Abstract
Reviewed by: David Rose, School of Psychology, University of Surrey, Guildford, UK
There is an old joke about a traveller who asks for directions and is told ‘Well, if I wanted to go there, I wouldn’t start from here’. So too with any journey: If you want to understand visual illusions, should you start from illusions or from an overall theory of perception? Which illusion will tell you most about vision, or which theory will best illuminate illusions?
The two conceptual portions of this book represent these complementary (and, hopefully, mutually converging) alternatives. Thus, in their Introduction, the Editors say the first 10 chapters delve into foundational principles such as the nature and definition of illusions and what if anything unites and explains them all (Part I, a full 20% of the book). They briefly raise the questions of realism in perception and of the philosophy of science (such as how observations always seem to induce multiple explanations based on differing principles, leading to persisting disagreements and the ignoring of empirical counterexamples). After their overview of the field, its history and currently thriving state, they point out that illusions, while still enigmatic, are important because they guide physiological research – as well as being stimulating in their own right. The remaining 80% of the book (Parts II–XI) comprises 105 brief (3–12 pages) chapters which describe a plethora of specific examples and types of illusion, in some cases with a degree of explanatory investigation and elaboration.
Let’s first consider the foundational section in some depth. Science usually starts by categorizing systematically the phenomena to be explained, and indeed the field of illusion science is clearly well endowed with such ordering schemes – as witness those listed in the historical reviews (by the Editors in their Introduction and by Wade in Chapter 1) and the additional examples presented by Gillam, by Fermüller and by Koenderink in Chapters 4, 6 and 8. These typically incorporate putative explanatory causes or levels at which illusions arise, such as physical or optical, physiological or neural, psychological or cognitive, biological or environmental, informational, mathematical or algorithmic. But classification schemes imply each type or class of illusion may have a different explanation – that is, there is no single explanation for all ‘illusions’. Is that true? What do they have in common that they should all have the same label?
The commonest definition (given explicitly by many authors, or assumed implicitly by most) is as deviations or errors in the perception of physical reality (and distinct from hallucinations or delusions). But fundamental questions then arise, such as what we mean by ‘reality’ (kittens, quarks, colours, space-time, information, subjective experiences and/or minds?), how we know what reality is (e.g., by empirical research, mathematical modelling, introspection and/or a priori reasoning), and what ‘veridical’ perception of reality would, should or could conceivably be like. In consequence, some instead define illusions as discrepancies between perception and our conception or knowledge of physical reality (e.g., the Editors; Gillam, Chapter 4; Bulatov, Chapter 11) or claim that the very concept of ‘illusion’ is meaningless, redundant or unnecessary (Koenderink, Purves et al., Rogers and sometimes Grossberg, Chapters 7–10). Although the Editors admit that no strict, clear-cut definition is possible, this volume shows that most researchers nevertheless use the standard definition with productive effect.
The detailed discussion of the nature of illusions begins in Wade’s Chapter 1, with his excellent, in-depth review of the history of the very concept of illusion. He takes a realist approach (‘most of perception is veridical’, p. 3) and suggests that illusions are deviations or errors in perception of the physical attributes of the stimulus. They are revealed when an object’s properties appear different under different circumstances but that object itself can be assumed not to have changed. Although scientific knowledge about the physical world has grown over time, yet the experienced illusions have remained constant for thousands of years. Whether theories help us interpret illusions or vice versa remains an open question.
(I won’t elaborate on Chapters 2, 3 and 5 here because they do not discuss the definition of illusions per se, but note they cover well the theories, histories and relatively scarce and fragmented empirical studies in their very important fields, i.e., culture, age, gender, animals and actions.)
In Chapter 4, Gillam characterizes illusions as visual effects we do not understand, that are unaffected even if we do have cognitive knowledge about them and (for geometric illusions) as not reflecting even a non-Euclidean map of visual space. They are deviations from perceptual realism that are useful for testing different theories of vision. She then gives an exemplary reasoned critique of the various classes of putative explanation for geometrical illusions, concluding that no single one suffices and ‘Illusions are very difficult to explain’ (p. 72)!
