Symposium on Letheby and Gerrans, “Self unbound: ego dissolution in psychedelic experience”

I am delighted to announce the next symposium in our series on articles from Neuroscience of Consciousness.  Neuroscience of Consciousness is an interdisciplinary journal focused on the philosophy and science of consciousness, and gladly accepts submissions from both philosophers and scientists working in this fascinating field.

We have two types of symposia.  For primarily theoretical articles, we will have several commentators from a variety of theoretical perspectives.  For novel empirical research, we will have single commentators whose goal is to bring out the theoretical challenges and import of the results.  This symposium is based on Chris Letheby and Philip Gerran’s fascinating paper, “Self unbound: ego dissolution in psychedelic experience.” We have excellent commentaries from John Michael, Inês Hipólito, and Raphaël Millière. These are followed by a response from Letheby and Gerrans.


Letheby and Gerrans’ paper, “Self unbound: ego dissolution in psychedelic experience,” combines three distinct research topics: the longstanding conundrum of the nature of the self, the predictive processing framework of perception and cognition, and the puzzling phenomenon of ego-dissolution during psychedelic experience.  According to the authors, the predictive processing approach supports fictionalism about the self – the self is fundamentally an explanatory posit used by the brain to explain the unified aspects of autobiographical experience.  But there is no unitary entity that this posit successfully refers to.  Hence the self is a fiction.  Letheby and Gerrans use this account to further explain ego-dissolution, in which users of psychedelics experience a loss of their sense-of-self.

Binding of features is a ubiquitous phenomenon in perception and cognition, and it is fundamentally abductive.  According to the predictive coding account, object-to-feature binding in perception is achieved through a top-down process of prediction and error correction.  The result is a confirmed or disconfirmed prediction about which features are bound to which objects.  The “self-model,” according to Letheby and Gerrans, works similarly.  Here, what is bound together is information that is “salient” for the organism – the salience system tracks information as it is vital for attaining an organism’s goals.  A self-model is posited to explain a variety of “self-binding” effects, in which self-related information is processed more quickly than non-self-related information.  Letheby and Gerrans further this perspective by suggesting that the self-model is a hierarchical predictive coding structure, which predicts the salience of information for the organism over time, and includes its history and goals at multiple scales. 

The notion of the self that results is more robust than “narrative” views of self-representation, because it posits an indivisible, persistently identical (“Cartesian”) self which unifies autobiographical information in the self-model.  But since there is no such object – the self is simply a posit made by the self-model – there is in fact no self.  Thus, Letheby and Gerrans disagree with other predictive-coding based views of the self, such as the one defended by Jakob Hohwy and John Michael, on which the causal influences exerted by the self-representation in cognition substantiate the notion of a “real” self.

Turning to ego-dissolution, Letheby and Gerrans argue that the phenomenon is due to psychedelic interruption of the self-model.  On this view, the salience system still signals the salience of environmental information, but no longer as it relates to the organism.  This explains the character of ego-dissolution, on which objects are seen as having profound significance, but in a way that is severed from their importance to the subject’s aims.  The dissolution of the self-model can come in degrees, and is not interrupted entirely during psychedelic experience.  Instead, the “coherence” of the self-model and the salience system is interrupted.  Hence, the authors claim, the predictive coding approach has the resources both to answer philosophical questions about the self and account for fascinating empirical phenomena surrounding psychedelic experience.


Thanks very much to our contributors for participating.  Thanks also to Jakob Hohwy and the other editors of Neuroscience of Consciousness, and to Oxford University Press.  Please feel free to comment in the discussion board below!


John Michael:  Commentary on Letheby and Gerrans (2017)

John Michael — University of Warwick;

Letheby & Gerrans (2017, henceforth L&G) offer a cogent and compelling instrumentalist account of the self – i.e, as a useful but ultimately false representation of a ‘simple and enduring substance to which attributes are bound which serves to integrate and unify cognitive processing across levels and domains’ (p. 2). They use this account to synthesize and interpret recent empirical findings bearing upon self-awareness and in particular ego-dissolution under the influence of psychedelics. In the following brief commentary, I will identify a few questions and indicate a few possible directions in which L&B’s ideas might fruitfully be developed further.

