-dissolution induced by drugs and meditation; see alsoin this special issue).
Recent years have seen a resurgence of interest in the therapeutic potential of psychedelics. Several studies have found preliminary evidence that with administration in controlled circumstances psychedelics can be both safe and therapeutic, with an emphasis on the importance of context in achieving therapeutic outcomes. Interestingly, the positive therapeutic effects seem to scale with "peak" or mystical experience in the psychedelic stateas adopting shallow policies (such as addictive behaviours) may appear adaptive in the short term, rather than risking policies with greater expected free energy due to low precision or uncertainty. It has recently been proposed that psychedelics "relax" high-level priors in the generative model, allowing for the (context-dependent) revision of pathological high-level beliefs. Both psychological insight and peak-experience in the psychedelic state appear to be predictors of long-term positive prognoses.
The REBUS -"RElaxed Beliefs Under pSychedelics" -model of psychedelic function, offers a preliminary but promising model of psychedelic action where psychedelics, through 5-HT 2A agonism, "relax" high-level priors or beliefs. Here, the focus will be on how this mechanism may be cast under the hierarchical predictive processing framework as modulating precision-weighting. To bring this into focus, this section will review how precision-weighting sets a variable Bayesian learning rate in order to highlight certain features relevant to understanding the effects of psychedelics within this framework. Christoph Mathys and colleagues have recently developed a mathematical tool for modelling Bayesian inference modulated by expectations of volatility known as the hierarchical Gaussian filter. The hierarchical Gaussian filter Mathys posits allows a system to optimally balance the influence of prediction errors in changing environments -in other words, to adjust its learning rate.
To recap, the predictive processing framework asserts that the brain instantiates "generative models" of the causes of incoming sensory data, iteratively updating these predictive models in light of incoming "prediction error". This predictive inference is thought to occur across a hierarchy of inferred causes, where high levels track causes and regularities operating over deeper spatial and temporal scales, and lower levels track regularities over shallower spatial and temporal scales. The picture of the living or cognitive system as one which needs to optimise its own learning rate emerges out of the operationalization of Bayesian inference in predictive processing, namely in terms of predictions and precision-weighted prediction errors. According to predictive processing the prediction is given by the prior probability (which itself comes from the previous posterior) and the prediction error is given by the difference between the prediction and the incoming sensory evidence. Prediction error is weighted according to the relative precisions of the prior and the prediction error (where precision is equivalent to the inverse variance of each probability distribution). Intuitively, highly precise prediction error will drag the posterior closer to the distribution of the sensory evidence and further from the prior, and in cases of low precision weighting of the prediction error, the inference relies more on the prior. This determines the learning rate: The more certain we are that the prior hypothesis is correct, the less we should be influenced by the prediction error (the evidence), which means that the learning rate is low. Conversely, the better the precision on the prediction error, the higher the learning rate; that is, the more we trust the quality of the evidence the more we should learn from itIn other words, the lower the learning rate, the greater the influence of top-down modulation from priors; the higher the learning rate, the greater the influence of the sensory evidence on the resulting posterior. Here, precision-weighting is the key mechanism -heavily weighted prediction errors drive a higher learning rate. In order to approximate Bayesian inference over time, it is essential for sensory systems to balance the learning rate appropriately. Over-reliance on priors means the system will fail to learn from experience, whereas over-reliance on sensory evidence (which may be noisy) will lead the system to "overfit". On this picture, Bayesian perceptual inference that minimises prediction error on the appropriate timescale -that is, not overfitting or underfitting the model -needs to have a means of regulating the learning rate. This is implemented by building models of precision, or expected uncertainty, where higher-level priors track longer-term regularities that inform the relative precisions of more basic priors. Optimising the learning rate, and in-so-doing minimising prediction error over time, is a critical challenge the brain faces. This is equivalent to selecting a time frame over which to minimise prediction error. Minimising prediction error over too short a timescale -overfitting -runs the risk of increasing prediction error in the long run. Conversely, failing to accommodate new evidence will lead to underfitting, a failure to update predictions in light of new sensory evidence.
