The necessity of theoretical constructs: a rebuttal of the behaviourist approach to memetics
- a commentary on Gatherer's paper: Why the `Thought Contagion' Metaphor is Retarding the Progress of Memetic

Derek Gatherer's (1998) thesis that the "thought contagion" metaphor has slowed down progress in memetics is clearly exposed, well-argued and based on what is obviously a broad knowledge of the most recent literature relevant to memetics. However, the epistemology underlying his argumentation is totally outdated. Although Gatherer bases his thesis on Watson's behaviourism, which was developed during the 1920's and 1930's, his philosophy that only observable entities should be part of a scientific theory goes back much further, to Comte's 19^th century positivism. It was elaborated in the beginning of the 20^th century by the logical positivists (or logical empiricists) of the Vienna Circle.

His view that "scientists are generally constrained by the requirement that their models should mirror nature" (my emphasis) is older still. It was already criticized by some of the positivists themselves, and is now completely outdated (cf. Turchin 1993). At least, Gatherer seems to be aware that this mirroring conception is problematic when he notes that quantum physicists were forced to abandon it because of "the overwhelming weight of empirical evidence".

But positivism and behaviourism too have by now been completely abandoned and discredited, the first as a methodology for science in general, the second as a methodology for psychology in particular. Popper (1959), Wittgenstein and other philosophers of science have irrefutably shown why a positivist approach cannot work. Cognitive psychology has irreversibly taken over from behaviourist psychology (although it must be said that in psychology the behaviourist position still has some influence). There is no space here to review all arguments against positivism.

Let me just mention two facts that should be obvious:

(1) all scientific models have unobservable, theoretical constructs at their core;

(2) all empirical research, to be in any way meaningful or useful, presupposes - explicitly or implicitly - a theoretical construct.

As evidence for proposition (1), it suffices to look at the fundamental theories of physics, the most established and empirically best supported of all sciences. Essential concepts such as quarks, electromagnetic fields, wave functions, black holes or space-time geometries are all in principle unobservable. You do not even need to consider weird, non-classical theories such as quantum physics to find unobservable entities: Newtonian mechanics, the most basic of all scientific theories, is built on the concept of "force", a mysterious, non-material entity that can only be observed indirectly through its effects on the movement of objects.

This does not mean that such theories lack empirical support. Popper (1959) has made it clear that no theory or law can ever be verified or proven by observations. However, observations are necessary to refute theories that make incorrect predictions. Thus, a theory can be considered reliable if its implications have been repeatedly tested by experiments, without being refuted. As Turchin (1993) notes, models or theories are recursive generators of predictions. The hypothesis that nucleons consist of quarks does not generate any immediately testable predictions, since quarks can by definition never be observed outside of a nucleon. However, indirectly the quark hypothesis entails other hypotheses, which entail further hypotheses, and so on, until we come to the level where a hypothesis can be tested by observation. Of course, we would like to make the path from theoretical construct to empirical test as short as possible, but if we insist on too short paths, we will be stuck with theories that have a very restricted domain of application (Turchin, 1993).

As to proposition (2), all scientists will agree that observations are only useful if they can be repeated, and if repeated observations produce similar results. But this presupposes that the different observations and their respective results are all recognized as instances of the same phenomenon. If each observation is unique, there is no regularity and thus no scientific law to be inferred. However, the only way we can decide that different observations confirm the same law is because we have a theoretical conception of what such a law could be.

Let me illustrate this principle with an example that is mentioned by Gatherer as representative of the more `behaviourist' field of social contagion: Phillips' (1974) test of the hypothesis that subjects who hear about others committing suicide are more likely to commit suicide themselves. At first sight, this seems a good illustration of Gatherer's view of a meme as a replicating behaviour: suicide in one person leads to suicide in another person. No need to postulate unobservable constructs, such as a "mnemon" for suicide, you might think.

The first difficulty with this view is that there is no simple, unique "suicide behaviour". You can commit suicide in millions of different ways: by drinking poison, inhaling gas, shooting yourself with a gun, jumping from a cliff, or driving at high speed into a wall. All these types of behaviour look utterly different to the observer. Yet, we would all classify them as "suicide". You might argue that they all lead to a person's heart stopping to beat, an unambiguous observation. But so do millions of other events which we do not classify as suicide. What distinguishes suicide from accidental death is the intention of the subject to make an end to his or her life - another theoretical construct which should be rejected outright by any genuine behaviourist.

