|| Human Nature Review ISSN 1476-1084 | Table of Contents | What's New | Search | Feedback | Daily News | Submit A Manuscript ||
PDF of this article Download Adobe Acrobat Reader Email the reviewer Reviewer's web site Search for papers by Benzon, W. L. Author's web site Search for papers by Aunger, R. Send a response to this article Search the web for related items Contact the Editors
Human Nature Review 2002 Volume 2: 454-462 ( 17 October )
URL of this document http://human-nature.com/nibbs/02/benzon.html
Colorless Green Homunculi
William L. Benzon
The Electric Meme: A New Theory of How We Think
by Robert Aunger
The Free Press, 2002. ISBN: 0743201507
Though he wasn’t the first to propose that culture evolves by a Darwinian mechanism—F. T. Cloak (1973) was earlier—the idea has been associated with Richard Dawkins ever since he coined the term “meme” in The Selfish Gene. As genes are the fundamental replicating units of biological evolution, so memes are the replicators of cultural evolution. In Dawkins’ words (p. 192):
Examples of memes are tunes, ideas, catch-phrases, clothes fashions, ways of making pots or of building arches. Just as genes propagate themselves in the gene pool by leaping from body to body via sperms or eggs, so memes propagate themselves in the meme pool by leaping from brain to brain via a process which, in the broad sense, can be called imitation.
Dawkins had little to say about just where one might look in the brain to find memes and other memeticists have been content to follow him in that. Robert Aunger, an anthropologist at the University of Cambridge, believes that it is time memeticists end their agnosticism on this matter. To that purpose he has written The Electric Meme.
It is best viewed as two books. One book is comprised of the first six chapters, which are preparatory in nature, reviewing the current state of memetics, alternative analyses of human cultural evolution, types of replicator (DNA, prions, computer viruses), and the physical nature of information. This book is competent, interesting, and thought provoking.
The second book sets forth Aunger’s new theory of neuromemetics. It differs from the first as night from day. For reasons that I cannot fathom, when Aunger begins constructing his neural model he sheds the discipline of Dr. Jekyll in favor of the random muttering of Mr. Hyde. This second book is a failure. Aunger’s ideas are vague, incoherent, and contradictory. Trying to infuse his phrases and sentences with meaning made me feel like Hercules fighting the many-headed Hydra. Every time Hercules cut one head off, two or three others sprung up to replace it. Every time I’ve tried to patch one of Aunger’s coherence-leaks I’ve had to entertain a handful of improbable assertions to make the one patch plausible. I cannot recommend this second book to anyone for any purpose.
That is a harsh judgment; as such it requires detailed justification. Because the long seventh chapter contains Aunger’s central statement, I will concentrate on it. Aunger’s succeeding chapters assume the validity of those ideas. If they are nonsense, then those later ideas are airy nothings, perhaps full of sound and fury, but signifying very little.
Finally, in the interests of full disclosure, I note that I am on record has holding a view that, if memes are to be found anywhere, they will be found in the external world, not in the brain (Benzon 2001). Thus I was not expecting Aunger to produce a theory that convinced me. But I wasn’t prepared for the incoherent farrago he delivered.
Neuromemetics in Brief
Aunger begins the seventh chapter by reviewing some very basic information about nervous systems. He then argues that, if we are to find memes in the brain, we should look, not in permanent structures (neurons and their parts), but in evanescent states taken by those permanent structures “in the context of a few neurons at most, working in concert” (p. 192). Aunger regards "this definition of a meme in terms of a state rather than a material substrate as the key conceptual move in neuromemetics" (p. 198). This is an important point as it “means the qualities of a meme don’t depend on the unique features of the wiring diagram of an individual brain, which reflects the sequence of local environments it experienced during its development and learning activity” (p. 199). I am sympathetic with Aunger on this point, but on little else.
Aunger uses three types of entity to construct his model: 1) nodes, 2) their states, which include memetic and non-memetic states, and 3) the signals by which nodes communicate with one another. Noting that we still have much to learn, Aunger tells us that nodes might be individual synapses, or neurons, or even groups of neurons. When a node in some state receives a spike, it typically goes into some other state. It sometimes happens that a node in a non-memetic state receives a spike from a node in a memetic state and then, in consequence, goes into that same memetic state. Beyond noting that these states have to do with chemicals in and around neurons Aunger says little about them.