In line with the REBUS model, the relaxing of high-level priors under classical (serotonergic) psychedelicsmeans the system adopts a very high Bayesian learning rate -that is, it is in a highly plastic state, in accordance with research showing an increase in plasticity under psychedelics. This picture casts the perceptual effects of psychedelics -"tripping"as rampant overfitting of the sensory data, resulting from a loss of the usual con-straint exerted by higher-levels on lower-levels of the inferential hierarchy. This "rampant overfitting", resulting from diminished influence from contextualising high-level priors tracking regularities on longer timescales, means the model fits a very short temporal scale, rapidly cycling through candidate models to account for the incoming sensory signal. It is worth highlighting a compatibility of the high Bayesian learning rate approach with other accounts of the mechanism of action of psychedelics in the predictive brain. The REBUS model posits the mechanism of action of psychedelics as reduced precision at high levels rather than increased precision at the sensory peripheries, as psychedelics appear to disrupt functioning via stimulation of 5-HT 2A receptors on deep pyramidal neurons, thought to encode high level priors or beliefs. In contrast, Philip Corlett and colleagues have suggested that psychedelics preserve normal priors and act by increasing sensory noise through enhanced AMPA signalling. On this approach, if the relaxation of high-level priors is indeed an effect of psychedelics, it could be understood to be the result of the fact that "the persistence and strength of the sensory signal suggest that there is something to be explained". Arbitrating between these two mechanistic accounts and disentangling causation -whether the relaxation of high-level priors causes the reduction in sensory gating, or reduction in sensory gating eventually lowers precision at high levels -becomes very difficult here, and it is not clear a simplistic causal account is the right approach. While identifying the mechanisms of action is a key empirical and theoretical project, one potential advantage of the high Bayesian learning rate hypothesis is that it doesn't distinguish between high precision at low levels and low precision at high levels, and as such remains agnostic over the mechanism of action while preserving the useful theoretical features of both accounts that will inform the theoretical account of ego-dissolution that follows.
A high Bayesian learning rate is concordant with the enhanced neural plasticity observed in individuals in a psychedelic state. While an impairment to highlevel cognition is found under psychedelics, in line with the high Bayesian learning rate hypothesis, low-level learning (including extinction learning) and processing appears to be unaffected or enhanced in the psychedelic state. Further evidence for a high Bayesian learning rate under psychedelics is provided by a study looking at the effect of psilocybin on Kanisza trianglesperceptual objects where the brain "fills in" illusory contours using prior expectations -which found reduced filling in and a reduction in the related evoked potentials, concordant with the fact that a high Bayesian learning rate will reduce the effect of sensory history on current perception. In binocular rivalry studies -where different images are presented to each eye simultaneously, and are typically experienced as switching from one percept to the other -reduced switch rates and increased likelihood of the percept being a fusion of the two images has been observed under psilocybin, suggestive of less influence of priors on constraining current perception. Oddball paradigms are also suggestive of a weakened influence of priors on perception under psychedelics. In a sequence of tones, an "oddball" tone (unexpected given prior experience and context) generates a "mismatch negativity", an evoked brain response which has been interpreted in predictive coding terms as prediction error violating the expectations of the sequence. Under LSD, the surprise response to oddball stimuli is blunted, suggestive of a weakened influence of prior expectations. Arguably, there is also phenomenological evidence for the high Bayesian learning rate hypothesis. Perhaps most eloquently articulated by Aldous Huxley: "Visual impressions are greatly intensified and the eye recovers some of the perceptual innocence of childhood, when the sensum was not immediately and automatically subordinated to the concept". This observation lends itself to a straightforward translation into the terms of predictive processing, where "subordinated to the concept" can be understood as "constrained by higher-level priors". More generally, psychedelic phenomenology such as dynamic distortions of spatial dimensions, where things change dramatically in size and shape can be understood as a failure of high-level priors to canalise and constrain lower level predictions.
