Beyond the Cartesian Theater: More Better Models and Metaphors

Before we can have a theory about anything, a theory that makes quantitative, falsifiable claims, we need some kind of model. A model is less rigorous than a theory, really just a way of thinking about something. Climbing down the ladder of rigor still another rung we get to metaphors. A metaphor, like a model, is just a way of thinking about something, but unlike a model, it doesn't pretend to be complete but merely lacking detail; rather it is completely flimsy, and if you look at it from the wrong angle, or push just a little too hard, the whole thing collapses.

Right now we have no idea how to think about the mind, but we have to think of it in some way. We can't help but think in terms of metaphors, and if you choose the wrong ones, you end up channeling your thinking in certain directions and not in others, making assumptions without justification and overlooking whole classes of possibilities. No matter how well we know that we are speaking loosely, our mental images and metaphors have a creeping influence on our subsequent speculation. As you speak, so you will think. Moreover, the more deeply you want to go with your investigations and speculations, the more this kind of thing matters. You can have a somewhat wrong model in mind and still come up with perfectly accurate empirical predictions. When you are trying to rethink things in a fundamental way, however, when you are trying to make conceptual leaps and nail down what might be true, what must be true, and what could not possibly be true about the really big things like consciousness, the wrong metaphor, or simply the lack of the right one, can really derail you.

What we would like is a large and varied repertoire of metaphors to draw from, with some awareness of each of their limitations. We've already seen the Cartesian Theater, which Daniel Dennett (Dennett 1991) is right to dismiss as a valid metaphor for the mind. He is also right in saying that it is very difficult to exorcise it from our thinking about minds. In this essay, I am not going to make a coherent knock-down argument for any particular point of view, but rather I will kick around some models and metaphors. These are just fragments, images and half-baked ideas that for one reason or another, keep coming to my mind as I think. As with the proverbial blind men and the elephant, I think there are grains of truth in each of them, but limitations as well. I hope to extend the range of discussion of the mind by adding to the collection of idioms. Even the wrong ones (and they are all wrong, at some level of detail - that's why they are just models and metaphors) at least help us talk about minds. Inevitably, as I discuss what each model gets right and gets wrong, this exercise will lay bare some of my own intuitions and opinions about the mind; but then again, that's kind of the idea. At some point in the future, some of them, or parts of some of them, may get stitched together into an actual theory that is cognitively, phenomenologically, and biologically plausible.


These days we can't help but think of the mind in terms of computers. In spite of everything I have written about consciousness, I still do it all the time, and I even think it is often useful to do so, as long as you are aware of the caveats that go along with it. The computer is simply the dominant metaphor for minds (and lots of other things as well) just as the clock and the steam engine were dominant metaphors in eras past. (A pretty good discussion of the ascendancy of the computer model of the mind can be found in P. Cisek's article in the Nov/Dec 1999 issue (Reclaiming Cognition) of the Journal of Consciousness Studies.)

In a sense, the more you actually know about how computers work the longer the period of intellectual detox must be. You just can't help phrasing things in terms of live processor vs. dead data, a centralized point of control, search and sort algorithms and the like. I really like Dennett's pandemonium/fame model of mind as an antidote to this (more on this later). Dennett makes the point that the serial von Neumann type machine (a fancy way of talking in an idealized, abstract sense about "the computer" as we understand computers) is highly artificial even in our own minds. That is, our minds, as they evolved, did not start out thinking that way. Serial, von Neumann type thinking was a great trick, but one we learned relatively late in the evolutionary game - it is not the natural architecture of minds. As Dennett puts it, our mind is a serial machine simulated on a massively parallel substrate. It is this type of linear, symbolic thought that we get presented to us when we try to introspect "how we think". We went on to build computers to mimic this naive impression we have of our own thought processes, and obviously it has worked out quite well for us in lots of ways. But to imagine that we can capture the essence of thought itself that way is a beguiling mistake.

Let me say up front that I am aware that I am putting up something of a straw man argument against computers here. There are many, many different computer languages, and many, many different computing environments, and I am lumping everything together as if it were the 1950's, and "computer" means straightforward, old-school linear programming, of the type we deal with in languages like C or assembly language. I'm old-school - this is largely how I think of computers. At their core, (or these days, their cores) computers darn well are straightforward, old-school machines. In such a device, there is a linear thread of control, laid down by the human programmer. Control may pass to a subroutine, which may do all kinds of complicated things, including passing control to still other subroutines, but when it is done it issues a "return" statement of some kind, and control pops back up to the calling code. (Yes, it really is called popping. Popping the stack, as a matter of fact).

I assume that I need not go into too much detail as to why the computer is such a compelling metaphor. It certainly seems as though they do a lot of the stuff that minds do; after all, we designed them to. They process information, they remember stuff, they seem to know stuff in some sense or another, and you can program them to do seemingly any cognitive task you can imagine, if you just have enough memory and CPU cycles. I remember, upon first learning to program, feeling the same confidence that the early researchers in the 1950's felt that I should be able to whip up a functioning AI. There are, however, some problems with computers. Of course there are the arguments against functionalism, and the whole Hard Problem what-is-it-like-to-see-red thing. But there are other reasons why a computer, at least a traditionally programmed one, is not quite the model we are looking for.