  1. To begin, I would like to raise a few questions pertaining to the differences between L&G’s account and the account offered by Hohwy & Michael (2017). As L&B note, Hohwy and Michael (2017) argue that the self-model qualifies as a self because it performs many of the functions attributed to the self, such as being a ‘hidden cause’ of thought, emotion, and behaviour.’ L&B agree that the self-model performs these functions but resist the suggestion to think of it as referring to the self.  As they put it: ‘…it does not follow from the existence of a robust, causally efficacious, and multi-layered self-model that the entity represented by this model exists’ (p.8). Instead, they maintain, the self-model is ‘more like an existential placebo than a successfully self-referential model’ (p.8). This is because the self-model does not satisfy a set of criteria which L&B have identified as central to the concept of the self (at least to the self-model’s representation of the self), namely ‘indivisibility, substantiality, and strict identity over time, as well as distinctness from and ownership/authorship of experiences, thoughts, and feelings’ (p.2). One question arises at this stage: why identify these particular criteria as central to the self? In other words, what are the reasons for thinking that the self-model postulates a Cartesian self rather than some other kind of self? Wouldn’t, for example, a Humean bundle theory fit well with the binding account offered here? On such a view, the self could be identified with a patterned bundle of affective experiences, thoughts and memories rather than with any substantive entity that has those experiences, etc.
  2. In fact, some of the work on self-prioritization by Sui & Humphreys (2015) and others (Woźniak, Kourtis, & Knoblich, 2018) – which L&G draw upon – seems to indicate that what the self-model treats as related to the self has a dynamic and gradual rather than a clear and stable boundary. For instance, information related to friends and family members gets prioritized in much the same way as information related to the self, and even arbitrary shapes can gain access to prioritized processing if they are associated with the self (or more specifically with the word ‘You’) or with friends (i.e. with words referring to friends). If the self-model postulates a stable and clearly demarcated self and prioritizes information related to that postulated entity, why should we expect this graded prioritization of information according to how closely it is bound up with one’s own experiences, memories, body, name, etc.?
  3. If the self-model does postulate the self as a particular entity which is the bearer or substrate of experiences, thoughts, memories, etc., does it necessarily follow from this that the self-model postulates a single, continuous self? In other words, are their compelling reasons to think that the self-model attributes autobiographical memories from childhood and current sensations to the same self? To be honest, I’m not sure how we would know whether this is the case. As a starting point in thinking about this issue, however, one consideration that may speak against the idea that the self-model posits a continuous self is that we tend to devalue or neglect past or future emotions compared to present ones – i.e. we are inaccurate at recalling the affective character of experiences that lie in the past (Levine 1997) and at forecasting our emotions in the future (Wilson and Gilbert, 2005), and we discount the value rewards that lie in the future Benhabib, Bisin, & Schotter (2010).
  4. If one thinks of reference in at least partially causal-realist terms, as Hohwy & Michael suggest, then the self-model may be taken to refer to a nexus of hidden causes of thought, emotions and behavior by virtue of having causally interacted with those hidden causes in such a way as to shape them over development to match the model. On such a view, the self-model could successfully refer to the self even if it turned out to be erroneous in very central respects (for example, as L&G persuasively argue, the self-model may refer to endogenous causes of behavior etc. but not successfully refer to any enduring, indivisible particular). The causal-realist view may have the drawback of making it too easy for the self-model to successfully refer (is it possible on such a view for the self-model to fail to refer?), but L&G may have the opposite problem: Is it not possible on their view to be wrong about the properties of the entity one is experiencing/referring to but still to be experiencing/referring to the entity?
  5. It strikes me that L&G’s view may be well-placed to account for certain pathologies of the self, for example cases in which individuals have the experience of shifting between different personalities or selves (e.g. dissociative disorder or borderline personality disorder; Foote et al., 2006). If the self-model represents the self as a stable and continuous entity, then we might indeed expect that when very different emotions or thoughts are experienced at different times, the self-model would attribute them to distinct selves. It would be interesting to know whether L&B have given much thought to such cases and if so, what their account may enable them to say to illuminate them.


Benhabib, J., Bisin, A., & Schotter, A. (2010). Present-bias, quasi-hyperbolic discounting, and fixed costs. Games and Economic Behavior69(2), 205-223.

Foote, B., Smolin, Y., Kaplan, M., Legatt, M., Lipschitz, D., Sar, V., & Dogan, O. (2006). Axis I dissociative disorder comorbidity in borderline personality disorder and reports of childhood trauma. Journal of Clinical Psychiatry67, 1583-1590.

Hohwy, J., & Michael, J. (2017). 16 Why Should Any Body Have a Self?. The Subject’s Matter: Self-Consciousness and the Body, 363.

Letheby, C., & Gerrans, P. (2017). Self unbound: ego dissolution in psychedelic experience. Neuroscience of Consciousness2017(1), nix016.

Levine, L. J. (1997). Reconstructing memory for emotions. Journal of Experimental Psychology: General126(2), 165.

Sui, J., & Humphreys, G. W. (2015). The integrative self: How self-reference integrates perception and memory. Trends in Cognitive Sciences19(12), 719-728.

Wilson, T. D., & Gilbert, D. T. (2005). Affective forecasting: Knowing what to want. Current Directions in Psychological Science14(3), 131-134. Woźniak, M., Kourtis, D., & Knoblich, G. (2018). Prioritization of arbitrary faces associated to self: An EEG study. PloS one13(1), e0190679.

Inês Hipólito: Why Would Any Predictive System Need a Self?

Inês Hipólito, University of Wollongong;

Whether something like a self has a part to play in the hierarchical predictive processes, and if so, how to conceive of it – as a representational entity, or phenomenal – is debatable. Defenders of the representational self contend that the self is a causally efficacious entity underlying the hierarchy of integrative processes. That is, there is a causal power exerted by the object(s) of self-representation, such as beliefs or bias (Hohwy and Michael, 2017). Against this ‘causal realist interpretation of the reference of the self’, are those like Metzinger (2004) arguing for a conception of the self as a ‘phenomenal avatar’.

In their stimulating target article, Letheby and Gerrans (henceforth L&G) agree with Metzinger’s view of the self as a ‘phenomenal avatar’. L&G reject that anything, including the self-representation, possesses the right attributes to qualify as a self. Here, I will call their proposal the ‘phenomenal self’ to conceptually contrast it with Hohwy and Michael’s (2017) representational self. Specifically, L&G make the case that the self, although playing a role in the hierarchical predictive processes, does not exist[1]. Instead, they claim, the self is a fundamental cognitive strategy. It is important to note that they side with Hohwy and Michael (2017) in that the self plays an integrative role in predictive coding; but where they disagree is that the the right attributes to qualify as a self, L&G argue, are not instantiated by anything actual.

According to L&G, then, although the “binding of representations of stimuli is incorporated into the self-model, and occurs via top-down predictions generated by that model” (p. 8); the self-model is Cartesian, i.e., a “substance with the ontological claim that no such substance exists” (p. 9). The self, they claim, “does not entail the existence of an object to which attributes are bound – though it does require the representation of an object, to which representations of attributes are bound” (p. 2). While the self does not exist, positing a fictional self helps them in explaining the integration of exogenous causes and how they are bound to a pre-existing representation. So, they argue that the self plays a fundamental role in the binding and integration of information (i.e., organising the different levels and domains of processing), into pre-existing representations, but is not causally efficacious. 

It would surely be exciting to explore how, on L&G’s account, the phenomenal self is instantiated or otherwise underpinned by the hierarchical processing. Is the self postulated also in explaining the construction of representations? Or is it postulated only to orchestrate the binding of contents to and within pre-existing representations? This would be important to know because it could perhaps shed some light on the question of how these representations are constructed in first place. The task of answering said question is a well-known challenge[2] to the cognitivist reading of predictive processing[3]. The interesting question is, how are representations, to which the self is supposed to bound contents, formed in first place? Before assuming that representations are conveniently available to the phenomenal self, it would be helpful to clarify the nature and source of said representations.

