Emergence without magic: the role of memetics in multi-scale models of evolution and behaviour

Richard Dawkins effectively defined the field of memetics by introducing the term "meme" as "a unit of cultural transmission or a unit of imitation" [4]. Memes are defined to behave as "self-replicators", as part of the process of their transmission between human "hosts" [18]. This mechanism is loosely inspired from epidemiology, and particularly virology. Despite the variations of the definition of memetics in the literature, the original definition is sufficient for the purposes of this paper.

Section 2 motivates the importance of certain theoretical issues in the scientific understanding of "complex" and "emergent" phenomena, and how these issues are relevant to memetics. Section 3 introduces a philosophical framework from which an idealised view of scientific methodology is developed. This view allows a clear analysis of the issues described in Section 2 as they concern existing theories, but the methodology cannot be immediately applied to build new theories relating to complexity and emergence. However, Section 4 summarises the adjustments that are needed in order to remedy this. As a caveat, it is remarked that the technical baggage of a strict methodology is often avoided in model making, and thus it is important to be clear about the limitations of a model when discussing its scientific validity. Section 5 argues that these inherent limitations must be identified in memetic theory. This allows the major philosophical criticisms to be assessed under the view taken in Section 3. Section 6 suggests that the criticisms could be avoided, under a suitable reinterpretation of the role of memetic theory as part of a "multi-level" view of human behaviour. Such "unification" with other theories of culture and cognition could provide a more complete scientific picture. A reinterpretation of memetics is outlined, using ideas from the methodology. To conclude, it is suggested that the methodology might be applied to other complex and emergent phenomena.

The science of the objects and interactions of normal human experience is not unified in its foundations. There can be significant incompatibility between scientific theories having overlapping domains but which use language based on different representations of natural features. For example, the fact that the mind-body problem remains unresolved means that theories of human behaviour and neural behaviour still use very different elementary objects and interactions. Explaining the diversity of organic life based on an analysis of pre-biotic chemistry has similar difficulties. It is appealing to consider the possibility of an "ontological unity" [13] under a "reductionist" assumption, so that, for instance, a theory with one set of basic representations could ultimately underlie all accounts of the observations at the levels of both neurobiology and social behaviour.

There are two main reasons why such problems of unity remain unresolved. Firstly, a natural process is involved which works simultaneously at very different scales of time, or space, or both. Secondly, an observer is able to appreciate the variety of form and behaviour exhibited by that process at the different scales taken individually, but has difficulty relating them precisely in an appreciation of the process as a whole. Thus an analysis is needed of the philosophical inconsistencies between traditional single-scale models that ultimately represent aspects of the same process.

The general advantages of ontological unity are essentially twofold: obtaining logical consistency between theories having non-identical domains of applicability, and obtaining a coherence which can "transcend the narrow borderlines of each [traditional] discipline" [13, p. 262]. Together, these goals motivate the idealised scientific methodology developed in the next section. However, a full account of the argument advocating a search for ontological unity in the social sciences can be found in Ref. [21].

There is continued under-appreciation of the role of methodology in making scientific progress: however, "methodological well-foundedness has been constitutive of, rather than tangential to, the most important appraisals of theories." [16, p. 59]. In this section, a view of scientific methodology is summarised. It has many similarities to those discussed in the "complex systems" and cybernetics literature [3, 10, 11].

Central to the proposed methodology is a form of scientific instrumentalism, similar to that used by d'Espagnat [5]. In this view, an observer has a certain, limited, perceptual ability, and a certain set of physical instruments (having intrinsic limits of resolution). Theories are seen as syntheses of measurements and the consequent perceptions of form that an observer is capable of. As seen in the examples of Section 2, scientific theories relate to perceptions of structure and function in natural phenomena that can differ greatly according to the scale at which the observer measures. Clearly some uniformity is necessary between the instruments and perceptions of different observers, in order that observations can be corroborated and theories can be generally accepted. However, as a methodology no assumptions of "absolute scientific truth" or "independent physical reality" are required, so that the methodology may be almost universally applicable throughout the natural sciences.

