From Bacteria to Bach and Back: The Evolution of Minds

by Daniel C. Dennett

evolution is a process that depends on amplifying things that almost never happen.

What Darwin and Turing did was envisage the most extreme version of this point: all the brilliance and comprehension in the world arises ultimately out of uncomprehending competences compounded over time into ever more competent—and hence comprehending—systems.

Distribution of expertise or understanding of this sort is a hallmark of human creative projects, and it is clearly essential for today’s high-tech artifacts, but not for all earlier artifacts. A lone artificer can make a spear, or even a kayak or a wooden wagon or a thatched hut, understanding every aspect of the design and construction, but not a radio or an automobile or a nuclear power plant.

At the height of operations over 130,000 people worked full time on the project, and only a tiny percentage of them had any idea at all what they were making.

Darwin’s “strange inversion of reasoning” and Turing’s equally revolutionary inversion were aspects of a single discovery: competence without comprehension. Comprehension, far from being a Godlike talent from which all design must flow, is an emergent effect of systems of uncomprehending competence: natural selection on the one hand, and mindless computation on the other. These twin ideas have been proven beyond a reasonable doubt, but they still provoke dismay and disbelief in some quarters, which I have tried to dispel in this chapter. Creationists are not going to find commented code in the inner workings of organisms, and Cartesians are not going to find an immaterial res cogitans “where all the understanding happens.”

“How else?” one might well ask. Intelligent design of this sort starts with a goal (which may well be refined or even abandoned along the way) and works top-down, with the designers using everything they know to guide their search for solutions to the design problems (and sub-problems, and sub-sub-problems …) they set for themselves.

Evolution, in contrast, has no goals, no predefined problems, and no comprehension to bring to the task; it myopically and undirectedly muddles along with what it has already created, mindlessly trying out tweaks and variations, and keeping those that prove useful, or at least not significantly harmful.

a process with no Intelligent Designer can create intelligent designers who can then design things that permit us to understand how a process with no Intelligent Designer can create intelligent designers who can then design things.

The intermediate steps are instructive. What about the clam rake gives away its artifactual status? Its very simplicity, which indicates its dependence on something else for its ability to defy the Second Law of Thermodynamics, persisting as uniform and symmetrical collections of atoms of elements in

improbable juxtapositions. Something gathered and refined these collections....

affordances: “What the environment offers the animal for good or ill.” Affordances are the relevant opportunities in the environment of any organism: things to eat or mate with, openings to walk through or look out of, holes to hide in, things to stand on, and so forth.

The chief difference is that the design of the elevator circuits was done by intelligent designers who had worked out descriptions of the problems, and representations of reasoned solutions, complete with justifications. In the R&D history of the bacteria, there was no source code, and no comments were ever composed, to provide hints of what Mother Nature intended.

How does Nature debug its designs? Since there is no source code or comments to read, there can be no debugging by brilliant intellectual explanation; design revision in Nature must follow the profligate method of releasing and test-driving many variants and letting the losers die, unexamined. This won’t necessarily find the globally optimal design but the best locally accessible versions will thrive, and further test-driving will winnow the winners further, raising the bar slightly for the next generation.20 Evolution is, as Richard Dawkins’s (1986) memorable title emphasizes, the Blind Watchmaker, and given the R&D method used, it is no wonder that evolution’s products are full of opportunistic, short-sighted, but deviously effective twists and turns—effective except when they aren’t!

The so-called intelligence in trees and sponges and insects is not theirs; they are just brilliantly designed to make smart moves at the right time, and while the design is brilliant, the designer is as uncomprehending as they are.

The total weight of all life on the planet now—the biomass—is currently estimated as more than half made up of bacteria and other unicellular “robots,” with “robotic” plants making up more than half the rest. Then there are the insects, including all the clueless termites and ants that outweigh the huge human population celebrated by MacCready. We and our domesticated animals may compose 98% of the terrestrial vertebrate biomass, but that is a small portion of life on the planet. Competence without comprehension is the way of life of the vast majority of living things on the planet and should be the default presumption until we can demonstrate that some individual organisms really do, in one sense or another, understand what they are doing.

