Surfaces and Essences

by Douglas Hofstadter

  We Go as Deep as We Can Go

  No, we are not constantly gulled by surface appearances — virtually never, in fact. But defending this viewpoint requires some explanation. We have just seen that the troubling and counterintuitive “fact” of superficiality trumping depth has been experimentally demonstrated only in the limited case of subjects’ domains of incompetence, and this already greatly diminishes the finding’s impact. But we might further ask why it is that novices are so often seduced by surface-level features. Are we humans really so shallow that we are interested only in what glitters? That’s most doubtful. In order to understand these findings, then, we will need to go a bit more deeply into the nature of what is commonly said to be “superficial”. Indeed, so far we have used the term in a rather casual fashion, relying on people’s usual views about its meaning. But now we have to be more specific about what we mean.

  A superficial feature is an aspect of a situation that can be modified without touching the core of that situation. Thus the color of my gearshift — originally black, it was recently repainted yellow — has no effect on how it works as a gearshift. Color is quite obviously a surface-level feature for the category gearshift. Likewise, when it comes to problems on a test, superficial features are those that can be modified without affecting the problem’s goal or the pathways allowing it to be solved. For example, if a new problem has to do with shopping in a mom-and-pop grocery store as opposed to shopping in a supermarket, or with going to the dentist’s office rather than going to the lawyer’s office, its core is unlikely to be affected. On the other hand, when features are crucial to a category — when their modification changes the category itself — then one speaks of structural features. Thus a car whose motor has been removed, or worse, a car that has been compressed into one cubic foot at the wrecking yard, loses its car-ness. In the case of problems to be solved, structural features are those whose alteration would change the goal of the problem or the pathways to solving the problem. They are those features that one needs to pay attention to in order to find the solution, whereas superficial features are those that one can ignore.

  Even if these definitions seem to be totally reasonable, they give rise to a couple of paradoxes concerning the phenomenon of superficiality’s dominance over depth. The first one comes from the fact that, by definition, a novice in a particular domain cannot tell what the essence of a concept in the domain is. In other words, the distinction between surface-level features and deep features doesn’t apply to novices, because to them any trait that they perceive could equally plausibly be shallow or deep. (This brings to mind the case of desks, discussed in Chapter 4, where attributes that once seemed essential to deskness, such as being bulky and covered with papers and having pull-out drawers, were revealed to be superficial only when virtual desks came on the scene.) Given that novices have no clue as to what’s deep and what’s shallow, how could their remindings always involve shallow features? How could it be the case that a person whose cognitive system can’t even distinguish between shallow and deep features would invariably “smell” and be fatally lured by the shallow ones when reminding is involved? This seems perverse. Experiments have shown that surface-level features guide the retrieval of analogies, but the reason behind this cannot be that they are surface-level as opposed to deep features, because this distinction doesn’t exist for the novice. And so it falls to us to explain the findings in some other manner, which we will do below.

  That’s the first paradox associated with the thesis of the dominance of the superficial in memory retrieval. The second is that the thesis amounts to a bizarre principle of cognitive anti-economy. Normally, principles of economy, which describe the efficiency of our behavior under the various and sundry constraints imposed on us by life, are taken to be inviolable. More specifically, if we agree that superficial features are defined by their irrelevance to a situation or to the goal of a problem, then how could a principle of cognitive economy explain the fact that the sources in analogical memory retrieval are selected precisely because of the irrelevance of their features to the given situation? Why would memories be triggered by irrelevant rather than relevant similarities? Again, this seems like a perverse if not fatal strategy for supposedly thinking beings to be guided by.

  Fortunately, the resolution of these two paradoxes is quite simple. Novices have not built up the deeper categories of the domain, hence they don’t perceive them. In other words, it’s not the case that novices perversely and paradoxically favor shallow aspects over deeper ones, systematically snubbing what really counts. Rather, novices try as best they can to recognize what is important and relevant in a new situation, but lacking crucial knowledge, they most often cannot do so, and therefore they have to settle for shallow and most likely irrelevant features. As a result, the analogies they draw to prior situations tend to be shallow rather than deep, but this is only because the deeper analogies are not available to them, given the current state of their conceptual repertoires.

  Rather than positing that memory retrieval — a process as ceaseless and as indispensable for cognition as the beating of one’s heart — is archaic and useless (which is essentially what the above-cited evolutionary interpretation implies), we might more reasonably posit that the mechanism of memory retrieval is, in fact, perfectly efficient, as long as one takes into account certain constraints that we will now clarify.

  The fact that novices are incapable of distinguishing surface-level features from deep ones, that novices have no little bird whispering to them, “This is a relevant feature, but that one is not!”, leads us to adopt a different viewpoint about novices’ memory retrieval. What, then, do novices do? What do we all do? Among all the diverse memories to which we are led by our knowledge (or lack of knowledge) of a given situation, we rely on whichever one is the most salient.

  In short, what we use to guide our retrieval of memories when we are in an unfamiliar situation is not what is most superficial, but what is most salient to us. That is to say, the features that guide our retrieval of a specific memory are chosen not because they reside on the surface, but because — quite to the contrary — among all the potential retrieval cues to which we have access, they are the deepest ones.