So before we dismiss computers altogether (and I emphasize again that dismissing them for good would be nearly impossible as well as counter-productive, since we will always be drawn to the extensive vocabulary of computers when talking about minds) let's think about them for a bit. Ignoring for the moment all of the what-is-it-like-to-see-red arguments, why are computers not like minds? In particular, if we were to design a brand new computer language, or a computer environment (like an SDK - Software Development Kit) that would allow people to tinker with AI, what would it look like?

Mental history is written by the mental winners. Thought only seems logical, rigorous and inevitable in retrospect. I think, however, that thought is marked by constant distractions. Some are dismissed, some are pursued, and some fall in between, nudging or coloring other threads of thought. We are interrupt driven, and interrupts come not only from outside, but from inside as well. When we drill down or call a subroutine, we might take off in a whole new direction, and never return (or never quite return) to the original thread. Somehow, "low-level" routines can derail the "high-level" routines, but not willy-nilly. The system works, after all, and I can drive to work and bake cookies and not forget what I am about. Somehow, in minds, there isn't this distinction between low-level routines and high-level ones, such that high-level routines call the low-level ones, which are bound to do exactly as they are told and return to the caller. High-level and low-level in minds is a bit of a two-way street. This suggests that the mind is more like a fractal - patterns get applied at high and low levels, and there is no such thing as a low-level utility routine, only patterns that can be applied at any level. Or even, perhaps, patterns that opportunistically apply themselves at any level.

It also seems to me that for a computer to behave like a mind, it would have to possess more of a memory, not just a memory of the type that computers already possess. Computers store numbers in memory in a very deliberate way, treating memory like a chest of drawers. Put the red socks in this drawer, take them out of that drawer later. In a computer that acted like a mind, however, the machine would actually maintain the vapor trails of past thoughts themselves. This sounds a little like a CPU cache, but whereas the cache is invisible to the software by design, this would not be. Past thoughts, including their flailings and dead ends, would be stored as objects that can later be held up to the light and examined.

It is misleading to think of the mind as a tiny spark of active CPU, that accesses opaque memory by address and plucks the datum stored there and loads it into a register. The way a computer uses memory is like a warehouse filled with crates, and a big overhead claw that can grab any crate and carry it to a central platform. That's just not the way memory and mind work at all. In a mind, reference is not opaque - in some funny way that I do not pretend to have my arms around yet, a reference subsumes the thing referred to.

In a mind, as new ways of thought are cobbled together with routines, with threads or submodules invoking each other, the pattern of linkages is abstracted from the particular problem at hand and stored as a whole new invocable algorithm. Somehow, what is stored must not be just the beginning state and the ending state of a particular function or transformation, but the process of the transformation itself, or what it is like to perform it. We must blur the distinction between algorithm and data, so that our system works with blobs of algorithmic data stuff. In fact, I think that there is likely no distinct line we can draw in actual minds between data, algorithm, and the execution engine (CPU) itself. In a mind, data gets applied to new situations; data gets applied to other data; algorithms get applied to each other; algorithms are treated as raw material for other algorithms. A mind gets to see an algorithm all at once, to comprehend it, which is a far different thing than merely executing it. We humans, for instance, have a much easier time with the Halting Problem than computers do, usually.

It is the hard and fast distinction between memory, data, and CPU that is the most insidious effect of drawing on the computer metaphor. I do not know much about neural nets, but I suspect that they help to dissolve the data/algorithm distinction, although I also suspect that they throw the baby out with the bathwater, in that a neural net is crippled in its algorithm-generating capabilities. I do not believe that any neural net could have written the quick sort algorithm, for instance.

It seems that for a computer to think at all like a mind, it would have to have lots of processes going at once, and these processes would trade off control between them, and they would have a high degree of visibility into one another's workings. They might help each other out, but they might exact a price for such help; the helpers might demand more of a say, coloring the thoughts thought or tugging some control away. Actually, phrased this way, the whole thing sounds less like a computer environment and more like a parliamentary government with lots of parties.

Moreover, the units of processing (threads/routines/processes) would be able to connect with each other with a higher probability of producing something coherent than today's computers would. That is, a computer executing a program walks a knife edge of coherence surrounded by a sea of chaos. Drop one bit, or walk off the end of an array, and you are simply done (General Protection Fault, blue screen of death). Computers are very chaotic systems (neighbors do not map to neighbors) - change one thing about a computation and you get drastically different results. This is not generally how organic things work, including brains. I think that if we were to create a development environment/language for true AI, the fundamental bricks out of which the system created algorithms would be hardier. They could knock around, invoke or connect with each other if not randomly, at least somewhat randomly, with a greater probability of creating something that would run.

