Daniel C. Dennett

HOW ARE HALLUCINATIONS POSSIBLE?

1. THE BRAIN IN THE VAT

Suppose evil scientists removed your brain from your body while you slept, and set it up in a life-support system in a vat. Suppose they then set out to trick you into believing that you were not just a brain in a vat, but still up and about, engaging in a normally embodied round of activities in the real world. This old saw, the brain in the vat, is a favorite thought experiment in the toolkit of many philosophers. It is a modern-day version of Descartes's (1641) (Dates in parentheses refer to works listed in the Bibliography.) evil demon, an imagined illusionist bent on tricking Descartes about absolutely everything, including his own existence. But as Descartes observed, even an infinitely powerful evil demon couldn't trick him into thinking he himself existed if he didn't exist: cogito ergo sum, "I think, therefore I am." Philosophers today are less concerned with proving one's own existence as a thinking thing (perhaps because they have decided that Descartes settled that matter quite satisfactorily) and more concerned about what, in principle, we may conclude from our experience about our nature, and about the nature of the world in which we (apparently) live. Might you be nothing but a brain in a vat? Might you have always been just a brain in a vat? If so, could you even conceive of your predicament (let alone confirm it)?

The idea of the brain in the vat is a vivid way of exploring these questions, but I want to put the old saw to another use. I want to use it to uncover some curious facts about hallucinations, which in turn will lead us to the beginnings of a theory an empirical, scientifically respectable theory of human consciousness. In the standard thought experiment, it is obvious that the scientists would have their hands full providing the nerve stumps from all your senses with just the right stimulations to carry off the trickery, but philosophers have assumed for the sake of argument that however technically difficult the task might be, it is "possible in principle." One should be leery of these possibilities in principle. It is also possible in principle to build a stainless-steel ladder to the moon, and to write out, in alphabetical order, all intelligible English conversations consisting of less than a thousand words. But neither of these are remotely possible in fact and sometimes an impossibility in fact is theoretically more interesting than a possibility in principle, as we shall see.

Let's take a moment to consider, then, just how daunting the task facing the evil scientists would be. We can imagine them building up to the hard tasks from some easy beginnings. They begin with a conveniently comatose brain, kept alive but lacking all input from the optic nerves, the auditory nerves, the somatosensory nerves, and all the other afferent, or input, paths to the brain. It is sometimes assumed that such a "deafferented" brain would naturally stay in a comatose state forever, needing no morphine to keep it dormant, but there is some empirical evidence to suggest that spontaneous waking might still occur in these dire circumstances. I think we can suppose that were you to awake in such a state, you would find yourself in horrible straits: blind, deaf, completely numb, with no sense of your body's orientation.

Not wanting to horrify you, then, the scientists arrange to wake you up by piping stereo music (suitably encoded as nerve impulses) into your auditory nerves. They also arrange for the signals that would normally come from your vestibular system or inner ear to indicate that you are lying on your back, but otherwise paralyzed, numb, blind. This much should be within the limits of technical virtuosity in the near future perhaps possible even today. They might then go on to stimulate the tracts that used to innervate your epidermis, providing it with the input that would normally have been produced by a gentle, even warmth over the ventral (belly) surface of your body, and (getting fancier) they might stimulate the dorsal (back) epidermal nerves in a way that simulated the tingly texture of grains of sand pressing into your back. "Great!" you say to yourself: "Here I am, lying on my back on the beach, paralyzed and blind, listening to rather nice music, but probably in danger of sunburn. How did I get here, and how can I call for help?"

But now suppose the scientists, having accomplished all this, tackle the more difficult problem of convincing you that you are not a mere beach potato, but an agent capable of engaging in some form of activity in the world. Starting with little steps, they decide to lift part of the "paralysis" of your phantom body and let you wiggle your right index finger in the sand. They permit the sensory experience of moving your finger to occur, which is accomplished by giving you the kines-thetic feedback associated with the relevant volitional or motor signals in the output or efferent part of your nervous system, but they must also arrange to remove the numbness from your phantom finger, and provide the stimulation for the feeling that the motion of the imaginary sand around your finger would provoke.

