Monday, 4 May 2015

Is Autism a Deficit in Invariance Detection?

If ASD is a problem detecting invariants, the world would remain a 'blooming buzzing confusion' and lead to the behaviours we see in children with ASD, claims a new paper. 
A new paper in Frontiers in Psychology (Hellendoorn, Wijnroks & Leseman, 2015) has proposed that autistic spectrum disorders might be the developmental consequence of a low level, domain general perceptual deficit, specifically the detection of invariants. They explicitly ground this hypothesis in Gibson's ecological approach and theories of embodied cognition that emphasise the key role perception plays in behaviour. This seemed like something I should evaluate, so thanks to Jon Brock for sending this my way on Twitter.

While I am very sympathetic to the basic idea, this particular implementation is too flawed to get off the ground. The authors make a critical conceptual confusion. They mix up invariant features of the world with invariant features of perceptual arrays that might serve as information for the world, and this stops the paper in it's tracks. I think an interesting exercise might be to fix this problem and then simply repeat the paper with the more careful grounding to see where you end up. 

In this post I've briefly reviewed the claims in the order in which they came up in the paper. I've focused my attention on the central hypothesis about invariant detection because that underpins everything else. I've also briefly summarised some of the cited evidence and implications as laid out by the authors, and commented on any issues I saw. This bit is briefer, because my knowledge of the specifics of ASD are limited. I am also considering a comment to Frontiers on this paper, so feedback on this welcome. If you want in on a comment or reworking of the paper, let me know!

The hypothesis (scattered over page 2) is this:
Our central hypothesis is that children with ASD are born with a deficit in detecting invariant structures and that differences in the ability to detect invariants might explain the heterogeneity in the ASD phenotype....Central to our hypothesis is that we propose that from birth on infants with ASD experience difficulties in detecting invariants and as a result the core symptoms of ASD will gradually emerge. We propose that the core symptoms of ASD emerge after many experiences with the environment during which even simple invariants and affordances may not be detected or very slowly....As outlined above, a deficit or delay in invariance and affordance detection at birth in ASD, may lead to a cascade of consequences. Our hypothesis suggests that it will lead to various degrees of impairments in every developmental domain, from sensorimotor development to language and social development. This hypothesis may explain the wide variety of ASD impairments in different developmental domains instead of being limited to only on a subset of behaviors. Moreover, we suggest that the severity of autistic symptoms is determined by the degree of impairment of this particular learning process. That is, the less well the process is functioning, the more severe the symptoms will be and vice versa.
I am in principle very sympathetic to this approach to something like ASD. Perception is at the root of all cognition and behaviour, and variation in perceptual access to the world will lead to people living very different lives. For example, there are also theories about visual perception differences leading to schizophrenia. The effects of this kind of deficit would indeed be widespread and develop over time, two hallmarks of ASD that current theories struggle with. So, how well does it go?

Invariance, but where?

The authors run into immediate trouble, unfortunately, by losing track entirely over whether they mean invariance in information or invariance in the environment. An example:
Invariance detection means selective attention to the relatively stable patterns or structural regularities in the changing stimulus array (Gibson and Pick, 2000; Gogate and Hollich, 2010). Invariance detection helps people to perceive and experience order in their surrounding world which has changing and invariant properties. An example of such an invariant is constancy of size.
pg 2 
The first two sentences are about information; higher order relations in perceptual arrays that remain invariant-over-transformation (to use Gibson's phrase). These invariants are the kinds of thing that can a) serve as targets for perceptual learning, because they hang around and b) be specific to properties of the environment. So these invariants are both useful and learnable; all good. 

The last sentence is about the world. Size constancy refers to the fact that an object remains the same actual size despite variation in it's optical size as it and the observer move relative to one another, and that we perceive this constancy. It is a fact of the environment to be specified. EcoPsych101: keep the world (dynamics) and information (kinematics) separate in your analysis, and, in general, the word invariant always refers to information.

This paragraph is just an example of the confusion that runs through the paper, and this is a problem. Their hypothesis is actually that individuals develop ASD because of difficulty detecting invariant information, and a lot of what they discuss is indeed perceptual. But this underlying conceptual confusion hurts their argument and must be fixed before this theory can guide research. 