Given this picture of the action of psychedelics within a predictive processing framework, and the characterisation of self-models in terms of allostatic control, how should states of psychedelic-induced ego dissolution be conceptualised? The proposal here is that a loss of precision on high-level priors results in a flattening of temporal depth of the affordance landscape for the organism -precisely because it is high-level priors tracking longer timescales that structure temporally deep generative models. Recall, under active inference, lower and higher hierarchical levels encode regularities that unfold at faster and slower timescales respectively, such as the expected consequences of action both for proximal and distal goals. Adopting a high Bayesian learning rate is equivalent to changing the time frame over which prediction error is minimised to fit very short timescales. As a result, the deep temporal models that typically guide action and policy selection collapse, and the faster timescales correspond-. Dissolving the self: Active inference, psychedelics, and ego-dissolution. Philosophy and the Mind Sciences, 1(I), 2.©The author(s).ISSN: 2699-0369 ing to lower levels are modelled in a much finer degree of detail (Pink-Hashkes, Rooij, & Kwisthout, 2017). On the account presented in this paper, the self-model is constructed and bolstered in relation to affordances in the environment on several interlocked timescales, where high-levels contextualise and canalise the levels below and allow for motivational orientation to action opportunities pertaining to distal outcomes. Under psychedelics, the relaxation of high-level priors and the corresponding high Bayesian learning rate results in a collapse in the temporal thickness of deep generative models, and a collapse in the temporal depth of the corresponding self-model, which is understood as being is bolstered according to counterfactually rich expectations of the consequences of action on multiple timescales. The collapse in temporal thickness can be understood as occurring due to a failure of sensory attenuation, occurring due to low precision at high-levels and a correspondingly high Bayesian learning rate. Similar stories about aberrant precision at high-levels of the hierarchy corresponding to inferences about affordances and agency have been proposed to underpin hallucinations and delusions in psychosis. Distinguishing between endogenous and exogenous causes -that is, distinguishing between perceptual inputs caused by oneself and those caused by the world -is vital for an agent to be able to effectively move through action space. Corollary discharges -predictions about the sensory consequences of actions -allow the system to do this by withdrawing precision from self-generated movements, and are thought to underpin experienced agency of actions. The failure to predict the consequences of movement due to a failure of sensory attenuation is thought to result in an inability to attribute agency; for instance, a failure of corollary discharge has been thought to cause the attribution of inner speech to an external source in voice-hearing. Importantly, for present purposes, corollary discharge can be understood as a kind of prior, and low-precision priors have been associated with the severity in psychotic symptoms and disturbances of agency in people with schizophrenia. A reduction of precision on highlevel priors in the psychedelic state means that the corollary discharges that would usually cancel out the expected consequences of actions fail to do so, generating an increase in prediction error at lower levels. These unexpected consequences are then attributed to external rather than internal causes, as the more prediction error is generated, the more likely an action (or thought) has exogenous rather than endogenous causes. This echoes similar themes in the autism literature. In autism, the failure of sensory attenuation "leads to the hypervigilant attention to sensory detail at the expense of a hierarchically deep explanation for sensations"leading to what has been termed a "loss of central coherence". Attribution to exogenous rather than en-dogenous causes could result in a loss of "perceptual mineness" -the background feeling that my experiences are "mine" -if, as has been argued, perceptual mineness is underpinned by anticipation of changes in perceptual inputs in relation to movements.Ego-dissolution is not, however, confined to a loss of agentive control over immediate action outcomes, but may be characterised by a more profound dissolution of