Suddenly, they are faced with a problem that will quickly get out of hand, for just how the sand will feel depends on just how you decide to move your finger. The problem of calculating the proper feedback, generating or composing it, and then presenting it to you in real time is going to be computationally intractable on even the fastest computer, and if the evil scientists decide to solve the real-time problem by pre-calculating and "canning" all the possible responses for playback, they will just trade one insoluble problem for another: there are too many possibilities to store. In short, our evil scientists will be swamped by combinatorial explosion as soon as they give you any genuine exploratory powers in this imaginary world. (The term combinatorial explosion comes from computer science, but the phenomenon was recognized long before computers, for instance in the fable of the emperor who agrees to reward the peasant who saved his life one grain of rice on the first square of the checkerboard, two grains on the second, four on the third, and so forth, doubling the amount for each of the sixty-four squares. He ends up owing the wily peasant millions of billions of grains of rice (2M-1 to be exact). Closer to our example is the plight of the French "aleatoric" novelists who set out to write novels in which, after reading chapter 1, the reader flips a coin and then reads chapter 2a or 2b, depending on the outcome, and then reads chapter 3aa, 3ab, 3ba, or 3bb after that, and so on, flipping a coin at the end of every chapter. These novelists soon came to realize that they had better minimize the number of choice points if they wanted to avoid an explosion of fiction that would prevent anyone from carrying the whole "book" home from the bookstore.)

It is a familiar wall these scientists have hit; we see its shadow in the boring stereotypes in every video game. The alternatives open for action have to be strictly and unrealistically limited to keep the task of the world-representers within feasible bounds. If the scientists can do no better than convince you that you are doomed to a lifetime of playing Donkey Kong, they are evil scientists indeed.

There is a solution of sorts to this technical problem. It is the solution used, for instance, to ease the computational burden in highly realistic flight simulators: use replicas of the items in the simulated world. Use a real cockpit and push and pull it with hydraulic lifters, instead of trying to simulate all that input to the seat of the pants of the pilot in training. In short, there is only one way for you to store for ready access that much information about an imaginary world to be explored, and that is to use a real (if tiny or artificial or plaster-of-paris) world to store its own information! This is "cheating" if you're the evil demon claiming to have deceived Descartes about the existence of absolutely everything, but it's a way of actually getting the job done with less than infinite resources.

Descartes was wise to endow his imagined evil demon with in-finite powers of trickery. Although the task is not, strictly speaking, infinite, the amount of information obtainable in short order by an inquisitive human being is staggeringly large. Engineers measure information flow in bits per second, or speak of the bandwidth of the channels through which the information flows. Television requires a greater bandwidth than radio, and high-definition television has a still greater bandwidth. High-definition smello-feelo television would have a still greater bandwidth, and interactive smello-feelo television would have an astronomical bandwidth, because it constantly branches into thousands of slightly different trajectories through the (imaginary) world. Throw a skeptic a dubious coin, and in a second or two of hefting, scratching, ringing, tasting, and just plain looking at how the sun glints on its surface, the skeptic will consume more bits of information than a Cray supercomputer can organize in a year. Making a real but counterfeit coin is child's play; making a simulated coin out of nothing but organized nerve stimulations is beyond human technology now and probably forever. (The development of "Virtual Reality" systems for recreation and research is currently undergoing a boom. The state of the art is impressive: electronically rigged gloves that provide a convincing interface for "manipulating" virtual objects, and head-mounted visual displays that permit you to explore virtual environments of considerable complexity. The limitations of these systems are apparent, however, and they bear out my point: it is only by various combinations of physical replicas and schematization (a relatively coarse-grained representation) that robust illusions can be sustained. And even at their best, they are experiences of virtual surreality, not something that you might mistake for the real thing for more than a moment. If you really want to fool someone into thinking he is in a cage with a gorilla, enlisting the help of an actor in a gorilla suit is going to be your best bet for a long time.)

One conclusion we can draw from this is that we are not brains in vats in case you were worried. Another conclusion it seems that we can draw from this is that strong hallucinations are simply impossible! By a strong hallucination I mean a hallucination of an apparently concrete and persisting three-dimensional object in the real world as contrasted to flashes, geometric distortions, auras, afterimages, fleeting phantom-limb experiences, and other anomalous sensations. A strong hallucination would be, say, a ghost that talked back, that permitted you to touch it, that resisted with a sense of solidity, that cast a shadow, that was visible from any angle so that you might walk around it and see what its back looked like.

Hallucinations can be roughly ranked in strength by the number of such features they have. Reports of very strong hallucinations are rare, and we can now see why it is no coincidence that the credibility of such reports seems, intuitively, to be inversely proportional to the strength of the hallucination reported. We are and should be particularly skeptical of reports of very strong hallucinations because we don't believe in ghosts, and we think that only a real ghost could produce a strong hallucination. (It was primarily the telltale strength of the hallucinations reported by Carlos Castaneda in The Teachings of Don Juan: A Yaqui Way of Knowledge [1968] that first suggested to scientists that the book was fiction, not fact, in spite of his having received a PhD in anthropology from UCLA for his 'research' on Don Juan.)