Things this deficit might cause problems with
The authors then review some areas of changed function in ASD that might be explainable with an invariance detection deficit. These areas are 
  • Motor functioning and imitation
  • Perception and sensory sensitivities
  • Repetitive and stereotyped behaviours
  • Language and communication
  • Social interaction
Now, I'm a little biased, in that I think perception of invariant structure in energy arrays (information) is the basis of all this stuff, so a deficit in detecting those invariants should impact on these tasks. However, I will flag up that there is not a lot of empirical work that is specifically looking at invariant perception in these tasks in ASD because no one has thought to look yet. The research cited in support, therefore, is not a set of tests of the hypothesis, they are results that can be interpreted as supporting the hypothesis. This would be ok if it wasn't for the fact that their hypothesis has that massive and critical confusion about where the relevant invariance is; the motor functioning section is riddled with confusion, for example. This problem is obviously fixable by tidying the hypothesis, but it's going to remain a problem for this particular paper. 

Neurobiological mechanisms of this deficit
This section has a nice review of a few relevant papers and findings. The basic result is that 'invariance detection' is not something that happens in a particular location in the brain and so it's not simple a matter of looking for localised structural of functional differences. They therefore look to work focused more on connectivity and synchrony, which at least seems like a sensible place to go. They also of course mention mirror neurons. Again this section is fairly speculative but it is at least a little more focused on perception.

Developmental cascades and severity
They point out an important fact about ASD, namely that it is a developmental disorder. It develops over time and therefore is unlikely to just be the result of one broken thing in charge of, say, social interactions. There is also variation in the severity of the disorder (it's a spectrum) and in combination, this supports the idea that the root cause is a low level problem whose consequences show up as a cascade of knock-on effects. This is all well aligned with a developmental, dynamical systems style approach (e.g. Thelen & Smith, 1994). 

Relationship to other theories of ASD
There are three other main theories of ASD; Theory of Mind, executive dysfunction and weak central coherence. These are all fundamentally cognitive theories so problems getting to any kind of meaning are explained as deficits in the internal construction of meaning. This theory, grounded (in principle) in Gibson assumes meaning is perceived, not constructed. There are apparently also some more perceptual hypotheses but the main difference is that this theory proposes a specific mechanism that would lead to differences, rather than just a way of describing perceptual problems. 

There is also some debate about whether ASD is a single syndrome, or whether it can be split out into problems with social communication and restricted/repetitive behaviours and interests (the opaquely named 'fractionable triad' hypothesis; Happé & Ronald, 2008). If this were the case then this invariance detection hypothesis would have problems; the authors review some of the contradictory findings just to point out this claim is not set in stone. 

Is this specific to ASD?
The authors briefly discuss that an invariant detection deficit might lead to other effects, but are very cautious because there's just no data.

Testing and applying the hypothesis
This section should really have been longer. They propose one idea for testing the hypothesis of a domain-general perceptual learning deficit in ASD: category learning tasks in which people classify novel abstract shapes into categories. It is very unclear how this taps invariant detection, although the various features (novelty, abstractness, etc) supposedly rule out the other explanations of ASD. 

Weirdly, at no point do they avail themselves of the experimental literature in ecological psychology; nothing about biological motion perception (I know of work in Aberdeen looking at this in ASD and finding interesting differences) or any action tasks known to be underpinned by the perception of invariant variables in perceptual arrays (e.g. coordinated rhythmic movement). 

The authors suggest some implications for their hypothesis, namely that ASD individuals may benefit from structured environments that would help bootstrap their learning of invariant relations (something known to work quite well, although never quite described in these terms. They also suggest deep brain stimulation and transcranial magnetic stimulation might be useful for various specific things associated with ASD.

I think this idea is viable and worth exploring, but not as laid out in this paper. Their argument is currently grounded in a confusion and this will simply never lead to anything useful. As I say, there might be mileage in simply fixing the problem and repeating the analysis. But there are other weaknesses as well; there should be more detailed predictions made that could guide future work, and I'd like to see some consideration of any evidence from the ASD literature that this hypothesis can't, at first blush, account for. Those effects and findings would then be the thing to focus your initial attention on, to see if you can generate a coherent ecological explanation given the time and space to do so. 

If the authors, or any other ASD researchers were interested in trying to do this work, drop me a line. I think getting this right would be valuable, regardless of whether or not it worked out. 

