Introspection Explained

by rsbakker

So I couldn’t get past the first paper in Thomas Metzinger’s excellent Open MIND offering without having to work up a long-winded blog post! Tim Bayne’s “Introspective Insecurity” offers a critique of Eric Schwitzgebel’s Perplexities of Consciousness, which is my runaway favourite book on introspection (and consciousness, for that matter). This alone might have sparked me to write a rebuttal, but what I find most extraordinary about the case Bayne lays out against introspective skepticism is the way it directly implicates Blind Brain Theory. His  defence of introspective optimism, I want to show, actually vindicates an even more radical form of pessimism than the one he hopes to domesticate.

In the article, Bayne divides the philosophical field into two general camps, the introspective optimists, who think introspection provides reliable access to conscious experience, and introspective pessimists, who do not. Recent years have witnessed a sea change in philosophy of mind circles (one due in no small part to Schwitzgebel’s amiable assassination of assumptions). The case against introspective reliability has grown so prodigious that what Bayne now terms ‘optimism’–introspection as a possible source of metaphysically reliable information regarding the mental/phenomenal–would have been considered rank introspective pessimism not so long ago. The Cartesian presumption of ‘self-transparency’ (as Carruthers calls it in his excellent The Opacity of Mind) has died a sudden death at the hands of cognitive science.

Bayne identifies himself as one of these new optimists. What introspection needs, he claims, is a balanced account, one sensitive to the vulnerabilities of both positions. Where proponents of optimism have difficulty accounting for introspective error, proponents of pessimism have difficulty accounting for introspective success. Whatever it amounts to, introspection is characterized by perplexing failures and thoughtless successes. As he writes in his response piece,  “The epistemology of introspection is that it is not flat but contains peaks of epistemic security alongside troughs of epistemic insecurity” (“Introspection and Intuition,” 1). Since any final theory of introspection will have to account for this mixed ‘epistemic profile,’ Bayne suggests that it provides a useful speculative constraint, a way to sort the metacognitive wheat from the chaff.

According to Bayne, introspective optimists motivate their faith in the deliverances of introspection on the basis of two different arguments: the Phenomenological Argument and the Conceptual Argument. He restricts his presentation of the phenomenological argument to a single quote from Brie Gertler’s “Renewed Acquaintance,” which he takes as representative of his own introspective sympathies. As Gertler writes of the experience of pinching oneself:

When I try this, I find it nearly impossible to doubt that my experience has a certain phenomenal quality—the phenomenal quality it epistemically seems to me to have, when I focus my attention on the experience. Since this is so difficult to doubt, my grasp of the phenomenal property seems not to derive from background assumptions that I could suspend: e.g., that the experience is caused by an act of pinching. It seems to derive entirely from the experience itself. If that is correct, my judgment registering the relevant aspect of how things epistemically seem to me (this phenomenal property is instantiated) is directly tied to the phenomenal reality that is its truthmaker. “Renewed Acquaintance,” Introspection and Consciousness, 111.

When attending to a given experience, it seems indubitable that the experience itself has distinctive qualities that allow us to categorize it in ways unique to first-person introspective, as opposed to third-person sensory, access. But if we agree that the phenomenal experience—as opposed to the object of experience—drives our understanding of that experience, then we agree that the phenomenal experience is what makes our introspective understanding true. “Introspection,” Bayne writes, “seems not merely to provide one with information about one’s experiences, it seems also to ‘say’ something about the quality of that information” (4). Introspection doesn’t just deliver information, it somehow represents these deliverances as true.

Of course, this doesn’t make them true: we need to trust introspection before we can trust our (introspective) feeling of introspective truth. Or do we? Bayne replies:

it seems to me not implausible to suppose that introspection could bear witness to its own epistemic credentials. After all, perceptual experience often contains clues about its epistemic status. Vision doesn’t just provide information about the objects and properties present in our immediate environment, it also contains information about the robustness of that information. Sometimes vision presents its take on the world as having only low-grade quality, as when objects are seen as blurry and indistinct or as surrounded by haze and fog. At other times visual experience represents itself as a highly trustworthy source of information about the world, such as when one takes oneself to have a clear and unobstructed view of the objects before one. In short, it seems not implausible to suppose that vision—and perceptual experience more generally—often contains clues about its own evidential value. As far as I can see there is no reason to dismiss the possibility that what holds of visual experience might also hold true of introspection: acts of introspection might contain within themselves information about the degree to which their content ought to be trusted. 5