That, more or less, is the core of Aunger’s theory. Note that meme replication takes place within a single brain. This is quite different from conventional memetics, in which memes replicate from one brain to another. Aunger allows for this as well, devoting his next chapter to it. But in his view this is a derived activity, not a basic one. If Aunger’s account of the basic process fails, then his account of the derived process has no starting point. For that reason I see little need to discuss it.
Aunger wants bench neuroscientists to look for memes in neural tissue, and I suspect he would welcome the attention of computational investigators as well. These people need a clear and consistent theoretical model to start from. My fundamental criticism is just that Aunger fails to provide that. He confuses nodes with states and talks about signals in the most fantastic way.
What’s a Spike?
Aunger generally refers to signals passing between nodes as spikes, or, occasionally, as spike trains. “Spike” is a common term in neuroscience and it refers to an impulse in a single neuron. Given that Aunger’s nodes can be individual synapses or groups of neurons as well as single neurons, his usage of the term “spike” is problematic. This is not a matter of “mere semantics” that isn’t worth our notice. Rather, it reflects a profound confusion about neural processes. To see why we have to review just what a spike is.
Neurons generate outputs in the form of millisecond-long impulses that move along a neuron’s axon from its root (at the cell body) to its terminal. Those impulses are often called spikes. If the axon branches, as is often the case, then the impulse travels down both branches. When the spike reaches the synaptic terminals it causes neurotransmitters to be released into the synapse where they then flow to post-synaptic neurons. As Aunger informs us, each neuron is connected to thousands of other neurons. This implies that each spike will affect many post-synaptic neurons; it also implies that each spike reflects integration over inputs from many neurons. The net effect of this many-to-many connectivity is that it makes no sense to talk of a spike moving from one neuron to the next. Spikes originate and dissipate within a single neuron.
When Aunger talks about a node “emitting a spike” (p. 195) he implies that the node is a single neuron, for it is single neurons that generate spikes. But that is only one of the three cases he considered in defining nodes. In the case where a node is a synapse, it is incorrect to say that they emit spikes. The electrochemical events at the synapse may put in motion by spikes, but they are not themselves spikes. In the case where a node consists of some unspecified group of neurons, it isn’t clear what he means by saying it emits a spike. Does Aunger mean all of the individual spikes considered as a single object? If so, must they be synchronized with one another or can they be spread over some interval? Or perhaps Aunger is imagining that a number of neurons are arranged in a local circuit such that the spikes of one neuron can be considered the output of the entire circuit. He doesn’t specify which, if either, is the case. The upshot is that when Aunger talks of nodes emitting spikes his meaning is not clear. Synaptic events, single spikes, and the actions of groups of neurons are different (though related) processes and must be treated as such.
Aunger’s conception of a spike is further muddied when he refers to a spike’s “ability to find the right kind of target cell in which to deposit its message” (pp. 220). That makes no sense. Spikes have no choices to make, nor the power to make them. They simply travel from one end of an axon to the other, following all branches. Similarly, it doesn’t make sense to talk of a spike carrying a message which it can “deposit” anywhere. Spikes are simple impulses; the only thing they can be said to carry is an electric charge.
What is so peculiar about this muddled conception of spikes is that it contradicts the account Aunger sketched in the opening pages of the chapter. That discussion is basically correct but he seems to forget it when it comes time to outline his model. In Aunger’s world when textbook neuroscience conflicts with the demands of speculative memetics it must give way.