Dennett's Pandemonium

In "Consciousness Explained", (Dennett 1991) Daniel Dennett proposed a so-called pandemonium model, that partially echoes Marvin Minsky's Society Of Mind idea (Minsky 1985). It is a good one, worth reprinting here:

There is no single, definitive "stream of consciousness" because there is no central headquarters, no Cartesian Theater where "it all comes together" for the perusal of the Central Meaner. Instead of such a single stream (however wide), there are multiple channels in which specialist circuits try, in parallel pandemoniums, to do their various things, creating Multiple Drafts as they go. Most of these fragmentary drafts of "narrative" play short-lived roles in the modulation of current activity but some get promoted to further functional roles, in swift succession, by the activity of a virtual machine in the brain. The seriality of this machine (its "von Neumannesque" character) is not a "hard-wired" design feature, but rather the upshot of a succession of coalitions of these specialists.

The basic specialists are part of our animal heritage. They were not developed to perform peculiarly human actions, such as reading and writing, but ducking, predator-avoiding, face-recognizing, grasping, throwing, berry-picking, and other essential tasks. They are often opportunistically enlisted in new roles, for which their native talents more or less suit them. The result is not bedlam only because the trends that are imposed on all this activity are themselves the product of design. Some of this design is innate, shared with other animals. But it is augmented, and sometimes even overwhelmed in importance, by microhabits of thought that are developed in the individual, partly idiosyncratic results of self-exploration and partly the predesigned gifts of culture. Thousands of memes, mostly borne by language, but also by wordless "images" and other data structures, take up residence in an individual brain, shaping its tendencies and thereby turning it into a mind.

I have already argued against adopting this model as the truth, lock stock and barrel, but I like aspects of it. I tend to agree with Dennett that thought is a lot less linear and single-threaded than we think it is, and that there are a lot of competing/cooperating specialist circuits at work (elsewhere Dennett evocatively calls these "demons". This actually hearkens back to the idea of computer processes, as certain types of these are called "daemons" in UNIX and Linux. Note also that "pandemonium" literally means "demons everywhere").

Voltron I accept that there could be lots of demons in my mind, perhaps that entirely make up my mind. It is certainly a great metaphor. But again, we can't let it limit us, and we must think hard about the ways the things in our minds might not be like individual, you know, demons, or anything subject to the constraints that actual biological beings might live and die under. How many demons are there? What delineates a demon? How do new ones come into existence, and do they ever die? To what extent do they compete, and what happens to the losers? To what extent do they cooperate? Some people use terminology that suggests that the mind is a "Darwinian memosphere". Does natural selection work on demons in the same way that it does on species? Can demons mate and produce offspring? Or can they simply merge together like Voltron? Do individual demons change over time, or adapt?

Dennett uses the term "coalitions" (there's that parliamentary thing again). After allying themselves to accomplish something, does the coalition fall apart into exactly the same set of individual demons that went into it, or does the Voltron/coalition demon get to stick around, added to the menagerie along with its component demons? Perhaps a demon can sort of will itself to have new powers, unlike biological beings. Maybe they reproduce in a way that is more like mitosis than sexual reproduction. To what extent do patterns of relations between demons harden and become new demons themselves?

What do demons want? To the extent that they compete, what resources are they competing for? How, if at all, do they stack, or nest, or apply themselves to one another? What are the channels of interaction between them? Do they all have mutual visibility, as if they are sitting in a huge stadium watching each other? Does any one demon, or coherent coalition, hold the floor at any given time? It may well be the case that the channels of communication are not instantaneous, and not all global, and that the longer a given signal sticks around, the more broadly it gets propagated. Stability and consistency of a signal have a way of focussing the mind. Are the signals themselves between the demons just more demons? If not, how should we think about them? Or is it really more like a jungle, in which they happen across one another from time to time, interacting sporadically and opportunistically?

And what are demons, anyway? Are they kind of like algorithms, (like how to bake a cake) or are they individual episodic memories? Probably yes and yes. I guess they are more or less memes. It also may well be the case that my general knowledge of how to bake a cake is not merely synthesized from, but actually composed of, the collection of individual memories I have of baking cakes, reading about baking, etc. It could be something in between, in which the "how to bake a cake" demon is not entirely independent of the individual memories, but is not simply a collage of them either. There could also be lots of levels making this whole thing up with individual memories forming intermediate demons, which then form the "how to bake a cake" demon. Perhaps there is no fixed, individual "how to bake a cake" demon until needed, at which point the demons synthesize it on demand.

Phenomenological Plausibility of Demons

Why does this whole demon/memosphere idea seem even vaguely plausible in the first place from a first-person point of view? As we go about through our lives, we have a sense that our mind is not monolithic, that there are parts of it working away offline. Not only do results of these offline computations pop into our main stream of consciousness, but there is a definite sense, in me anyway, of a whole train of thought, in all of its what-its-like-to-see-red glory, being plugged into whatever else I was thinking about or experiencing. Such trains of thought come complete with a sense that they didn't just come into existence at the moment "I" became aware of them, but that they had been developing on their own for some time.