In clarifying this, it might also be useful to know explicitly what theoretically vindicates a non-causally efficacious phenomenal self in a probabilistic system. Supposedly, Hohwy and Michael’s (2017) representational self is justified because there needs to be a self to embody the beliefs and bias that determine inferences about the world. As to the phenomenal self, on the contrary, it is not clear whether it is also justified by the system’s need of preservation, with the exception that, in this case, the self is ‘made up’ by the hierarchical system?

It seems this is what L&G have in mind when they say that, “the self is postulated by higher-level processes as an entity to facilitate the binding or integration of information” (p. 1). The question then is how a postulate ‘knows’ what and how to correctly bind contents to pre-existing representations. Do the correctness conditions come from the self embodying bias and beliefs that ‘govern’ the binding, or from the sensory data itself? If the latter, then the sensory data cannot be as ambiguous as the cognitivist reading of predictive processing posits. If the former, then L&G need to help out Hohwy and Michael (2017) explain the authority of the self in the interpretation of sensory data without it being a homunculus. If sense data are the ‘objects of perception’ then how can you not have a homunculus posit? That’s a thought worth considering. Hohwy (2007) claims that “perception is indirect […] what we perceive is the brain’s best hypothesis, as embodied in a high-level generative model, about the causes in the outer world.” (Hohwy 2007, 322). But, if the objects of perception are our brain’s best hypothesis of the world then this seems to scream out for a homunculus in the exact same sense that the sense datum theory did.[4]

Surely not every system that generates hypotheses must also implicitly posit a ‘person in the head’. But if perceiving is the interpretation of data analogous to what a scientist does, then, since both Hohwy (2013[5]) and Gerrans (2014[6]) endorse the brain as a scientist analogy (along with a self that does the cognitive work), they both need to deal with the old homunculus. The analogy is a controversial feature[7]in predictive processing precisely because however it is endorsed might lead to old homuncular problems for two reasons. First, the brain, like a scientist in the interpretation of the data of her experiments, is supposed to have some kind of higher-level authority in the determination of what is the case. If in addition to endorsing the brain as a scientist, it is also claimed that there is a self, representational or phenomenal, then one would have to place the self as the scientist in the head governing the cognitive experiments and drawing conclusions[8]. The second problem is the supposition that mental representations conveniently exist in advance, that is, the analogy with what is already known by science. The analogy allows to simply assume that mental representations and representational content, like all the information already known in science, are conveniently available, ready to be used and eventually updated. So, instead of explaining how representations are constructed by the hierarchical processing, the analogy perpetuates the assumption that since there is information already available, the scientist in the head can posit contenful hypothesis and make sense of the sense data results.

Things can however look promising to overcome this, if it is possible to philosophically explain how the non-causally efficacious phenomenal self gets a role in the construction and update of mental representations without it being a higher-level authority. If so, then L&G’s phenomenal self might be preferable to Hohwy and Michael’s (2017) representational self.

There is yet a further worry L&G have to deal with. That is, if representations are scientific posits, generated by hierarchical Bayesian inference, and phenomenology needs to be explained, then they need to bridge a difficult gap. That of how to establish an intelligible link between the purely information-processing notion of the hierarchy and the phenomenological reading of the self. This is a contemporary emergence of the well-known explanatory gap between the phenomenal and the descriptions of neural activity.

In conclusion, L&G’s paper stimulates questions that motivate new ways of thinking about controversial features of the predictive processing framework, such as the role of the self in the hierarchical predictive processing. Their view is challenged by still open questions. Why would any predictive system need a self? Could the self be higher-level without being causally efficacious? Could the self be a postulate and at the same time have influence over the binding of contents within the representations? If the self determines the binding of contentful representations, could mental representations be constructed without the self? How to establish the intelligible link between informational processings and the phenomenal self?


Thanks to Ian Robertson and Robert Clowes for their insights in the discussion of these ideas.


Anderson, M. L. (2017). Of Bayes and bullets: An embodied, situated, targeting-based account of predictive processing. Johannes Gutenberg-Universität Mainz.

Clark, A. (2016). Surfing uncertainty: Prediction, action, and the embodied mind. Oxford University Press.

Clowes, R. W. (2015). The Reality of the Virtual Self as Interface to the Social World. In J. Fonseca & J. Gonçalves (Eds.), Philosophical Perspectives on Self (pp. 221-276). Lisbon: Peter Lang.

Colombo, M., Elkin, L., & Hartmann, S. (2018). Being Realist about Bayes, and the Predictive Processing Theory of Mind. British Journal of Philosophy of Science. axy059,

Drayson, Z. (2018). Direct perception and the predictive mind. Philosophical Studies, 175(12), 3145-3164.

Gerrans, P. (2014). The measure of madness. MIT press.

Gładziejewski, P. (2016). Predictive coding and representationalism. Synthese, 559–582.

Gregory, R. L. (1980). Perceptions as hypotheses. Philosophical Transactions of the Royal Society B: Biological Sciences 290(1038): 181–97.

Helmholtz, H. (1860/1962). Handbuch der physiologischen Optik (Southall, J. P. C. (Ed.), English trans.), Vol. 3. New York: Dover.

Hipolito, I. (forthcoming/2019). Perception is not always and everywhere inferential. Australasian Philosophical Review.

Hohwy, J. (2007). Functional integration and the mind. Synthese, 159(3), 315–328.

Hohwy J. (2013). The predictive mind. Oxford: Oxford University Press.

Hohwy, J., & Michael, J. (2017). Why Should Any Body Have a Self?. The Subject’s Matter: Self-Consciousness and the Body, 363.

Hutto, D. D. (2018). Getting into predictive processing’s great guessing game: Bootstrap heaven or hell?. Synthese, 1-14.

