A NUMBER OF QUESTIONS can be raised about the relationship between the conception of emergence presented in Chapter 3 and other conceptions of emergence. My purpose in this appendix is to review these issues, to sharpen my account of emergence, and to rebut some well-known and influential objections to emergence that have appeared in the philosophical literature.
Discussions of emergence are structured by the classical opposition between reductionism and emergentism (Silberstein 2002). These positions have epistemological and ontological components. Epistemological reductionism states that the best understanding of a system is to be found at the level of the structure, behavior, and laws of its component parts and their relations; ontological reductionism states that the relations between the parts of the system are all determined without remainder by the intrinsic properties of the most basic parts. Epistemological emergentism states that the best understanding of a system is to be found at the level of the structure, behavior, and laws of the whole system, and ontological emergentism that a whole is more than the sum of its parts and their intrinsic properties.
As one might gather from my discussion of emergence in Chapter 3 this opposition between parts versus whole, or between basic features (conceived as intrinsic properties of microscopic particulars) versus emergent features, is part of the problem, not part of the solution.
Beyond the classical opposition between reductionism and emergentism tricky issues arise because there are now so many different senses of “reduction” and “emergence” in the scientific and philosophical literature. Many different kinds of phenomena have been considered emergent, and different epistemological and/or ontological criteria of emergence have been used to classify them as emergent.1 There seems to be no good reason, therefore, to search for any single and neatly analyzable concept of emergence independent of particular explanatory contexts. The focus here will accordingly be on the notion of emergence implicit in the theory of autonomous systems and on the application of this notion to biological and cognitive phenomena by the enactive approach.
Let me begin by setting up a framework in the form of the following provisional definition of emergence in the case of complex systems (Thompson and Varela 2001). This definition is meant to capture the main features of emergence that researchers in complex systems theory seem to have in mind when they talk about emergence.
Definition: A network, N, of interrelated components exhibits an emergent process, E, with emergent properties, P, if and only if:
(1) E is a global process that instantiates P, and arises from the coupling of N ’s components and the nonlinear dynamics, D, of their local interactions.
(2) E and P have a global-to-local (“downward”) determinative influence on the dynamics D of the components of N.
And possibly:
(3) E and P are not exhaustively determined by the intrinsic properties of the components of N, that is, they exhibit “relational holism.”
Although the term emergent property is widespread, I prefer emergent process. Strictly speaking, it does not make sense to say that a property emerges, but only that it comes to be realized, instantiated, or exem-plified in a process or entity that emerges in time. Emergence is a temporal process, but properties (whether considered as universals or as linguistic abstractions) are atemporal. For instance, the property of being alive did not emerge when life originated on Earth; rather, it came to be instantiated as a result of the emergent process of autopoiesis that constitutes living cells. This example also points to the importance of the causal characteristics of emergent processes: the emergent network of autopoiesis constitutes a biological individual (a cell) that brings about changes in the external environment. It also creates a structured context in which new kinds of events can take place, such as protein synthesis and RNA/DNA replication, which cannot occur apart from or outside of the protected intracellular environment.
The emergent processes covered by this definition occur in networks whose coupled elements have nonlinear interactions. The distinction between nonlinear and linear interactions provides one way of distinguishing between systems that have emergent processes and systems that do not (Campbell and Bickhard 2002). Linear interactions are additive or proportional. They typically give rise (except in the quantum domain) to systems that are “aggregative” (Wimsatt 1986)—systems whose causal features derive from aggregating the properties of the components. Nonlinear interactions are nonadditive or nonproportional. They give rise (by definition) to systems whose activities cannot be derived aggregatively from the properties of their components. It is therefore tempting to say, borrowing the terminology of the classical British emergentists (Samuel Alexander, C. Lloyd Morgan, and C. D. Broad), that nonlinear processes generate “emergents,” whereas linear processes produce only “resultants.”
Nonlinearity results from positive and negative feedback relationships. These relationships bring about patterns of behavior, which can be described as constrained alternatives in the space of all possible global states of the system (as attractors in phase space). To understand such activity patterns we need to model them in the language of collective variables and order parameters, while at the same time showing how these collective variables and order parameters are biophysically realized in natural phenomena.
