Ellie Shaw – 01/01/2021
In this essay, it will be argued that applying a Predictive Processing Framework (henceforth PPF) to psychotic disorders provides a meaningful link between our visual perception of the world and the development of hallucinations, taken here as a symptom of psychosis. This link takes the form of a necessary access to the visual world, such that the lack thereof prevents one from making predictions about that world. To motivate this claim, I draw on the fact that congenital blindness (blindness from birth) is protective against the development of Schizophrenia and other psychotic disorders (Morgan, et al., 2018), whilst cortical blindness is not. This, I argue, is because cortically blind individuals have prior access to the visual world from which to make predictions, unlike congenitally blind individuals. Consequently, hallucinations and other psychotic symptoms can occur in the former case but not the latter.
I conclude that, for the following reasons, a PPF provides the link between vision and psychosis: (i) Vision is prime in generating predictions and prediction errors (henceforth PE) such as hallucinations. (ii) Hallucinations, within the context of psychotic disorders, occur as a result of faultily stored PEs, (iii) thus, access to the visual world is a necessary condition for the development of psychotic disorders. To my awareness this is a novel project. However much work has been done in the phenomenology of psychotic disorders (Sass & Parnas, Explaining schizophrenia: the relevance of phenomenology, 2007) (Krueger, 2020) (Sass & Parnas, Phenomenology of Self-Disturbances in Schizophrenia: Some Research Findings and Directions, 2001) (Borda & Sass, 2015), and on the PPF (Griffin & Fletcher, 2017) (Sterzer, et al., 2018) whose views I draw on to inform my theory.
Visual Perception and Schizophrenia, A Predictive Processing Framework
I begin by providing an account of the PPF and explicating how it can be applied to psychosis and visual perception. The PPF “understands the brain as a prediction -action machine that constantly predicts sensory input and aims to minimize error in its predictions” (Schwengerer, 2019, p. 565). On this model the brain is understood as a computational system which aims to model one’s experience of the world by making predictions about “the causes behind sensory inputs” (ibid, p.566). PPF accepts that there is an external world containing events, objects and entities, and which interacts with the brain through our sensory experiences. This sensory information is used to generate predictions about the individual’s current experience of the world. “The prediction about the world is then tested by predicting what the brain’s next sensory inputs will be. Based on whether this prediction is correct, the model is updated.” (ibid, p.566) One important feature of this account is the top-down model of cognition according to which, “each layer of neural processing is thus trying to predict the current response at the layer below (except for the bottom layer, such as the retina, which transduces some energetic signal). Under these conditions the brain can self-supervise its own learning, as the lower levels do indeed come to occupy sequences of states that the high levels can (in the processing sense) access.” (Clark, 2012, p. 759)
Thus, the brain is seen to be a complex hierarchical system of neural networks (Friston, 2008) which allows for predictions to be made about one’s perceptual experience. “Predictions are encoded at higher levels of a hierarchical system and are sent as predictive signals to lower levels” (Sterzer, et al., 2018, p. 636). In some cases, an individual’s incoming information about the world may contravene the predictions made by the system. In these cases, “a prediction error signal is sent to update the predictive model at higher levels. […] Formalized as Bayesian inference, predictions (prior) and sensory data (likelihood) are represented in the form of probability distributions. The posterior results from the combination of prior and likelihood […] weighted by their respective precisions […] and updates the predictive model.” (Sterzer, et al., 2018, p. 363)
In simple terms, when there is a mismatch between the expected input and the actual input an error signal is transmitted which then updates the system such that it accounts for the error. The system then ‘learns’ from this error, and in ordinary cases takes the past error into account when making future predictions. Predictions are thought of in terms of probability of occurrence, such that if a correct prediction occurs this is then integrated into the system and the next time I find myself in the same situation, that prediction can be applied. In what follows, I apply a PPF to visual perception.