Active percepts, not just past memories, are demons. You tune out the actual shapes of the trees on the side of the road as you drive to work each day, the colors of the houses on your street, etc. Your eyes pick up all these details (that is, the corresponding photons do actually strike your retinas), and somewhere there is a perceptual demon who, according to this way of thinking, is exquisitely conscious of all that stuff, but "you" aren't aware of it, unless there is a conscious effort at attention to such details.

Now of course this sense could be an illusion. Dennett makes a good point about this. He says that qualophiles like me first argue that the target of our speculation, subjective consciousness itself, is in principle not amenable to third-person scientific techniques of investigation. But then we claim that some aspects of it are also outside our capacity for first-person introspection as well. How can we possibly posit with a straight face something that can not be detected by third person or first person methods? ("Does anyone, on reflection, really want to say that?", he asks. "Putative facts about consciousness that swim out of reach of both "outside" and "inside" observers are strange facts indeed.") In what way is this different at all from the infamous luminiferous ether?

But still, there is this sense - the sense that I (whatever "I" am) am peripherally aware, then sometimes acutely aware, of whole other trains of thought/experience that have a time sense associated with them. A sense that even though "I" wasn't aware of the train of thought/experience a moment ago, it was churning away on its own, and now that I am aware of it, I am patched into it, complete with its sense of its own conscious history. This is where Dennett and I part ways. He denies the consciousness of the demons. I think that even if his pandemonium model is roughly correct, there is no reason the individual demons themselves could not be conscious. He, of course, believes that consciousness just is "fame in the mind", i.e. the collective activity of lots of (unconscious) demons.

Have you ever been listening to an oldies station, and heard a song that you have not heard in years or decades, but had the distinct sense that the very same song was going through your head sometime in the past week? Of course you have. I have had this sense suspiciously often. Often enough, in fact, that I have a hard time believing that I actually just happened to be replaying all those songs consciously in the few days before I heard them on the radio. I think that all of my song memories are playing all the time, but "I" am not aware of them. And if song memories, what other memories are on hot standby in the same way? Is there a "Dancing In The Moonlight" demon, who just sings that song all the time, forever until you die? I can't rule it out. It may be that all of our old moments of consciousness are still in there, as some kind of standing waves.


How do you ever get a thought in edgewise, with all these demons singing? Not to mention the ones thinking, remembering, and sensing your shoes through the soles of your feet. I suspect that you (or perhaps I should go with the scare quotes, "you") tune them out. Like the jackhammer outside your window, it's not as if the demons go away or stop, but after a short while they just don't impinge upon "you" anymore, unless it would be a good idea for them to do so. But what are the forces that determine how good or bad an idea it is for them to seize the stage (there's that Cartesian theater again)?

I think it is entirely likely that all of the demons are conscious to some extent or another, whether or not "I" am conscious of what they are conscious of. At one time it took a lot of concentration for me to tie my shoes, but now I could almost do it in my sleep. I constructed a tying-shoes demon, and when I tie my shoes, somewhere in my mind, it is hard at work, concentrating like mad (although I can willfully focus my attention on the act of shoe tying and make it conscious). There may be lots and lots of willful and creative consciousness in my mind that "I" don't know about.

I have a strong suspicion that a great deal of the mind's activity is inhibitory. We spend an awful lot of effort shutting down streams of information, and channeling activity. If you sense another metaphor coming on, you are right. It strikes me that, to borrow an image from the memeticists, the mind is like an organism under constant assault by viral memes (demons). We tune out the singing demons by quickly developing antibodies to them. If the "Dancing In The Moonlight" demon sings the same song in the same way for too long, we jam the signal by installing a counter-signal, a counter demon. We handicap; we compensate. It doesn't stop, but we accommodate it by adjusting the state of our consciousness to account for its constant presence. And of course, even though I speak of singing demons, this goes for the remembering-my-childhood-cat-Fluffy demon as well, and the demons that notice the trees along the highway. The demon and its meme-jamming anti-demon are locked in a self-canceling embrace forever, leaving the mind as an intricate balance of tensions, like a bicycle wheel.

Pringle's potato chip If you have ever trued a warped bicycle wheel, you will know exactly what I mean. If you haven't, each spoke can be tightened or loosened by twisting the bushing at its outer end with a truing wrench. if you loosen or tighten the spokes in one area, the wheel goes out of true, and warps. You try to bring it back into true by tightening and loosening spokes, but it is tricky, prone to all sorts of overcorrection and misbalancing of the tensions across the whole wheel. You bring one area into true, then spin the wheel a bit and discover that a quarter turn away from your original area is now bent out like a Pringle's potato chip. But the counter-balanced demon is still always poised, ready to make a play for the Stage if it gets the right input, or if it can figure out how to adapt itself to new circumstances.

