Autism at the cognitive level: Domain-general information processing models
SO FAR, THE cognitive models reviewed in this book, whether taking a developmental perspective or not, have largely focused on explaining the social features of autism. While attempts have been made to extend the explanatory power of these models to encompass non-social aspects of the condition, as we have seen, these are rarely adequate. In addition, there are open questions about the position of what we call “social cognition” within development — are our social interactions with others a test-bed in which we hone our domain-general skills, such as attention and executive functions? Or do these domain-general skills come first and our social interactions draw on these underlying processes?
In response to the incompleteness of social-focused theoretical accounts, and in recognition of the prevalence and importance of non-social components of autism, a number of theorists have proposed domain-general cognitive interpretations. We group these under the umbrella term ’information processing’ because they all emphasise the different ways in which autistic people seem to take in, organise and respond to input from the world. Crucially, these models focus on aspects of input that are common to both social and non-social contexts, often at the level of perceptual features of the environment.
1. Perceptual process accounts: weak central coherence theory
The ’weak central coherence’ account was the first domain-general theory to attempt to explain strengths, as well as difficulties, in autism. Uta Frith coined the term ’central coherence’ to refer to the tendency neurotypical people have to draw information together, integrating details in context to form a gestalt (i.e. an organised whole that is more than the sum of its parts), see the bigger picture and get the overall meaning (Frith & Happé, 1994). Her early and pioneering studies of autistic strengths in block design (see Figure 8.1) and embedded figures tests, with her then-PhD student Amitta Shah, formed the bedrock for this theory, which proposed that in autism, the neurotypical drive for ’central coherence’ is reduced, resulting in better attention to and memory for local details, but lessened global processing (Shah & Frith, 1983, 1993).
Figure 8.1 The block design test
In fact, Kanner also noticed the eye for detail in autism, writing in 1943 of the children’s
inability to experience wholes without full attention to the constituent parts… . A situation, a performance, a sentence is not regarded as complete if it is not made up of exactly the same elements that were present at the time the child was first confronted with it. If the slightest ingredient is altered or removed, the total situation is no longer the same and therefore is not accepted as such.
(Kanner, 1943, p. 246)
Fragmented perception and thinking is also described in autobiographies: Donna Williams has described her senses as “chaotic, fragmented and constantly shifting and fluctuating” (Williams, 2009),1 and Gunilla Gerland writes,
Every little bit of fact seemed to land in its own compartment in my head and refused to be linked with any other. I tried poking into details. I dissected them and hoped a unified whole would appear, but it rarely did.
In the original version of this book, an anecdote (shown in Figure 8.2) was given to illustrate the notion of weak central coherence, something that the great Lorna Wing had told Francesca when she was still a PhD student. This is perhaps still the easiest way in which to convey the idea of weak central coherence:
Lorna Wing, when assessing a bright autistic boy, presented him with a toy bed, and asked the child to name the parts. The child correctly labelled the bed, mattress and quilt. Lorna then pointed to the pillow and asked, ’And what is this?’ The boy replied, ’It’s a piece of ravioli’.
The child in the anecdote was not joking, nor was his sight impaired — indeed Lorna commented that the pillow did indeed look just like a piece of ravioli, if taken out of context. However, she would never have seen it that way, because, like most neurotypical people, her interpretation of information was constrained by the context. The central coherence theory suggests that autistic perception, attention and memory are free from such contextual constraints.
Context affects not only visual, but also verbal and auditory processing; when we read homographs (words with two meanings and pronunciations but one spelling; see Figure 8.3), we rarely notice the potential ambiguity, because we automatically use the sentence context to identify the meaning; e.g. ’In her eye there was a big tear’ versus ’In her dress there was a big tear’. Not using the sentence context to determine the correct pronunciation of homographs has been shown in autism, but importantly, this is a default only; if asked to read for meaning or alerted to the ambiguous words, autistic people have no difficulty integrating the sentence. This, and other evidence (reviewed in Happé & Frith, 2006), suggests weak coherence is a cognitive style and represents a bias not a deficit. Just as neurotypical people can, if necessary, memorise unconnected information without meaning (e.g. cramming for an exam, remembering a pin number), so autistic people can, if necessary, integrate local details into global meaning, but it seems neither comes naturally or without some effort.