The ideal scientific "explanation" of a natural phenomenon would be a theory that could predict the observations possible under all scientifically valid experimental conditions. This would include pathological cases such as the extremes of the time or space scales of the observer. The view that in principle a "single theory" could achieve this may be termed methodological reductionism (MR) [15] (similar to its use in Ref. [Kup], or the "ontological reductionism" as used in Ref. [Sto]). By a "single theory" it is meant that causal explanations of all observable features of the phenomenon can be provided by a set of formal models which can be precisely related in a unified, multi-scale theory. A theory with these qualities will be referred to as a comprehensive theory, and is the basis of MR explanation as used here. (A clearer definition will be made once the rest of the methodology is introduced.) The existence of a comprehensive theory of a natural phenomenon will provide the instrumentalist meaning of "ontological unity" (or simply "unity", for convenience).

Formalism affords logical clarity and power in accounting for the structures and mechanisms in a complex system. In MR, it is common that formalisms are given the primary role in the explanation of a natural phenomenon. Examples of formalisms are differential or integral equations, maps, and symbolic calculi, which have different inbuilt notions of time, space, and state. In this methodology the term "model" will be associated with a formal dynamical system that is given an explicit interpretation. The interpretation gives meaning to the model's elementary atomic components and formal mechanisms, by relating them to the properties of a more concrete model (that reflects a more detailed view of the phenomenon). The interpretation can be seen as an interface between the pair of levels. The higher and lower levels of the pair may then be called the emergent and underlying levels, respectively. Thus the form of a model is a set of representations of some chosen structural and functional aspects of the observed process. The representations are of a more or less abstract (or conversely, concrete) character, which motivates the idea of a model's level of representation (subsuming the intuitive notion of "scale").

An interface defines the domain of applicability of the emergent model relative to the underlying model. Three aspects of this domain are distinguished here: the temporal, spatial, and state sub-domains of the underlying model. These are mathematically precise concepts. For convenience, the three sub-domains together will be called a spatio-temporal-state domain (STS-domain), where it is understood that it is in fact a sub-domain of the underlying level's model.

The logical consistency of the interface is the degree to which the elementary components and mechanisms of the emergent model (seen there as a set of axioms) can be derived from the dynamic properties of the underlying model's restriction to the STS-domain. A (totally) logically consistent pair of models have an interface that consists entirely of theorems relating the underlying level's features to the formal elements of the emergent level model.

It is now possible to define a comprehensive theory of a natural phenomenon as one that can demonstrate a logically consistent hierarchy of mathematical models at all levels of representation.

Some adjustments are necessary in order to make the idealistic methodology practicable. This is because (a) it has no a priori base level of analysis; and (b) it asserts that "true explanations" can be achieved only by comprehensive theories, which potentially involve model hierarchies of arbitrary "height" and "density".

The progress of scientific explanation involves the testing of working hypotheses, leading to their gradual sophistication or replacement. This justifies the temporary use of a root level in a model hierarchy, until developments in instrumentation or mathematical tools permit a more accurate and lower-level model to be put below it. This resolves problem (a), whilst acknowledging the practical sense of the term "explanation" as being relative to the contemporary model of "reality".

Intrinsically dominant scales are evident in the perception of natural processes. Thus it is sufficient to select a small set of appropriate levels of representation above the root level, thereby resolving problem (b). A general view of levels is motivated by Foster [6], and an illustration of it is found in Lister and Weingarter [17, Ch. 1].

Finally, the form of efficient models in practice is seldom totally determined by an underlying level. If it were, the models would often be cumbersome, losing their descriptive flexibility and conceptual clarity. Thus it is unavoidable that some features will be introduced into the models that are not present in a more concrete view of the phenomenon. These may be called artefacts. The ideal of total logical consistency is therefore not attainable. Nevertheless, a suitable degree of consistency may be agreed upon, when the inconsistencies due to the artefacts are minimised by suitably restricting the domain of applicability of the model.

The issue of conceptual problems in current memetic theory can now be discussed. A conceptual problem in a theory is internal if there is logical inconsistency or vagueness of its basic categories of analysis, and external if there is a logical incompatibility with another theory that has an overlapping domain of application [16].

Consider the following statements:

In order to judge their validity, the statements will be recast in the language of the proposed methodology, where each point will be treated separately.