Vision is replete with what might be called ‘information information,’ features that indicate the reliability of the information available. Darkness, for instance, is a great example, insofar as it provides visual information to the effect that visual information is missing. Our every glance is marbled with what might be called ‘more than meets the eye’ indicators. As we shall, this analogy to vision will come back and haunt Bayne’s thesis. The thing to keep in mind is the fact that the cognition of missing information requires more information. For the nonce, however, his claim is modest enough to acknowledge his point: as it stands, we cannot rule out the possibility that introspection, like exospection, reliably indicates its own reliability. As such, the door to introspective optimism remains open.

Here we see the ‘foot-in-the-door strategy’ that Bayne adopts throughout the article, where his intent isn’t so much to decisively warrant introspective optimism as it is to point out and elucidate the ways that introspective pessimism cannot decisively close the door on introspection.

The conceptual motivation for introspective optimism turns on the necessity of epistemic access implied in the very concept of ‘what is it likeness.’ The only way for something to be ‘like something’ is for it to like something for somebody. “[I]f a phenomenal state is a state that there is something it is like to be in,” Bayne writes, “then the subject of that state must have epistemic access to its phenomenal character” (5). Introspection has to be doing some kind of cognitive work, otherwise “[a] state to which the subject had no epistemic access could not make a constitutive contribution to what it was like for that subject to be the subject that it was, and thus it could not qualify as a phenomenal state” (5-6).

The problem with this argument, of course, is that it says little about the epistemic access involved. Apart from some unspecified ability to access information, it really implies very little. Bayne convincingly argues that the capacity to cognize differences, make discriminations, follows from introspective access, even if the capacity to correctly categorize those discriminations does not. And in this respect, it places another foot in the introspective door.

Bayne then moves on to the case motivating pessimism, particularly as Eric presents it in his Perplexities of Consciousness. He mentions the privacy problems that plague scientific attempts to utilize introspective information (Irvine provides a thorough treatment of this in her Consciousness as a Scientific Concept), but since his goal is to secure introspective reliability for philosophical purposes, he bypasses these to consider three kinds of challenges posed by Schwitzgebel in Perplexities, the Dumbfounding, Dissociation, and Introspective Variation Arguments. Once again, he’s careful to state the balanced nature of his aim, the obvious fact that

any comprehensive account of the epistemic landscape of introspection must take both the hard and easy cases into consideration. Arguably, generalizing beyond the obviously easy and hard cases requires an account of what makes the hard cases hard and the easy cases easy. Only once we’ve made some progress with that question will we be in a position to make warranted claims about introspective access to consciousness in general. 8

His charge against Schwitzgebel, then, is that even conceding his examples of local introspective unreliability, we have no reason to generalize from these to the global unreliability of introspection as a philosophical tool. Since this inference from local unreliability to global unreliability is his primary discursive target, Bayne doesn’t so much need to problematize Schwitzgebel’s challenges as to reinterpret—‘quarantine’—their implications.

So in the case of ‘dumbfounding’ (or ‘uncertainty’) arguments, Schwitzgebel reveals the epistemic limitations of introspection via a barrage of what seem to be innocuous questions. Our apparent inability to answer these questions leaves us ‘dumbfounded,’ stranded on a cognitive limit we never knew existed. Bayne’s strategy, accordingly, is to blame the questions, to suggest that dumbfounding, rather than demonstrating any pervasive introspective unreliability, simply reveals that the questions being asked possess no determinate answers. He writes:

Without an account of why certain introspective questions leave us dumbfounded it is difficult to see why pessimism about a particular range of introspective questions should undermine the epistemic credentials of introspection more generally. So even if the threat posed by dumbfounding arguments were able to establish a form of local pessimism, that threat would appear to be easily quarantined. 11

Once again, local problems in introspection do not warrant global conclusions regarding introspective reliability.