Kiefer, A. B. (2017). Literal perceptual inference. Johannes Gutenberg-Universität Mainz.

Kiefer, A., & Hohwy, J. (2018). Content and misrepresentation in hierarchical generative models. Synthese, 195(6), 2387-2415.

Kirchhoff, M. D., & Robertson, I. (2018). Enactivism and predictive processing: a non-representational view. Philosophical Explorations, 21(2), 264-281.

Letheby, C., & Gerrans, P. (2017). Self unbound: ego dissolution in psychedelic experience. Neuroscience of Consciousness, 2017(1), nix016.

Metzinger, T. (2004). Being no one: The self-model theory of subjectivity. MIT Press.

Rao, R. P., & Ballard, D. H. (1999). Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects. Nature Neuroscience, 2(1), 79.

Orlandi, N. (2016). Bayesian perception is ecological perception. Philosophical Topics, 44(2), 327-351. Wiese, W., & Metzinger, T. (2017). Vanilla PP for philosophers: A primer on predictive processing. In T. Metzinger & W. Wiese (Eds.). Philosophy and Predictive Processing: 1. Frankfurt am Main: MIND Group. doi: 10.15502/9783958573024.

[1] See how Clowes (2015) asks in what sense (virtual) selves that do modelling work are real.

[2] See Hutto 2018; Orlandi 2016.

[3] There is debate about whether predictive processing posits representations, and if so how best to describe them. The debate resonates around the question of how to interpret content, either as internal (Keifer and Hohwy 2018; Gladziejewski 2016; Hohwy 2013; Helmholtz 1925; Gregory 1980; Rao and Ballard 1999), external (Clark 2016); or enactive (see SI “Radical views on cognition”, forthcoming in Synthese; SI “Predictive Brains and Embodied, Enactive Cognition” by Synthese, 2018; see also Kirchhoff and Robertson 2018; Hutto 2018; Anderson 2017). The cognitivist account, however, is that “it is useful to describe the estimates posited by PP [predictive processing] as representations” (Wiese and Metzinger 2017, p. 1)

[4] See Drayson 2018.

[5] “Scientific hypothesis testing is paradigmatically a matter of experimentation, that is, active intervention by the scientist in causal chains in order to reveal causal relations. If perception is like hypothesis testing, we should expect a similar notion of intervention in perception” (2013, p. 77, my emphasis)

[6] “A scientist explaining some discrepant evidence is doing the same thing as the oculomotor system controlling the trajectory of a limb” (2014, pp. 46-47, my emphasis).

[7] See Kiefer (2017) on realism about Bayesian inference. See how Colombo et al. (2018) dispute this by claiming agnosticism as to whether Bayesian models are true.

[8] I make this point in Hipolito forthcoming/2019.

Raphaël Millière: From Self-Binding to Self-Consciousness

Raphaël Millière — University of Oxford;

In their rich and thought-provoking article, “Self unbound: ego dissolution in psychedelic experience”, Chris Letheby and Philip Gerrans articulate two main claims: a psychosemantic claim, according to which the sense of self that accompanies ordinary conscious experience is the product of a representation of the self as of an enduring substance; and an error theory of selfhood, according to which there is no actual entity corresponding to this representation of the self as of an enduring substance. They argue that these two claims are partially supported by two recent developments in psychology and neuroscience: first, new evidence suggesting that there is a bias for self-relevant information in cognitive binding; and second, new data on the phenomenology and neurophysiology of drug-induced ego dissolution, a complex effect of psychedelic drugs typically described as a breakdown of the sense of self. According to Letheby and Gerrans, these results suggest that the sense of self pervading ordinary experience is underlain by “self-binding” processes, namely hierarchical processes binding self-relevant information to generate the sense of an enduring self, and that these processes can break down during ego dissolution, revealing that there is no such enduring self in reality.

This article has the notable merit of shedding light on new evidence from recent studies of drug-induced states that had not previously attracted much attention in philosophy of mind, thus paving the way for future philosophical discussions. Moreover, I am very sympathetic with many of the authors’ claims, which fit nicely with the account of drug-induced ego dissolution I have put forward in recent works (Millière, 2017; Millière et al., 2018). Nonetheless, the article raises interesting questions that I would like them to address.

First, I have a lingering concern regarding the meaning of what they call “self-binding”, a concept borrowed from recent psychological work by Sui and Humphreys (2015). Indeed, one might worry that the article conflates two distinct notions within this construct, namely (a) “the preferential enhancement of cognitive binding for self-relevant information” (p. 2), and (b) “the binding of representations of stimuli which are incorporated into [a] self-model” (p. 8). The first notion of self-binding corresponds to Sui and Humphreys’ work on the “self-reference effect”, which refers to increased performance in the professing of self-related stimuli (such as words and faces) compared to other-related stimuli. Indeed, a number of recent studies have found that self-reference enhances performance in the recollection of adjectives (Leshikar, Dulas, & Duarte, 2015) and faces (Cunningham, Brebner, Quinn, & Turk, 2014), in the binding of visual features in the perception of faces (Keyes & Brady, 2010; Keyes, 2012), and in the ability to switch from a prior association with shapes to new associations (Wang, Humphreys, & Sui, 2016). As Sui and Humphreys note, these effects might be partially explained by the familiarity of the stimulus, its inherent reward value and/or its emotional valence. Nonetheless, they emphasize that these factors do not completely account for their results, and that there must be something special about the treatment of self-relevant information. As Sui puts it in a follow-up paper, such evidence suggests that “self-reference operates as a perceptual glue to bind external stimuli together when facing a complex environment.” (Sui, 2016, p. 482).