Suppose I am out walking late one night, and I hear a strange rustling noise coming from the bushes. I derive, from my previous experiences of walking late at night, that the noise may be a threat. I see a figure walking some distance behind me, my prediction of a threat is confirmed, and the subsequent response is to hold my housekeys between my fingers and get my mobile out. The “prediction gets confirmed and its assigned probability increases” (Schwengerer, 2019, p. 566). Therefore, next time I hear a noise and see a figure, I will predict based on that earlier experience, that the noise was caused by the figure and given the late hour, view the figure as a threat.
Now let us assume that I hear the rustling noise in the bushes and turn to look, assuming that there is some threat. However instead of seeing a figure behind me, I see a fox running out of the bushes. I then predict that this is not a threat and therefore leave my keys and mobile in my pocket. In this case, the prediction ‘there is a threat’ is registered as incorrect by my brain. What we have then is a PE that ‘there is a threat’ and a subsequent updating of the system to account for the error. The next time I hear a noise late at night I view it as equally probable that the cause was an animal compared to another person. In order for the brain to make successful predictions, it generates associations, such as ‘this fruit that grows on this tree is edible and readily available to harvest and eat’. Naturally these associations require some degree of flexibility, for example that fruit may be seasonal, rotten, or eaten by deer. Consequently, these associations “must be flexible in the face of change but robust in the face of noise”. (Griffin & Fletcher, 2017, p. 269) Noise in this context equates to distractions or errors in the statistical sense.
Predictive Processing Accounts of Psychotic Disorders
Many scholars (Corlett, 2018) (Fletcher & Frith, 2008) (Wilkinson, 2014) have postulated that “the early emergence of delusions and hallucinations could be conceptualized in terms of associative learning models, andmore particularly, in terms of a disruption to prediction-error dependent updating” (Griffin & Fletcher, 2017, p. 267). The story that a PPF advocate tells of psychosis is one in which abnormal PEs would cause disturbances to the overall system. The result of this is that the subject is not able to form accurate predictions about the objects of their perception. As a result of this fault, the system produces defective and unreliable results, affecting the world modelled off these results such that it appears strange. This fault in the system produces the delusions and hallucinations that characterise psychotic disorders.
Using Schizophrenia as an example of a psychotic disorder which arises from “a disruption to prediction-error dependent updating” (ibid, p.267), I develop and maintain my thesis that visual access to the world is a necessary condition for the onset of hallucinations and delusional beliefs which characterizes psychosis. This, I take it, explains why congenitally blind individuals are immune from developing psychotic disorders such as Schizophrenia (Morgan, et al., 2018) whilst cortically blind individuals are not. I understand hallucinations in this context to refer to “perception-like experiencesthat occur without an external stimulus” (American Psychiatric Association, 2013, p. 87). Schizophrenia is understood in accordance with the following criterion laid out in the DSM-V (ibid): “1. Delusions 2. Hallucinations 3. Disorganised speech […] 4. Grossly disorganised or catatonic behaviour 5. Negative symptoms (e.g. di- minished emotional expression or avolition)” (American Psychiatric Association, 2013, p. 99)
To cross the diagnostic threshold, individuals must have “two (or more) of the following, each present for a significant portion of time during a 1-month period […] At least one of these must be (1), (2), or (3).” (ibid). “Signs of the disturbance” (American Psychiatric Asso- ciation, 2013, p. 99) will affect the individual for a period of up to six months, included within this six month peri- od is “at least 1 month of symptoms” (ibid) as laid out in (1)-(5)i . To be considered clinically significant, these symptoms should have a significant impact on an indi- vidual’s ability to carry out day to day activities.
From a PPF standpoint then, PEs which are not updated and stored in the correct manner can cause the onset of psychosis and psychotic disorders. This in turn gives rise either to hallucinations, delusions, or a combination of the two, depending on how the system attempts to store and make sense of the initial PE. I focus here on hallucinations and psychosis, arguing that it is an error in the processing and storing of visual input which leads to the presentation of such symptoms. Without access to the visual input, then, these symptoms cannot occur.