  For different people, the salience of a given feature depends on their expertise in the given domain. The research of psychologist Michelene Chi long ago confirmed that categories constructed by experts in a given domain are very different from those constructed by novices, and that experts don’t rely on the same cues as novices do in categorizing new situations. Novices pay attention mostly to superficial aspects, simply because to them, those are the most salient qualities. For example, novices in physics will tend to group together all problems involving pulleys, or all problems involving springs, whereas people who have more experience in physics will tend to group together problems that involve the same physical principles, such as the conservation of energy. Thus an expert, instead of labeling a particular problem “a pulley problem”, as would a typical novice, will call it “an energy-conservation problem”. Why is this? It is because in their attempt to find connections between problems, novices, unaware of the key features to look for, rely on features that, to an expert, seem irrelevant; experts, by contrast, have already built up categories that critically depend on the key features.

  Only as an individual’s knowledge of a particular domain increases can there be a gradual evolution from categorization based on obvious features to categorization based on more abstract ones. When people eventually attain a high enough level of abstract knowledge of the important types of situations that frequently arise in a domain, then they are able to reliably apply this abstract knowledge by analogy to new situations whose façades are very different from the situations that they encountered during their learning phase.

  This explains why a chess novice can easily see surface-level facts in a chess situation (“My queen can take this pawn!”), but will not be able to see deeper things (“This fork is t
hreatened by my opponent’s knight”), for such subtleties, transcending the novice’s perceptual and conceptual repertoire, escape the novice’s view. In order to acquire the ability to see such things, one has to gain expertise. Novices can easily be enticed by the surface-level traits that they perceive in situations, since (at least to their untrained eyes) there is nothing deeper around to lure them away from those traits. The distinction between surface-level and deep does not apply to them, even though it may well exist objectively. This is why we are casting doubt on the psychological relevance of such a distinction for novices. The reason novices are drawn to objectively superficial features is simply that those are the only features they are able to perceive.

  Suppose that a researcher were to give non-speakers of Bengali a hundred poems in Bengali and then asked them to sort the poems into a few sensible categories. The piles that would result would of course be determined by surface-level attributes, such as a poem’s overall length, a poem’s average line-length, whether a poem looks typographically uniform or highly irregular, whether a poem uses indented lines or not, and perhaps a few others, but that’s about as far as it goes, because those sorts of categories — visual or geometric, and unrelated to meaning — are the only ones that are available to the non-Bengali speaking “novice”. By contrast, if the subjects of the experiment were university-educated native speakers of Bengali, we would see extremely different categories emerge. These would have to do with what a poem talks about, the era that it exudes, its use of formal or informal language, its tone and register, its classical illusions or lack thereof, its grammatical complexity or simplicity, its use of colorful language, and on and on. All of these categories are of course utterly invisible to the “novice”.

  There is thus no paradox about the novices’ categorization behavior, because it’s not the case that the novices are perversely opting for shallow categorizations over deep ones; they are simply making the deepest possible new categories they are able to make, given their inability to read a single word of Bengali. Meanwhile the experts, who in theory could see and pay attention to all the same shallow features as the novices are using, are ignoring those features, almost as if they didn’t see them, just the way seasoned drivers ignore the color of the gearshift they are using; moreover, the deeper semantic and stylistic features that guide the native speakers of Bengali jump to their eyes every bit as rapidly and effortlessly as the shallower features jump to the eyes of the non-speakers of Bengali. For an expert in the domain, the deep features are not elusive or hidden; rather, they are the most salient features!

  Let’s also note that this phenomenon does not just depend on knowing the Bengali language. It is plausible that poetry specialists who knew no Bengali whatsoever could, on looking at a sampler of Bengali poems, discover some insightful categories that even an average native speaker of Bengali might never think of. Of course these would have to involve form rather than content, but they could still be of interest to Bengali-speaking poetry lovers. We are thus brought back once again to the idea that the features people tend to notice in any domain are as deep as their perception allows, which is a function of the set of categories they have evolved over their lifetime.

  It’s for this reason that the deepest clues available are what guides a person who is searching for analogies between fresh situations and ones in memory. If a novice is not guided by deeper features, it’s because novice-level knowledge doesn’t afford glimpses of such features. It’s not what is superficial but what is salient that catches one’s attention, and this applies equally to novices and to experts. The difference, however, is that as one gradually acquires greater expertise in any domain, the identifying features of deeper categories gradually grow more salient. This is a genuine principle of cognitive economy, for it states that no matter what one’s level of expertise is, when searching in memory for an analogous case, one goes as deep as one’s expertise allows. And we should point out that when we speak of expertise, we do not have in mind just the narrow and technical kind of expertise possessed by a handful of specialists in some arcane domain, but also the expertise that each of us has picked up concerning our everyday environment. We are all pretty much experts in what surrounds us, and that’s a lucky thing, since it’s exactly the kind of expertise that comes in handiest in life.