In their article, Letheby and Gerrans often talk about self-binding as the set of cognitive processes that represent the self as the hidden cause of self-related stimuli. They speculate that such processes are needed “to parse experiences into internally and externally caused” (p. 8). However, it is unclear that the notion of self-binding introduced by Sui and Humphreys really pertains to the distinction between endogenous and exogenous stimuli. Rather, it pertains to the detection of self-referential information within exogenous stimuli, such as words and faces. Furthermore, the kind of binding process that Sui and Humphreys’ notion of self-binding refers to is presumably the binding of exogenous stimuli into perceptual wholes experienced as external objects, rather than the binding of endogenous stimuli into a representation of the self as their hidden enduring cause. Indeed, there is a difference between experiencing one’s body as one’s own, for example, and perceiving one’s own face on a computer screen. Both experiences may involve a form of self-representation, but the latter is not a plausible candidate for the basis of the sense of self that allegedly pervades ordinary experience. When talking about self-binding, Letheby and Gerrans claim that “self-awareness is the experience of cognitive processes in which these binding processes are intact” (p. 2). This claim sounds rather implausible if the relevant kind of process is the perceptual binding of self-relevant external stimuli investigated by Sui and Humphreys; at the very least, there seems to be more to the notion of self-awareness than the awareness that some external stimuli such as words and faces are related to oneself. In fact, Letheby and Gerrans agree that the notion of self-awareness they are interested in includes both de se thought and bodily self-awareness, neither of which are directly related to Sui & Humphreys’ notion of self-binding. This ambiguity in the notion of self-binding becomes apparent for example when the authors write that “the salience system is constantly creating the illusion of substantial selfhood by binding information into a representation not of the world in itself, but of the world as it matters to the organism” (p. 3). Self-awareness is not simply the awareness of what matters to me in the world: it is the awareness of myself as myself (or of something as myself, if one wishes to remain neutral on the metaphysics of selfhood). A slightly different way to put this point is to say that self-awareness is not just a matter of self-related processing, but also crucially involves self-specific processing, which pertains to the representation of the “self” as the subject of thoughts and actions (Legrand & Ruby, 2009). One might worry that Letheby and Gerrans’ notion of self-binding equivocates between self-relatedness and self-specificity.

Another related concern regards the relationship between self-binding processes and consciousness. Sui and Humphreys are cautiously silent on this topic, as it remains unclear to what extent the “self-reference effect” has downstream consequences on self-consciousness. One might be quicker at binding together perceptual features into meaningful wholes when perceiving self-relevant stimuli, even if this process does not necessarily underlie a conscious awareness of the stimuli as relevant to oneself. In other words, self-binding need not involve conscious self-representation. This remark also calls into question the claim that “self-awareness is the experience of cognitive processes in which these binding processes are intact” (p. 2). Indeed, one may be quicker at binding together visual features when one perceives what happens to be a picture of oneself, without being conscious that what one is looking at is in fact one’s own face. To be fair, experimental results discussed by Sui and Humphreys crucially rely on the participants’ ability to report whether what they are seeing is self-related or other-related. Nonetheless, one may wonder whether the enhancement of binding for self-relevant stimuli does occur even when the self-relevance of stimuli is non-consciously processed. Sui and Humphreys argue that “empirical measures of self-biases that can be sued [sic] as a proxy for self-representation” (Sui & Humphreys, 2017, p. 1); whether such self-biases only occur for conscious self-representation remains an open question.

A final question regards Letheby and Gerrans’ claim that the kind of self-awareness that allegedly pervades ordinary experience is an awareness as of an enduring Cartesian self. They argue that “our normal experience of unity compels the inference not just that we are a self, but that the self is a Cartesian substance” (p. 2); indeed, “the content of the self-model is… of a substance or an entity which has the properties and experiences” (p. 9). This claim fits with Letheby and Gerrans’ account of self-binding in a predictive processing framework, consistent with Michael and Hohwy (2017), according to which the brain generates a hierarchical predictive model of an enduring self as the hidden cause of endogenous stimuli. On this account, self-representation results from a form of inference that minimizes prediction errors generated by endogenous stimuli by modelling the self as their hidden cause. There is a latent ambiguity, however, in the claim that “our normal experience of unity compels the inference… that the self is a Cartesian substance”. Letheby and Gerrans suggest that such inference underlies an experience of the self as of a Cartesian substance. According to an alternative proposal, however, conscious or non-conscious processes might ground a belief that one is an enduring Cartesian substance. There need not be anything in experience that actually corresponds to such a belief – any conscious representation of the self as of an enduring substance. Indeed, an alternative account of self-awareness could emphasize that there are different kinds of conscious self-representation (e.g. thinking about oneself, experiencing one’s body as one’s own, or experiencing one’s location in egocentric space as one’s own), none of which necessarily represents the self as an enduring Cartesian substance, even if they can ultimately ground the belief that one is such a substance (Forstmann & Burgmer, 2017). In other words, rather than claiming that there is an actual “illusion of substantial selfhood” (p. 3) in experience, one could simply argue that self-awareness grounds the “Cartesian intuition” (p. 9) shared by some populations. There is much to be said on this topic; but it is worth noting that there is empirical evidence suggesting that belief in the existence of the self as an enduring substance independent from the body is variable across cultures, calling into question the very idea that Human beings are intuitive Cartesians (Hodge, 2008; Watson‐Jones, Busch, Harris, & Legare, 2016; Becker et al., 2018).


Becker, M., Vignoles, V. L., Owe, E., Easterbrook, M. J., Brown, R., Smith, P. B., … Lay, S. (2018). Being oneself through time: Bases of self-continuity across 55 cultures. Self and Identity, 17(3), 276–293.

Cunningham, S. J., Brebner, J. L., Quinn, F., & Turk, D. J. (2014). The Self-Reference Effect on Memory in Early Childhood. Child Development, 85(2), 808–823.

Forstmann, M., & Burgmer, P. (2017). Antecedents, Manifestations, and Consequences of Belief in Mind–Body Dualism. In C. M. Zedelius, B. C. N. Müller, & J. W. Schooler (Eds.), The Science of Lay Theories: How Beliefs Shape Our Cognition, Behavior, and Health (pp. 181–205). Cham: Springer International Publishing.