But if realJy strong hallucinations are not known to occur, there can be no doubt that convincing, multimodal hallucinations are frequently experienced. The hallucinations that are well attested in the literature of clinical psychology are often detailed fantasies far beyond the generative capacities of current technology. How on earth can a single brain do what teams of scientists and computer animators would find to be almost impossible? If such experiences are not genuine or veridical perceptions of some real thing "outside" the mind, they must be produced entirely inside the mind (or the brain), concocted out of whole cloth but lifelike enough to fool the very mind that concocts them.

2. PRANKSTERS IN THE BRAIN

The standard way of thinking of this is to suppose that hallucinations occur when there is some sort of freakish autostimulation of the brain, in particular, an entirely internally generated stimulation of some parts or levels of the brain's perceptual systems. Descartes, in the seventeenth century, saw this prospect quite clearly, in his discussion of phantom limb, the startling but quite normal hallucination in which amputees seem to feel not just the presence of the amputated part, but itches and tingles and pains in it. (It often happens that new amputees, after surgery, simply cannot believe that a leg or foot has been amputated until they see that it is gone, so vivid and realistic are their sensations of its continued presence.) Descartes's analogy was the bell-pull. Before there were electric bells, intercoms, and walkie-talkies, great houses were equipped with marvelous systems of wires and pulleys that permitted one to call for a servant from any room in the house. A sharp tug on the velvet sash dangling from a hole in the wall pulled a wire that ran over pulleys all the way to the pantry, where it jangled one of a number of labeled bells, informing the butler that service was required in the master bedroom or the parlor or the billiards room. The systems worked well, but were tailor-made for pranks. Tugging on the parlor wire anywhere along its length would send the butler scurrying to the parlor, under the heartfelt misapprehension that someone had called him from there a modest little hallucination of sorts. Similarly, Descartes thought, since perceptions are caused by various complicated chains of events in the nervous system that lead eventually to the control center of the conscious mind, if one could intervene somewhere along the chain (anywhere on the optic nerve, for instance, between the eyeball and consciousness), tugging just right on the nerves would produce exactly the chain of events that would be caused by a normal, veridical perception of something, and this would produce, at the receiving end in the mind, exactly the effect of such a conscious perception.

The brain or some part of it inadvertently played a mechanical trick on the mind. That was Descartes's explanation of phantom-limb hallucinations. Phantom-limb hallucinations, while remarkably vivid, are by our terminology relatively weak; they consist of unorganized pains and itches, all in one sensory modality. Amputees don't see or hear or (so far as I know) smell their phantom feet. So something like Descartes's account could be the right way to explain phantom limbs, setting aside for the time being the notorious mysteries about how the physical brain could interact with the nonphysical conscious mind. But we can see that even the purely mechanical part of Descartes's story must be wrong as an account of relatively strong hallucinations; there is no way the brain as illusionist could store and manipulate enough false information to fool an inquiring mind. The brain can relax, and let the real world provide a surfeit of true information, but if it starts trying to short-circuit its own nerves (or pull its own wires, as Descartes would have said), the results will be only the weakest of fleeting hallucinations. (Similarly, the malfunctioning of your neighbor's electric hairdryer might cause "snow" or "static," or hums and buzzes, or odd flashes to appear on your television set, but if you see a bogus version of the evening news, you know it had an elaborately organized cause far beyond the talents of a hairdryer.)

It is tempting to suppose that perhaps we have been too gullible about hallucinations; perhaps only mild, fleeting, thin hallucinations ever occur the strong ones don't occur because they can't occur! A cursory review of the literature on hallucinations certainly does suggest that there is something of an inverse relation between strength and frequency as well as between strength and credibility. But that review also provides a clue leading to another theory of the mechanism of hallucination-production: one of the endemic features of hallucination reports is that the victim will comment on his or her rather unusual passivity in the face of the hallucination. Hallucinators usually just stand and marvel. Typically, they feel no desire to probe, challenge, or query, and take no steps to interact with the apparitions. It is likely, for the reasons we have just explored, that this passivity is not an inessential feature of hallucination but a necessary precondition for any moderately detailed and sustained hallucination to occur.