Competence without comprehension is the way of life of the vast majority of living things on the planet and should be the default presumption until we can demonstrate that some individual organisms really do, in one sense or another, understand what they are doing.

Often, the female cuckoo will roll one of the host eggs out of the nest—in case the host parents can count. And as soon as the cuckoo chick is hatched (and it tends to incubate more quickly than the host eggs), the little bird goes to great efforts to roll any remaining eggs out of the nest. Why? To maximize the nurture it will get from its adoptive parents. The video

Look up

Bearing in mind that evolution can only provide for challenges encountered during R&D, we can predict that the more novel the artificial intrusions in the bird’s Umwelt are, the less likely it is that the bird will interpret them appropriately, unless the bird’s lineage evolved in a highly varied selective environment that obliged natural selection to settle on designs that are not entirely hard-wired but have a high degree of plasticity and the learning mechanisms to go with it.

Learning can take over where natural selection left off, optimizing the individuals in their own lifetimes by extracting information from the world encountered and using it to make local improvements.

The rule of attribution must be then, if the competence observed can be explained without appeal to comprehension, don’t indulge in extravagant anthropomorphism. Attributing comprehension must be supported by demonstrations of much more intelligent behavior.

How could experiments support the verdict of comprehension? By showing that the animals can do what we comprehenders can do with variations on the behavior.

The behavior is neither a simple “knee-jerk” reflex inherited from her ancestors nor a wily scheme figured out in her rational mind; it is an evolution-designed routine with variables that respond to details in the circumstances, details that the sophisticated soliloquy captures—without excess—in the rationale of that

Comprehension is not the source of competence or the active ingredient in competence; comprehension is composed of competences.

comprehension comes in degrees. At one extreme we have the bacterium’s sorta comprehension of the quorum-sensing signals it responds to (Miller and Bassler 2001) and the computer’s sorta comprehension of the “ADD” instruction. At the other extreme we have Jane Austen’s comprehension of the interplay of personal and social forces in the emotional states of people and Einstein’s comprehension of relativity. But even at the highest levels of competence, comprehension is never absolute. There are always ungrasped implications and unrecognized presuppositions in any mind’s mastery of a concept or topic. All comprehension is sorta comprehension from some perspective.

Where understanding is concerned, we all depend on something like a division of labor: we count on experts to have deep, “complete” understanding of difficult concepts we rely on every day, only half-comprehendingly.

We idealize everybody’s thinking, and even our own access to reasons, blithely attributing phantom bouts of clever reasoning to ourselves after the fact. We tend to see what we chose to do (a chess move, a purchase, parrying a blow) to have been just the right move at the right time, and we have no difficulty explaining to ourselves and others how we figured it out in advance, but when we do this we may often be snatching a free-floating rationale out of thin air and pasting it, retrospectively, into our subjective experience.

fixed, created by the R&D of evolution by natural selection. They are born “knowing” all they will ever “know”; they are gifted but not learners. Each generation generates variations which are then tested against Nature, with the winners copied more often in the next round. Next come the Skinnerian creatures, who have, in addition to their hard-wired dispositions, the key disposition to adjust their behavior in reaction to “reinforcement”; they more or less randomly generate new behaviors to test in the world; those that get reinforced (with positive reward or by the removal of an aversive stimulus—pain or hunger, for instance) are more likely to recur in similar circumstances in the future. Those variants born with the unfortunate disposition to mislabel positive and negative stimuli, fleeing the good stuff and going for the bad stuff, soon eliminate themselves, leaving no progeny.