Figure 8.2 An example of de-contextualised perception
Reproduced by kind permission of the artist, Axel Scheffler.
Might an apparent local focus in autism just reflect difficulties with executive functions? For example, when drawing, autistic people may begin with details, which is unusual in neurotypical people; is this because of problems planning ahead? In line with the ’fractionated triad’ account described in Chapter 3, there is some evidence that detail focus and executive difficulties are distinct (Brunsdon & Happé, 2014); for example, boys with ADHD who have problems planning and inhibiting their actions don’t show detail focus, and detailed drawing style doesn’t correlate with poor planning during drawing (Booth et al., 2003).
Figure 8.3 The homograph task, ’lead and tears’
Reproduced by kind permission of the artist, Axel Scheffler.
A recent meta-analysis, albeit drawing on tasks that tend to conflate global and local processing styles, demonstrated that slower global processing was the only robust effect found across studies (van der Hallen et al., 2015). More recently, Francesca and her colleague Rhonda Booth have suggested that superior local processing on one hand and reduced global/integrative processing on the other may be distinct and somewhat independent in autism, even though most tasks used to date have conflated the two (Booth & Happé, 2018). They report tasks designed to tap global and local processing independently, and suggest that measuring these dimensions separately may help untangle heterogeneity of cognitive style within autism.
2. Other perceptual process models
Weak central coherence theory is accompanied by other models which draw on overlapping evidence, with subtle differences in interpretation. Laurent Mottron’s enhanced perceptual functioning model proposes that the perceptual systems of autistic people may out-perform their neurotypical peers, resulting in skills that can be interpreted as biases (Mottron et al., 2006). The theory predicts superiority in local processing, without a parallel difficulty at the global processing level. Evidence includes studies using drawing tasks in which people with autism are more capable of drawing an impossible figure compared with adults without autism. More recently, this account has inspired an investigation of whether estimates of the intellectual ability of autistic children may be more accurate when drawing on measures that capitalise on perceptual strengths. The authors found that using a battery of measures selected in this way (e.g. Raven’s progressive matrices) resulted in much higher estimates of ability than a traditional IQ test (Courchesne et al., 2015).
Kate Plaisted-Grant and colleagues have proposed another variant, which suggests that the specific features affected in autism are discrimination and generalisation. These are framed as an opposing pair of skills, the former enhanced and the latter reduced in autism. This model is particularly based on data from visual search tasks which reveal that autistic people show particular strengths for the most difficult, conjunctive search tasks, requiring excellent discriminatory abilities (Plaisted et al., 1998a, 1998b; O’Riordan et al., 2001). A recent analysis of the locus for superior visual search in autism concluded that autistic adults “excel in non-search processes, especially in the simultaneous discrimination of multiple visual stimuli” (Shirama et al., 2017).
How are we to make sense of these multiple, subtly different explanations of the perceptual profile in autism? While they may make different predictions regarding performance on carefully designed experimental tasks, it is not clear to what extent distinguishing between these accounts would impact on, for example, classroom practice or home life. A sensitive psychological battery might provide a personalised profile of an autistic individual’s perceptual processing — perhaps distinguishing between people who struggle with global processing and those with especially powerful local perceptual skills. But in the meantime, what we can draw from this literature is twofold. First, the emphasis on describing and explaining autistic strengths provides a welcome relief from the majority deficit-focused psychological literature on autism. Second, the research offers clues about how we might enhance teaching of children with autism and provide suitable employment opportunities for adults on the spectrum. Of course, such recommendations should not be allowed to drown out what autistic people themselves tell us directly about their needs, or the very real existence of individual differences. Nevertheless, these findings can provide a useful starting place, especially when working with those who may appear at first glance to have a significant learning disability. For this reason, it is essential that this work is shared with practitioners and translated into relevant guidelines.