the sense of being a self or "I" distinct from the outside world. On the view presented in this paper, pre-reflective self-consciousness arises not just through modelling control over the most immediate sensory consequences of actions, but is bolstered by inferences about endogenous control over the distal sensory consequences of allostatic action and action policies. Under a high dose of a psychedelic, the temporary suspension on the gating mechanism on incoming sensory data, described in this paper in terms of a high Bayesian learning rate, render both the proximal and distal sensory consequences of actions highly unpredictable, and the system ceases to have the sense of their being an agent which can (and should) be controlling sensory outcomes. Several authors have emphasised the psychedelic experience is a dynamic process as opposed to a firmly designated state, and different types of ego-dissolution might occur both over the course of the experience and at different dosages. For example, inferences on the boundaries of the body) might be increasingly blurred due to a failure to attenuate the flurry of low-level prediction error. Aspects of the self-model corresponding to longer timescales may break down due to a sustained failure of high levels to attenuate prediction error from low levels due to highly volatile prediction errors, consistent with the fact that bodily ego-dissolution tends to precede dissolution of narrative self. This fact is also perhaps suggestive, in opposition to the high-levels posited by the REBUS model, that ego-dissolution could be seen as the result of the high-levels failing to contextualise the upsurge of prediction error from across the cortex. The fact that the highest level of the self-model are "increasingly abstract, complex and invariant", may explain why higher levels of the self-model are going to be less perturbed by prediction error and perhaps only reliably altered at high dosages. Empirical exploration of these possibilities might be a fruitful avenue for future work, in particular through bridging the neurocomputational mechanisms posited here to both the dynamics of the experience as uncovered through "microphenomenological" interviews, and to the underlying neural correlates of the experience. The account of ego-dissolution in terms of a collapse in the temporal thickness of the affordance landscape presented here should also apply to the concept of a "cognitive affordance" landscape, where the "central function of autonomous activity in the mind wandering network is to create a constant stream of affordances for cognitive agency, a continuing internal competition among possible cognitive actions". Metzinger argues that mental actions -such as the volitional control of endogenous attention, or retrieval of an episodic memory -have epistemic rather that pragmatic goal states. On the allostatic control model of selfhood, the self-model would be constructed and bolstered relative not only to the expectations of the control of the sensory consequences of actions, but also the consequences of mental actions, where the consequences of a mental action might be epistemic and also interoceptive (consider a case where a memory triggers an autonomic response which subsequently acts as the afferent input to an interoceptive inference underpinning a felt emotion). Under psychedelics, loss of control of the expected outcomes of mental actions (as well as a loss of the pragmatic concerns usually driving which epistemic actions to take) might then also be fundamental to the experience of ego-dissolution. This idea is consistent with the fact that under psychedelics mental phenomena "take on the character of objective reality", where the ownership of mental phenomena seems to subside and "the individual may feel like a bystander watching the mental activity of another person". It is worth mentioning a potential implication of this view for consciousness science more broadly. The psychedelic experience and ego-dissolution are often described as an "expansion" of consciousness.argues that not only self-consciousness, but consciousness itself, is underpinned by temporal thickness: "consciousness is nothing more than inference about my future; namely, the selfevidencing consequences of what I could do". States of egodissolution, understood as collapse in the temporal thickness of the generative model, suggest that while temporal thickness very much structures our normal waking experience, it is not clear that temporal thickness ought to be equated with consciousness per se (see also.