Similarly, imagine some object, like a ball, with many small hooks, and a rubber band attached to each hook and stretched taut (maybe the rubber bands are attached at the other end to some larger frame, or room). The ball in the center is suspended, held in place by the sum of the tensions in the rubber bands, pulling in all directions. If you loosen, tighten, or cut any of the rubber bands, the ball in the center would move, just a little bit. Likewise, the tension in all of the remaining rubber bands would adjust slightly as well. I wonder if, in the pandemonium model, the demons might all be in contact with one another like the bicycle spokes or the rubber bands.

The demons want to dominate, to grab center stage. In order to maintain a thought or percept at all, you must be good at developing and maintaining antimemes, of countering all of the different assaults and distractions from the different demons. Imagine the little Dutch boy of the fable, with a million fingers in a million holes in the dyke. Or, you wait until the whole system reaches some kind of equilibrium for some amount of time. If some demon really wants to apply itself, maybe you just let it.

This idea of demons/antidemons (or memes and antimemes) respects a couple of ideas. Most importantly, it respects a sense of holism in the mind. The mind, according to this conception of it, really is one unified thing, with a balance of tensions keeping much of it more or less inert at any given time. This image helps make sense of some of the conscious-but-not-conscious scenarios people have devised over the years: the ticking clock you are not aware of until it stops, the pattern of the design on the carpet, the sensation of your socks against your ankles. Your "peripheral" awareness of such things is in there, and part of your overall conscious field, but neutralized by an antimeme.

Carried further, it may well be the case that our whole mind, every sensation and even every memory, is just like that: always right there, as part of your all-at-once now, but tuned out. The cacophonous demons are not off somewhere, each in its own soundproof room. They are all there, all the time, fully patched in. We actively exert ourselves to cancel them out, and this exertion is a collective exertion, performed by other demons. The mind is a self-policing pandemonium.

Darwinian Memosphere of Demons

I like the idea of active demons, even if most of them live in the shadows most of the time, rather than the mind as presiding over a mountain of static data. The mind is clearly great at parallel processing, but even that understates the situation, I think. When we are given a fact, say, that contradicts something we know, even somewhat indirectly, it is remarkable how quickly we notice. If we learn something new and surprising about cars, it is hardly plausible that we run through all the thoughts and memories involving cars (individual cars as well as general car knowledge) in our minds and adjust them accordingly. Facts, memories, are recalled as needed, as if by magic. It certainly seems as if the old fact jumps up, as if offended, to take on the newcomer. Old thoughts are less like dead data waiting to be accessed, searched, sorted, or applied, than like little sparks of mind themselves, capable of asserting themselves.

I suspect that there is no single burly stagehand guarding the Stage, screening demons who want their moment in the spotlight. Demons can jump on and apply themselves to any detail of a new or developing stimulus (thought or percept) that catches their fancy. In this way, they get to flesh out the "focused-upon" detail more fully. However, uppity demons get smacked down. Demons can jump on the Stage, applying themselves whenever they want, but there is a cost. If they are just spamming, grabbing the spotlight when they have nothing to contribute, they either strengthen the counter-demon response and get tuned out extra in the future, or they get corrupted somehow. In order to survive intact over the long term, demons must tiptoe through the minefield of existing demons on the way to the stage without stepping on anyone else's tail or hoof.

You know how sometimes you remember an event from the distant past, and you are not sure if you are actually remembering the event or remembering your subsequent remembering of it on other occasions? Your memorable recalling of it in the past has effectively jammed the original memory. Any toehold or reference tag that would have triggered the original memory will also now trigger the memory of the memory. The original has been masked. Was Fluffy yellow? I always thought of him as yellow. But Mom has a photo and he's black. Oops. If you have to bribe too many stagehands to get onstage, you may find that you have nothing left once you get there. Demons who cry wolf get ignored later (or countered more vigilantly)

I speculate that different demons have different niches in the memosphere. Some are swaggering alphas, that apply themselves broadly and promiscuously to whatever processing that needs doing, while some are rarely seen, and just stay in their tiny niche, with very specific criteria for activation. According to this notion, a swaggering alpha's identity may be so smeared out and indeterminate that it hardly has an identity left, just the barest shape of one, a tone or coloring it can impart. (The notions of causation or object permanence might be such demons.)

While at the other end of the spectrum, the den-dwelling, seldom-seen demons get to keep their specificity in sharp detail (like specific episodic memories, or particular skills). Perhaps the alphas are more appropriately seen as eager beavers, willing to trade quality and specificity for sheer quantity and frequency, whereas the den-dwellers make the opposite call. As in nature, different demons employ different strategies and make different evolutionary compromises, until just about every conceivable niche is filled.

Just as it may weaken or corrupt a demon to apply itself overly broadly, demons may be similarly insulted by allowing other, incompatible demons to pass onto the Stage. There may be, for instance, a demon that enforces or embodies what we think of as a valid chain of logical inference, and it will not tolerate another demon that violates its criteria for a valid chain of inference to ascend. To allow such a thing would be to make it less likely that the valid inference demon would be allowed to ascend in the future. In this way, demons collectively constitute rules or constraints on each other. A truth I am certain of or perhaps even a symbol I know how to interpret may be a rock-solid demon that will simply always win competitions with other demons.