Returning to the purely scientific perspective, we note that these explanatory models do not propose a specific mechanistic account. In each case, differences in the way that autistic people take in and respond to sensory input are evident, but the underlying process that governs these differences has not been elucidated. In Section 5, we review one new model of brain function which may have relevance here.
3. Integration and complexity
Another attempt to characterise autism from an information processing perspective comes from investigations of differences in processing and integrating multiple sources of information (Minshew et al., 1997; Minshew & Goldstein, 1998). These models do not focus on the perceptual level — approaching and responding to input — but instead explore the ways in which information is integrated in the brain, between the input and response phases. For example, Neumann and colleagues (2006) used a computational modelling approach to propose that the looking patterns to faces shown by many autistic people (looking more at mouths and less at eyes, compared with non-autistic people) result from differences in the top-down modulation of attention, rather than from bottom-up perceptual processes. Much evidence for this theory comes from fMRI studies which reveal differences in functional brain responses to multi-modal input and structural differences in circuits considered necessary for information integration (e.g. Castelli et al., 2002; Bird et al., 2006). One meta-analysis concluded that, while sample sizes are often small and thus results may not be robust, findings to date suggest differences in integration of different brain regions, rather than on any localised ’deficit’ (Philip et al., 2012).
At the behavioural level, specific differences have been found in the tendency to integrate multiple sensory inputs — such as visual, auditory and tactile information (Iarocci & McDonald, 2006). Relatedly, some studies have found that autistic people need a larger quantity of motion cues to detect movement (Milne et al., 2005) — though this finding is not consistent (Foss-Feig et al., 2013). One interpretation of these data is that insensitivity to motion cues reflects a difficulty combining the multiple sources of information — such as relative position of foreground figure to background — required to detect movement. This work has been used to interpret the finding, discussed in Chapter 7, that infants who later receive an autism diagnosis attend more to areas of a stimulus that display audio-visual synchrony (Klin et al., 2009). Applying an information processing account here suggests that this attention to multi-modal synchrony is a reflection of effortful processing of multi-sensory input, rather than a simple attentional preference. The same interpretation can be used to frame findings from the eye-tracking literature, which consistently show the greatest group differences between participants with and without autism when stimuli are multi-modal (i.e. audio and visual input) and moving (Chevallier et al., 2015).
Another theoretical account which might fall under the broad heading of integration and complexity is the interest-based theory, monotropism (Murray et al., 2005). This theory, developed by autistic academics, posits that the defining feature of autism is atypical allocation of attention. The difference between autistic and non-autistic people is characterised as follows: “It is the difference between having few interests highly aroused, the monotropic tendency, and having many interests less highly aroused, the polytropic tendency” (ibid., p. 140). Consequently, this model places causal primacy on the intense focus apparent in the diagnostic domain of RRBIs, with other diagnostic features following from this underlying difference. To the extent that social interaction requires diffuse and distributed attention, autistic people are not well suited to that activity. Monotropic theory, which awaits empirical testing, provides a vivid description of the autistic experience of novelty and change, giving a valuable insight into the autistic experience of a crisis, or “meltdown”:
To a person in an attention tunnel every unanticipated change is abrupt and is truly, if briefly, catastrophic: a complete disconnection from a previous safe state, a plunge into a meaningless blizzard of sensations, a frightening experience which may occur many times in a single day.
(Ibid., p. 147)
All of the theoretical approaches listed in this section share a high level of relevance to aspects of the lived experience of autistic people, not always well captured by cognitive models. For example, difficulties integrating sensory information could explain sensitivity to sensory input. Failure to integrate could make separate sensory signals overwhelming (hyper-sensitivity) or might cause some sensory input to be ignored because it is processed in isolation, instead of being boosted by concurrent input signals (hypo- sensitivity). Differences in how movement is processed and perceived could lead to atypical gross motor patterns and to dyspraxia, which is common among autistic people. Monotropic attention would lead to the development of specialised skills but also difficultly dealing with change. All of these features would impact on social cognitive abilities, by inhibiting the online, rapid processing of social information which often incorporates multiple inputs — e.g. speech, gesture, facial expression. However, this theory family is severely limited by a lack of data directly testing the predictions of the model. Rather, the model has largely been derived from post-hoc interpretations of patterns of neurological and behavioural data, and is currently lacking robust experimental support.