Objects and their States
Given Aunger’s confusion about one entity in his theoretical triumvirate, it is not surprising that he is confused about the other two as well. In the following examples Aunger confuses objects (nodes) with their states (memes):
1) one meme can fire without the other doing so (p. 199)
2) the meme is telling its neighbor down the line (p. 201)
3) memes produce spikes that express their “intention” (p. 221)
4) the same neuronal node can be a replicator one minute and not the next (p. 215)
5) memes will be able to adopt a wide range of states (p. 216)
6) Robust replicators must be able to take on many more states than they currently do (p. 216)
In the first three cases Aunger talks of memes sending signals. As memes are defined as states of nodes, that makes no sense. It is the nodes which send signals. In example 4 he talks of nodes being replicators when it is memes that are defined to be the replicating elements—perhaps he’s using “replicator” in a sense derived from the Star Trek device, which is a machine capable of “replicating” any and all foodstuffs. Example 5 tells us that memes, defined as states, can themselves have states; though perhaps Aunger really meant to say that it is the nodes that “will be able to adopt a wide range of states.” Example 6 is even more confusing and so I leave its explication as an exercise to the reader.
In each of these cases, and many more scattered throughout the chapter, one can supply some plausible interpretation for Aunger’s confused wording. But that shouldn’t be necessary; it is Aunger’s responsibility to be clear and direct. If we skip forward in the book about one hundred pages, however, we find a statement that clearly asserts memes to be neural structures, not states. In the course of rethinking the nature of replicators Aunger tells us they can all be modeled by ball-and-stick constructions. After mentioning the physical model that was so important to Watson and Crick he asserts that “for memes . . . the balls are cells (neurons), while the sticks are synaptic connections” (p. 321). That is quite clear: memes are things, physical structures.
What happened to “the key conceptual move in neuromemetics,” that memes are states? How is it that, after having criticized Dawkins for asserting that memes are patterns of neural connections (p. 193) Aunger adopts that very same position? Either he’s playing games, daring us to call him out, or he doesn’t know how to use his own theoretical tools. Neither alternative engenders confidence in his work.
As a way of appreciating the depth of Aunger’s confusion, recall Chomsky’s famous example sentence: colorless green ideas sleep furiously. Though grammatically well-formed, the sentence is meaningless—that was Chomsky’s point, the independence of syntax from meaning. The notion of “colorless green” is contradictory while that of “furious sleep” is a bit odd. The deepest problem, however, and the one that interests us, is that ideas are not the kind of thing that can accept color as an attribute nor can they participate in events like sleeping.
It makes no more sense to say that memes emit spikes or that spikes choose their targets than it does to say that ideas sleep. Philosophers call this species of error a category mistake. The errors in Chomsky’s sentence are readily apparent because his terms designate familiar things. Aunger’s errors are less obvious because he is talking about matters that are both conceptually abstract and unfamiliar. But they are errors nonetheless and they undercut his attempts to write intelligible prose.
A theory which is confused about the nature of its basic constructs and that contradicts its central assertion cannot be much of a theory. Such incoherence is a consistent and persisting feature of Aunger’s thinking, not a small set of momentary lapses. The cumulative effect of these many lapses is a theory in which anything can happen. Let us examine some of these memetic sports.
I am going to begin by quoting a long paragraph in full. Then I will repeat the paragraph with added commentary.
Aunger is exploring how memes could be subject to selection inside a single brain. He is imagining that various behavioral plans undergo an unconscious competition with the winner becoming conscious. Here’s how that selection might take place (p. 219):
We can elaborate somewhat on how an unconscious process of selection among alternative ideas might operate. The brain signals responding to a single event can be mirrored at many different places (probably in the same region of the brain). It’s as if the brain enlists neurons from many precincts to “vote” on each of its actions. Thus spike trains representing all of the different interpretations of a stimulus compete for the job of directing the organism’s behavioral response to that stimulus. The mutual connectivity that appears so endemic in the brain—node 1 being wired up to node 2 and vice versa—may be a way in which signals from multiple information-processing streams are synchronized, to coordinate behavior responses based on a convergence of the “best thinking” the brain has done. Synchronization could line up the alternative interpretations of a situation for better assessment of their relative merits, producing a “fair” competition in the sense that each of the spiking “racers” sets off when the same starting gun fires. This way of reining in spike trains also suppresses cheaters, who could otherwise sneak through the tournament without actually recruiting new members for their team based on merit: Head-to-head competition prohibits the raiding of competing teams. This is important because the goal of the race is to achieve the largest pool of support and thus a better chance of attracting attention. The possibility that gets voted the winner in such a competition actually gets acted upon and may also bubble up into consciousness as well. Attention is thus a kind of top-down selection system among racing spike trains.