Although Chapter 6 (by Fermüller) is entitled ‘Motion illusions in man and machine’, it is actually about general computational processing and modelling, rather than about mobile robot vision as you might expect. It begins by stating that ‘vision cannot be veridical’ (p. 79) and allows that some illusions may be cognitive. It then takes a constructivist approach to perception which involves scene modelling, the inverse optics problem and Marr’s levels. Motion illusions are suggested to arise at the algorithmic level with errors or biases in the estimation of optic flow parameters. For some illusions, self and eye movements interact with these internal algorithmic mechanisms. The exposition here is quite mathematical (compared to the rest of the book) but presumably will be clear to other computational modellers.
In Chapter 7, Grossberg presents at length yet again his comprehensive set of neuro-computational models of the entire visual system (‘the Grossberg school of modeling’; p. 110) and how their functioning accounts for all classes of illusion. Illusions are defined most generally as unfamiliar combinations of edges and surfaces (which are processed in parallel according to his models). Amodal boundaries are illusions; their completion is necessary so we don’t see the gaps caused by the blind spot or retinal veins (no mention of arteries: ‘veins nourish the eye’, p. 94). Therefore, we can’t tell which parts of visible edges are real or illusory (filled in). (This despite the argument he made on the same page 95 that eye movements compensate for image occlusion by veins.) All object motion perception is illusory; so too afterimages. Our percepts, both apparently real and illusory, are the brain’s optimal interpretation of the stimulus arising from the settling of resonant (iterative bottom-up and top-down) neural circuits. This model has ‘developed Helmholtz’s intuitions about unconscious inferences into rigorous science’ (p. 113). In conclusion, he says both bottom-up and top-down processes can cause illusions.
The philosophy is ontologically realist, yet at times he opines that all perception is illusory, since it is all a construction of the brain. Moreover: ‘percepts … are … illusions that just happen to match the structure of the world’, p. 93 (what a fortunate coincidence for our survival!). Nevertheless, his account of visual brain function could still be considered to lie within the conventional mainstream. Importantly, however, he also regards all borders as invisible (and thus illusory) while consciously perceived colours ‘flow’ or fill-in as surfaces between them, a process controlled by illusory global shapes. This flow sounds to me rather like the ‘mental paint’ of the idealists – but here it is just simply said to occur, in an ontologically unspecified space that is (in some mysterious way) the same for both phenomenal and neural processes. Towards the end of the chapter, the exposition of Grossberg’s admittedly complex, all-encompassing multiple models becomes rushed and jargon-filled (and Figure 7-3 is incorrect, while a simple typo on p. 113 causes a lot of confusion; hint: ‘are’ should read ‘and’). But his distinction between border and filling-in processes is cited favourable several times elsewhere in the book.
Next, Koenderink begins Chapter 8 by agreeing that conventional definitions of illusions involve misperceptions of reality and erroneous representations. He organizes the chapter around two extreme views of perception. The first involves inverse optics, Bayesianism, algorithms, symbols and Marr-style bottom-up information processing to form veridical representations. This extreme, however, he condemns as projective, objective and anthropocentric. It omits meaning and thus leads either to behaviourism, mysticism (meaning is already in the optical data) or magical emergentism (of consciousness). At the opposite extreme, which he clearly favours, is the belief that perception is top-down ‘controlled hallucination’ that uses bottom-up ‘optical structure’ to create information, by a process akin to scientific hypothesis testing or analysis-by-synthesis. But awareness itself is ‘an aspect of reality’ (p. 123) that is beyond physics and science! Koenderink points out that lots of intermediate models are of course possible. It all depends on definitions – but we just can’t agree on these, even on what is meant by ‘physical world’, for example. Successful visuomotor actions indicate near-veridical representations may exist sub-consciously (although he says other explanations are available). The top-down theory says evolution leads to fitness not veridicality, and thus ‘evolution breeds illusion’ (p. 124); either all is illusion or nothing is (though he admits some illusions can be described objectively, p. 124). He then proposes a taxonomy of illusions based on levels of ontology (with subcategories) and lots of nice artworks as examples. In conclusion, illusions can be used as tools to unravel psychoanatomy but not neuroanatomy, since a solution to the mind-body problem remains obscure.