(1) This statement suggests a problem similar to the classic "symbol grounding problem" [9]. Such problems rely on a distinction between the interpretation of an elementary component of a model (a) as a direct representation of a feature present in the root level -- i.e. the instrumentalist interpretation of "physical reality"; or (b) as a simplifying, abstract representation of a complex feature present in a lower level. Under interpretation (a), a meme would be a symbolic entity having an independent existence of its own. Ideas and their interactions are presumed to have a neuropsychological substrate, and acts of communication involve a physical mechanism. Thus interpretation (a) can be rejected, and the symbolic nature of a meme is merely an artefact. However, memetics often makes no claims about its neuropsychological substructure. For instance, consider a memetic formalisation due to Lynch [18]. The basic objects of his model are "mnemons" (memory abstractions) and human "hosts", and the basic interaction between host and meme is a "replication event". Lynch aims for a "non-metaphoric memetics" by avoiding explicit analogy between memes and the replicators found in biology (genes, viruses) or computer science (software viruses). To do this a symbolic calculus is used to describe replication events. This yields a clear description of memetic mechanisms, avoiding reference to those found in the analogous domains, and so this aspect of memetics is internally logically consistent. However, formalisation cannot rid the basic representations of their intrinsic vagueness, and so an internal conceptual problem is present. Lynch [18] claims:

The admission that the mnemon, the basic concept of the model, has no direct interpretation in a lower-level theory of culture or cognition means that it has no precise meaning, according to the definitions of Section 3. Unfortunately, what meaning remains is largely due to the analogy with a gene, and for memetics to mature this is clearly undesirable.

(2) The second statement demonstrates the external type of conceptual problem. Tooby and Cosmides [21] argue strongly that culture cannot be acquired socially without the presence of some prior knowledge which can provide a context. For a new-born child this "knowledge" has to be due to genetically-based, and content-specific, cognitive faculties. Also, the form of some social behaviours appear to be "evoked" by features of the local environment, rather than explicitly learnt [21]. Memetic theories are often incompatible with this, when occurrences of evoked culture are either interpreted purely as a transmitted form, or are not recognised at all. Thus memetics may have a more limited domain of applicability, and should acknowledge possible exceptions to its basic tenets. Further examples of exceptions are seen in Section 6.

(3) The conclusion of (1) implies that the level of memetics is neither identical to, nor coherently interfaced with, any traditional root level. Thus the third statement claims that by itself memetics has no explanatory power -- under the MR view of explanation. In this sense, therefore, one might call memetics "sterile".

Under the proposed methodology the suggested response to the above criticisms is (a) to treat memetics as a single level in a hypothetical hierarchy of levels, and (b) to reinterpret its basic assumptions in the context of its neighbouring levels. The hierarchy would make up a comprehensive theory of culture as it relates to social and individual psychology, and ultimately so on down through levels of brain function. This would provide an ontological unity between memetics and other, related theories, and consequently the criticisms would no longer apply. Also, the ease and conciseness inherent in memetic descriptions of cultural transmission would be seen as having a role in providing intuition and an overview of high-level features present in the cultural process. Without this, one might never "see the wood for the trees".

A suggested reinterpretation of memetics is now outlined. Although the following sketch of the cultural process is simplistic (in particular, the role of evoked culture is ignored), it should be sufficient to indicate how a serious effort might proceed.

Humans have a propensity for learning the organisation of social relations and activities that they observe. They form groups that undertake such organised activities for the achievement of a variety of goals in a generalised form of social "contract", or "exchange" [2]. These exchanges, and their associated physical props, can be seen as schemas that reside in both a psychological and a physical domain. Groups are able to adhere to schemas during their execution, and maintain their form in a culture over lifetimes or generations -- all in spite of a constantly changing environment. This makes the schemas relatively invariant forms of cultural information that appear to propagate through society. When the propagation is due to a schema's utility, the Darwinian model of memetic transmission is easily justified [1]. But there are many less obvious psychological and social mechanisms underlying the dynamics of schemas which may be in competition with the basic model. These are discussed further in Ref. [1]. It has been suggested here that memetics should be seen as only a partial account of culture. However, contemporary memetics sometimes tries to force the basic model of meme transmission to patch over these apparent "exceptions" by introducing unfounded, ad hoc assumptions. For instance, by endowing memes with the intentionality necessary in order to "induce" laziness, etc., in individuals. This has prompted accusations of mysticism and vitalism, which might have been avoided.