  Thus if we (presumably experienced drivers) get into an unfamiliar car to drive it somewhere, we are immediately able to use its gearshift quite efficiently by making analogies to the relevant aspects of other gearshifts we have used before, and we are not deflected from its main functions by such superficial features as its color, its size, its shape, its degree of dirtiness, or where it is located in the car. All those qualities may be seen or felt, but they are instantly abstracted away, leaving as its most salient features the number of gears it has, and where they are located in space relative to each other. Those are the things that matter for driving, and it’s no coincidence that they are also the most salient features to seasoned drivers. For someone to be a seasoned driver means precisely that what matters for driving pops right out to them effortlessly.

  Essences Are Revealed by Surfaces

  Superficial and deep seem to be diametrical opposites. A quick trip to the dictionary will tell you that the words “surface” and “depth” are antonyms, and the concrete imagery associated with these words reinforces this sense of opposition. Specifically, an object’s surface is apparent and accessible; it is where the object meets the rest of the world. Think, for instance, of the surface of the earth, or the surface of a house, or of the ocean, of a stone, of a ball, of a mattress, of a wall, of a body, of a plant. Conversely, the depths of an object are internal, hidden, far from the object’s “skin”. Think of the depths of the earth, of a forest, of a lake, of a house, of a wound, and so forth. If one interprets these words at a more abstract level, their opposition remains every bit as strong; this is clear from their standard usages. Thus, “staying on the surface” means limiting oneself to appearances, to what is immediately obvious; this contrasts starkly with “going into depth”, which means being profound, transcending appearances, getting at the meat of things, seeing beyond the surface, not being distracted by what first meets the eye.

  The adjective “superficial”, which we have often used as a synonym for “surface-level”, reveals that “surface” has a second meaning. Not only does “superficial” refer to what lies on the surface and is thus immediately accessible, but it also frequently expresses a value judgment — indeed, a negative one. To call a book, a film, a piece of music, or a scientific idea “superficial” is clearly to imply that it is just “fluff”, neither deep nor serious. This idea that “superficial” and “deep” are opposites gives surfaces a bad reputation.

  Words, however, are often misleading, and we would suggest that the apparent contrast between surfaces and depths is, in fact, merely a surface-level contrast. Although dictionaries tell us that they are antonyms and our intuition tends to uphold this prejudice, the fact is that this opposition is only apparent. The royal road to the depths of a thing, to its core, to its essence, is precisely what lies at its surface. The surface gives us clues — deep clues — as to what is hidden inside, affording glimpses of the depths at the core. Surface and depth are thus related as are glove and hand, the former constituting the outermost layer of the latter.

  The idea is not difficult to grasp. Psychologist Eleanor Rosch has described how our categories respect the correlational structure of the world. That is, not all combinations of properties are equally likely; rather, certain properties tend to co-occur in our environment. For example, the property of flying is correlated with the properties laying eggs and having a beak. In other words, when we perceive surface-level features, that activates in our minds other features that are correlated with those first features. These secondarily activated features are ones that our experience tells us tend to be present when the first ones are, but in themselves they are not instantly perceptible. I
n this way, what we see on something’s surface leads us to its hidden depths, and thus allows us to draw meaningful, insight-lending analogies with what we have known before.

  Our physiological senses react to surfaces, and our brain uses this input to activate certain categories, which give us clues to the gist, the crux, the core, the essence of what we are dealing with. Psychologist Myriam Bassok has carefully studied this notion of “induced structure”, focusing on how it applies to the way that students learn in schools. Furthermore, the relevance of her findings goes well beyond the educational system; indeed, they apply to the way that we relate to the world around us. Thus a feather is likely to be light; a peach is likely to have a pit; something round stands a fair chance of being able to roll; and kicking something that looks like an anvil is not, in general, highly advisable.

  Psychologist James Gibson proposed the notion of an object’s affordances, meaning the possible actions naturally suggested to a person who perceives the object. Thus a doorknob might tempt us to turn it, a button would cry out to be pushed, a bag would invite us to place things in it or remove things from it, a switch would suggest “Flip me!”, and so on. When an artefact’s perceptual affordances are consistent with the purposes for which it was designed, then the artefact will be easy for people to use. This idea is well known to designers, who use the term “transparent” to designate an artefact whose mode of use is apparent from its design. Thus an important design principle would be that of making an artefact’s instantly available surface reveal its deeper nature. The basic idea is that surface and essence should, ideally, be closely related.

  To be sure, the world is full of traps that lead people to insist that appearances are deceiving and to be very wary of assuming that wherever there’s smoke there’s fire. There are all sorts of ways of being deceived by appearances. Indeed, this theme lies at the core of the notion of fauxthenticity, which we discussed in Chapter 3, and it also gives rise to con games, and even to the false belief, held by teachers who are tired of grading, that they can evaluate their students’ understanding by very superficial rapid-reading techniques. But these are rare cases; in most cases the connection between surface and depth is not misleading. It is a minority of cases that are misleading, although of course the consequences of such cases can be huge. In most situations, the surface-level cues that we pick up quickly furnish a reliable guide to the situation’s essence. This is why we survive and even manage quite well in the world.