4. Systemising and empathising
An influential account, growing out of Simon Baron-Cohen’s notion of a ’male brain’ type in autism (though see Fine, 2010, for a robust critique of apparent gender differences in neuroscience research) suggests that autism can be understood as a combination of poor ’empathising’ and good ’systemising’. Both constructs are measured as trait dimensions using self-report checklists; the Empathising Quotient asks about social skills and preferences, while the systemising quotient asks how much one likes to impose or discover order, or work out how things work. The extent to which these constructs are orthogonal dimensions or trade-off against one another remains uncertain (Wheelwright et al., 2006).
The systemising quotient attempts to capture the tendency to engage in systematic activities and thinking styles, apparent in rule-based, predictable and logical contexts — for example, common degree subject choices for someone with a high systemising quotient might be engineering, mathematics or computer sciences (Manson & Winterbottom, 2012). In the context of autism, systemising does not contradict previously described perceptual accounts. In fact, systemising presupposes low-level characteristics such as an eye for detail, though it remains an open question whether a detail-focused perceptual style leads to a preference for systemising, or whether systemising activities hone relevant perceptual skills. There are consistently high systemising quotient scores among the autistic population and also in their close relatives, though the field is weakened by reliance on self-report. Lawson et al. (2004) created a Physical Prediction Questionnaire, using items taken from the Vincent Mechanical Diagrams Test, to assess understanding of physical systems — but an ideal test of systemising would present individuals with a wholly novel system to ’crack’ (and hence avoid confounds with prior experience/interests/teaching). Harvey and colleagues (2016) found a correlation between self- reported systemising and performance on a code-breaking task in a small group of hackers and a larger unselected group, but to our knowledge this ability hasn’t been tested in autistic people. One study that tested ’foraging’ found no evidence of a more systematic approach in children with autism compared to children with typical development (Pellicano et al., 2011). Further work, ideally presenting novel challenges that can be solved using either systematic or non-systematic solutions, would push forward the evidence base for this dimensional theory.
Turning to the Empathising Quotient, here the nomenclature of “empathising” is highly problematic. The dictionary definition of empathising is “to understand and share the feelings of another”, and in general discourse, empathy is often used to mean simply caring about or resonantly responding to other people’s feelings. We have seen in Chapter 6 that many social cognitive models attempting to explain autism suggest that autistic people have a difficulty representing other people’s mental states — but knowing what other people are thinking (ToM) is not the same as caring what they are feeling (emotional empathy). Whether emotional empathy is affected at all in autism is hotly debated both by scientists (Bird & Viding, 2014) and self-advocates — with the latter often remarking that their lived experience is of feeling too much, rather than too little, of others’ emotions.
Many autistic advocates have pointed out the damage which has been done to their community by the spread of the idea that autistic people are incapable of empathy. In particular, this construct is offensive to a group of people who may be working harder than most to work out others’ mental state and to respond as expected by social norms. To label any mismatch between the expected behaviour in a particular social context and what an autistic person actually does, as resulting from a lack of empathy, betrays a significant lack of empathy on the part of the non-autistic observer. In Chapter 9, we extend this discussion of empathy and autism with reference to the double empathy problem proposed by Milton (2012).
5. Bayesian accounts
A relatively new approach to understanding the brain, at the time of writing, is the Bayesian, or predictive coding model. This is fiendishly hard to understand in detail, but in essence the model suggests that our brains operate by (rapidly, constantly) making top-down predictions about the world and comparing these against incoming perceptual evidence. The goal is to minimise mismatch or ’prediction error’ — so predictions (known as “priors”) get more and more accurate at matching with incoming information, and this guides behaviour. A key component of the system is that experience contributes to the formulation of more precise priors.