What about the mind's current state makes it enticing for the next demon or coalition of demons to make a play for the Stage? What sorts of current thoughts create a hospitable niche for subsequent thoughts? I suspect that the answer is far from deterministic, or rather, that it is chaotic: you never know what details or seemingly unimportant aspects of a thought or percept will strongly grab hold of a demon's fancy and take you in a whole new direction. In particular, it is not necessarily the overall big idea or perceived direction a current thought is going in that subsequent thoughts hook into but those aforementioned distractions, even if most of the distractions do not go anywhere interesting and wind up being dead ends. Moreover, I think that demons do not necessarily have a preference for high-level deployment as opposed to low-level filling out the detail of some thought or percept - they just like a good fit.

So: the demons that are maximally compatible with all the existing demons are allowed to take the Stage. Each new moment of consciousness is new and unique, however. So it is not the case that old demons simply get to relive their glory days in the spotlight; more likely they get to inform the creation of a new demon - they get to be the primary parent, or chief architect. Each incumbent demon is like a craftsman, or a specialized muscle that shapes a new demon. Although some are more like specific memories, some are more like general facts or general strategies. Some are more algorithmic/prescriptive, and others more data/descriptive, on a sliding scale. Each has a bit of "what is it like?" and each has a bit of "what does it do?". Indeed, it is hard to separate the two aspects. When I drive down my street and see an object in my field of vision that ultimately resolves to "house", it is probably not the case that my "house" demon simply grabs the spotlight; more likely it helps spawn a new moment of consciousness, a new, yet distinctly housey demon.

We often speak of our minds containing models: models of reality, models of self, models of my cat, etc. What sense can we make of such talk if our minds are constituted by demons? Are there models at all, if each new moment of consciousness is whipped up on the fly dynamically? I feel comfortable saying yes. Any model is a black box with an interface. You ask certain questions in the right way, and the model gives you consistent answers. The model may be implemented by a static table of bits or a database, with a relatively mechanical query engine, or it may be implemented by a raucous parliament. Our "models" may not be as model-like as we suppose.

This pandemonium image blurs the distinction between immediate sensations and memory, which to my way of thinking is one of its virtues. Memory is smarter and more active than is generally supposed. Memories are not in cold storage, off in a file cabinet, but right in your mind now, pressing on your consciousness.

Finally, this image bolsters an intuition that the ability to get bored is essential for intelligence. As you are assaulted by the same thoughts, you tune them out. You cease to be conscious of what isn't novel. Could autism be somehow related to meme immune deficiency disorder?

So what are the selection criteria for letting demons on the Stage? Which demons do get promoted to the inner circle? Whoa - what inner circle? Alright, yes, there is no Cartesian theater, not exactly, but even Dennett acknowledges that there is something like a consensus that forms (pretty quickly at that) about what the narrative center of gravity is (or was) at any moment in my mind. This may be an artifact of the narrative-spinner demon, the chatterbox, and may not mean as much as it seems with regard to what "I" am thinking, but there is something to the notion that I thought about Fluffy today, but had not in the week before today. There is something like a spotlight of attention on certain trains of thought, even though (as I suspect) there are lots and lots of other trains of thought going on at the same time. So with my invocation of the Cartesian theater, here is an example of my using a discredited metaphor because, dammit, it helps me say what I want to say.

The Spotlight of Attention

While I am here, I should just say that the proverbial spotlight of attention is a bad metaphor, even though I just used it. Attention is actively created, not passively observed. The spotlight metaphor wrongly suggests that the thing attended to in the mind already exists, in all its detail, in the dark before the spotlight is shone on it. It is really a bad metaphor for the same reason that the whole Cartesian Theater is. In fact, it really is an aspect of the same metaphor, come to think of it. Where was the last time you saw a spotlight, drawing your attention? Probably the theater. Rather than imagining that all of our thoughts, percepts, memories, etc. are all there, fully realized, but in the dark until their moment in the spotlight, it is more likely the case that we function as a sort of just-in-time mental reality generator, creating things on the fly as we "turn our attention to" them.

The idea of demons having to pay a cost for inappropriate activation may help improve the "spotlight of attention" metaphor. As a demon, you get to create the spotlight any time you want, making other demons conform to you. Seizing attention is really a way of corralling or bullying other demons into trying to apply themselves to you, even at a cost to themselves of less-than-appropriate activation. Depending on the situation, seizing attention is like issuing an "all hands on deck" with more or less urgency. Attention, then, isn't some spotlight being shone on a particular demon, but is the collective combinations of lots of demons, perhaps with one at the center as a ringleader or catalyst. Things like pain or a threat tend to focus the attention. This may be a way of having one imperative light a fire under all of the demons, in effect shouting at them, "I don't care if this doesn't fit your criteria of applicability! Find a way to apply yourselves to this situation, however suboptimally to yourselves!"