Bayne takes a similar tack with Schwitzgebel’s dissociation arguments, examples where our naïve assumptions regarding introspective competence diverge from actual performance. He points out the ambiguity between the reliability of experience and the reliability of introspection: Perhaps we’re accurately introspecting mistaken experiences. If there’s no way to distinguish between these, Bayne, suggests, we’ve made room for introspective optimism. He writes: “If dissociations between a person’s introspective capacities and their first-order capacities can disconfirm their introspective judgments (as the dissociation argument assumes), then associations between a person’s introspective judgments and their first-order capacities ought to confirm them” (12). What makes Schwitzgebel’s examples so striking, he goes on to argue, is precisely that fact that introspective judgments are typically effective.

And when it comes to the introspective variation argument, the claim that the chronic underdetermination that characterizes introspective theoretical disputes attests to introspective incapacity, Bayne once again offers an epistemologically fractionate picture of introspection as a way of blocking any generalization from given instances of introspective failure. He thinks that examples of introspective capacity can be explained away, “[b]ut even if the argument from variation succeeds in establishing a local form of pessimism, it seems to me there is little reason to think that this pessimism generalizes” (14).

Ultimately, the entirety of his case hangs on the epistemologically fractionate nature of introspection. It’s worth noting at this point, that from a cognitive scientific point of view, the fractionate nature of introspection is all but guaranteed. Just think of the mad difference between Plato’s simple aviary, the famous metaphor he offers for memory in the Theaetetus, and the imposing complexity of memory as we understand it today. I raise this ‘mad difference’ for two reasons. First, it implies that any scientific understanding of introspection is bound to radically complicate our present understanding. Second, and even more importantly, it evidences the degree to which introspection is blind, not only to the fractionate complexity of memory, but to its own fractionate complexity as well.

For Bayne to suggest that introspection is fractionate, in other words, is for him to claim that introspection is almost entirely blind to its own nature (much as it is to the nature of memory). To the extent that Bayne has to argue the fractionate nature of introspection, we can conclude that introspection is not only blind to its own fractionate nature, it is also blind to the fact of this blindness. It is in this sense that we can assert that introspection neglects its own fractionate nature. The blindness of introspection to introspection is the implication that hangs over his entire case.

In the meantime, having posed an epistemologically plural account of introspection, he’s now on the hook to explain the details. “Why,” he now asks, “might certain types of phenomenal states be elusive in a way that other types of phenomenal states are not?” (15). Bayne does not pretend to possess any definitive answers, but he does hazard one possible wrinkle in the otherwise featureless face of introspection, the 2010 distinction that he and Maja Spener made in “Introspective Humility” between ‘scaffolded’ and ‘freestanding’ introspective judgments. He notes that those introspective judgments that seem to be the most reliable, are those that seem to be ‘scaffolded’ by first-order experiences. These include the most anodyne metacognitive statements we make, where we reference our experiences of things to perspectivally situate them in the world, as in, ‘I see a tree over there.’ Those introspective judgments that seem the least reliable, on the other hand, have no such first-order scaffolding. Rather than piggy-back on first-order perceptual judgments, ‘freestanding’ judgments (the kind philosophers are fond of making) reference our experience of experiencing, as in, ‘My experience has a certain phenomenal quality.’

As that last example (cribbed from the Gertler quote above) makes plain, there’s a sense in which this distinction doesn’t do the philosophical introspective optimist any favours. (Max Engel exploits this consequence to great effect in his Open MIND reply to Bayne’s article, using it to extend pessimism into the intuition debate). But Bayne demurs, admitting that he lacks any substantive account. As it stands, he need only make the case that introspection is fractionate to convincingly block the ‘globalization’ of Schwitzgebel’s pessimism. As he writes:

perhaps the central lesson of this paper is that the epistemic landscape of introspection is far from flat but contains peaks of security alongside troughs of insecurity. Rather than asking whether or not introspective access to the phenomenal character of consciousness is trustworthy, we should perhaps focus on the task of identifying how secure our introspective access to various kinds of phenomenal states is, and why our access to some kinds of phenomenal states appears to be more secure than our access to other kinds of phenomenal states. 16

The general question of whether introspective cognition of conscious experience is possible is premature, he argues, so long as we have no clear idea of where and why introspection works and does not work.

This is where I most agree with Bayne—and where I’m most puzzled. Many things puzzle me about the analytic philosophy of mind, but nothing quite so much as the disinclination to ask what seem to me to be relatively obvious empirical questions.