But things get even worse for the behaviourist interpretation of suicide contagion. The hypothesis which Phillips confirmed is called the "Werther effect". The main observation is that people who were exposed to stories of suicide (originally, Goethe's novel "The suffering of the young Werther") are more inclined to commit suicide. Most of the victims of the Werther effect never actually witnessed somebody committing suicide. So, what was replicating was not actual suicidal behaviour, but the idea that suicide is a good way out of an apparently unsolvable problem. The only way to explain the variety of observations gathered by Phillips is to assume the existence and replication of an unobservable, mental entity: the intention to end one's life.

Of course, as Gatherer might reply, we can never actually observe the replication of the suicide idea. But as I argued earlier, this is a general problem with all scientific concepts. We can merely observe certain indirect effects of the postulated phenomenon, just like astronomers can only observe the presence of a black hole through its effect on nearby matter, not by seeing it directly. Such indirect effects can be misleading, as Gatherer illustrates with his example of the man making Windsor knots for his colleagues: our observations may make us think that these people each know how to make a Windsor knot. But Gatherer is wrong when he concludes from this example that it would be more fruitful to study the replication of the Windsor knot itself, rather than of the ability to make Windsor knots. This is similar to the argument that astronomers should better concern themselves with the study of the light rays that enter their telescopes rather than with the study of stars, galaxies and planets.

Let me suggest an alternative example of a copying machine containing a particular page. The machine may produce 1, 10, 1000, or one million copies of that page. Let us assume that the page contains information that could revolutionize society. For example, it could describe a method to produce unlimited amounts of energy through cold fusion. However, in how far the information on that page will affect the evolution of culture and society does not depend on the actual number of copies that come out of the machine. It depends on whether other people will read that information, whether they will understand it, believe it, communicate it to others, be willing to apply it, etc. These are all mental processes, which are not directly observable. The actual number of physical copies ("artefacts") is irrelevant if those copies are merely stored in a drawer, dropped in a trashbin, or blown away by the wind. It is the cognitive interpretation that determines whether the meme will affect other people's behaviour.

Gatherer also uses his Windsor knot example to argue that the situation in memetics is essentially different from the one in Mendelian genetics. Since one "mnemon" can produce hundred observable instantiations ("phenotypes"), there is no one-to-one relation between memotype and phenotype. But the same problem exists in genetics: one gene can produce dozens of feathers, antlers, branches, leaves or flowers. If you find different pieces of wood, shells or feathers in a geological deposit, you may be misled to think that they were produced by different organisms. In both cases, memetic and genetic, we have to be very careful how we infer the presence of an unobserved, but hypothesized, entity from observable phenomena. Instead of counting the number of Windsor knots, we should better test people on their ability to make a Windsor knot.

This ambiguous correspondence between observed and hypothesized phenomena is a general problem for all scientific modelling, not a specific difficulty for memetics. The problem is particularly acute in psychology, where the hypothesized, "mental" entities tend to be especially complex and variable. This explains why positivism, in the guise of behaviourism, has survived so long in that domain. Yet, by now psychology has developed a wide range of methods to empirically determine the presence of theoretical constructs. The different tests used in personality psychology to assess traits such as intelligence, introversion, dominance, etc. can serve as examples. There is no direct way to observe intelligence. Yet, the scores on different standardized IQ tests provide reliable predictions of a subject's chances for success in other intelligence-related activities, such as academic achievement. The simplest way to explain these observed regularities is to assume that "intelligence", although a purely mental, unobservable entity, objectively exists.

In a similar way, we could devise tests for different memes or features of memes, and then make predictions that people who "have" these memes, as measured by the test, will behave differently from people who score negatively on the test. For example, you could test subjects' knowledge of a hypothetical suicide meme, and then predict that the suicide rate in people with such knowledge will be higher than in those without such knowledge.

The basis of most psychological tests is natural language: you can simply ask people whether they consider suicide as an option. Of course, we know that such questions and their answers can be very subjective, and that there are plenty of spurious factors that may influence the results. However, that is insufficient reason to dismiss linguistic communication as a means to detect the presence of memes. A priori rejecting this method simply means rejecting the possibility of any kind of communication or transfer of information via language. And that would mean rejecting the idea that reading and writing papers, the activity we are engaged in at this moment, can contribute to the development of scientific knowledge.