Hodge, K. M. (2008). Descartes’ Mistake: How Afterlife Beliefs Challenge the Assumption that Humans are Intuitive Cartesian Substance Dualists. Journal of Cognition and Culture, 8(3), 387–415.

Hohwy, J., & Michael, J. (2017). Why should any body have a self? In A. Alsmith & F. Vignemont (Eds.), The Subject’s Matter: Self-Consciousness and the Body (pp. 363–391). Cambridge, MA: MIT Press.

Keyes, H. (2012). Categorical perception effects for facial identity in robustly represented familiar and self-faces: The role of configural and featural information. Quarterly Journal of Experimental Psychology, 65(4), 760–772.

Keyes, H., & Brady, N. (2010). Self-face recognition is characterized by “bilateral gain” and by faster, more accurate performance which persists when faces are inverted. Quarterly Journal of Experimental Psychology, 63(5), 840–847.

Legrand, D., & Ruby, P. (2009). What is self-specific? Theoretical investigation and critical review of neuroimaging results. Psychological Review, 116(1), 252–282.

Leshikar, E. D., Dulas, M. R., & Duarte, A. (2015). Self-referencing enhances recollection in both young and older adults. Aging, Neuropsychology, and Cognition, 22(4), 388–412.

Millière, R. (2017). Looking For The Self: Phenomenology, Neurophysiology and Philosophical Significance of Drug-induced Ego Dissolution. Frontiers in Human Neuroscience, 11.

Millière, R., Carhart-Harris, R. L., Roseman, L., Trautwein, F.-M., & Berkovich-Ohana, A. (2018). Psychedelics, Meditation and Self-Consciousness. Frontiers in Psychology, 9.

Sui, J. (2016). Self-Reference Acts as a Golden Thread in Binding. Trends in Cognitive Sciences, 20(7), 482–483.

Sui, J., & Humphreys, G. W. (2015). The Integrative Self: How Self-Reference Integrates Perception and Memory. Trends in Cognitive Sciences, 19(12), 719–728.

Sui, J., & Humphreys, G. W. (2017). The ubiquitous self: what the properties of self-bias tell us about the self. Annals of the New York Academy of Sciences, 1396(1), 222–235.

Wang, H., Humphreys, G., & Sui, J. (2016). Expanding and retracting from the self: Gains and costs in switching self-associations. Journal of Experimental Psychology. Human Perception and Performance, 42(2), 247–256.

Watson‐Jones, R. E., Busch, J. T. A., Harris, P. L., & Legare, C. H. (2016). Does the Body Survive Death? Cultural Variation in Beliefs About Life Everlasting. Cognitive Science, 41(S3), 455–476.

Chris Letheby and Philip Gerrans:  Response to Commentaries

Chris Letheby — University of Western Australia;

Philip Gerrans — University of Adelaide;

Our respondents share the idea that a predictive “self-model” is a crucial aspect of cognition whose nature can be illuminated by psychedelic ego dissolution. Many of the challenges they raise concern how we characterise the model: as a representation of a unified persisting entity; a Cartesian substance; a fixed point in attractor space that explains the experience of synchronic and diachronic unity despite constant changes in physical and psychological properties.

One important point is that the self-model is hierarchical. As well as integrating information horizontally at different levels (e.g. in rapid sensorimotor integration or interoception) the self-model vertically integrates these layers. Thus the highest, symbolic level of conceptual self-modelling known as the “narrative self” sits above layers of perceptual, interoceptive, emotional/affective and sensorimotor processing each of which exploits a proprietary self-model.  Unless integrated with these other layers, the narrative self would be a merely verbal expression of propositional knowledge about one’s personal history and propensities. Indeed, pathologies such as dissociative or depersonalisation disorders exhibit this phenomenology: subjects report facts about themselves consistent with their thin verbal narrative, while feeling affectively, agentially, or somatically dissociated. They know the experience is theirs while not experiencing it as theirs. Conversely, in dementia lower levels of selfhood are preserved while narrative functions disintegrate. The self-model has a hierarchical structure revealed by conditions of disintegration, including psychedelic experience.

A second point is that our aim was to provide an explanation of the phenomenology of ego dissolution that would link it to neural correlates at a perspicuous level of explanation. Psychedelics’ 5-HT2A agonist action is well established. However why that action should lead to reports of ego dissolution cannot be answered absent a theory of how the sense of self is produced by the affected circuitry. It is here that our account has potential. It fits well with many results implicating the default mode network (DMN) in a range of high-level self-referential processes. That the DMN subserves the narrative self by producing a sense of self-awareness is another well-grounded empirical hypothesis, as is the idea that the salience network underpins the embodied self ultimately via interoceptive predictive processing. Psychedelics weaken functional coupling between nodes of these networks, while increasing their coupling to external networks. Our thought was that at the neural level psychedelics disrupt the mechanisms that implement the self-binding function identified by Sui and Humphreys.

These considerations are not decisive regarding the representational content of the predictive self-model—e.g. whether the self is represented as a “patterned bundle” or an underlying entity. Our argument, however, picked up a point also made by Hohwy and Michael (H&M): that object representation as an integrative strategy seems obligatory at low levels of cognition and ubiquitous at higher levels of perception and sensorimotor integration. Predictive coding considerations suggest that object representation at higher levels would also exploit this strategy. Recent advances in AI (e.g. classification and complex pattern recognition by convolutional neural networks) also seem to exploit the ability of multilayer networks to learn to represent individuals as bearers of properties. Thus one advantage of these networks is that they do not only represent objects as bundles of properties but as substances that bear the properties. Those higher-level object representations then function as predictive constraints on the way lower level properties associate.

The use of this strategy in self-modelling would explain the otherwise puzzling phenomenology of ego dissolution. If this model was of a patterned bundle, it is not clear why subjects should report the experience of dying, disintegrating, or dissolving into the cosmos. Consider Michael Pollan’s experience on 5-MeO-DMT:

I felt a tremendous rush of energy fill my head… Terror seized me—and then, like one of those flimsy wooden houses erected on Bikini Atoll to be blown up in the nuclear tests, “I” was no more, blasted to a confetti cloud by an explosive force. Whatever this was, it was not a hallucination. A hallucination implies a reality and a point of reference and an entity to have it. None of those things remained. (Pollan 2018).