Better still is the next grade, the Popperian creatures, who extract information about the cruel world and keep it handy, so they can use it to pretest hypothetical behaviors offline, letting “their hypotheses die in their stead” as the philosopher of science Karl Popper once put it. Eventually they must act in the real world, but their first choice is not random, having won the generate-and-test competition trial runs in the internal environment model. Finally, there are the Gregorian creatures, named in honor of Richard Gregory, the psychologist who emphasized the role of thinking tools in providing thinkers with what he called “potential intelligence.” The Gregorian creature’s Umwelt is well stocked with thinking tools, both abstract and concrete: arithmetic and democracy and double-blind studies, and microscopes, maps, and computers.

The merely Darwinian creature is “hard-wired,” the beneficiary of clever designs it has no need to understand. We can expose its cluelessness by confronting it with novel variations on the conditions it has been designed by evolution to handle: it learns nothing and flounders helplessly. The Skinnerian creature starts out with some “plasticity,” some optionality in a repertoire of behaviors that is incompletely designed at birth; it learns by trial-and-error forays in the world and is hard-wired to favor the forays that have “reinforcing” outcomes. It doesn’t have to understand why it now prefers these tried-and-true behaviors when it does; it is the beneficiary of this simple design-improvement ratchet, its own portable Darwinian selection process. The Popperian creature looks before it leaps, testing candidates for action

against information about the world it has stored in its brain somehow. This looks more like comprehension because the selective process is both information-sensitive and forward-looking, but the Popperian creature need not understand how or why it engages in this pretesting. The “habit” of “creating forward models” of the world and using them to make decisions and modulate behavior is a fine habit to have, whether or not you understand it. Unless you were a remarkably self-reflective child, you “automatically” engaged in Popperian lookahead and reaped some of its benefits long before you noticed you were doing it. Only with the Gregorian creature do we find the deliberate introduction and use of thinking tools, systematic exploration of possible solutions to problems, and attempts at higher-order control of mental searches.

To summarize, animals, plants, and even microorganisms are equipped with competences that permit them to deal appropriately with the affordances of their environments. There are free-floating rationales for all these competences, but the organisms need not appreciate or comprehend them to benefit from them, nor do they need to be conscious of them. In animals with more complex behaviors, the degree of versatility and variability exhibited can justify attributing a sort of behavioral comprehension to them so long as we don’t make the mistake of thinking of comprehension as some sort of stand-alone talent, a source of competence rather than a manifestation of competence.

The task of a nervous system is to extract information from the environment to use in modulating or guiding successful behavior.

instances of semantic information, since we identify the information of interest to us on a particular occasion by specifying what it is about (events, conditions, objects, people, spies, products …).

even the dross that sticks in our heads from the flood that bombards us every day has its utility profile. Much of it sticks because it is designed to stick, by advertisers and propagandists and other agents whose interests are served by building outposts of recognition in other agents’ minds, and, as Sterelny notes, much of the rest of it sticks because it has some non-zero probability (according to our unconscious evaluations) of being adaptive someday. People—real or mythical—with truly “photographic memories” are suffering from a debilitating pathology, burdening their heads with worse than useless information.

when to flower.31 We can think about potential utility if we like: Suppose a man with a chain saw is approaching, visible to any organism with eyes but not to the tree. Eyes are of no use to a tree unless it also has some way of using the information (if not to run and hide, perhaps to drop a heavy limb on the lumberjack, or secrete some sticky sap that will gum up the saw). The presence of the information in the light might someday “motivate” a trend in the direction of eyesight to trees, if a behavioral payoff were nearby! Unlikely,

In all these cases, semantic information about how best to fit in has been mindlessly gleaned from the cycle of generations, and notice that it is not encoded directly in the organisms’ nervous systems (if they have any) or even in their DNA, except by something akin to pragmatic implication. Linguists and philosophers of language use the term pragmatics to refer to those aspects of meaning that are not carried by the syntax and “lexical” meanings of the words but conveyed by circumstances of particular utterances, by the Umwelt, in effect, of an utterance.