Koenderink usefully defines some terms from phenomenology and the structure of perception that entails: ‘Illusions occur in immediate visual experience following immediate visual awareness. Awareness may be called “presentation” so as to differentiate it from “representation.” … Presentations as such cannot be illusionary because they are beyond true or false. The perception, or visual experience, that immediately follows the presentation … refers to physical reality. Consequently, it may turn out to be nonveridical, in which case one speaks of an “optical illusion.”’ (p. 119). Whether you understand and agree with Koenderink’s concisely argued view depends also on whether you accept his presuppositions (in the continental phenomenological tradition) that conscious experience is meaning, that ‘thoughts … are actions of a person’ (p. 119), his ontological dualism between subjective and objective, his explicit social constructivism (he says that veridicality exists only if other people agree with you, p. 132) and his assumption that veridicality and fitness do not coexist or even correlate.
In an appropriately brief Chapter 9, Purves, Wojtach and Lotto also take a radical antirealist stance, claiming vision’s biological successfulness is not due to veridicality, and ‘the distinction between veridical and illusory is false’ (p. 141). Vision is not about representing objects (though physical objects exist and have influenced evolution to give us ‘behaviourally useful qualities, … [not] correct or incorrect representations of the physical world’ (p. 141). Vision is like pain (which, the authors simply assert, is without representations – as if bodily tissue damage is not part of the physical world). Thus, they conclude, the very concept of illusions is unnecessary. They don’t, however, say why usefulness and representations (even if only roughly veridical) are mutually incompatible, or how adaptive, creative, flexible, planned behaviour in novel, complex, unpredictable environments can occur without the ability to restructure and reorganize general purpose mental models that represent the most likely prior state of the world (see also Chapters 4 and 28; Hickok, 2015; Maniatis, 2015).
In a studiously reasoned Chapter 10, Rogers too challenges the usefulness of distinguishing between veridical and illusory perception. First, he asks what ‘the stimulus’ is, arguing by means of counterexamples that it cannot be physical objects or the retinal image, but (following Gibson) information in the patterns of light. However, what is relevant is not all the information but only that which is selectively extracted for use by our evolved perceptual systems. This gives us ‘for most of the time … more or less correct answers as to what is out there in the world’ (p. 149). The ‘discrepancies’ we normally refer to as illusions do not occur between perception and the stimulus but merely between conflicting or contradictory conclusions derived from different selections of the informational features of the optic array (or between information from different sensory modalities, or between feature analysis and cognitive knowledge). What we call illusions are merely effects we do not understand yet, cases where we do not know how the visual system selects and processes information. When we have ‘a complete and correct description of how our perceptual system works’ (p. 148), then ‘there can be no discrepancies and hence no illusions’ (p. 149).