Let’s leave the first two sentences alone. Here are the third and fourth sentences (I’ve inserted identifying numbers):
(3) It’s as if the brain enlists neurons from many precincts to “vote” on each of its actions. (4) Thus spike trains representing all of the different interpretations of a stimulus compete for the job of directing the organism’s behavioral response to that stimulus.
The general idea is reasonable enough. The idea of competitive processes in the brain goes back at least to Kilmer, McCulloch and Blum’s 1969 model of the reticular activating system. Such schemes have been extensively explored through computer simulations of neural nets, and they are central, for example, to Minsky’s well-known The Society of Mind account of the mind. They are also at the heart of practical technologies, such as optical character recognition and speech recognition.
Sentence four, however, is a little warped. Presumably the spike trains are competing to be chosen as the best interpretation of the current stimulus. That is the interpretation that will, in turn, be used in formulating a response. But the winning spike train isn’t going to direct that response; rather it will be incorporated into that response. The response itself will be directed by some node or nodes, which will emit a cascade of spikes that will drive the motor system, and so forth.
We continue with sentences five and six:
(5) The mutual connectivity that appears so endemic in the brain—node 1 being wired up to node 2 and vice versa—may be a way in which signals from multiple information—processing streams are synchronized, to coordinate behavior responses based on a convergence of the “best thinking” the brain has done. (6) Synchronization could line up the alternative interpretations of a situation for better assessment of their relative merits, producing a “fair” competition in the sense that each of the spiking “racers” sets off when the same starting gun fires.
Yes, there is a great deal of mutual connectivity between neurons (the literature doesn’t talk about Aungerian nodes), and this connectivity exists at all scales, from immediate neighbors, to neurons that are several centimeters apart. I don’t have any particular objection to asserting that this may be a way of synchronizing “processing streams,” but the notion is too vague to carry much conceptual weight.
Sentence six is more problematic. Presumably Aunger has introduced racing as an analogy to give concrete expression to the idea of competition. But I don’t see how to apply the terms of that analogy to the physical processes taking place in the brain. The racers seem to be the spikes themselves. As spikes do travel along axons from one place to another in a neuron one can imagine races between spikes in different neurons. A spike’s velocity, however, depends on the diameter of the axon and on whether or not the axon is myelinated. That means that, in order to win such a race, a spike has to travel the fattest, best myelinated, and shortest axon it can find. As spikes have no choice over which axon they travel, that suggests that it is the meme that makes the choice. That implies that the meme needs to know the physical characteristics of nodal neurons in order to pick one from which it can generate a winning spike. How does a meme, which is defined to be some state a node can take, get that information and set about selecting which node it wants to occupy?
While I can continue this particular line of fantasizing, it does not seem very promising. Alternatively, we could begin a line of speculation that sees the race as a competition to generate the largest number of spikes in a fixed period of time. To make this work we must, of course, also provide some mechanism for counting these spikes. In either case, we must also provide a neural mechanism for firing a “starting gun.”
This is not illuminating. Surely we want a neural memetics that replaces the self-propelled scheming memebots of the popular literature with a replicating apparatus that is, in principle, as specific and mechanistic as the cellular and molecular apparatus of DNA replication. Instead Aunger has provided the homuncular meme with two companions, the homuncular spike and the homuncular node. This is not progress.
Let us continue with the last four sentences in the paragraph. Rather than continuing in the tedious style above I will simply ask some questions as a way of indicating difficulties.
(7) This way of reining in spike trains also suppresses cheaters, who could otherwise sneak through the tournament without actually recruiting new members for their team based on merit: Head-to-head competition prohibits the raiding of competing teams. (8) This is important because the goal of the race is to achieve the largest pool of support and thus a better chance of attracting attention. (9) The possibility that gets voted the winner in such a competition actually gets acted upon and may also bubble up into consciousness as well. (10) Attention is thus a kind of top-down selection system among racing spike trains.