Allow me to return to the earlier example of walking alone at night and hearing a noise. An individual who is not suffering from a psychotic disorder may view the likelihood of the noise coming from an animal or a nearby person as equally probable given their past experiences and prior beliefs. However, the psychotic individual will place more emphasis on the latter, viewing it as more probable that there is an individual following them, and neglecting or downplaying the possibility that the noise was caused by an animal. PPF, then, can help explain many of the symptoms that characterize Schizophrenia and other psychotic disorders. This can be seen in the following phenomenological account of Schizophrenia which highlights the significance of PEs in generating visual hallucinations: “during periods of acute bombardment, paving stones transform into demonic faces, shattering in front of my petrified eyes. When I am in contact with people, they can become grotesquely deformed, their skin peeling away to reveal decomposing inner muscles and organs […] I am powerless to resist the frenzied subjection of alien beings as steel bars are suspended from the ceilings trapping me like a caged beast” (Bayley, 1996, p. 728)
Phenomenology can be a useful tool then, for understanding the ways in which these PEs manifest themselves with in a PPF. Whilst some scholars advocate for a purely phenomenological account of mental illness (Sass & Parnas, Explaining schizophrenia: the relevance of phenomenology, 2007) (Sass & Parnas, Phenomenology of Self-Disturbances in Schizophrenia: Some Research Findings and Directions, 2001) (Krueger, 2020), I differ from this and hold that phenomenology can be utilised to describe the experiential content of hallucinations; but that the cause of these is the faultily stored PE(s) which are incorporated into the system.
Furthermore, these first-person descriptions serve to illustrate the manner in which faultily stored PEs manifest within the individual, and how these distort their experiences of the world. This further supports my argument that a PPF can provide valuable insight into the link between vision and psychotic illnesses. In the above example of a Schizophrenic hallucination, the ‘deformed’ people described by the psychotic individual are examples of the ways in which faultily stored PEs manifest themselves at the conscious level. The input (visual experience of some person’s face) is incorrectly incorporated into the system, leading to a strange visu- al experience (the face appears to be rotting) that does not match the actuality of the situation (the person’s face is not rotting). In the remainder of this essay, I shall argue that visual input is a necessary condition in generating predictions, such that without visual input one cannot experience hallucinations. This, I maintain, explains why there are no cases Schizophrenia in congenitally blind individuals (Morgan, et al., 2018) I make explicit within this argument, that visual input is the most important form of sensory input.
On the link between vision and psychosis.
The predictive processing system can then be thought of as having a couple of individually necessary and jointly sufficient features. I take these to be (a) Sensory input (e.g. visual, auditory, olfactory, gustatory, tactile) of some x (b) The means of generating sensory input about that x. (c) The existence of some world which contains that x. (d) Neural mechanisms by which pre- dictions are made and outputs produced based on that x. (e) Some observable output in that x (e.g. behavioural or verbal).
For the present purposes I am interested in conditions (a) and (b), although each of these conditions is philo- sophical significant in their own right. I focus on visual input in my examination of (a) and (b). From (a) and (b) I argue that access to the visual world is a necessary condition for the development of psychosis. This explains why congenitally blind individuals are protected from developing Schizophrenia and other psychotic disorders. My argument is summarised here: (P1): On the Predictive Processing Framework, PEs which are not updated in the correct manner can and do generate psychotic symptoms (e.g. hallucinations). (P2): These PEs require some input (P3): The primary source of that input is vision. (P4): If one does not have access to that visual input, one can generate neither predictions nor PEs. (P4dem 1): Congenital blindness is protective for Schizophrenia and other psychotic disorders. (P4dem 2): Hallucinations can and do occur in cases of cortical blindness (e.g. Charles Bonnet Syndrome). (C): The link between vision and psychosis takes the form of visual access to the world from which predictions can be generated. I begin by defending (P3). “The retina is an active information processor, in which, among other things, early forms of transformation traditionally thought to occur only in visual cortex, take place. For example, in response to moving stimuli, cone photoreceptors produce motion blur.” (Silverstein & Rosen, Schizophrenia and the eye , 2015, p. 47)
This, in turn provides information about multiple visual features such as “orientation and direction of move- ment, […] changes in texture between stimulus regions, spatial frequency, and depth” (Silverstein & Rosen, Schizophrenia and the eye , 2015, p. 47). Consequently, ‘motion blur’ impacts the way in which individuals view and interpret more complex phenomena “such as facial expressions” (ibid, p.47) This finding is significant because it suggests that factors affecting the workings of the retina can “result in intensified, degraded or noisy input to higher levels of processing in […] [the] visual cortex” (Silverstein & Rosen, Schizophrenia and the eye , 2015, p. 47). Due to the fact that “the retina devel- ops from the same tissue […] as the brain” (ibid, p.47) it is plausible that changes in the retina could reflect changes in the structure of the brain.