The emergent processes of concern to the enactive approach occur in complex systems that need to be seen as autonomous, such as unicellular and multicellular organisms, the immune system, and the brain. An autonomous system has operational closure and accordingly dynamically maintains its own organization as the crucial invariant. “Organizational-operational closure” characterizes the system’s invariant form through structural and material change, and thus is a topological and morphodynamic notion (Deffuant et al. 1995; Petitot 1992). Yet because this invariant form is precisely the dynamic pattern of a circular network whose constituent processes operate under closure (every product of the network stays within the network), the morphodynamics in this case defines not merely a formal identity through time but a formal self-identity. This type of morphodynamics marks an important difference between emergence in autonomous systems and other oft-cited examples of emergence (such as convection rolls). In the case of autonomy (organizational-operational closure), what emerges is simultaneously a “self” or “individual” and a correlative niche (the domain or set of interactions possible for such a system given its organization and concrete structural realization). The exemplar and minimal case of this kind of emergence is the living cell (see Chapters 5 and 6).
Another way to characterize autonomous systems is in relation to the contrast between “decomposable” and “nondecomposable” systems (Bechtel and Richardson 1993; Simon 1969). Decomposable systems have a clear hierarchical organization. Each component or subsystem operates according to its own intrinsic principles independent of the others, making the system strongly modular. In a “nearly decomposable” system, higher levels interact with lower levels through top-down or feedback relations. The nature of these interactions determines the extent to which the system is nearly decomposable: “A system will be nearly decomposable to the extent that the causal interactions within subsystems are more important in determining component properties than are the causal interactions between subsystems” (Bechtel and Richardson 1993, p. 27). As the components of the system become less governed by intrinsic factors and more by the system’s organization, then the system is “minimally decomposable.” A nondecomposable system is one in which the connectivity and interrelatedness of the components give rise to global processes that subsume the components so that they are no longer clearly separable. In such a system, the distinction between preexisting parts and supervening whole becomes problematic. Not only does the whole emerge from the components, but also the components emerge from the whole.
An autonomous system is at least minimally decomposable, if not nondecomposable. More precisely, when one adopts an autonomy perspective, one ipso facto characterizes the system as at least minimally decomposable. The reason is that an autonomous system is an organizationally and operationally closed network; hence it is the connectivity of its constituent processes that determines its operation as a network.
Neural assemblies can be used to illustrate these ideas. With only a few exceptions, the brain is organized on the basis of a principle of reciprocity: if area A connects to area B, then there are reciprocal connections from B to A (Varela 1995; Varela et al. 2001). The traditional practice in neuroscience is to map these reciprocally interconnected brain areas onto a hierarchy of processing levels from peripheral (lower) to central (higher) areas (measured in terms of synaptic distance from sensory stimulation). The sensory end is taken as the starting point, and perception is described as proceeding through a series of feedforward or bottom-up processing stages. Top-down influences are equated with back-projections and feedback from higher to lower areas. This scheme treats the brain as a nearly decomposable or minimally decomposable system (depending on the extent to which the top-down influences are emphasized).
From a dynamicist point of view, however, the picture looks different. First, the dynamicist does not depict the brain as a processing hierarchy that starts at the sensory end. Strictly speaking, brain processes are always ongoing and do not start or stop anywhere. A better entry point for many purposes of analysis can be found in the brain’s own endogenous activity, as reflected in the organism’s states of preparation, expectation, emotional tone, and attention, which are necessarily active at the same time as the sensory inflow (Varela et al. 2001, p. 230; see also Engel, Fries, and Singer 2001; Freeman 1999a, 1999b, 2000; Lutz et al. 2002). Such activity arises far from the sensors—from the frontal lobes or limbic system, or in the middle of the whole network from temporal and associative cortices. There is considerable evidence that this kind of endogenous activity participates even in the early stages of sensory perception. Although this sort of activity is usually described as top-down or feedback, “top down” and “bottom up” are heuristic terms for what is in reality a large-scale network that integrates both incoming and endogenous activity. It is precisely at this network level that collective-variable dynamics and order parameters become important for characterizing the large-scale integration of widely distributed neuronal activities.