In nature, accuracy and reliability are expensive achievements, not gifts from above. Short of magic, metacognition requires physical access and physical capacity. (Those who believe introspection is magic—and many do—need only be named magicians). So when it comes to deliberative introspection, what kind of neurobiological access and capacity are we presuming? If everyone agrees that introspection, whatever it amounts to, requires the brain do honest-to-goodness work, then we can begin advancing a number of empirical theses regarding access and capacity, and how we might find these expressed in experience.

So given what we presently know, what kind of metacognitive access and capacity should we expect our beans to possess? Should we, for instance, expect it to rival the resolution and behavioural integration of our environmental capacities? Clearly not. For one, environmental cognition coevolved with behaviour and so has the far greater evolutionary pedigree—by hundreds of millions of years, in fact! As it turns out, reproductive success requires that organisms solve their surroundings, not themselves. So long as environmental challenges are overcome, they can take themselves for granted, neglect their own structure and dynamics. Metacognition, in other words, is an evolutionary luxury. There’s no way of saying how long homo sapiens has enjoyed the particular luxury of deliberative introspection (as an exaptation, the luxury of ‘philosophical reflection’ is no older than recorded history), but even if we grant our base capacity a million year pedigree, we’re still talking about a very young, and very likely crude, system.

Another compelling reason to think metacognition cannot match the dimensionality of environmental cognition lies in the astronomical complexity of its target. As a matter of brute empirical fact, brains simply cannot track themselves the high-dimensional way they track their environments. Thus, once again, ‘Dehaene’s Law,’ the way “[w]e constantly overestimate our awareness—even when we are aware of glaring gaps in our awareness” (Consciousness and the Brain, 79).  The vast resources society is presently expending to cognize the brain attests to the degree to which our brain exceeds its own capacity to cognize in high dimensional terms. However the brain cognizes its own operations, then, it can only do so in a radically low dimensional way. We should expect, in other words, our brains to be relatively insensitive to their own operation—to be blind to themselves.

A third empirical reason to assume that metacognition falls short environmental dimensionality is found in the way it belongs to the very system it tracks, and so lacks the functional independence as well as the passive and active information-seeking opportunities belonging to environmental cognition. The analogy I always like to use here is that of a primatologist sewn into a sack with a troop of chimpanzees versus one tracking them discretely in the field. Metacognition, unlike environmental cognition, is structurally bound to its targets. It cannot move toward some puzzling item—an apple say—peer at it, smell it, touch it, turn it over, crack it open, taste it, scrutinize the components. As embedded, metacognition is restricted to fixed channels of information that it could not possibly identify or source. The brain, you could say, is simply too close to itself to cognize itself as it is.

Viewed empirically, then, we should expect metacognitive access and capacity to be more specialized, more adventitious, and less flexible compared to that of environmental cognition. Given the youth of the system, the complexity of its target, and the proximity of its target, we should expect human metacognition will consist of various kluges, crude heuristics that leverage specific information to solve some specific range of problems. As Gerd Gigerenzer and the Adaptive Behaviour and Cognition Research Group have established, simple heuristics are often far more effective than optimization methods at solving problems. “As the amount of data available to make predictions in an environment shrinks, the advantage of simple heuristics over complex algorithms grows” (Hertwig and Hoffrage, “The Research Agenda,” Simple Heuristics in a Social World, 23). With complicated problems yielding little data, adding parameters to a solution can compound the chances of making mistakes. Low dimensionality, in other words, need not be a bad thing, so long as the information consumed is information enabling the solution of some problem set. This is why evolution so regularly makes use of it.

Given this broad-stroke picture, human metacognition can be likened to a toolbox containing multiple, special-purpose tools, each possessing specific ‘problem-ecologies,’ narrow, but solvable domains that trigger their application frequently and decisively enough to have once assured the tool’s generational selection. The problem with heuristics, of course, lies in the narrowness of their respective domains. If we grant the brain any flexibility in the application of its metacognitive tools, then the potential for heuristic misapplication is always a possibility. If we deny the brain any decisive capacity to cognize these misapplications outside their consequences (if the brain suffers ‘tool agnosia’), then we can assume these misapplications will be indistinguishable from successful applications short of those consequences.