Studies have shown a connection between the struc- ture of the eye and the development of Schizophrenia. Individuals with a Schizophrenia diagnosis were shown to have “wider retinal venules, suggesting abnormality” (Meier, et al., 2013, p. 1451). In simple terms, the blood vessels in the retina were shown by Meier et al to be wider in Schizophrenic individuals than in the control group, with the results of this study being successfully replicated two years later (Meier, et al., 2015). This abnormality is taken to play a causal role in the development of Schizophrenia by the original authors.
I agree to a certain extent with this conclusion, the abnormal widening of the blood vessels in the retina could play a role in increasing the likelihood of faulty perceptions (PEs) because this would cause altered or abnormal visual inputs are integrated into the system in the improper manner. This leads the agent to believe that they are perceiving x in one manner, when actually they are perceiving either x in anther manner, or some y which they perceive as some x due to the widening of the blood vessels causing an odd perceptual experience (i.e., they experience either a hallucination or a delusion). Further studies on the relation between vision and psychosis support this conclusion, with several purporting that “visual distortions were more sensitive to later conversion to psychosis than abnormalities in other domains, including auditory distortions or thought disorder.” (Silverstein, Visual Perception Dis- turbances in Schizophrenia: A Unified Model, 2016, p. 78)
From a philosophical sense then, it is easy to see why vision is prime in generating priors under PPF. If a seeing individual were to walk through a zoo on a crowded day, their perceptual priors would be constantly updating and the individual would be able to attribute the incoming sensory information to the correct sources (for example seeing a lion, hearing a distant roar and attributing the roar to the lion). If that same individual were then to be blindfolded, they would be assaulted by a mass of confusing sensory information (they would be unable to attribute the roar to a lion, as compared to a tiger). Sensory information quickly becomes confused in this case and produces PEs which are faultily stored within the PPF. It is clear then, that vision plays a significant role in the ways individuals organise their experiences of the world. Thus, access to visual information from which to generate input, is a necessary feature of the development of psychotic disorders. That visual input is also prime compared to other forms of sensory input, such as auditory input, in the production of faultily stored PEs. This is confirmed by the fact that congenital blindness is protective of Schizophrenia (Morgan, et al., 2018) but congenital deafness is not (du Feu & McKenna, 1999)
From here, I move to defending (P4) of my argument, if one does not have access to visual input, then one cannot generate PEs about that input. To do so, I draw on two cases; Case 1 is that congenital blindness is protective of developing Schizophrenia and other psychotic disorders (Morgan, et al., 2018) and Case 2 demonstrates that psychotic symptoms can and do occur in the cortically blind by drawing on the example of Charles Bonnet Syndrome (henceforth, CBS).
Case 1: Congenital Blindness and Schizophrenia
The premise to be defended then, is that congenital blindness is protective of developing Schizophrenia because it prevents one from making predictions about the visual world, but that cortical blindness (that is blindness which develops later in life) is not protective because cortically blind individuals (such as those who suffer from CBS) can draw on previous access to the visual world and use this to generate faultily stored PEs. I begin with Case 1. There are no instances of Schizophrenia in the congenitally blind due to the fact that they lack access to the visual world, and thus have nothing on which they can make predictions.