How does a meme cheat in such a competition? What are these teams, what are the members? What or who’s attention is being attracted in sentence eight? Those judges, the ones voting in sentence nine, what or who are they? How is it that attention becomes a “top-down selection system”? What’s the top, where’s down?
“So,” you say, “the analogy is not very good. Still, we get the idea.” And that, I’m afraid, just won’t do. While it is common to employ analogy in explicating difficult technical matters, that isn’t how Aunger is using this analogy. He doesn’t offer any technical argument at all. Rather, this analogy is itself his argument. His failure to provide an analogy that can reasonably be put into correspondence with neural processes thus becomes a failure to provide a coherent account of selection among memes. He would have been much better off using one of the many existing models of competitive processing in nervous systems. That he hasn’t done so suggests either that he isn’t aware of these models; is aware but doesn’t understand them; or understands them, but doesn’t see how to interpret at least one of them as implementing competition among rival replicators.
Incredibly enough, Aunger ends this section by saying that such a model “if true, would bring psychology fully in line with the ‘hard’ sciences and provide a solid foundation for a science describing social groups in terms of such Darwinian agents” (p. 230). Is he trying to tell us that the “hard” sciences have been built on the kind of irresponsible and incoherent theorizing exhibited in The Electric Meme?
At this point it looks like Aunger is using neuromemetics as an excuse to make grand statements about the neural underpinnings of thinking without bothering to review the neural literature. Since that literature doesn’t talk about memes, it follows that it is irrelevant to Aunger’s enterprise. The way is thus free and clear for him to create, as his subtitle suggests, “a new theory of how we think.”
Given that Aunger asserts his central tenet to be both true and not true, it should come as no surprise that his hypothesized memes turn out to be meaningless. There is no point in summarizing this particular argument since only the conclusion matters. We can make do with a smattering of phrases.
As we have been told, memes produce spikes and those spikes “represent things.” As meaning is “a function of a firing pattern . . . you can’t equate meaning with memes. Meaning is always context-dependent, and so unrepeatable.” However, “one can say that a meme implicitly represents something, because it has the capability of producing a spike with that quality.” The vital point is that “what a spike represents depends on where it is found in the brain” (all quotes from p. 221). Since memes are quite mobile, it follows that they can generate spikes all over the place. From this it follows “that a particular spike train may mean ‘double-handled pot’ on this occasion, but ‘aspect of hair-line on forehead’ then next time around, and ‘sense of grief over a dead pet’ the next” (p. 224). Thus memes have no inherent meaning.
What hath Aunger wrought? If memes are inherently meaningless, then how can they be the basis of human communication and culture? For all his invocation of neural detail, his sixty-four pages of neuromemetics amounts to little more than a gaggle of colorless green homunculi spiking furiously.
Aunger’s definitions and descriptions are so very loose that there is no appreciable difference between a meme-induced spike and externally applied electrical impulses. Thus we can hypothesize that when investigators use micro-electrodes to stimulate the brains of their subjects they are in fact injecting artificial memes into those brains. The electrical pulses are inherently meaningless, just like memes, and their effects vary according to what brain structures are stimulated, just like meme-induced spikes. Similarly, we might speculate that epileptic seizures are caused by the sudden replication of memes over wide areas of the brain, with grand mal seizures occurring when replication tournaments engulf the entire brain.
This isn’t theorizing. It can hardly even be characterized as thinking. At best it is a vigorous raid into the neurosciences for phrases and sound bytes to be used in weaving complex and intricate memetic designs. The neuroscientific details exist only to camouflage and ornament Aunger’s meaningless memes.
What About Neuromemetics?
Aunger’s theorizing is so inept that his failure cannot be taken as evidence against the possibility of a neural memetics. A skilled and experienced theorist might succeed where Aunger has failed. Judging from the seven blurbs on its cover, however, The Electric Meme has been received with such approbation by such substantial thinkers that one wonders whether or not these scholars are willing to see Aunger’s theorizing for the failure it is. Two of them go so far as to suggest that Aunger may well have founded a new discipline. This is not the way to encourage an intellectually serious attempt to identify the neural foundations of cultural evolution.