In other words, this picture of human metacognition (which is entirely consistent with contemporary research) provides an elegant (if sobering) recapitulation and explanation of what Bayne calls the ‘epistemic landscape of introspection.’ Metacognition is fractionate because of the heuristic specialization required to decant behaviourally relevant information from the brain. The ‘peaks of security’ correspond to the application of metacognitive heuristics to matching problem-ecologies, while the ‘troughs of insecurity’ correspond to the application of metacognitive heuristics to problem-ecologies they could never hope to solve.

Since those matching problem-ecologies are practical (as we might expect, given the cultural basis of regimented theoretical thinking), it makes sense that practical introspection is quite effective, whereas theoretical introspection, which attempts to intuit the general nature of experience, is anything but. The reason the latter strike us as so convincing—to the point of seeming impossible to doubt, no less—is simply that doubt is expensive: there’s no reason to presume we should happily discover the required error-signalling machinery awaiting any exaptation of our deliberative introspective capacity, let alone one so unsuccessful as philosophy. As I mentioned above, the experience of epistemic insufficiency always requires more information. Sufficiency is the default simply because the system has no way of anticipating novel applications, no decisive way of suddenly flagging information that was entirely sufficient for ancestral problem-ecologies and so required no flagging.

Remember how Bayne offered what I termed ‘information information’ provided by vision as a possible analogue of introspection? Visual experience cues us to the unreliability or absence of information in a number of ways, such as darkness, blurring, faintness, and so on. Why shouldn’t we presume that deliberative introspection likewise flags what can and cannot be trusted? Because deliberative introspection exapts information sufficient for one kind of practical problem-solving (Did I leave my keys in the car? Am I being obnoxious? Did I read the test instructions carefully enough?) for the solution of utterly unprecedented ontological problems. Why should repurposing introspective deliverances in this way renovate the thoughtless assumption of ‘default sufficiency’ belonging to their original purposes?

This is the sense in which Blind Brain Theory, in the course of explaining the epistemic profile of introspection, also explodes Bayne’s case for introspective optimism. By tying the contemplative question of deliberative introspection to the empirical question of the brain’s metacognitive access and capacity, BBT makes plain the exorbitant biological cost of the optimistic case. Exhaustive, reliable intuition of anything involves a long evolutionary history, tractable targets, and flexible information access—that is, all the things that deliberative introspection does not possess.

Does this mean that deliberative introspection is a lost cause, something possessing no theoretical utility whatsoever? Not necessarily. Accidents happen. There’s always a chance that some instance of introspective deliberation could prove valuable in some way. But we should expect such solutions to be both adventitious and local, something that stubbornly resists systematic incorporation into any more global understanding.

But there’s another way, I think, in which deliberative introspection can play a genuine role in theoretical cognition—a way that involves looking at Schwitzgebel’s skeptical project as a constructive, rather than critical, theoretical exercise.

To show what I mean, it’s worth recapitulating one of the quotes Bayne selects from Perplexities of Consciousness for sustained attention:

How much of the scene are you able vividly to visualize at once? Can you keep the image of your chimney vividly in mind at the same time you vividly imagine (or “image”) your front door? Or does the image of your chimney fade as your attention shifts to the door? If there is a focal part of your image, how much detail does it have? How stable is it? Suppose that you are not able to image the entire front of your house with equal clarity at once, does your image gradually fade away towards the periphery, or does it do so abruptly? Is there any imagery at all outside the immediate region of focus? If the image fades gradually away toward the periphery, does one lose colours before shapes? Do the peripheral elements of the image have color at all before you think to assign color to them? Do any parts of the image? If some parts of the image have indeterminate colour before a colour is assigned, how is that indeterminacy experienced—as grey?—or is it not experienced at all? If images fade from the centre and it is not a matter of the color fading, what exactly are the half-faded images like? Perplexities, 36

Questions in general are powerful insofar as they allow us to cognize the yet-to-be-cognized. The slogan feels ancient to me now, but no less important: Questions are how we make ignorance visible, how we become conscious of cognitive incapacity. In effect, then, each and every question in this quote brings to light a specific inability to answer. Granting that this inability indicates either a lack of information access and/or metacognitive incapacity, we can presume these questions enumerate various cognitive dimensions missing from visual imagery. Each question functions as an interrogative ‘ping,’ you could say, showing us another direction that (for many people at least) introspective inquiry cannot go—another missing dimension.