Passivity, however, is only a special case of a way in which relatively strong hallucinations could survive. The reason these hallucinations can survive is that the illusionist meaning by that, whatever it is that produces the hallucination can "count on" a particular line of exploration by the victim in the case of total passivity, the null line of exploration. So long as the illusionist can predict in detail the line of exploration actually to be taken, it only has to prepare for the illusion to be sustained "in the directions that the victim will look." Cinema set designers insist on knowing the location of the camera in advance or if it is not going to be stationary, its exact trajectory and angle for then they have to prepare only enough material to cover the perspectives actually taken. (Not for nothing does cine'ma verity make extensive use of the freely roaming hand-held camera.) In real life the same principle was used by Potemkin to economize on the show villages to be reviewed by Catherine the Great; her itinerary had to be ironclad.

So one solution to the problem of strong hallucination is to suppose that there is a link between the victim and illusionist that makes it possible for the illusionist to build the illusion dependent on, and hence capable of anticipating, the exploratory intentions and decisions of the victim. Where the illusionist is unable to "read the victim's mind" in order to obtain this information, it is still sometimes possible in real life for an illusionist (a stage magician, for instance) to entrain a particular line of inquiry through subtle but powerful "psychological forcing." Thus a card magician has many standard ways of giving the victim the illusion that he is exercising his free choice in what cards on the table he examines, when in fact there is only one card that may be turned over. To revert to our earlier thought experiment, if the evil scientists can force the brain in the vat to have a particular set of exploratory intentions, they can solve the combinatorial explosion problem by preparing only the anticipated material; the system will be only apparently interactive. Similarly, Descartes's evil demon can sustain the illusion with less than infinite power if he can sustain an illusion of free will in the victim, whose investigation of the imaginary world he minutely controls.(For a more detailed discussion of the issues of free will, control, mindreading, and anticipation, see my Elbow Room: The Varieties of Free Will Worth Wanting, 1984, especially chapters 3 and 4.)

But there is an even more economical (and realistic) way in which hallucinations could be produced in a brain, a way that harnesses the very freewheeling curiosity of the victim. We can understand how it works by analogy with a party game.

3. A PARTY GAME CALLED PSYCHOANALYSIS

In this game one person, the dupe, is told that while he is out of the room, one member of the assembled party will be called upon to relate a recent dream. This will give everybody else in the room the story line of that dream so that when the dupe returns to the room and begins questioning the assembled party, the dreamer's identity will be hidden in the crowd of responders. The dupe's job is to ask yes/no questions of the assembled group until he has figured out the dream narrative to a suitable degree of detail, at which point the dupe is to psychoanalyze the dreamer, and use the analysis to identify him or her. Once the dupe is out of the room, the host explains to the rest of the party that no one is to relate a dream, that the party is to answer the dupe's questions according to the following simple rule: if the last letter of the last word of the question is in the first half of the alphabet, the questions is to be answered in the affirmative, and all other questions are to be answered in the negative, with one proviso: a noncontradiction override rule to the effect that later questions are not to be given answers that contradict earlier answers. For example:

Q: Is the dream about a girl? A: Yes.

but if later our forgetful dupe asks

Q: Are there any female characters in it?

A: Yes [in spite of the final t, applying the noncontradiction override rule]. (Empirical testing suggests that the game is more likely to produce a good story if in fact you favor affirmative answers slightly, by making p/q the alphabetic dividing line between yes and no.)

When the dupe returns to the room and begins questioning, he gets a more or less random, or at any rate arbitrary, series of yeses and noes in response. The results are often entertaining. Sometimes the process terminates swiftly in absurdity, as one can see at a glance by supposing the initial question asked were "Is the story line of the dream word-for-word identical to the story line of War and Peace?" or, alternatively, "Are there any animate beings in it?" A more usual outcome is for a bizarre and often obscene story of ludicrous misadventure to unfold, to the amusement of all. When the dupe eventually decides that the dreamer whoever he or she is must be a very sick and troubled individual, the assembled party gleefully retorts that the dupe himself is the author of the "dream." This is not strictly true, of course. In one sense, the dupe is the author by virtue of the questions he was inspired to ask. (No one else proposed putting the three gorillas in the rowboat with the nun.) But in another sense, the dream simply has no author, and that is the whole point. Here we see a process of narrative production, of detail accumulation, with no authorial intentions or plans at all an illusion with no illusionist.

The structure of this party game bears a striking resemblance to the structure of a family of well-regarded models of perceptual systems.