However, when Rogers says ‘no illusions’ he actually means ‘no category of percepts called illusions’ – a matter of epistemology not ontology. He is eliminating the label not the phenomenon itself. The Müller-Lyer shafts will still look unequal, just as monochromatic red plus green metamers for yellow still look yellow even though (we think) we already know exactly the relevant information processing mechanisms. Overall, I think there are several arguments interwoven in this chapter. One begins with his reasoning that the relevant real stimuli are selected patterns of information. Rogers repeatedly says descriptions of these are ‘equivalent to’ (p. 154) descriptions of the perceptual mechanisms that select and process them. But there also seems to be an ontological identity: a ‘mutuality of information and how our perceptual systems work’ (p. 149). Since the ‘equivalence’ or ‘mutuality’ is perfect there can be no discrepancy between our percepts and the reality we see. (Presumably, that seen reality must be exactly the selected information, or its content or meaning; he specifically excludes qualia from discussion at one point.) A second argument (which can run independently of the first; see also Morgan, 1996) begins by asserting that, although metameric colour matches were traditionally classed as illusions, they have come to be seen as merely ‘an illustration of how the perceptual system works’ (p. 150). By analogy, we could infer, the same will (or at least should) happen for all the other ‘things we loosely call illusions’ (p. 154). Therefore, as perception science progresses, there will be piecemeal replacement of the various sub-categories of illusion (as traditionally understood, i.e., as unexplained discrepancies between perception and reality), until eventually, with perfect knowledge, there will be no more unexplained discrepancies and thus no category of phenomena to call ‘illusions’. That is, the discrepancies will remain but will no longer be unexplained. (And if the first argument is also correct there will be no discrepancies either.) Whether you accept these and Rogers’ other arguments depends whether you accept his development of Gibsonianism, the exclusion of illusion-defining discrepancies as between perception and the distal stimulus (despite his use of ‘what is out there in the world’, p. 149, as the relevant aim of vision), and the downplaying of cognition and qualia (which are only mentioned once each). After all, cognitive instructions to attend to the distal or the proximal stimulus can affect perception (e.g., Gilchrist, 2012), and it is not clear whether our ability to zoom attention in and out counts as part of ‘how the visual system works’ (remembering that Gibson, 1979, explained illusions as the results of processing too much or too little information). It is also dubious whether our explanatory successes will ever catch up with the continual discovery of new illusions – as can be seen in the rest of this book.
The chapters in Part I raise important, deep and interesting issues. However, even together they do not present all points of view (mainly, just those of their authors). While it is valid and important to bring these topics to the fore, there is no claim in this Compendium to present a comprehensive sampling of the many fundamental theories available in the literature on the philosophy of perception and what it means to say we perceive ‘reality’. These are important for the interpretation of illusions – but are outside the direct remit of this volume (e.g., Plato’s cave, Descartes’ demon, Eddington’s table, Dennett’s multiple drafts and so on). Philosophers are currently poking around in this area and hopefully the chapters in Part I will stimulate more of us to engage in such writing in the future.
There is clearly a lot of groundwork to sort out. Some of our confusions arise from the variety of implicit or covert definitions we assume, not merely of ‘illusion’ and ‘reality’ but also of the boundaries between ‘sensation’, ‘perception’, ‘cognition’, ‘awareness’, ‘experience’, ‘(sub)conscious’, and so on. Deadlock in our debates can also arise from the ad hoc cherry-picking of examples and counterexamples from very different categories of perceptual phenomena which may involve very different underlying processes or levels of explanation. It is clear that those who use the traditional definition of ‘illusions’ need to make it more precise, for example, as systematic (to exclude random errors or noise) misperceptions (to exclude delusions) of some of those qualities (to allow selective attention) of the distal stimulus (to exclude hallucinations) that are within the range of our senses (to exclude trivial counterexamples such as sub-threshold and beyond the visible spectrum) – and even that may not cover all possible counterexamples, such as non-linear intensity estimation, or perception under pathological conditions or mutations, which some people (you know who you are) may count as ‘illusions’. For importantly, illusory experiences are not arbitrary. Their deviations from ‘veridical’ constitute systematic distortions, which are not random, chaotic or meaningless, and provide useful clues to both mechanisms and functions. Moreover, even if all perception is a ‘grand’ illusion (or set of illusions – or hallucinations, metamers, etc.; e.g., Noë, 2002), this does not disprove the idea that veridical perception is the goal or ideal of the visual system – and thus the actual topic for psychology (Millikan, 1993, 2004). And veridical is useful, even if only sufficiently so to provide evolutionary advantage. Even illusions carry some information about external physical reality; they are not random hallucinations or noise. Finally, the unity of ‘illusions’ that justifies their shared title comes from the higher level function they are all imperfectly subserving (whether this is veridical representation, fitness, motor guidance or whatever, e.g., Zavagno, Daneyko, & Actis-Grosso, 2015), rather than from their lower level implementational mechanisms, which include (inevitably) multiple realizations of that function.