In the context of autism, one account derived from the Bayesian brain framework, suggests that the autistic brain relies on under-specified predictions because it does not take experience into account effectively (Pellicano & Burr, 2012). In some cases, such a neurological underpinning would result in greater precision — as when autistic people show reduced susceptibility to visual illusions (Happé, 1996). A Bayesian interpretation could be that autistic people view the illusion on its own terms because they do not (mis)apply expectations based on experience. However, in the majority of situations, prior experience helps non-autistic people to ’smooth’ their sensory input, classifying it according to their expectations so that they do not need to be distracted by variability at a detailed level. For example, prior knowledge allows neurotypical people to experience a familiar space as the same every time, even though in reality small details will have changed. Meanwhile, autistic people might experience that space as brand new, or at least full of distracting changes.
However, an alternative Bayesian account suggests autism is characterised by excessive precision in predictions (Lawson et al., 2004). According to this theory, autism is associated with predictions that are over- specified and detailed, resulting in a larger experience of error. In other words, by this account, autistic people do use prior experience, but these priors are very detailed and strong, resulting in ’overfitting’ and a lack of generalisation (Van de Cruys et al., 2013, 2014). This opposite hypothesis can explain exactly the same phenomenon as noted earlier — autistic people might struggle with minor changes in a familiar space because their prior expectation of that space is over-specified and includes fine details. Meanwhile, neurotypical people benefit from a prior that corresponds to an approximate, generalisable representation of the space.
These interpretations of autism, based on a Bayesian model of the fundamental workings of the brain, require a great deal more investigation before their value can be estimated. The Bayesian theories of the brain in general are relatively untested, and most evidence so far has focused on basic perceptual processes and has limited relevance to higher functions. Moreover, experimental tasks, while they may be coherent with a predictive coding interpretation, rarely if ever provide a robust test of this theory. In the case of autism, it would be encouraging if a Bayesian account could make a completely novel prediction about task performance in autism, rather than delivering new interpretations of existing patterns in the literature.
6. Information processing and the social domain
Just as social cognitive models of autism can be critiqued for failing to account for features outside the social domain, it is an open question whether the theoretical models of autism reviewed in this chapter can also explain the presence of specific social differences. One basis on which we might link perceptual and social differences is to characterise social interactive experiences as requiring holistic processing, integrated across multiple input streams and without a systematic, rule-based structure. If this holds, we could argue that problems in the social domain are based on disruptions to the specific type of processing required to capture the social world. Some evidence which aligns with this interpretation comes from work arguing that atypical social attention is most apparent when stimuli are complex (Birmingham et al., 2012).
On the other hand, there is no agreed, evidence-based way to determine objectively the complexity of a stimulus. Additionally, human brains have evolved for certain types of task; we find easy and effortless what computer systems find hard (e.g. extracting objects in a 3-D scene), and vice versa (e.g. division of very large numbers). Therefore, it remains impossible to verify the statement that the social world is somehow uniquely or even particularly ’complex’ in relation to other types of information, despite the appeal of this notion. Information-processing and perceptual theories are also hard to apply to findings of ’fine cuts’, — autism-specific differences between social and closely matched non-social tasks. For example, can domain-general processing differences explain why so many children with autism can conjure a meta- representation in relation to an out-of-date photograph, but that the same task in relation to an out-of-date belief proves impossible?
7. Current debates
A number of theoretical models move beyond the social domain in an attempt to describe non-social diagnostic features and experimental findings, including autistic strengths. However, these efforts to capture the unique information processing styles of autistic people have not yet provided a coherent account that can encompass the wide variability in findings. A number of models have been proposed which are relatively similar in their predictions and interpretations (e.g. weak central coherence versus enhanced perceptual functioning). In other cases, separate theories are not mutually exclusive: e.g. differences in predictive coding might give rise to detail focus which is then manifest as a preference for systemising. The area still suffers from a reliance on small samples in individual experimental studies with too little replication and large-sample work, making it even harder to distinguish between competing accounts or derive practical lessons from them.