“where does all the information in the DNA come from?” and Darwin’s answer is simple: it comes from the gradual, purposeless, nonmiraculous transformation of noise into signal, over billions of years. Innovations must (happen to) have fitness-enhancing effects from the outset if they are to establish new “encodings,” so the ability of something to convey semantic information cannot depend on its prior qualification as a code element.

We see the acquisition of both know-how and factual information as learning, and it is always a matter of using the base of competence/knowledge you already have to exercise quality control over what you acquire.

Since Aesop we’ve known that the boy who cries wolf stops commanding attention and credence after a while. Batesian mimicry (such as a nonpoisonous snake with markings that mimic a poisonous variety) is a similar kind of parasitism, getting a benefit without going to the cost of manufacturing poison, and when the mimics outnumber the genuinely poisonous snakes Aesop’s moral takes hold and the deceitful signal loses its potency.

The claim that I defend is that human culture started out profoundly Darwinian, with uncomprehending competences yielding various valuable structures in roughly the way termites build their castles, and then gradually de-Darwinized, becoming ever more comprehending, ever more capable of top-down organization, ever more efficient in its ways of

searching Design Space. In short, as human culture evolved, it fed on the fruits of its own evolution, increasing its design powers by utilizing information in ever more powerful ways.

It is true that the brain’s architecture is massively parallel, with a vision system about a million channels wide, for instance, but many of the brain’s most spectacular activities are (roughly) serial, in the so-called stream of consciousness, in which ideas, or concepts or thoughts float by not quite in single file, but through a von Neumann bottleneck

of sorts. You can simulate a virtual serial machine on a parallel architecture—that’s what the brain does, as I showed in Consciousness Explained—and virtual parallel computers can be implemented on serial machines, to any breadth you like, at a cost of operating speed.

while the controller (of a ship, a city, a body, an oil refinery) needs to use energy at some interface (to push the rudder, utter the decree, turn up the temperature), the energy required to run the control system itself is ad lib—whatever you like, and not very much.

It is important, Deacon claims, that a brain be made of cells that are themselves autonomous little agents with agendas, chief of which is staying alive, which spawns further goals, such as finding work and finding allies.

If a region is underutilized, its neighbors soon recruit the cells there to help with their own projects. Neurons that are destroyed are not, in general, replaced the way skin cells and bone cells and blood cells (among others) are. Nerve “regeneration” is still mainly a dream of bioengineers, not a normal feature of nervous systems, so the plasticity observed in many experiments must be due to neurons being reassigned to new tasks or required to take on additional workloads.

Economists have shown why centrally planned economies don’t work as well as market economies, and centrally planned (top-down) architectures are ineffective brain-organizers for much the same reasons.

Neurons, in contrast, are all different; they come in a variety of quite clearly defined structural types—pyramidal, basket, spindle, and so on—but even within types, no two neurons are exactly alike. How can such a diverse population get organized to accomplish anything? Not by bureaucratic hierarchies, but by bottom-up coalition-formation, with lots of competition.

The simplest moving parts within neurons, the motor proteins and microtubules and the like, really are motiveless automata, like the marching broomsticks in The Sorcerer’s Apprentice, but neurons themselves, in their billions, play more enterprising and idiosyncratic roles than the obedient clerks I was imagining them to be, and that fact has major implications for the computational architecture of brains.

What could a neuron “want”? The energy and raw materials it needs to thrive—just like its unicellular eukaryote ancestors and more distant cousins, the bacteria and archaea. Neurons are sorta robots; they are certainly not conscious in any rich sense—remember, they are eukaryotic cells, akin to yeast cells or fungi. If individual neurons are conscious, then so is athlete’s foot. But neurons are, like yeast and fungi, highly competent agents in a life-or-death struggle, not in the environment between your toes but in the demanding environment between your ears, where the victories go to those cells that can network more effectively, contributing to more influential trends at the levels where large-scale human purposes and urges are discernible.