Second (and to anticipate the discussion of Proposition 2), this large-scale dynamics can modulate local neuronal activity by entraining or “pulling” the behavior of individual neurons into a particular pattern of global activity (see the discussion of the neurodynamics of epilepsy in Chapter 3). This dynamic-systems form of global-to-local influence neither requires nor is equivalent to top-down control in a sequential hierarchy of processing stages (Engel, Fries, and Singer 2001; Thompson and Varela 2001).
To look at the brain in this way is to characterize it as at least a minimally decomposable system—a system in which the behavior of the components is determined largely by the system’s organization. Yet it is also important to consider that the brain—in its dynamic operation as a large-scale network—might need to be characterized as a nondecomposable system. No doubt it is useful for certain explanatory purposes to characterize the brain as a (nearly or minimally) decomposable system. Yet problems can arise when one assumes that the explanatory strategies of decomposition and localization (differentiating a system into separable components and assigning responsibility for specific tasks to those components) are adequate to characterize brain activity (Bechtel and Richardson 1993; Uttal 2001). In particular, decomposition and localization are insufficient to characterize the operational closure of the brain as a dynamic neuronal network. In Maturana and Varela’s words:
Since, due to its constitution as a network of lateral, parallel, sequential, and recursive interactions, the nervous system closes on itself at all levels, the mutilations that it may suffer generally leave a closed neuronal network with a changed structure. Accordingly, the organization of the nervous system is essentially invariant under mutilations, while its domain of possible states, which depends on its structure, and, hence, on its architecture, is not. Yet, due to its closed organization, whatever is left of the neuronal network after a partial ablation necessarily operates as a different whole with different properties than the original, but not as a system from which some properties have been selectively subtracted . . . There is intrinsically no possibility of operational localization in the nervous system in the sense that no part of it can be deemed responsible for its operation as a closed network, or for the properties which an observer can detect in its operation as a unity. However, since every nervous system has a definite architecture, every localized lesion in it necessarily produces a specific disconnection between its parts and, hence, a specific change in its domain of possible states. (Maturana and Varela 1980, p. 129)
This characterization implies that the brain—understood operationally as a dynamic network of processes or “brainweb” (Varela et al. 2001)—is a nondecomposable system. We still lack a theoretical language for expressing the complex behaviors of such systems in dynamic-systems terms (see Le Van Quyen 2003). In the brain case, nondecomposability would mean that the brainweb generates global processes that subsume their components so that they are no longer clearly separable as components. At this dynamic level, the distinction between preexisting parts and supervening whole has no clear application: One might as well say that the components (local neuronal activities) emerge from the whole as much as the whole (dynamic patterns of large-scale integration) emerges from the components.
“Nondecomposability” and “decomposability” are heuristic, epistemological categories, not ontological ones. It is not my intention to argue for a metaphysical thesis of ontological holism on the basis of nondecomposability (which is not to say that nondecomposability is merely epistemological in the sense an ontological reductionist would assert). At the moment my point is rather to call attention to the non-decomposability perspective in order to correct a strong bias in much of classical and contemporary neuroscience toward a modular (localizationist) view of the brain (see also Uttal 2001). Ultimately, it is the interplay between these heuristic categories within and across various explanatory contexts that is important, not one heuristic versus the other. As Le Van Quyen states:
Following Simon (1973), it is important to consider this interplay as a “loose vertical coupling,” permitting the distinction between levels, and a “loose horizontal coupling,” allowing the separation between subsystems at each level. While the word “loose” suggests “decomposable,” the word “coupling” implies resistance to decomposition. In my view, the characterization of this “loose coupling” represents one of the essential challenges for future developments. (Le Van Quyen 2003, p. 84)
Complex systems theorists, as we have seen, appeal to the idea of circular or reciprocal causality, by which they mean that global patterns both arise from local interactions and govern or constrain those interactions. In synergetics, a branch of complex systems theory, a vivid but unappealing metaphor is used to describe this global-to-local influence. The global, collective-variable dynamics is said to influence local behavior by “enslaving” the network elements into a particular dynamic régime (Haken 1983).