Eörs Szathmáry’s (2002) review in Nature is a more serious matter, both because it is a formal review and because that review is in such a prestigious and visible journal. Szathmáry, an evolutionary theorist, is generally approving and notes that Aunger has some “truly novel and fascinating” ideas. In particular, the “idea that memes are primarily neural phenomena at the millisecond scale should open up a field of relevant neurobiological investigation.” Novel it is, and fascinating too, in the manner of disasters and cataclysms. That Aunger’s neuromemetics is incoherent and self-contradictory seems to have escaped Szathmáry’s notice, though perhaps that’s what he was responding to when he characterized it as “telegraphic.”
Whatever these scholars may have had in mind when they penned their praise I suspect they will reconsider if and when any of their graduate students start spiking to the beat of Aunger’s neuromemetic drummers. Intellectual specialists lacking neuroscientific knowledge might well be deceived on their first reading of Aunger’s prose, especially if they read him generously and assume that he knows what he’s talking about. But they will not remain deceived once they study his words carefully. No, despite this superficial praise, I am not yet worried about the specialist community.
The Electric Meme, however, has been published as a trade book directed at the educated public. Judging from the comments posted at Amazon.com, for example, some of these readers have taken Aunger’s ideas at face value and are quite pleased with them. That is not surprising. The general idea of memes has been a seductive one; people want to believe it. Readers are thus willing to believe that any difficulties they experience in reading The Electric Meme reflect their own ignorance.
One of the attractions of writing for a general audience is that one has an opportunity to speculate more freely than one can in the refereed literature. At the same time, your audience is less likely to detect any mistakes you make as they lack the specialized intellectual skill required. Balancing speculative freedom against your responsibility to a vulnerable audience is difficult. Aunger took the freedom but misjudged the responsibility.
If we allow others to follow Aunger¹s example I fear that we will play into the hands of those who would diminish the cultural authority of science and rationality. Aunger¹s inability to escape the pull of meme-speak puts his work squarely in the sights of postmodern and deconstructive critiques of science (and everything else). Make no mistake, I think that those critiques of science are unwarranted. I believe that, using appropriate methods, we can obtain objective knowledge of the world. But Aunger¹s neuromemetic thinking does not exhibit those methods. Truth has been forgotten.
I would like to thank Susan Blackmore, Walter Freeman, and Timothy Perper for their comments. I take full responsibility for the final review.
Benzon, W. L. (2001) Beethoven's Anvil: Music in Mind and Culture. New York, Basic Books.
Cloak, F. T. (1973) Elementary Self-Replicating Instructions and Their Works: Toward a Radical Reconstruction of General Anthropology through a General Theory of Natural Selection. Paper presented at the Ninth International Congress of Anthropological and Ethnographical Sciences. Available on the world wide web at: http://www.thoughtcontagion.com/cloak1973.htm
Dawkins, R. (1989). The Selfish Gene, New Edition. Oxford, New York, Oxford University Press.
Kilmer, W. L., W. S. McCulloch and J. Blum (1969) "A Model of the Vertebrate Central Command System." International Journal of Man-Machine Studies 1: 279-309.
Minsky, M. (1985) The Society of Mind. New York, Simon and Schuster.
Szathmáry, E. (2002) Replication at the Speed of Thought. Nature 418: 370-371.
Buy The Electric Meme from Amazon.com Amazon.co.uk Amazon.fr Amazon.de Amazon.co.jp Amazon.ca
Computer-generated translation of this page français deutsch español português italiano - also try this alternative fast translation service.
© William L. Benzon.
William L. Benzon, Ph.D. , a cognitive scientist, is Senior Scientist with Metalogics Incorporated and an associate editor of The Journal of Social and Evolutionary Systems. He is also co-founder of the musical ensemble AfroEurasian Connection. He lives in Jersey City, New Jersey. Dr Benzon is author of Beethoven's Anvil: Music In Mind And Culture, Basic Books, 2001.
Benzon, W. L. (2002). Colorless Green Homunculi. Human Nature Review. 2: 454-462.