Under the present methodology schemas can be modelled using concepts from dynamical systems theory (in mathematics) and systems theory (from cybernetics). Although developed in more detail elsewhere [1], some benefits are: (a) A concise and unified account of both exceptions and artefacts in formal models by referring to precise properties of their STS-domains; (b) The cultural evolution of schemas is interpreted in terms of the changing stability of "attractors" of behavioural patterns in a state space; (c) Schemas do not need to have an artificially closed, atomic structure; and (d) Intentionality is not required, since schemas have a straightforward interpretation within traditional psychology. Thus schemas may provide a step towards understanding the concrete mechanism of information storage in the brain.

A similar approach could be applied to other "complex" and "emergent" phenomena, since analyses of these in the literature, such as that based on the "meta-system transition" [10], involve some form of multiple levels. Motivation for the mathematical aspects of this approach may be found in Jackson [12, Epilogue] and Kauffman [14], and also in the success of the techniques used to analyse spatially distributed systems comprising of coupled differential equations [7, 8, 11, 19]. Also, such an approach would suggest a way to end the problems of metaphorical representations, and thereby accelerate the achievement of logical consistency between theories. For instance, Küppers [15] discusses the relevance of both methodological reductionism and formal models of pre-biotic chemistry to theories of the origin of life.

[1] Clewley, R.H., 1998, Reinterpreting Memetics in a Multi-Level View of Evolution and Behaviour. Preprint. Available at my research page.

[2] Cosmides, L., and Tooby, J., 1992, Cognitive Adaptations for Social Exchange, Ch. 3 of "The Adapted Mind: Evolutionary Psychology and the Generation of Culture", eds. J. Barkow, L. Cosmides, J. Tooby, Oxford University Press, New York.

[3] Crutchfield, J.P., 1994, Is Anything Ever New? - Considering Emergence, in "Integrative Themes", eds. G. Cowan, D. Pines, D. Melzner, Santa Fe Institute Studies in the Sciences of Complexity XIX, Addison-Wesley.

[4] Dawkins, R., 1976, The Selfish Gene, Oxford University Press, Oxford.

[5] d'Espagnat, B., 1989, Reality and the Physicist, Cambridge University Press, Cambridge.

[6] Foster, C.L., 1992, Algorithms, Abstraction and Implementation, Academic Press, London.

[7] Golomb, D., et al., 1992, Clustering in Globally Coupled Phase Oscillators, Physical Review A, Volume 45, Number 6, pp. 3516-3530.

[8] Haken, H., 1983, Synergetics, 3rd ed., Springer-Verlag, Berlin.

[9] Harnad, S., 1990, The Symbol Grounding Problem, Physica D 42, pp. 335-346.

[10] Heylighen, F., 1991, Modelling Emergence, in "World Futures: the Journal of General Evolution", special issue on Creative Evolution, ed. G.Kampis.

[11] Hoppensteadt, F.C. and Izhikevich, E.M., 1997, Weakly Connected Neural Networks, Springer-Verlag, New York.

[12] Jackson, E. Atlee, 1991, Perspectives of Nonlinear Dynamics, Vol. II, Cambridge University Press, Cambridge.

[13] Kanitscheider, B., 1991, Unification as an Epistemological Problem, in "Advances in Scientific Philosophy", eds. G. Schurz, G. Dorn, Rodopi B.V., Amsterdam.

[14] Kauffman, S.A.,1990, Requirements for Evolvability in Complex Systems: Orderly Dynamics and Frozen Components, Physica D 42, pp. 135-152.

[15] Küppers, B.O., 1990, Information and the Origin of Life, MIT Press, Cambridge, Massachusetts.

[16] Laudan, L., 1977, Progress and its Problems, Routledge and Kegan Paul, London.

[17] Lister, R.G. and Weingarter, H.J., 1991, Perspectives on Cognitive Neuroscience, Oxford University Press, New York.

[18] Lynch, A., 1998, Units, Events and Dynamics in Memetic Evolution, Journal of Memetics - Evolutionary Models of Information Transmission, 2.

[19] Nicolis, G. and Prigogine, I., 1989, Exploring Complexity: An Introduction, W.H. Freeman, N.Y.

[20] Stöckler, M., 1991, Reductionism and the New Theories of Self-Organization, in "Advances in Scientific Philosophy", eds. G. Schurz, G. Dorn, Rodopi B.V., Amsterdam.

[21] Tooby, J. and Cosmides, L., 1992, The Psychological Foundations of Culture, Ch. 1 of "The Adapted Mind: Evolutionary Psychology and the Generation of Culture", eds. J. Barkow, L. Cosmides, J. Tooby, Oxford University Press, New York.