What is the relation between the social features of autism and the basic information processing style which (presumably, theoretically) governs all incoming data? Have autistic people just been subject to a mysterious, coincidental combination of differences in both social interaction and information processing styles? Why should these two things have co-occurred in so many? Or are there cohorts of un-identified people who experience each of these broad symptom domains independently?
How can these accounts, or any accounts, explain the sensory hyper- and hypo-sensitivities apparent in autism? Bayesian models look promising in this regard, but as yet no robust experimental tests have been developed, and research pertaining to autism specifically is in its infancy.
Is it necessary to arrive at a single, accepted account of the information processing style of autistic people? Might we simply allow all of the theoretical accounts described earlier to co-exist, each providing a subtly different emphasis or level of interpretation in relation to the shared question of how autistic people (uniquely) take in and process information. This proposal relates to the practical significance of the work covered in this chapter — perhaps it is enough to have evidence-based models that legitimise the differing sensory experiences, perceptual skills and preferences of autistic people, by illustrating aspects of their fundamental underpinnings? Or is there benefit to autistic people to be gained by having a more fine-grained understanding of the precise ways in which autistic information processing differs from the mainstream norm?
COMMUNITY CONTRIBUTION, JON ADAMS — ARTIST AND AUTISTIC ADVOCATE
I’ve always asked questions; ever since I can remember, I’ve asked about the world around me. I picked up a stone aged five on a holiday in Wales and on seeing fossils asked what they were. This one act set me on a quest for all the rest of my life. I had to understand the world below me, around me and above. Now I can see that this didn’t necessarily mean socially. I had a thirst for detail but was also equally compelled to see where this fitted on the global scale, be it in ’time’ or positioning. I was always asking “Where did this fit?”, I needed the bigger picture as the detail without this reference was useless.
For example, the fossils I’d found had a certain definable taxonomy and age. I needed to know what they were and their positioning in the geological and evolutionary timeline. I constructed this 3-D larger view quite naturally, made out of these fragments both by reading widely and practical experience. It’s never linear, always a cloud not a spectrum. I seemed to have no problem with time as a concept, partly I think because being synaesthetic, I can reach out and touch it. It wasn’t the same with people, as there seemed to be an unwritten rogue element within their patterning I had to factor in. I couldn’t join them up in the way I could in sciences or history.
When I was seven, reaching deep within me, I said I’d be an artist. I coped at school for a few years until my ability to cover my differences failed. Schooling in the late 1960s was not fun, and I soon fell afoul of the differing demands and systems. I was rescued by my abilities with facts and drawing. Then one of my pictures was torn up by the teacher in front of the class, as I’d spelt my name wrong, and I decided to hide more of myself. I didn’t go to art college but rather university to train to be a palaeontologist. I had the skills to find, describe and recognise fossils’ positioning in the bigger picture of geological time. I could apply these skills to other subjects, including contemporary arts where I now create and direct socially engaged art projects.
I was apprehensive in the run up to my autism diagnosis as I didn’t seem to fit criteria around detail/bigger picture as my lived experience seemed to contradict. Yes, I systemise well, but I feel what if maybe the current theories don’t wholly fit ’our bigger picture’ rather than autistic people fitting within them?
Gerland, G. (2003). A real person: Life on the outside. London: Souvenir Press.
Happé, F., & Frith, U. (2006). The weak coherence account: Detail-focused cognitive style in autism spectrum disorders. Journal of Autism and Developmental Disorders, 36, 5—25.
Murray, D., Lesser, M., & Lawson, W. (2005). Attention, monotropism and the diagnostic criteria for autism. Autism, 9(2), 139—156.
Pellicano, E., & Burr, D. (2012). When the world becomes ’too real’: A Bayesian explanation of autistic perception. Trends in Cognitive Sciences, 16(10), 504—510.