It is widely held that human vision, for instance, cannot be explained as an entirely "data-driven" or "bottom-up" process, but needs, at the highest levels, to be supplemented by a few "expectation-driven" rounds of hypothesis testing (or something analogous to hypothesis testing). Another member of the family is the "analysis-by-synthesis" model of perception that also supposes that perceptions are built up in a process that weaves back and forth between centrally generated expectations, on the one hand, and confirmations (and disconfirma-tions) arising from the periphery on the other hand (e.g., Neisser, 1967). The general idea of these theories is that after a certain amount of "preprocessing" has occurred in the early or peripheral layers of the perceptual system, the tasks of perception are completed objects are identified, recognized, categorized by generate-and-test cycles. In such a cycle, one's current expectations and interests shape hypotheses for one's perceptual systems to confirm or disconfirm, and a rapid sequence of such hypothesis generations and confirmations produces the ultimate product, the ongoing, updated "model" of the world of the perceiver. Such accounts of perception are motivated by a variety of considerations, both biological and epistemological, and while I wouldn't say that any such model has been proven, experiments inspired by the approach have borne up well. Some theorists have been so bold as to claim that perception must have this fundamental structure.

Whatever the ultimate verdict turns out to be on generate-and-test theories of perception, we can see that they support a simple and powerful account of hallucination. All we need suppose must happen for an otherwise normal perceptual system to be thrown into a hallucinatory mode is for the hypothesis-generation side of the cycle (the expectation-driven side) to operate normally, while the data-driven side of the cycle (the confirmation side) goes into a disordered or random or arbitrary round of confirmation and disconfirmation, just as in the party game. In other words, if noise in the data channel is arbitrarily amplified into "confirmations" and "disconfirmations" (the arbitrary yes and no answers in the party game), the current expectations, concerns, obsessions, and worries of the victim will lead to framing questions or hypotheses whose content is guaranteed to reflect those interests, and so a "story" will unfold in the perceptual system without an author. We don't have to suppose the story is written in advance; we don't have to suppose that information is stored or composed in the illusionist part of the brain. All we suppose is that the illusionist goes into an arbitrary confirmation mode and the victim provides the content by asking the questions.

This provides in the most direct possible way a link between the emotional state of the hallucinator and the content of the hallucinations produced. Hallucinations are usually related in their content to the current concerns of the hallucinator, and this model of hallucination provides for that feature without the intervention of an implausibly knowledgeable internal storyteller who has a theory or model of the victim's psychology. Why, for instance, does the hunter on the last day of deer season see a deer, complete with antlers and white tail, while looking at a black cow or another hunter in an orange jacket? Because his internal questioner is obsessively asking: "Is it a deer?" and getting NO for an answer until finally a bit of noise in the system gets mistakenly amplified into a YES, with catastrophic results.

A number of findings fit nicely with this picture of hallucination. For instance, it is well known that hallucinations are the normal result of prolonged sensory deprivation (see, e.g., Vosberg, Fraser, and Guehl, 1960). A plausible explanation of this is that in sensory deprivation, the data-driven side of the hypothesis-generation-and-test system, lacking any data, lowers its threshold for noise, which then gets amplified into arbitrary patterns of confirmation and disconfirmation signals, producing, eventually, detailed hallucinations whose content is the product of nothing more than anxious expectation and chance confirmation. Moreover, in most reports, hallucinations are only gradually elaborated (under conditions of either sensory deprivation or drugs]. They start out weak e.g., geometric and then become stronger ("objective" or "narrative"), and this is just what this model would predict (see, e.g., Siegel and West, 1975).

Finally, the mere fact that a drug, by diffusion in the nervous system, can produce such elaborate and contentful effects requires explanation the drug itself surely can't "contain the story," even if some credulous people like to think so. It is implausible that a drug, by diffuse activity, could create or even turn on an elaborate illusionist system, while it is easy to see how a drug could act directly to raise or lower or disorder in some arbitrary way a confirmation threshold in a hypothesis-generation system.