The Burbling Spring

Erector set

According to naive, classical AI, we build our thoughts, our world, logically, using existing rules and facts like girders in an erector set. This does not seem right to me. Perhaps we humans have a certain creative wellspring within us. The stuff of which our thoughts and percepts are made just bubbles up from the ground. As with an actual spring, left on its own, this stuff just lies on the ground in a puddle. But when channeled appropriately, it can become something intricate and powerful. This metaphor appeals to my sense that a lot of the tricky stuff the mind does is inhibitory. It is the way we constrain our thoughts, forcing them at great pressure where we want them, that makes us smart, and not necessarily (or exclusively) the gushing of the spring itself. The more tightly we constrain our thoughts, the more things we rule out, the more powerful the jet of thought becomes. Einstein was a genius not because he had greater water pressure in his spring than the rest of us, but because of the narrowness of his pipes: the integrity of his reality-model, and all the huge spaces of possibility he ruled out, left only very narrow channels through which his water could spurt.

This image counters the natural image of thoughts being structures built out of smaller elements like Legos or girders. Instead, we build (or train) channels, or forms, like those wooden forms construction workers make into which they pour concrete, and the thoughts, percepts, mentality, flow through the forms. The more tightly we build the forms or channels, the greater the force with which the thoughts spurt through, and the farther they get. These forms are built of the sorts of things we would expect to constrain thought - beliefs, habits, facts, etc.

Tape vs. Blob

I think it is seductively easy to assume that we are perfectly integrated all the time. That is, we can say concretely that "I have belief p" or "I know that p". I suspect that while in general we know what we know, think what we think and believe what we believe, it often takes some effort to comb through our reality and tease out the tangles. Not all of us make this effort, and none of us does it perfectly. The underlying image that I take exception to is that of a single, unified reality model, like a great big blob or structure, with any new facts/thoughts/percepts having to be integrated into it.

In contrast, I wonder to what extent our mind is like a tape that we keep adding to our entire conscious lives, recording our facts/thoughts/percepts as memories. In this view, there is no reason at all why p and ~p could not perfectly well coexist at different places on the same tape. Obviously, there is some integration, but the image of the tape respects the sense in which our knowledge and beliefs might be context-specific strategies or algorithms rather than some kind of Pure Knowledge, fully abstracted, globally applied, and globally accessible.

Synthesis/Analysis Feedback Loop

One one hand, there is a sense that parts of my mental processing are farmed out to submodules, or performed by somewhat autonomous demons. On the other hand, there is a strong sense that I am intimately connected to my thoughts and percepts, that there is some kind of deep holism at work. Thoughts and percepts have a sort of all-at-once quality, to the point where it seems that in some sense, I am my thoughts and percepts. Assuming we have good reasons for taking these conflicting intuitions seriously, how do we reconcile them? The following is a guess at what might be happening with certain mental operations.

It seems to me that as I am forming a thought, or taking in a complex percept, I have to synthesize perceptual fragments into some kind of whole. Different sense modalities get bound; tables, chairs, pine-scented air fresheners get recognized as such individually as well as belonging in the larger context. But as soon as something comes together as a single, unified thought, it gets broken down again, analyzed, almost as if it were new raw data. Lather, rinse, repeat.

Once a thought or percept is put together into a whole, the mind seems to strive to see, given the whole and without regard to the pieces that gave rise to it, what pieces it can make out of it. Then it tries to assemble the pieces into a coherent whole again, possibly winding up with a somewhat different end result from that which it started with. I imagine this coalescing/fragmenting loop happening as many times as it has to until the whole thing stabilizes: the thing you synthesize each time through the loop, and the collection of parts you analyze it into, are the same in successive iterations. The survivors of this process, the things that live long enough, are what we recognize as thoughts or percepts.

It may be that this corresponds somewhat to Dennett's "fame in the brain" idea, the longer you sustain the loop, the more iterations, the more you attract the attention of other demons, and the more famous the process becomes. Perhaps it is the thoughts that are quickly resolved that are automatic, and never get very famous, and the tricky cases, the ambiguous ones, are those that take a while to settle down, and have to draw on demons far and wide to work out a stable configuration.

This image gives voice to the e pluribus unum, the one-yet-many sense we (or I, at least) get from consciousness. It also incorporates the idea of feedback loops, which I think must be present in minds. The parts influence the whole, which influences the parts, etc.


Multi-dimensional hyperspaces are an almost ridiculously pregnant metaphor for all kinds of things. I once read that one of the reasons for this is that a hyperspace is, to put it in modern terms, non-chaotic. Neighbors map to neighbors, always. If you have n variables, and you plot a point in an n-dimensional hyperspace, and then if you vary any of the variables continuously, your resulting point in the hyperspace will also vary continuously. On a normal two dimensional grid, you can only get away with this for two variables. If you try to represent more than two variables, at some point you will get discontinuous jumps as you continuously vary some of the variables.