This suggests the pharmacological disruption to Pollan’s brain undermined its ability to model the existence of an entity distinct from the stream of experience—an assumption necessary to experience anything as a hallucination. Millière cites evidence that reflective intuitions do not uniformly favour Cartesianism cross-culturally, but this is consistent with the idea that the brain represents the self as Cartesian at a deeper level; as Metzinger (2005) argues, this would account for the cross-cultural prevalence of out-of-body experiences.

Michael raises further concerns about the content of the self-model, citing evidence that it has a gradual boundary. This seems to clash with our claim that the model represents a sharply bounded entity. Millière’s comments about the ambiguity of ‘self-binding’ are relevant here: the work of Sui and Humphreys shows that information deemed self-relevant is bound preferentially, but this seems distinct from binding information into a model ofthe self.

The solution is that an adequate model of any entity must represent not just the entity and its intrinsic properties, but also its extrinsic or relational properties. An adequate model of the dining table will specify not just size, shape, colour etc. but spatial position, load-bearing capacity, and financial and sentimental value. Likewise, mutatis mutandis, for the self. To model the putative entity underlying the stream of experience the brain must represent its innumerable extrinsic properties: owner and inhabitant of this body, vulnerable subject of environmental threat, voter, spouse, parent, etc.

On our view the self-model has a hierarchical and graded structure: the self is represented as bearing various more-or-less essential properties, with probable interpersonal variation. Dennett’s centre of narrative gravity requires a complex narrative; our centre of representational gravity requires a complex hierarchical representation. Adequately representing any individual requires representing its career and fortunes, which involve standing in relations that vary over time and admit of degrees. Nonetheless, the focal point of the representation is the postulated particular that stands in the relations and bears the properties. Indeed, the postulation of the particular is the glue that gives the complex model its coherence and structure.

This point connects with recent evidence that psychedelic therapy engenders a restored sense of ‘connectedness’ to aspects of oneself, the senses, the body, other people, nature, and the world (Carhart-Harris et al. 2018). By disrupting self-binding processes psychedelics allow alternative patterns of binding to be explored, providing an opportunity to ‘reset’ the self-model (Carhart-Harris et al. 2017). This connects with Michael’s comments about the multiplicity of self-modelling in various disorders, with which we agree. This malleability is harnessed beneficially in psychedelic therapy.

Millière is correct to distinguish self-relatedness from self-specificity conceptually. Our conjecture is that the same basic mechanism underlies both effects: the binding of multimodal stimuli into an integrated model of a unified self that bears properties and stands in relations. Self-related processing is thus underwritten by self-specific processing; computationally, the parsing of information into self-relevant and self-irrelevant involves the postulation of a particular. The hierarchical structure of the self-model is relevant: we are suggesting that a solution evolved by the mind to the problem of distinguishing internally- from externally-caused stimuli has been extended to the processing of more abstract relations (e.g. relevance) between the organism and patterns in its inputs. This conjecture entails a testable prediction: If Sui and Humphrey’s experiments were repeated on psychedelically-intoxicated subjects, the binding advantage they identify for self-reference should be diminished, and this diminution should correlate with (i) subjective ratings of ego dissolution and (ii) alterations in the neural substrates of self-binding (integrity of the DMN and Salience network.)

These remarks shed light on Michael’s question about the diachronic persistence that we claim is built into the content of the self-model. It is true that mental time travel and affective forecasting are unreliable and that events in the distant past often feel like “something that happened to someone else”. Of course, discounting of future rewards has limits: we typically save for our own retirement, not our neighbours’. But the point is well taken and speaks again to the graded, hierarchical, imperfect nature of self-modelling. Take the truism that it is easier to believe intellectually in one’s mortality than grasp it emotionally. Here we would say the attribute of mortality is bound with relative ease into higher, reflective layers of the self-model but less so into affective and motivational layers, creating a degree of dissonance and incoherence. The fact that the self-model postulates a particular does not imply maximal consistency or coherence, synchronic or diachronic; it is a Joycean and not an Austenian machine.

Michael raises the issue of reference and whether the self-model qualifies as a self. He asks whether, on our view, a representation can successfully refer to an entity despite getting some of its properties wrong. The point cuts both way, as Michael acknowledges: H&M would not say that typical causes of God-representations qualify as God. On our view there is a sufficient disparity between (i) the properties of the model itself and (i) the properties of the entity it postulates to warrant elimination rather than revision. But resolving this dispute would require delving deeply into the theory of reference. Perhaps the simplest thing to say is that, under psychedelics, we discover that we are not a substance but a virtual avatar; that we are radically wrong in what we ordinarily and unthinkingly take ourselves to be. This basic point is compatible with agnosticism about whether what remains truly qualifies as a self.

Hipolito’s comments provide an opportunity to clarify certain aspects of our proposal. She contrasts our “non-causally efficacious phenomenal self” with the causally efficacious “representational self” of H&M—but this is a false dichotomy. Our phenomenal self just is H&M’s representational self as it appears in phenomenal consciousness. On our view, the phenomenal experience of substantial selfhood results from predictive modelling (representation) of an entity underlying the flow of experience, just as the phenomenal experience of substantial tablehood results from predictive modelling of an entity in which properties are unified. We agree with H&M that the self-model qua neurally-implemented predictive model plays a causal role in cognitive processing. What we deny is that in virtue of playing this causal role, the self-model qualifies as a self. On our view representations of the self are (in these respects) like representations of gods: both play a causal role in cognitive processing and contribute to the contents of phenomenal consciousness, but the former representations are not actual selves any more than the latter representations are actual gods.