To cope with such problems, psychologists have devised numerous techniques that minimize the subjectivity inherent in asking questions, for example: repeating the same question with different words, asking questions to which the answer is already known in order to test the subject's sincerity, extracting hidden, but reliable patterns from the list of responses, correlating answers to questions with independent observations, etc. All these techniques can be useful for memeticists who wish to operationalize their concept of a meme as a mental entity that influences behaviour.

To conclude my criticism of Gatherer's paper, I would like to ask him why the existing fields of social contagion and diffusion studies, which he holds up as an example of how memetics should develop, have produced so little understanding of cultural evolution. He suggests that this is merely due to an anti-evolutionary prejudice among social psychologists. To me it seems that the problem is more fundamental: it is the behaviourist approach itself which precludes the development of evolutionary models.

Indeed, if the only thing you study is observable behaviour and the way it replicates, then how can you ever account for variation and selection, the basic components of any evolutionary process? Variation in cultural evolution is not normally caused by the copying errors made by a printing press or photocopying machine: it is produced by the people who introduce certain changes in the behaviour they copy. Similarly, behaviours are not generally selected by observable transmission mechanisms such as printing or broadcasting. Behaviours or artefacts are replicated because people for some personal reason consider these behaviours to be worth copying. Thus, we can only get a real understanding of memetic variation and selection if we understand what goes on inside people's head when they listen to a song, read a book, or hear a story.

I have proposed a number of selection criteria that would allow us to predict which memes will survive or be eliminated (Heylighen, 1993, 1997, 1998). These criteria are mostly (though not exclusively) internal to the person or to the meme. I have also suggested that these criteria, although not directly observable in the behaviourist sense, can be operationalized and empirically tested. That seems to me the way to go if we want to create a real science of memetics: to develop clear theoretical constructs, with a broad explanatory power, which are then operationalized so that they become empirically testable. I agree with Gatherer that many of the theoretical constructs used by memeticists, such as Lynch's (1998) distinction between "belief" and "awareness", are still problematic. But that does not mean that we should stop theorizing. On the contrary, we should reflect more profoundly on what a meme is and how it can be modelled in the most general and yet most precise and concrete way possible.

In conclusion, Gatherer's paper is useful because it reminds us of the importance of operationalization, and the danger of remaining stuck in vagueness and abstraction. However, it is downright misleading in its suggestion that in order to achieve operationalization we should throw out the essential theoretical construct of meme as memory unit, only because it is not directly observable.

References

Gatherer, Derek (1998). Why the `Thought Contagion' Metaphor is Retarding the Progress of Memetics, Journal of Memetics - Evolutionary Models of Information Transmission, 2. http://cfpm.org/jom-emit/1998/vol2/gatherer_d.html

Heylighen F. (1993). Selection Criteria for the Evolution of Knowledge. In: Proceedings 13^th International Congress on Cybernetics (Association International. de Cybernétique, Namur), 524-528. ftp://ftp.vub.ac.be/pub/projects/Principia_Cybernetica/Papers_Heylighen/Knowledge_Selection_Criteria.txt

Heylighen F. (1997). Objective, subjective and intersubjective selectors of knowledge. Evolution and Cognition, 3:63-67. http://pespmc1.vub.ac.be/papers/knowledgeselectors.html

Heylighen F. (1998). What makes a meme successful? In: Proceedings. 16^th International. Congress on Cybernetics (Association International. de Cybernétique, Namur), in press. http://pespmc1.vub.ac.be/Papers/MemeticsNamur.html

Lynch, Aaron (1998). Units, Events and Dynamics in Memetic Evolution, Journal of Memetics - Evolutionary Models of Information Transmission 2. http://cfpm.org/jom-emit/1998/vol2/lynch_a.html

Phillips, D.P. (1974) The influence of suggestion on suicide: substantive and theoretical implications of the Werther Effect. American Sociological Review 39:340-354.

Popper K. (1959). The Logic of Scientific Discovery. London: Hutchinson.

Turchin V. (1993). On Cybernetic Epistemology. Systems Research, 10:3-28. ftp://ftp.vub.ac.be/pub/projects/Principia_Cybernetica/Papers_Turchin/Cybernetic_Epistemology.tex