Despite the doubts expressed by some authors in Part I about the word ‘illusion’, it is nevertheless in universal use. (For example, even some of these authors use it freely in the titles of their own subsequent chapters in the volume.) The enormous number of illusions (however defined) outlined in the 105 chapters of Parts II to XI give an eclectic overview of most of the recent work in the field. Thus, illusions seem to be ubiquitous in all categories of experience, including (to use the Part headings) geometry, brightness/lightness/colour, motion, faces, grouping/organization, attention, binocularity, adaptation, conflicting information, and multisensory perception. These parts vary widely in size, from 2 to 26 chapters, with each chapter focussing in depth on one illusion or class of illusions. Both classical and new cases are included. Some you will almost certainly not have encountered before, and some have only been published previously in non-English language journals (e.g., Morikawa, Chapter 26), at conferences (e.g., Lingelbach, Chapter 52) or not at all (e.g., Shapiro, Chapter 112). However, as befits a Compendium, many will already be well known to you from the published literature – although with several having the added benefit here of being accessible in colour and/or as videos (on the website). For importantly, the book is backed up by a website containing high-resolution versions of all of the figures (well, almost all – those for Chapters 60 and 62 and the correct Figure 7-3 haven’t been uploaded yet). There are also videos (some of them interactive) of most of the demonstrations that require a dimension of time or motion to show the effect (though unfortunately over a dozen chapters are lacking in this respect). The site is easy to access and use, although the videos are in various formats so you many need to download extra plugins to watch them all on your browser (and for Chapter 14’s files, I needed to get a new video converter app).
There is a small degree of repetition of coverage, but remarkably all the chapters are interesting and informative. It would be invidious to pick individual highs and lows, but personally, I found the faces section to have many of the most entertaining pictures, Koenderink’s Chapter 8 to have the best artworks and the motion demonstrations on the website to be unmissably vivid and striking. The chapters are (on the whole) readable without talking down. Although computational modelling is (according to Zanker: Chapter 85, p. 602) the aim of vision science – and indeed many quantitative models are described in the book – only a few chapters contain actual equations. On the other hand, many technical terms are used without definition (such as Michelson contrast, gamma function, symmetry breaking, CIE, … ) and there is no Glossary. However, the Index is good, although not comprehensive, lacking some terms of potential interest such as camouflage, learning, aging and eccentricity.
The book would not claim to be all-encompassing, so it may omit some of your favourite illusions (e.g., the flash-lag effect) and a few topics are underdeveloped, reflecting a genuine shortage of research (such as cross-modal influences, action, ontogeny and phylogeny). While only five chapters cite articles published since 2015, the Compendium does nevertheless demonstrate how much progress has been made in studying visual illusions in recent decades (cf. also Pinna & Reeves, 2017): our ‘jolly cottage industry of static and dynamic illusions’ (Zanker, p. 605) is indeed thriving.
For your money, you get a physically massive volume (difficult to prop up while reading; given its price and weight you’d expect a free lectern to be included). However, in these days many, far smaller books are in the £50 to £100 range (hardback), and large handbooks even more, hence its cost is, unfortunately but realistically, commensurate with current market conditions. For the price, one obtains a huge and colourful tome, its large glossy pages bearing copious illustrations in colour or black and white, hence equally at home on the coffee table or in the library. It is a pleasure to browse through as well as to read in detail or to consult as a source book and guide to particular illusions and their academic status. All this should go some way to offsetting the infamy that the book has already generated by its enormous cost and its yukky putrid off-yellow cover.
The Editors have done an impressive job. They have constructed a Compendium of immense scope written to a consistent standard by almost 150 of the leading authorities in visual illusion research. It is comprehensive, eclectic and well-organized. Although inevitably it has some limitations, in practice unavoidable, these certainly do not obviate its scientific value. This volume shows not just that vision is the most advanced of any field of research in its coordination of phenomenal, mathematical and biological aspects but also that the study of illusions is a key part of that process. It is therefore a major contribution to one of the most intellectually fascinating and important areas of enquiry.