The model of hallucination generation inspired by the party game could also explain the composition of dreams, of course. Ever since Freud there has been little doubt that the thematic content of dreams is tellingly symptomatic of the deepest drives, anxieties, and preoccupations of the dreamer, but the clues the dreams provide are notoriously well concealed under layers of symbolism and misdirection. What kind of process could produce stories that speak so effectively and incessantly to a dreamer's deepest concerns, while clothing the whole business in layers of metaphor and displacement? The more or less standard answer of the Freudian has been the extravagant hypothesis of an internal dream playwright composing therapeutic dream-plays for the benefit of the ego and cunningly sneaking them past an internal censor by disguising their true meaning. (We might call the Freudian model the Hamlet model, for it is reminiscent of Hamlet's devious ploy of staging "The Mousetrap" just for Claudius; it takes a clever devil indeed to dream up such a subtle stratagem, but if Freud is to be believed, we all harbor such narrative virtuosi.) As we shall see later on, theories that posit such homunculi ("little men" in the brain) are not always to be shunned, but whenever homunculi are rung in to help, they had better be relatively stupid functionaries not like the brilliant Freudian playwrights who are supposed to produce new dream-scenes every night for each of us! The model we are considering eliminates the playwright altogether, and counts on the "audience" (analogous to the one who is "it" in the party game) to provide the content. The audience is no dummy, of course, but at least it doesn't have to have a theory of its own anxieties; it just has to be driven by them to ask questions.

It is interesting to note, by the way, that one feature of the party game that would not be necessary for a process producing dreams or hallucinations is the noncontradiction override rule. Since one's perceptual systems are presumably always exploring an ongoing situation (rather than a fait accompli, a finished dream narrative already told) subsequent "contradictory" confirmations can be interpreted by the machinery as indicating a new change in the world, rather than a revision in the story known by the dream relaters. The ghost was blue when last I looked, but has now suddenly turned green; its hands have turned into claws, and so forth. The volatility of metamorphosis of objects in dreams and hallucinations is one of the most striking features of those narratives, and what is even more striking is how seldom these noticed metamorphoses "bother" us while we are dreaming. So the farmhouse in Vermont is now suddenly revealed to be a bank in Puerto Rico, and the horse I was riding is now a car, no a speedboat, and my companion began the ride as my grandmother but has become the Pope. These things happen.

This volatility is just what we would expect from an active but insufficiently skeptical question-asker confronted by a random sample of yeses and noes. At the same time, the persistence of some themes and objects in dreams, their refusal to metamorphose or disappear, can also be tidily explained by our model. Pretending, for the moment, that the brain uses the alphabet rule and conducts its processing in English, we can imagine how subterranean questioning goes to create an obsessive dream:

Q. Is it about father?

A. No.

Q. Is it about a telephone?

A. Yes.

Q. Okay. Is it about mother?

A. No.

Q. Is it about father?

A. No.

Q. Is it about father on the telephone?

A. Yes.

Q. I knew it was about father! Now, was he talking to me?

A. Yes....

This little theory sketch could hardly be said to prove anything (yet) about hallucinations or dreams. It does show metaphorically how a mechanistic explanation of these phenomena might go, and that's an important prelude, since some people are tempted by the defeatist thesis that science couldn't "in principle" explain the various "mysteries" of the mind. The sketch so far, however, does not even address the problem of our consciousness of dreams and hallucinations. Moreover, although we have exorcised one unlikely homunculus, the clever illusionist/playwright who plays pranks on the mind, we have left in his place not only the stupid question-answerers (who arguably can be "replaced by machines") but also the still quite clever and unexplained question-poser, the "audience." If we have eliminated a villain, we haven't even begun to give an account of the victim.

We have made some progress, however. We have seen how attention to the "engineering" requirements of a mental phenomenon can raise new, and more readily answerable, questions, such as: What models of hallucination can avoid combinatorial explosion? How might the content of experience be elaborated by (relatively) stupid, uncomprehending processes? What sort of links between processes or systems could explain the results of their interaction? If we are to compose a scientific theory of consciousness, we will have to address many questions of this sort.

We have also introduced a central idea in what is to follow. The key element in our various explanations of how hallucinations and dreams are possible at all was the theme that the only work that the brain must do is whatever it takes to assuage epistemic hunger to satisfy "curiosity" in all its forms. If the "victim" is passive or incurious about topic x, if the victim doesn't seek answers to any questions about topic x, then no material about topic x needs to be prepared. (Where it doesn't itch, don't scratch.) The world provides an inexhaustible deluge of information bombarding our senses, and when we concentrate on how much is coming in, or continuously available, we often succumb to the illusion that it all must be used, all the time. But our capacities to use information, and our epistemic appetites, are limited. If our brains can just satisfy all our particular epistemic hungers as they arise, we will never find grounds for complaint. We will never be able to tell, in fact, that our brains are provisioning us with less than everything that is available in the world.

So far, this thrifty principle has only been introduced, not established. As we shall see, the brain doesn't always avail itself of this option in any case, but it's important not to overlook the possibility. The power of this principle to dissolve ancient conundrums has not been generally recognized.

(from Daniel C. Dennett: Consciousness Explained (Penguin books, 1991.)