Hellendoorn A, Wijnroks L and Leseman PPM (2015) Unraveling the nature of autism: finding order amid change. Frontiers in Psychology 6:359. doi: 10.3389/fpsyg.2015.00359


  1. Thanks for this Andrew. Really useful.

    I must confess to becoming a little bored with all these new "theories of autism". Autism is so big and messy, and there are so many contradictory findings, it's easy to find studies to back up any story you want to tell.

    The difficulty for any such theory is that it has to account for both the broad scope of autism and the huge variability within autism. For example, it needs to explain why many people with autism have motor problems, but it also needs to allow for the fact that some people with autism have very good motor skills. Same goes for perception. Same goes for language. And I think it's pretty much impossible to do both without a LOT of handwaving.

    This "Invariance" account reminds me a lot of the Bayesian and Predictive coding accounts that are very much flavour of the month. Like these accounts, I wonder whether the Invariance account might be more usefully thought of as a framework for investigating the symptoms associated with autism (and the individual variation, and the overlap with other diagnoses) rather than an explanation of autism itself.

    1. I feel your pain on the 'theory of the month' problem. ASD seems particularly prone to this!

      I will say that I do think difficulties perceiving ecological information is the kind of the problem that could have repercussions that range this widely. But I take the point that this particular account doesn't give you the kind of explanatory power you need to say why person X has set Y of effects and someone else has set Z.

      Do you think it's worth commenting at Frontiers on? Or is this just another grand unified theory of autism that will fade away on it's own?

    2. Jon,
      If it is any consolation, a "theory" of this type, if done properly would not be a theory about what causes autism in any sense. It would, instead, be an improved description of the problem awkwardly labeled "autism."

      It is almost certainly worth a comment. Among other things, the hypothesis is way too broad, as most any "psychological" or "behavioral" disorder can be defined as a deficit in properly responding to certain invariants.

      If we are talking about autism in particular...I might add, in comment, that most of the analysis should flow smoothly from my series of papers on Eco Psych and Social Psychology. Either there are real things the individual could be detecting and responding to, or their are not. If there really are higher order invariants out there at a "social" level, just waiting to be detected, it should be obvious that some people will be better at doing that than others. It should also be obvious that some people could have broad deficits in detecting them.

  2. How cool that somebody is actually running with this idea! My friend and I have had some similar ideas. We were discussing the social aspect of ASD (we are both critical of the whole “theory of mind” concept, so we were thinking about interactive alternatives) and came up with a motivational explanation built on a similar principle as the paper you reviewed. The idea was that people on the spectrum have a difficult time managing suboptimal interactive contingencies, which if I understand Gibsonian information, has a similar connotation to detecting informational invariances. The usual way to think about motivation is with some kind of drive model, which assumes stasis as a default and tries to explain why an organism does anything at all. But we know that the thermodynamics of living systems preclude even the possibility of stasis. You can’t not act. So the question shifts to why an organism selects one action rather than another. Since organisms have a (functional) stake in establishing successful, reproducible interactions with their environments, they will tend to avoid interactions that consistently lead to failure and pursue those that lead to success. Obviously, interacting with people involves a greater than usual number of suboptimal contingencies (inadequately specifying information?), and thus would lead to a greater degree of failure in those with that deficit. So people on the spectrum (depending, obviously, on the character of the deficit) would avoid many of the common interactions with people that those not on spectrum take for granted, which, as the authors of the paper you reviewed suggested, results in a cascade of other problems. I think this also explains a lot of the repetitive and savantic behaviors. If your perceptual world is too noisy, discovering portions of an interactive space that reliably lead to success could be quite enjoyable (or at least relieving), especially if it opens up a space interactive competence that can be explored and filled out. This is just like with anyone who finds a new hobby or fun activity, but the intensity and narrow focus seen with some savants might result from inhabiting an otherwise noisy perceptual world.

    I agree with Jon that this account is more a description or a framing of ASD symptoms than an explanation of the disorder. What causes the deficit? I think, though, that if you can figure out what causes the deficit, you might find that different manifestations of that cause will result in different kinds of deficits, some more specialized and some more general.

  3. I will say that I do think difficulties perceiving ecological information is the kind of the problem that could have repercussions that range this widely.Friv 10000