How does this help us talk about minds? What if everything we know, every memory, every fact, is a blob or shape in a hyperspace? The number of variables that characterize the objects in the space, and thus the number of dimensions in the hyperspace, correspond to some set of to-be-determined primitives. Some shapes may overlap without harm, like the circles in a Venn diagram, but others are incompatible. Different shapes react in different ways when new shapes of various kinds step on them. In this view, our whole mental reality exists as some kind of n-dimensional modern art sculpture park. Some shapes are big, some are small, many enclose others, some are relatively featureless, while others have all kinds of strange appendages and protuberances. Some may even be more amorphous, and others harder, in order to accommodate (or not) intersection with others. Since we can always entertain thoughts that violate any of our existing beliefs, the rules about intersection and incursion of shapes into each other are not absolute.

The nice thing about the idea of the mind-hyperspace is that it takes some of the work out of knitting our thoughts and percepts together into a coherent single reality. You just add the new (thought, percept, etc.) to the space. If it trips over or bumps into or intersects with an existing thought, there is an incompatibility that needs to be worked out. We are thus freed of the computer model in which, in order to assimilate a new thought, we must walk through all the old thoughts, calculating their logical consequences, to see if the new thought can coexist with the old ones. Note that the hyperspace image is kind of like the burbling-spring-with-channels image, with existing thoughts, by virtue of their sitting there in the hyperspace, constraining the space into which new thoughts may form. Only instead of water (or cement) constrained by pipes, channels or plywood forms, new thoughts are yet more shapes, molded out of whatever the old thoughts are made of (kind of like plastic).

This image also respects the idea that a lot of the mind's work is inhibitory, that the tricky thing is setting up constraints. Existing habits, established long term goals, routine expectations, etc. limit the sorts of thoughts and perceptions we will have, but not absolutely: they constrain, but do not dictate precisely, and perhaps the constraints themselves are a bit squishy.

Another thing I like about this model is that it suggests a way to deal with things like this: imagine a broken appliance. Now when I did that, I immediately and effortlessly envisioned my trusty hand mixer (I've always preferred those to the counter-hogging, trendy, high-tech Kitchen Aid ones) shorting out and throwing sparks out its little air vents in the back. Now this has never happened to me, so this is not a memory I am replaying. How is it that we can so easily take a vague specification and come up with a very specific image that falls into that class? If the class is defined by a shape or blob in hyperspace, and all examples are somehow enclosed within it, all we have to do is throw a dart into the class-shape - choose a random point within it - and we come up with the hand mixer throwing sparks.

As an extra wrinkle, it may be that the shapes in the hyperspace do not have a distinct surface, but are somehow fuzzy - delineated probabilistically. Sometimes things can violate a shape's personal space, or get away with an incursion that it "shouldn't". Often, rules and boundaries can be broken, and sometimes broken a lot. It may be that the presence of an existing shape only makes it less likely that a new thought will be able to occupy that space. Certain shapes may be constituted to be more defensive of their integrity than others. As when we threw the dart to choose the broken mixer image, the generation of new thoughts may not be completely determined by the old ones, but merely constrained, so that the final placement of them in the hyperspace is determined by something else: perhaps it is random, perhaps it is some little creative tickle. Ultimately, when all other considerations are accounted for, it may be that certain patterns or linkages or contrasts or tensions simply please us.

As the metaphors go, this one contains even more hand-waving than most. What are the bases, that determine the number of dimensions? One dimension (thus one axis) for each primitive quale? For each new kind of idea? Can't the mind create new dimensions on the fly? How does that work, without incurring a combinatorial explosion? Surely there isn't a "broken appliance" shape in all of our minds; surely we don't create such a shape until asked the question, so aren't we right back where we were, having to walk through laboriously, gathering all the disparate examples of appliances that could be broken, breaking them, and tossing them into the newly created "broken appliance" shape? And how would this be manifested physically or computationally? The whole sculpture park also seems pretty static, too - it seems that the shapes should do more, or change moment to moment.

Yes, yes, and yes. It's just an image, but one I keep thinking in terms of. Your mileage may vary.


You may have noticed by now that some of these metaphors of the mind are mutually exclusive. The itch that motivated the tape vs. blob image is not respected very well by the sculpture park in hyperspace image. Moreover, both of them are pretty static - it is unclear who or what does any work in either. Demons and parliaments, however, do lots of work. To what extent do existing thoughts constrain the space in which new thoughts can occur, as opposed to actively constructing new thoughts, as opposed to being the raw material out of which new thoughts are constructed (possibly by the new thoughts themselves)?

I think it is not the case that old thoughts simply form the skeleton or scaffolding of which new ones are constructed. Perhaps there are ways of combining parts of the images (Darwinian, competing demonic ever-shifting shapes in a hyperspace? A tape on which is recorded a domain-specific structure corresponding to each moment's mental state?) How do thoughts fall along the descriptive/prescriptive, data vs. algorithm divide? And what is it like to see red, anyway? Not much help there. I promised very little at the beginning of this essay - just a grab bag of images and metaphors. The hope is that they add to our vocabulary, so that when we have to throw around a figure of speech colloquially, we have a broader range of such figures of speech to choose from, and we do not automatically rule out entire classes of models, hypotheses, and theories in our choices of words.