The question remains as to why the hypothesised collapse in the temporal thickness of the self-model under psychedelics can be both ecstatic and of enduring therapeutic value. To bring this into focus, it's worth recapitulating core features of the self-model provided earlier. Recall, interoceptive inference on states of the embodied self "attunes" organisms to their affordance landscape, where inferences about the state of the embodied self (e.g. hunger) prescribe certain prediction error minimising policies (e.g. finding food). Inferences pertaining to allostatic consequences on longer timescales may mean higher-level imperatives trump lowerlevel drives, such as choosing to abstain from chocolate cake to stay healthy. In the case of basic bodily needs, as described, these variables are controlledthrough action -active inference is deployed to bring the world into line with predictions, rather than adjusting predictions (via perceptual inference) to conform to the world -for instance eating when hungry. In just the same way that a hungry organism can act so as to harvest confirmatory evidence for the hypothesis "I am sated", hypotheses relating to higher-levels of the self-model geared towards control of outcomes on longer timescales act to constrain action in the present to bring downstream outcomes closer in line with the prior expectation. Overly precise priors driving action on a long timescale which are failing to be fulfilled, on this view, would be a persistent cause of suffering, due to the system consistently failing to meet (or align actions towards) the goal state. Under the model of psychedelic-induced ego-dissolution proposed, the high-precision highlevel priors geared towards control on multiple timescales cease to exert influence on the system due to the proposed lowering of precision of high-level priors under psychedelics. If action ordinarily arises from a process of minimising deviations between the organism's actual (inferred) and desired trajectory, the loss of precision on high-level priors means that, instead of driving action policies, they lose influence on the rest of the system and cease to structure pre-reflective self-consciousness to orient to action opportunities favouring their fulfilment. As these prior beliefs are relaxed, they instead become amenable to perceptual revision from the influx of (highly precise) interoceptive and exteroceptive information. The collapse in temporal depth in the psychedelic state is therefore not experienced as a loss of allostatic control, precisely for the reason that the priors pertaining to longer timescales are no longer asserting an influence on the system and constraining action (and perception) in their usual manner. This picture seems to align well with phenomenological reports of ego-dissolution: "It felt as if 'I' did no longer exist. There was purely my sensory perception of my environment, but sensory input was not translated into needs, feelings, or acting by 'me' " (unpublished online survey data quoted in. Peak experiences under psychedelics, then, could be understood as absence of prediction errors relating to allostasis due to a flattening of the temporal depth of the affordance landscape, resulting in the feeling of "oceanic boundlessness" -a sense of immense well-being and peace. Here, the "itinerant strategies" to stay within our "species-specific window of viability", are no longer necessary as the "first prior" -the expectation or imperative for existence -is being met without conditions. Following the TIBER model, many psychopathologies may be due to high precision on high-level priors. Peak psychedelic experience may act as a "reset" allowing for revision of entrenched high-level beliefs that structure pre-reflective self-consciousness (and, accordingly, the affordance landscape) -opening up new domains of salience and possibility for meaningful engagement with the world, through revised and retuned self-models. Increased bottom-up information flow (particularly from the limbic system), through a high Bayesian learning rate, may make entrenched high-level priors amenable to revision via perceptual inference rather than driving control via active inference. This lays the theoretical groundwork for why psychedelics may effectively treat depression: if depression is underpinned by a high precision prior of low allostatic self-efficacy, it follows that relaxation and revision of this prior should alleviate depressive symptoms. Finally, (and speculatively), if the account of "retuning" of self-models under psychedelics presented here generalises to the bodily self (which the experiential changes in bodily selfhood would suggest) this account is suggestive of a potential role for psychedelics in the treatment of chronic pain, and for phantom limb pain -for which there has already been promising results. The primary focus so far has been on "peak" experiences, due to the growing number of papers indicating they are central to positive long-term therapeutic outcomes. However, while generally psychedelics are thought to be very low risk, and there is evidence to suggest they are protective against mental health problems (Hendricks, Thorne,, acute and occasionally persistent adverse psychological reactions do sometimes occur. While "complete" ego-dissolution is described as a "state of complete surrender, associated bliss, and union with all things", "incomplete" ego-dissolutiondue to psychological resistance or an insufficient dose -can be characterised by intense fear, anxiety, or distress. On the account presented in this paper, this can be understood as resulting from psychological resistance, where psychological resistance here may be conceptualised as a high-precision prior on being able to control the experience, that is maintained though fear-driven endogenous attention. Failure to control the experience, in violating the highly precise prior for the goal state of control, is then experienced as a loss of allostatic control, bringing with it feelings of intense fear or distress. In therapeutic contexts, encouraging users to "let go" and "surrender" to the experience, could be understood in these terms, as discouraging the user from putting high (endogenous) precision on a prior for control that could result in adverse experiences when unfulfilled. These considerations highlight the essential importance of context in achieving therapeutic outcomes.