Hipolito questions whether, on our view, the postulation of a self plays a role in the construction of representations, or simply in the binding of contents “within pre-existing representations”. On our reading of cognitivist predictive processing, this too is a false dichotomy: as Hohwy (2013) argues, binding is not distinct from the construction of representations. Rather the binding of attributes and events is identical to the top-down postulation of an ontology. The cognitive binding problem dissolves on predictive processing because the brain minimises prediction error by postulating a coherent world of bound attributes from the top down.

In addressing Hipolito’s homuncular concerns our irrealism about the self qua distinct subject of experience is important. The cognitivist reading of predictive processing that we favour can be described in traditional philosophical terms: that internal models are the “objects of perception” or that “perception is indirect”, provided this is equivalent to the hypothesis that conscious experiences are intracranial processes and the brain does not ‘directly’ access extracranial entities in experience. As Metzinger and Revonsuo emphasise, the brain generates a virtual reality or world-simulation by performing (sub-personal, unconscious) Bayesian inferences on its sensory inputs.

But this picture only entails the necessity of a homunculus given the additional metaphysical premise that experiences require an ontically distinct experiencer. As irrealists about the self this is what we deny. Experiences happen, but not to anyone. There is not a movie playing in the Cartesian theatre with someone sitting and watching it. The whole theatre is a four-dimensional VR movie that usually includes the depiction of someone watching the screen; but in some scenes this character disappears, exposing the true nature of the situation. (To whom? The question is ill-posed.) All experience has the essential nature of Pollan’s trip: there is no entity having it. This is simply obscured in the usual case by the self-model.

This point is relevant to Hipolito’s question about whether the self fills the role of the “scientist in the head”. The Bayesian brain can be regarded fruitfully as a scientist, but this is a loose analogy for the fact that the contents of the brain’s predictive models are fixed by sub-personal processes which have the evolved function of minimising sensory prediction error, thereby approximating Bayesian inference in the long term. The self-model is not the scientist; it is just one of the many theories cooked up by the (neural, subpersonal) scientist to explain the sensory data. Again, this is not to deny that the self-model plays a causal role in cognitive processing: by representing the self as having various attributes, goals, interests, plans, and projects, the cortical midline networks implementing the self-model affect salience attribution, attention allocation, and many other functions. But this is all part of the unified strategy of prediction error minimization which operates at every level of the processing hierarchy. The postulated self does not play a role in cognitive processing; nor do gods. But the postulation of the self—the tokening of mental representations whose content is that the self exists—plays a causal role in the construction and updating of representations by forming part of the overall hierarchical model of the world implemented in the brain.


Carhart-Harris, R.L., Roseman, L., Bolstridge, M., Demetriou, L., Pannekoek, J.N., Wall, M.B., Tanner, M., Kaelen, M., McGonigle, J., Murphy, K. and Leech, R., 2017. Psilocybin for treatment-resistant depression: fMRI-measured brain mechanisms. Scientific reports7(1), p.13187.

Carhart-Harris, R.L., Erritzoe, D., Haijen, E., Kaelen, M. and Watts, R., 2018. Psychedelics and connectedness. Psychopharmacology235(2), pp.547-550.

Hohwy, J., 2013. The predictive mind. Oxford University Press.

Hohwy, J. and Michael, J., 2017. Why Should Any Body Have a Self?. In F. de Vignemont and A. J. T. Alsmith (eds.) The Subject’s Matter: Self-Consciousness and the Body. MIT Press.

Metzinger, T., 2005. Out-of-body experiences as the origin of the concept of a ‘soul’. Mind and Matter3(1), pp.57-84.

Pollan, M., 2018. How to Change Your Mind: What the New Science of Psychedelics Teaches Us about Consciousness, Dying, Addiction, Depression, and Transcendence. Penguin. Revonsuo, A., 2006. Inner presence: Consciousness as a biological phenomenon. MIT Press.


  1. Thanks Dan for organizing this, and thanks Chris and Phil for your reply! I have a quick comment regarding the empirical prediction of your theory. You write:

    “If Sui and Humphrey’s experiments were repeated on psychedelically-intoxicated subjects, the binding advantage they identify for self-reference should be diminished, and this diminution should correlate with (i) subjective ratings of ego dissolution and (ii) alterations in the neural substrates of self-binding (integrity of the DMN and Salience network.)”

    I doubt that any such experiment would be conclusive, because psychedelics are known to disrupt attention and impair performance on a number of cognitive tasks. Thus, one should expect performance on Sui & Humphrey’s experimental paradigms to be worse across the board, probably flatting the difference between self vs. other conditions. This is also why it is notoriously difficult to do psychophysics effectively in the psychedelic state.

  2. Professor A.Kanthamani

    Thanks to Dan for the illuminating symbosium.
    Letheby-Gerrans’s view of parsimonius self generates the following pairs:
    1a) self has hierarchical structure;
    2a) self has a layered structure;
    Ia and 1b are somewhat opposed; and, further on,
    1b) an integrated self binds under normative experience (DMN normatised);
    2b) a dissoluable self unbinds under psychedelic experience;
    Now, both of the above pairs are illumined by self-model (1) which is a predictive architecture(i.e. self exists); and self-model (2) self is an avatar(i.e.fiction).
    This will in turn generate that self is a narrative with abduction (with loose analogy with scientist); and self-referring and robust (with enhancing self-reference).
    Now this yields that it is a self model (i.e. one among others) compared with Metzinger’s being-no one self model (no self model, not one among others).
    and the self is backsliding and recessive as in Howhy.
    The authors claim that: ‘unless integrated with those others layers (i.e. perceptual, interoceptive, emotional/affective and sensori-motor/exteroceptive), the narrative self would be merely a verbal expression of propositional knowledge about one’s own personal history, plus body plus others (as Hohwy agrees).
    The contentious point is: which one is actually the narrative with a verbal expression as agreed in the above and narrative in the sense Metzinger uses in his layered and not a hierarchical (his is not a precursor of predictive architecture; such an architecture is only a ‘vanilla’ model which shares an agnosticism about theory or narrative.

Comments are closed.

Back to Top
%d bloggers like this: