William H. Calvin, "Ephemeral Levels of Mental Organization: Darwinian Competitions as a Basis for Consciousness" Tucson III Consciousness proceedings (Stuart Hameroff, Al Kaszniak and David Chalmers, editors), to be published by MIT Press in 1999. See also

Webbed Reprint Collection
This 'tree' is really a pyramidal neuron of cerebral cortex.  The axon exiting at bottom goes long distances, eventually splitting up into 10,000 small branchlets to make synapses with other brain cells.
William H. Calvin

University of Washington
Seattle WA 98195-1800 USA

copyright ©1998

William H. Calvin

Ephemeral Levels of Mental Organization:
Darwinian Competitions as a Basis for Consciousness



William H. Calvin

Department of Psychiatry and Behavioral Sciences
University of Washington
Seattle WA 98195-1800 USA


While I appreciate the scientific technique of picking away at pieces of a problem, I think that we greatly underestimate consciousness when we attempt to redefine it as merely awareness or selective attention. These are rather low-level capabilities in the neurologists pantheon of mental abilities.
    Consciousness seems closer to the higher intellectual functions such as structured language, planning ahead in novel situations, composing music, logical chains of inference, our fascination with games and rules, and our delight in discovering the hidden patterns in puzzles and humor -- all things that are seldom found even in our closest cousins, the great apes. A richer definition of consciousness must also encompass the transitions from one mental state to another, those dynamics that William James described in 1880:

"Instead of thoughts of concrete things patiently following one another in a beaten track of habitual suggestion, we have the most abrupt cross-cuts and transitions from one idea to another, the most rarefied abstractions and discriminations, the most unheard-of combinations of elements, the subtlest associations of analogy; in a word, we seem suddenly introduced into a seething caldron of ideas, where everything is fizzling and bobbing about in a state of bewildering activity, where partnerships can be joined or loosened in an instant, treadmill routine is unknown, and the unexpected seems the only law."

Just as intelligence has been described as "What you use when you dont know what to do," when no standard response will suffice, so too consciousness is prominently involved when the situation contains ambiguity or demands creative responses, ones that cannot be handled by a decision tree. Many mental activities can be handled by subroutines; consciousness helps to deal with the leftovers (and create new subroutines for next time).     But there is some method to all of this floundering around. The first key concept is that of levels of organization, and the second is that of darwinian approaches to creating and stabilizing new levels.

Comments to

William H. Calvin

Layers of middlemen are familiar from everyday economics, and our consciousness might well depend on how many there are. As Derek Bickerton (1990) noted:

[T]he more consciousness one has, the more layers of processing divide one from the world. . . . Progressive distancing from the external world is simply the price that is paid for knowing anything about the world at all. The deeper and broader [our] consciousness of the world becomes, the more complex the layers of processing necessary to obtain that consciousness.

Although it seems to have played little role so far in our modern investigations of consciousness, a hierarchy of levels of organization is a common concept in the sciences and in familiar technologies.
    Levels are best defined by certain functional properties, not anatomy. As an example of four levels, fleece is organized into yarn, which is woven into cloth, which can be arranged into clothing. Each of these levels of organization is transiently stable, with ratchet-like mechanisms that prevent backsliding: fabrics are woven, to prevent their disorganization into so much yarn; yarn is spun, to keep it from backsliding into fleece.
    A proper level is also characterized by "causal decoupling" from adjacent levels; its a "study unto itself." For example, you can weave without understanding how to spin yarn (or make clothing). Chemistry is a proper level: Mendeleyev discovered the patterns of the table of elements without knowing anything about the underlying patterns of electron shells (or the overlying patterns of the Krebs Cycle).
    Mental life can pyramid a number of levels. Some of the major tasks of early childhood involve discovering four levels of organization in the apparent chaos of the surrounding environment. Infants discover phonemes and create standard categories for them; six-month-old Japanese infants can still tell the difference between the English /L/ and /R/ sounds but after another six months of regular exposure to the Japanese phoneme that lies in between them in sound space, the baby will treat the occasional English sounds as mere imperfect versions of the Japanese phoneme. With a set of basic speech sounds, babies start discovering patterns amid strings of phonemes, averaging six new words every day. Between 18-36 months of age, they start to discover patterns of words called phrases and clauses, rules such as add -s for plural, add -ed for past tense. After syntax, then they go on to discover Aristotles rule about narratives having a beginning, middle, and end (and they then demand bedtime stories with a proper ending).
    We take great delight in discovering a hidden pattern, whether in a jigsaw puzzle or a subtle work of art; indeed, we often jump to conclusions, finding patterns where none exist. Ive often been fooled by the wind blowing around the house, hearing speech and assuming that a radio has been left on in another room. The delight in finding hidden patterns is part of what reinforces doing science (and, for that matter, astrology).
    This operates at both subconscious levels (those poorly-heard radio voices) and conscious ones ("seeing" the art). As Heinz Pagels pointed out in his posthumous 1988 book, The Dreams of Reason, consciousness likely involves operating at a high level:

Such a series of "causal decouplings" may be extraordinarily complex, intricate beyond our current imaginings. Yet finally what we may arrive at is a theory of the mind and consciousness  a mind so decoupled from its material support systems that it seems to be independent of them.... The biological phenomenon of a self-reflexive consciousness is simply the last of a long and complex series of "causal decouplings" from the world of matter.

But how are new levels created? What stabilizes some of them against unraveling? Jacob Bronowski (1973) preached a concept called stratified stability: "The stable units that compose one level or stratum are the raw material for random encounters which produce higher configurations, some of which will chance to be stable...." Randomness can, in the right circumstances, be creative.
    And the Darwinian process is particularly good at taking random variations and making new stable configurations, ones that would be highly improbable without the Darwinian algorithm. It can combine concepts such as horse and rhinoceros to yield imaginary concepts such as unicorn. Its intermediate stages ought to look something like your nighttime dreams, with their jumble of people, places, and occasions that dont really fit together very well. When we are awake, we are likely only aware of the highest-quality result.
    Ive mostly studied Darwinian processes for the higher intellectual functions (Calvin and Ojemann 1994, Calvin 1996ab, Calvin and Bickerton 1999); they are indeed capable of providing a Jamesian type of consciousness, with those sudden shifts and improbable associations, simply as a byproduct of copying competitions with playoffs. Here I am going to propose that Darwinian processes invent novelties, bootstrap their quality through a series of generations of improvements on the time scale of thought and action. This can happen at a number of levels of organization, constructing a "house of cards" which, while it may eventually collapse, can attain some heights in the meantime.
    Atop a level of codes for objects, one might have a level of relationships (such as simple sentences), and a level of relations among relationships (analogies) atop that. By temporarily stabilizing the intermediate levels, we are sometimes able to tack on an even higher level. A high-level outcome of such a process is what we might want to call consciousness (though the heights vary throughout the day; the relations level may be impossible until after morning coffee). To demonstrate how the cerebral cortex can operate a Darwinian process, and stabilize a new level, I will briefly introduce the essentials of any Darwinian process and then summarize how neocortex can implement one.

Comments to

William H. Calvin
Most disciplines within the natural sciences feature only one level of organization; there are, however, at least a dozen levels of organization within the neurosciences. While this makes for a rich subject, it also leads to much confusion and arguing at cross purposes (consider all the conflict over whether learning is a matter of gene expression, ion channel, synaptic, neuron, or circuit-level alterations).
    Each neuron has thousands of synapses, producing currents that summate to produce (via a nonlinear input-output transformation) an impulse train. But only rarely does the activity of a single neuron suffice to produce a perception or trigger an action. Mostly, neurons act as members of committees, what Hebb (1949) called a cell-assembly; just as in academia, one individual may function in different committees on different occasions. The best-known examples of hard-to-fathom committees, artificial neural networks (ANNs), lack the bi-level functionality of real neural networks, where both synaptic strength and impulse patterning are important; they usually do not incorporate the important wiring principles seen in cerebral cortex, such as the patterned sideways excitation capable of temporarily organizing large areas of cortex into clones (Calvin 1996b).
    Real neural networks are also in the business of producing spatiotemporal patterns, the coordinated sequence of activity that it takes to generate the body movements needed for manipulation and speech. Though spatial-only patterns can serve to categorize learned patterns from the input (as "connection strengths" of ANNs show), spatiotemporal patterns are needed to perform most movements or think about something.
    Macrocolumns such as the half-mm ocular dominance columns are the best-known intermediate-sized unit of cortical organization, and I earlier predicted (Calvin 1996b) a half-mm hexagonal-shaped module with characteristic spatiotemporal patterns associated with its synaptic strengths. Its defining characteristic is that it can make copies of its spatiotemporal pattern (rather like a little tune, were the cells of the hexagon to be mapped onto a musical scale), cloning it into adjacent patches of cortex. This can temporarily create a hexagonal mosaic of some centimeters square (rather like a plainchant choir, all singing the same melody). Because several mosaics may partially overlap when they encounter one another, composite codes may be created (and harmony-like issues arise), generating arbitrary concepts such as unicorns.
    Not only may a hexagons melody represent an object, but it can code for an action, a relationship, an analogy  even a structured sentence with phrases and clauses. A hexagonal mosaic is expansionistic, rather like a plainchant choir that recruits additional "lay" members in the manner of the expert choir singing the Hallelujah Chorus and recruiting the audience. Any level of organization can have a code and exhibit competition for territory.
    Though I originally investigated redundant codes because of their importance for precision timing, much needed for accurate throwing (Calvin 1983), another important consequence of cloning and competition for territory is that it can implement a Darwinian process, our finest example of how quality can be recursively bootstrapped. However, not everything called "Darwinian" has the possibility of recursion.

Simple carving and sorting processes show selective survival at work, e.g., beaches that progress from shingle to pebble to sand, all because heavier objects are not carried as far by flowing water. Similarly, one sees selection at work in processes that try a lot, then cull, then try some more, e.g., leaves may bud all over the tree, but only the ones on the sunny side survive a drought. Neural connections also have enormous sprouting early in life, then culling during childhood to create the adult patterns, e.g., there is a 70 percent drop in the axon count of monkey corpus callosum and a reduction of 30 to 50 percent in the number of synapses per neocortical neuron before maturity. Though we presently lack the techniques to measure it, there is likely a certain amount of weekly turnover: new synapses that are formed "on spec" but culled if they fail to "establish themselves."
    Such selection, while very important in wiring up and operating the brain, is not the Darwinian process that so impresses us from the accomplishments of species evolution and the immune response. So far as I can see (Calvin 1996b, 1997), there are six essential ingredients

1. Theres a pattern (genes, memes).

2. The pattern is copied (indeed, what can be semi-reliably copied tends to define the relevant pattern).

3. Variant patterns arise (via copying errors, or recombinations).

4. Populations of some variants compete for a workspace, e.g., bluegrass & crabgrass compete for my back yard.

5. There is a multifaceted environment that makes some variants more common (how often you cut, fertilize, water -- this is what Darwin called "natural selection").

6. The more successful variants are the most frequent center for further variants (Darwins inheritance principle).

Even non-biological processes may implement the six essentials, e.g., storytelling (Calvin 1997) as a basis for history. But if a process lacks any one of the six, it runs out of steam. Neural sprouting and pruning may create a pattern, but it doesnt clone itself to create populations that compete for territory, and so forth. Cloning and competition, if it lacks Darwins feature of centering the variants of the next generation on the more successful of the current generation, may just wander endlessly. It is unfortunate that Darwin named his theory for just one element of the six, because "natural selection" has misled many people into assuming that variation (#2) plus selection (#5) buys you the power of evolution -- but that recursive bootstrapping of quality that we find so impressive when new levels of organization are discovered, likely requires all six.
    There are a number of other things that, while not essential to the Darwinian algorithm in the strict sense, may stabilize it or speed it up, making them particularly important for any Darwinian process involved in thought and action. We often suffer from what the French call avoir l'esprit de l'escalier -- finally thinking of a witty reply, but only after leaving the party. There are fleeting "windows of opportunity" in more important kinds of behavior, too, whether prey or predator. So we might well expect a brain to find these five additional features of particular importance as well:

7. Stagnation may occur if variants cannot effectively escape a "well" and always backslide -- but that can also buy you stabilization.

8. Systematic recombination, e.g., sex doesnt leave variability to chance.

9. Fluctuating climate (more severe selection, more frequent culling -- and therefore more frequent opportunities when the climate improves again). Just as in stochastic resonance, noise helps to rapidly discover new niches.

10. Patchy subdivisions (island biogeography promotes inbreeding; also, a higher percentage live out on the habitats margins where making a living is, well, marginal). Fragment-then-reunite sequences serve to "pump" the central percentages of those variants capable of making a living on the margins.

11. Emptied niches to refill (with no competition for a few generations, rare variants get a chance to reproduce -- and they may have what it takes to survive the next crunch).

Comments to

William H. Calvin
Figure 1. Any sensory object (say, an apple) activates a number of feature detectors, each of which may form a triangular array. Hundreds of different ones are possible; the largest cortical area with one, and only one, member of each triangular array is a hexagon about 0.5 mm across. The spatiotemporal pattern within that hexagon can be considered a cerebral code for apple - perhaps not the most elementary code, but one capable of being copied and superimposed on other codes to form combination codes. Here an apple icon is used as a stand-in for the spatiotemporal pattern (upper left). Looking down on the cortex might reveal a territory of apple clones, another of banana or cherry clones, all competing for territory in association cortex. Their success depends not only on sensory inputs and drives, but on specific resonances in the synaptic connectivity of the neural networks; many hexagons will have multiple resonances (lower middle), just as the spinal cord's connectivity simultaneously supports a number of spatiotemporal patterns called the gaits of locomotion. Hexagonal codes can also serve as cortical codes for complex movements.

While space does not allow me to summarize the cortical circuitry that makes the hexagonal competitions possible (Calvin 1996ab), it does exhibit all six Darwinian essentials:

1. Theres a pattern (that spatiotemporal action pattern which fits into a 0.5mm hexagon).

2. The pattern is copied (recurrent excitation with express-train-like gaps produces mutual connections that entrain and then recruit).

3. Variant patterns arise (when hexagons escape conforming neighbors, or when patterns overlap).

4. Populations (hexagonal mosaics) of some variants compete for a workspace (there are "dueling choirs" in association cortex).

5. There is a multifaceted environment that makes some variants more common (sensory inputs and memorized resonances bias which pattern succeeds).

6. The more successful variants are the most frequent center for further variants (the bigger a territory, the more edge it has -- and thus more opportunities to escape the crystal-like conformity of central regions), giving rise to inheritance.

Success and quality are biased by the current sensory inputs and other aspects of the real-time environment, by the synaptic connectivity which has memorized the salient features of prior environments via learning experiences, and by the inborn patterning of cortical connections.

    Superficial neocortical circuitry also has all of the four optional catalysts (more later), and it produces in advance the spatiotemporal patterns that are needed for converting thought into action. So copying competitions of neocortex seem to have many appropriate ingredients for portraying the elements of an active mental life -- and the dynamics to quickly get from one mental state to a successor.

In this view, the subconscious immediately finds a home. There are many competitions running simultaneously, each with a local winner constituting a subconscious concept or thought (though most are of that low incoherent quality associated with dreams). Each winner can, via corticocortical connections, enter into competitions with more distant local winners. Thus one can have playoff competitions, for temporarily becoming the melody with the biggest choir (and perhaps this is what wed report as the content of our consciousness if asked, and if we could find means of expressing the concept). Winning doesnt last for very long, habituation being what it is. The content of consciousness shifts unpredictably when a concept choir from a different branch of the playoff tree finally gains the top slot.
    There is no "central place" for this kind of consciousness, in agreement with the neurology about what strokes can and cannot eliminate. Youd expect the winner to shift around to a series of places in both left and right brain, in front and rear. But theyre all in neocortex in this view, simply because subcortical structures and old-fashioned versions of cortex such as hippocampus arent known to have the cloning circuitry needed for the copying competitions. So far, thats a feature of the superficial layers of neocortex in many Brodmann areas, and in many mammalian species (rats are the notable exception).
    Some of these competitions involve codes (i.e., spatiotemporal firing patterns) that represent words, others relationships at various possible levels. It is tempting to treat consciousness as the current highest level of organization, not simply the one with the most numerous choir. When you first contemplate the toothpaste in the morning, the level of consciousness might not be very high, operating merely at the level of objects or simple actions, barely managing to use the level of relationships. The relations between relationships level (analogies) may require a double espresso. Poets, of course, have to compare metaphors, which requires a series of stage-setting preliminaries to construct and stabilize this ephemeral house of cards. Understanding such staging and scaffolding might someday allow us to spend more time at more abstract levels -- or even invent new levels (one can almost imagine meta-poets).

Comments to

William H. Calvin

Speed of operation likely involves cerebral versions of the four catalysts seen in more traditional evolutionary media. An animal species, to take a familiar example, has a range; imagine one irregular area, with much mobility and therefore mixing between center and periphery. As the climate worsens and the population numbers decline, this well-mixed population breaks up into a number of fragments. The subpopulations inbreed because theres nothing to eat in between subpopulations to support a journey; effectively, there are now many "islands" where there used to be a single "continent." And a higher percentage of individuals live near the edge of a subpopulation where conditions are marginal for making a living.
    If a subpopulation is sufficiently small, it will be atypical simply because of the "luck of the draw" (even if a jury pool is racially balanced, an individual jury drawn from it may be very lopsided). Things that depend on a critical mass of some infrequent characteristic (reciprocal altruism, for example; see Calvin and Bickerton 1999) may finally get a chance, simply because of this random fragmentation of the larger population.
    When climate later improves, some of these biased subpopulations may get a major chance to repopulate an abandoned territory, and so increase their percentage in the overall metapopulation. So pervasive fluctuations of any major resource have effects on evolution that go far beyond the caricature of "more severe selection." Furthermore, a series of population bottlenecks with subsequent re-expansion may occur because climate episodes repeat, allowing formerly rare variants to be pumped up until they are the majority type. 9;
    We might expect the brains "weather" (for which EEG rhythms might be an indicator) to do similar things to a neocortical Darwinian process. But with the ease of recombination seen where hexagonal mosaics overlap, the "species tree" is likely to look more like a web; binary trees can be maintained only if hybrids are typically sterile. And, in the cortex, there are all of the interrelationships between levels of organization (that hexagonal code can represent anything from a category to a metaphor) to consider. One levels competition could serve to bias the operation of another. Different levels might sometimes operate on different time scales. For example, a slow Darwinian competition ("agenda") could bias the faster competitions implementing thought and action  perhaps simply by varying the rate and extent of excitability fluctuations (furthermore, randomness is not really a requirement; fluctuations could easily average two little steps left for every big step right).

Backsliding needs to be slowed if new levels of organization are to maintain themselves. Memories are one way; as we know from semiotics and linguistics, symbols can be quite arbitrary and so even inadequate metaphors can be saved, enabling the copying competition to be restarted from some advanced position rather than starting from scratch each time. Ratchets may develop, simply because of the interaction of different time scales.
    A neocortical Darwinian regime may also have periods of "monoculture" where a particularly widespread mosaic develops featuring a uniform spatiotemporal firing pattern. In the musical-mapping analogy, this would correspond to a large plainchant choir. If this is maintained for some tens of minutes, it will prevent antecedent patterns from renewing their fading synaptic strengths (recalling something from short-term memory likely serves to restart the clock). Such a monoculture period might serve to allow for fresh starts afterward, with novel spatiotemporal patterns less likely to be captured by the resonances of the antecedent thoughts. Sleep may be one example; meditation with a meaningless mantra, long enough for temporarily enhanced synaptic strengths to fade, might similarly help to "clear the mind."

Comments to

William H. Calvin

Though our techniques are improving, no one has yet been able to look down on the cerebral cortex and see what Charles Sherrington predicted sixty years ago in his book, Man on his Nature:

"The brain is waking and with it the mind is returning. It is as if the Milky Way entered upon some cosmic dance. Swiftly the [cortex] becomes an enchanted loom where millions of flashing shuttles weave a dissolving pattern, always a meaningful pattern though never an abiding one; a shifting harmony of subpatterns.... Dissolving pattern after dissolving pattern will, the long day through, without remission melt into and succeed each other...."

Nor have we been able to see those cell-assemblies that Donald Hebb imagined fifty years ago in The Organization of Behavior. We only invented the microelectrode techniques, needed to study one cortical neuron at a time, about forty years ago and it is still a feat to study a dozen individual neurons at the same time. The picture that I have painted here predicts much more detail about those subpatterns (my hexagonal spatiotemporal pattern is essentially a candidate for the minimal cell-assembly that has one of everything but no redundancy), yet it too is beyond the bounds of what the microelectrode and imaging technologies can currently deliver.
    Fortunately, we have learned a great deal about the evolutionary process in the 160 years since Darwins discovery. When one sees the patterned cortical circuitry in the neuroanatomy, and combines it with what we currently know of the neurophysiology of neurons and synapses (and particularly the emergent properties of mutually exciting oscillators), we can at least say that similar dynamics ought to apply in cortical climates and landscapes, that island biogeography ought to operate there just as well as it operates in Hawaii, that repeated bottlenecks could operate in minutes as well as in millennia.
    The neocortex has what it takes to run Darwins recursive bootstrap for quality  and that it helps explain a great deal about how higher intellectual functions could operate. In particular, it gives us a way of thinking about consciousness that is free of dualism and miraculous leaps across a dozen levels of organization from quantum mechanics to consciousness. It is compatible both with the observed cortical localization of function and with the apparent lack of a location for consciousness.
     Is that all consciousness really is, just the current winner of a widespread copying competition in neocortex, biased by subcortical and sensory influences and the memories of past episodes? It certainly seems to be a good default explanation, the one against which you test your candidates for something more. While I see no reason to postulate higher controlling processes of greater intelligence, I would caution that we understand very little as yet about levels. (Is consciousness the highest level, or just an in-between level with widespread connections, perhaps capable of being spoken?) We know little about how a Darwinian success can be converted into a subroutine that obviates the copying competition after enough experience.
     But increasingly, consciousness is no longer a mystery (Dennett 1991), something that we simply dont know how to think about -- in the same way that the boundary between the living and the nonliving used to be a mystery, or the origin of the universe used to be a mystery. We can now imagine what the brain is doing with those express-train connections of neocortex. For those with enough acquaintance with biology to be able to think in population terms (Mayr 1994), there is now a rich evolutionary carryover that provides a framework for further investigation.

Comments to

William H. Calvin


Derek Bickerton (1990), Language and Species (University of Chicago Press).

Jacob Bronowski (1973), The Ascent of Man (Little, Brown), pp. 348-349.

William H. Calvin (1983). A stone's throw and its launch window: timing precision and its implications for language and hominid brains. Journal of Theoretical Biology 104:121-135.

William H. Calvin (1996a). How Brains Think: Evolving Intelligence, Then and Now (Basic Books).

William H. Calvin (1996b). The Cerebral Code: Thinking a Thought in the Mosaics of the Mind (MIT Press).

William H. Calvin (1997). The Six Essentials? Minimal Requirements for the Darwinian Bootstrapping of Quality. Journal of Memetics - Evolutionary Models of Information Transmission, 1, at

William H. Calvin (1998). Competing for consciousness. Journal of Consciousness Studies. 5(4)388-404.

William H. Calvin and Derek Bickerton (1999). Lingua ex machina: Reconciling Darwin and Chomsky with the Human Brain. MIT Press, to appear 1999.

William H. Calvin and George A. Ojemann (1994). Conversations with Neil's Brain: The Neural Nature of Thought and Language (Addison-Wesley).

Daniel C. Dennett (1991), Consciousness Explained (Little, Brown).

Donald O. Hebb (1949). The Organization of Behavior (McGraw-Hill).

William James (1880), "Great men, great thoughts, and the environment," The Atlantic Monthly 46(276):441-459.

Ernst Mayr (1994), "Population thinking and neuronal selection: metaphors or concepts?" In Selectionism and the Brain, edited by Olaf Sporns and Giulio Tonini (Academic Press), pp.27-34

Heinz Pagels (1988), The Dreams of Reason (Simon & Schuster).

Charles Sherrington (1940), Man on his Nature, the Gifford Lectures, Edinburgh, 1937-38 (Cambridge University Press), p. 178. || Home Page || Calvin publication list || The Calvin Bookshelf || revised 1 August 1998