In Defence of Illusions: A Reply to Braddick (2018)

The curious Lenin-inspired clickbait title of Braddick’s (2018) editorial provocatively compares illusion research to infantile disorders (and thus researchers of illusions to disordered infants). Although its stated target is the recent Oxford Compendium of Visual Illusions (Shapiro & Todorović, 2017), the editorial does not contain a review of the Compendium itself, nor does it mention its contents at all. Instead, it expresses the hope that a potential reviewer will address the question ‘is this a topic for a serious scientific publication?’. As if in reply, an actual reviewer wrote that ‘the study of illusions is a key part’ of vision studies (Rose, 2018). This assessment is in stark contrast to Braddick’s claim that illusions are ‘a category that is deeply unhelpful for science’. Nevertheless, Braddick does not want ‘to dismiss any particular “visual illusion” as a valuable subject for study.’ So how can illusions be valuable individually but unhelpful as a category? Why does the sum of potentially worthy parts not result in a scientifically serious whole? According to the editorial, it is OK if you confine particular illusion studies within separate domains of perception research, such as in ‘colour vision and spatial vision, rather than together’; but if you categorize them all as ‘illusions’, you face a terrible quandary: ‘what have we gained by putting them in the same box?’

The Compendium is indeed such an illusion ‘box’, and what we have gained is a publication similar to an Encyclopaedia or related reference work: Interested readers consult them to obtain concise, authoritative and up-to-date overviews about various special topics, written by experts in those topics. Thus, if you wish to know more about a phenomenon that is commonly labelled as a ‘visual illusion’, chances are that you might find relevant information somewhere among the 115 well-illustrated chapters on particular subjects, (co-)authored by 138 visual scientists. What if you don’t find the general category ‘visual illusion’ appropriate? Just skip the front page which carries that phrase and enjoy the remaining 800+ large-format pages dedicated to particular subjects, many in full colour. Along the way you may risk some ‘Wow-experiences’, but rest assured, no contribution commits the sin admonished against in the editorial, of just stopping at the ‘Wow’ like some first-year students. Whatever or whoever might be the object of the editorial’s wrath, the Compendium does not deserve Lenin’s alleged ‘scornful epithet’. ¹

The claim that we have gained nothing by using the notion of ‘illusions’ is illustrated in the editorial as follows: ‘If we had a good mechanistic theory of the Muller-Lyer illusion . . . would it be any help in deciding how colour contrast works?’ To answer this rhetorical question: probably not. But why is that relevant? Is there a general rule of which this is a particular case? Is the scientific legitimacy of a concept or field of study invalidated if you are able to point to two arbitrarily selected instances which may happen to lack some common features? Hopefully not, because this would be bad news for the legitimacy of almost any larger topic, including those mentioned by Braddick: for example, understanding one colour vision phenomenon may not necessarily help us understand how another colour vision phenomenon works; uncovering the mechanism underlying one spatial vision effect may not help us crack another; finding the cure of one infantile disorder may be useless to cure another; and so on.

Furthermore, insights from one instance or class of illusions may in some cases actually be ‘deeply helpful’ for thinking about other instances or classes. In other words, comparison of illusory phenomena that belong to different domains can, in principle, be a fruitful endeavour. For example, although colour contrast (or simultaneous lightness contrast) may indeed have little to do with a spatial effect such as the Müller-Lyer illusion, it still might have something in common with another spatial effect, the Ebbinghaus illusion: Somewhat simplifying, both effects involve central targets surrounded by opposite contexts, such that the contexts have opposite effects on the perception of the targets. This structural similarity of illusory effects may motivate the search for similarly structured neural mechanisms underlying them both; now, this particular idea may or may not turn out to be appropriate, but that’s a different issue. Or, illusions such as colour afterimages, motion afterimages and face aftereffects, which belong to different domains, also exhibit abstract across-domain similarities and could be accounted for on the basis of same general principles, such as neural fatigue and gradual recovery; or, it has been claimed by various researchers that several classical visual illusions may all be based on treating 2D displays as projections of 3D scenes; or, the apparent existence of the Müller-Lyer illusion not only in the visual but also in the tactual sense may uncover across-modality similarities; and so on. Again, regardless whether any of these approaches turn out to be adequate, the point is that they are all examples of cases in which potential general regularities would be missed if one were to think of illusions only as phenomena confined to single domains.

And what about the claim that the notion of illusions ‘is not a unifying principle that gets us anywhere’? Well, what are those unifying principles that do get us somewhere? And why exactly are illusions disqualified from this role? There is little in the current editorial to help answer such questions. However, here is a hint from a previous paper:

If our concern is to understand the mechanisms of perception – mechanisms which are presumably acting in the veridical just as much as in the non-veridical case – then the significant subdivisions of this subject are surely functional, rather than between the strikingly non-veridical cases and the rest. (Braddick, 1972)

Now, mechanisms are certainly paramount, but not all perception is about mechanisms. It is indeed likely that subdividing visual phenomena according to veridicality would generate a different classification than subdividing them according to function (though this notion would need to be elaborated); furthermore, the latter classification (if it existed) could be more conducive for understanding mechanisms, many of which are speculative. But if one classification is useful for some purposes, it doesn’t mean that another classification cannot be useful for other purposes! The contrast between veridicality and illusions, regardless how difficult it may be to define adequately (Todorović, 2016), arises rather naturally when we realize that whereas in most cases, our senses seem to serve as very sophisticated and precise guides through complex environments, in some cases, they appear to stumble stupidly when faced with some rather crude and simple stimulus constellations. We need a word to express this notion, and we have a good one: ‘illusion’; the terms ‘phenomenon’ or ‘effect’ do not express the concept adequately. There is indeed an underlying ‘unifying principle’ here, but it is epistemological (concerning perceptual success) rather than mechanistic (concerning perceptual process).

As another example of such a notion, consider the question of how it is that in spite of variable proximal stimulations we are able to recover constant distal features of objects, such as their size, shape or reflectance? There is a well-known word for that too: ‘visual constancy’. But just like illusions, constancies also arise in many different perceptual domains, and there is no reason to suppose that they all share the same mechanisms (Todorović, 2002). Does it follow then that constancies should go down the same ‘deeply unhelpful’, ‘not getting us anywhere’, ‘what have we gained by putting them in the same box’ drain that illusions are threatened with? As an example of the usefulness of the concepts and terms for constancies and illusions, note that without them we could not even conceive or formulate some intriguing general hypotheses, such as that illusions arise from misapplications of constancy mechanisms.

Putative perceptual mechanisms will come and go, but illusions are here to stay. Charged with pointlessness, irrelevance and even non-existence, they will nevertheless continue to provide both innocent infantile pleasures and fruitful headaches for adults for a long time to come.