So even though Bayne and Schwitzgebel draw negative conclusions from the ‘dumbfounding’ that generally accompanies these questions, each instance actually tells us something potentially important about the limits of our introspective capacities. If Schwitzgebel had been asking these questions of a painting—Las Meninas, say—then dumbfounding wouldn’t be a problem at all. The information available, given the cognitive capacity possessed, would make answering them relatively straightforward. But even though ‘visual imagery’ is apparently ‘visual’ the same as a painting, the selfsame questions stop us in our tracks. Each question, you could say, closes down a different ‘degree of cognitive freedom,’ reveals how few degrees of cognitive freedom human deliberative introspection possesses for the purposes of solving visual imagery. Not much at all, as it turns out.

Note this is precisely what we should expect on a ‘blind brain’ account. Once again, simply given the developmental and structural obstacles confronting metacognition, it almost certainly consists of an ‘adaptive toolbox’ (to use Gerd Gigerenzer’s phrase), a suite of heuristic devices adapted to solve a restricted set of problems given only low-dimensional information. The brain possesses a fixed set of metacognitive channels available for broadcast, but no real ‘channel channel,’ so that it systematically neglects metacognition’s own fractionate, heuristic structure.

And this clearly seems to be what Schwitzgebel’s interrogative barrage reveals: the low dimensionality of visual imagery (relative to vision), the specialized problem-solving nature of visual imagery, and our profound inability to simply intuit as much. For some mysterious reason we can ask visual questions that for some mysterious reason do not apply to visual imagery. The ability of language to retask cognitive resources for introspective purposes seems to catch the system as a whole by surprise, confronts us with what had been hitherto relegated to neglect. We find ourselves ‘dumbfounded.’

So long as we assume that cognition requires work, we must assume that metacognition trades in low dimensional information to solve specific kinds of problems. To the degree that introspection counts as metacognition, we should expect it to trade in low-dimensional information geared to solve particular kinds of practical problems. We should also expect it to be blind to introspection, to possess neither the access nor the capacity required to intuit its own structure. Short of interrogative exercises such as Schwitzgebel’s, deliberative introspection has no inkling of how many degrees of cognitive freedom it possesses in any given context. We have to figure out what information is for what inferentially.

And this provides the basis for a provocative diagnosis of a good many debates in contemporary psychology and philosophy of mind. So for instance, a blind brain account implies that our relation to something like ‘qualia’ is almost certainly one possessing relatively few degrees of cognitive freedom—a simple heuristic. Deliberative introspection neglects this, and at the same time, via questioning, allows other cognitive capacities to consume the low-dimensional information available. ‘Dumbfounding’ often follows—what the ancient Greeks liked to call, thaumazein. The practically minded, sniffing a practical dead end, turn away, but the philosopher famously persists, mulling the questions, becoming accustomed to them, chasing this or that inkling, borrowing many others, all of which, given the absence of any real information information, cannot but suffer from some kind of ‘only game in town effect’ upon reflection. The dumbfounding boundary is trammelled to the point of imperceptibility, and neglect is confused with degrees of cognitive freedom that simply do not exist. We assume that a quale is something like an apple—we confuse a low-dimensional cognitive relationship with a high-dimensional one. What is obviously specialized, low-dimensional information becomes, for a good number of philosophers at least, a special ‘immediately self-evident’ order of reality.

Is this Adamic story really that implausible? After all, something has to explain our perpetual inability to even formulate the problem of our nature, let alone solve it. Blind Brain Theory, I would argue, offers a parsimonious and comprehensive way to extricate ourselves from the traditional mire. Not only does it explain Bayne’s ‘epistemic profile of introspection,’ it explains why this profile took so long to uncover. By reinterpreting the significance of Schwitzgebel’s ‘dumbfounding’ methods, it raises the possibility of ‘Interrogative Introspection’ as a scientific tool. And lastly, it suggests the problems that neglect foists on introspection can be generalized, that much of our inability to cognize ourselves turns on the cognitive short cuts evolution had to use to assure we could cognize ourselves at all.