Psychedelics are known for their ability to profoundly alter consciousness and occasion so-called "mystical" experiences. The renaissance in psychedelic research in the past decade is beginning to shed light on the mechanisms underpinning the extraordinary states of consciousness induced by psychedelics. Within psychedelic phe-nomenology, experiences of ego-dissolution are of particular phenomenological, philosophical and therapeutic interest. This paper has given a preliminary account of how ego-dissolution under psychedelics can be understood in terms of predictive processing and active inference. The hypothesis here is that the action of psychedelics within the predictive processing framework is best understood as a "relaxation of high-level beliefs", and this can be unpacked in terms of a high Bayesian learning rate. Psychedelic-induced ego-dissolution, then, results in a collapse in temporal thicknessof the self-model as conceived within an active inference framework. The therapeutic effects of ego-dissolution, then, can be understood in terms of the relaxing and retuning of entrenched self-models, or a "resetting" or "opening" of the affordance landscape, allowing for the possibility of new modes of engagement with the world, oneself, and other people.
Create a free account to open full-text PDFs.
Carhart-Harris, R. L., Roseman, L., Haijen, E. C. H. M. et al. · Journal of Psychopharmacology (2018)
Carter, O. L., Hasler, F. ;., Pettigrew, J. D. et al. · Psychopharmacology (2007)
Griffiths, R. R., Johnson, M. W. · Journal of Psychopharmacology (2016)
Hendricks, P. S., Thorne, C. B., Clark, B. et al. · Journal of Psychopharmacology (2015)
Kometer, M., Cahn, B. R., Andel, D. et al. · Biological Psychiatry (2011)
Lebedev, A. V., L€ Ovd En, M., Rosenthal, G. et al. · Human Brain Mapping (2015)
Letheby, C., Gerrans, P. · Neuroscience of Consciousness (2017)
Ly, C., Greb, A. C., Cameron, L. P. et al. · Cell Reports (2018)
Lyons, T., Carhart-Harris, R. L. · Frontiers in Psychology (2018)
Millière, R. · Frontiers in Human Neuroscience (2017)
Milliere, R., Carhart-Harris, R. L., Roseman, L. et al. · Frontiers in Psychology (2018)
Moreno, F. A., Wiegand, C. B., Taitano, E. K. et al. · Journal of Clinical Psychiatry (2006)
Nour, M. R., Carhart-Harris, R. L. · British Journal of Psychiatry (2017)
Nour, M. R., Evans, J., Nutt, D. J. et al. · Frontiers in Human Neuroscience (2016)
Palhano-Fontes, F., Barreto, D., Onias, H. et al. · Psychological Medicine (2018)
Pink-Hashkes, S., Van Rooij, I., Kwisthout, J. · CogSci (2017)
Preller, K. H., Vollenweider, F. X. · Behavioral Neurobiology of Psychedelic Drugs (2016)
Ramachandran, V., Chunharas, C., Marcus, Z. et al. · Neurocase (2018)
Roseman, L., Nutt, D. J., Carhart-Harris, R. L. · Frontiers in Pharmacology (2018)
Scott, G., Carhart-Harris, R. L. · Neuroscience of Consciousness (2019)
Strassman, R. J. · Journal of Nervous and Mental Disease (1984)
Timmermann, C., Spriggs, M. J., Kaelen, M. et al. · Neuropharmacology (2018)
Papers in Blossom that reference this study
McAlpine, R., Kamboj, S. K. · Scientific Reports (2024)
Brasher, T., Rosen, D., Spinella, M. · International Journal of Wellbeing (2023)