If individuals lack the means to make predictions, it follows that they cannot make PEs, because the existence of PEs presupposes the existence of predictions. In simple terms there must be some prediction which is erroneous, if there is no such prediction, there is no such error. If the congenitally blind individual were to walk through the zoo, they would have a very different experience to our blindfolded individual. This is because the incoming information does not overwhelm the congenitally blind individual. Congenitally blind individuals are better able to distinguish between non-visual sensory information than their seeing counterparts (Hötting, Rösler, & Röder, 2004) because “vision most clearly provides the spatial scene within which sensory data from other modalities can be most efficiently contextualised and integrated; it enables the construction of multimodal or supramodal representations.” (Pollak & Corlett, 2019, p. 3) Without the visual representation of the world, then, psychotic symptoms (hallucinations) cannot occur.
Case 2: Cortical Blindness and Charles Bonnet Syndrome
Charles Bonnet Syndrome is characterized by “vivid and complex visual hallucinations in otherwise mentally healthy individuals with visual impairment.” (Lomo, Singh, & Peters, 2020, p. 1). These hallucinations are exclusively visual and “typically well defined ranging from abstract geometric patterns to complex, vivid images such as animals, people or scenes” (Hughes, 2013, p. 170). CBS is viewed as the result of muscular degeneration in the eye and often occurs in the geriatric population although onset can occur at any age or as the result of a traumatic brain injury. It is difficult to begin to make sense of what it is to have a visual hallucination and be unable to see. Therefore, to provide insight into CBS and maintain my conclusion, I shall draw on first person reports. Part of my philosophical project here involves taking these first-person accounts seriously, viewing them as an honest insight into that condition. However, it is difficult to make sense of what it means or what it is like to be blind and have a visual hallucination as a seeing individual. This is a tension I am not able to resolve at this time, but that nonetheless remains significant. One 86-year-old female with CBS (Lomo, Singh, & Peters, 2020, p. 2) “reported seeing a woman with long boots walking in front of her. She first experienced this hallucination while taking a walk with her husband. She mentioned telling her husband, how strange it was that the woman had been walking in front of them all the time but found out that her husband did not see any woman in front of them. The patient also reported seeing the face of a monkey, placed on her grandchild’s face, sitting on the sofa in front of her” (Lomo, Singh, & Peters, 2020, p. 2)
The individual reported that the hallucinations persisted for 5 months. How then, given the woman’s blindness can one account for her persistent visual hallucinations? I argue that because this woman had access to her past visual experiences before the onset of cortical blindness, she possesses the necessary data to generate hallucinations and thus meets the aforementioned conditions (a) and (b). These predictions are then processed and stored in the incorrect manner, because the brain interprets them as input. These predictions, however, could not get off the ground without previous visual experience. Fortunately, in the case of cortical blindness, individuals do have access to their past visual experiences which are stored as memories. Going back to the 86-year-old female, it seems plausible that she has memories of monkeys, her grandson and a woman wearing long boots. From these memories, she can generate hallucinations as the monkey will be taken as one instance of input, the grandchild another-together these form the visual hallucination of her grandchild with a monkey’s face. It is only because this woman meets the necessary condition of previous access to the visual world that she is able to generate such a hallucination, and indeed make sense of it in the first place. Consequently, I maintain that access to the visual world is a necessary condition for the onset of hallucinations in psychotic disorders. This can be seen through the fact that the cortically blind are able to generate hallucinations, but that the congenitally blind are not.
Concluding Remarks
In conclusion, it has been argued that access to the visual world is a necessary condition for the development of psychotic disorders. In order to motivate this conclusion I have utilised a PPF view of cognition, and demonstrated that congenitally blind individuals are protected from developing Schizophrenia and other psychotic disorders because they lack necessary access to the visual world from which to generate inputs. Cortically blind individuals, however, are not protective from developing psychotic illnesses because they have previous access to the visual world. This further motivates my claim that access to the visual world is a necessary condition for the development of psychotic disorders. Vision then is taken to be the primary means by which individuals make predictions about the world, and any disturbance in that vision or even the lack thereof can lead to the development of psychotic disorders.