The term downward causation is also often used to describe this sort of global-to-local influence. One of the earliest uses of the term in a scientific context was in an article by Donald Campbell (1974), called “‘Downward Causation’ in Hierarchically Organized Biological Systems.” Campbell’s view was that in hierarchically organized biological systems, downward causation from higher to lower levels of organization occurs in the form of natural selection: “Where natural selection operates through life and death at a higher level of organization, the laws of the higher-level selective system determine in part the distribution of lower-level events and substances . . . all processes at the lower levels of a hierarchy are restrained by and act in conformity to the laws of the higher levels” (p. 180). This idea that higher-level processes “restrain” lower-level processes so that they “act in conformity” to them corresponds to the idea that global processes (collective-variable dynamics) constrain or govern local interactions.
What exactly does “constraint” mean in this context? In complex systems theory, constraints can be understood as relational properties that the parts possess in virtue of their being integrated or unified (not aggregated) into a systemic network. “Constraint” is therefore a formal or topological notion (Deacon 2003). The form, configuration, or topology of a system limits or prevents certain possible behaviors the parts could have on their own, while simultaneously opening up new possibilities for them in virtue of the states the system as a whole can access (Juarrero 1999, pp. 132–133).
Let us examine this notion of constraints in more detail, beginning with the difference between “context-free constraints” and “context-sensitive constraints” (Gatlin 1972, as cited by Juarrero 1999, pp. 6–7, 131–140). A context-free constraint is one that is externally imposed and alters the probabilities of the available behavioral alternatives of the system’s components. For instance, a container filled with evenly diffused molecules of gas at room temperature is at thermodynamic equilibrium, but inserting a piston (a context-free constraint) into the container and moving the piston so that the molecules are compressed to one side imposes an orderly arrangement on them. If the pressure on the piston is removed, the system will move back toward equilibrium in accordance with the second law of thermodynamics. A context-sensitive constraint, on the other hand, is one that synchronizes and correlates previously independent parts into a systemic whole. Catalysis is a good example (Juarrero 1999, pp. 139–141). Imagine a primeval soup with several types of molecules (A, B, and C) randomly floating around in it. As a result of externally imposed, context-free constraints (e.g., the weather), there will be more of some molecules in certain areas than in others. Now imagine that A catalyzes the formation of B. This relationship between A and B imposes a context-sensitive constraint on both of them:
Once the probability that something will happen depends on and is altered by the presence of something else, the two have become systematically and therefore internally related. As a result of the operations of context-sensitive constraints and the conditional probabilities they impose, A is now part of B’s external structure. Because A is no longer “out there” independent of B, to which it is only externally related, the interdependence has created a larger whole, the AB system. Insofar as it is part of B’s new context or external structure, A has been imported into B. (Juarrero 1999, p. 139)
Juarrero calls this kind of context-sensitive constraint a “first-order contextual constraint” because it operates at the same level of organization as the individual components (A and B). A “second-order contextual constraint” is established when the organization of the whole system emerges as a constraint on the system’s components. Thus, in a simplified and idealized autocatalytic network, A catalyzes the formation of B, B catalyzes the formation of C, C of D, and D closes the loop and catalyzes the formation of A. The relationship between each pair of catalyzing and catalyzed molecules is a first-order contextual constraint. Once autocatalytic closure occurs, however, these first-order relationships become subject to the second-order contextual constraint of the organization as a whole.
A still more striking example is autopoiesis. A minimal autopoietic system corresponds not simply to an autocatalytic network, but to an autocatalytic network housed within and interdependently linked to a semipermeable membrane-boundary. Autocatalytic networks do not qualify as autopoietic systems because they do not self-produce their own topological boundaries (see Chapter 5).2 Autocatalytic networks either have no boundaries or their boundaries are set by an externally imposed, context-free constraint (such as the walls of an experimental container). In an autopoietic system, however, the membrane forms part of the second-order contextual constraint of the system’s organization. In this type of system, the entire self-producing organization of membrane-plus-internal-autocatalytic-network operates as a second-order contextual constraint from the whole to the parts. Furthermore, in multicellular organisms, the autopoiesis of the individual cells becomes subject to the higher-order contextual constraints of the multicellular organization (second-order autopoiesis). In other words, the autopoiesis of the individual cells becomes subordinated to the maintenance of the higher-order autopoiesis or autonomy of the multicellular organism (Maturana and Varela 1980, pp. 107–111; 1987, pp. 73–89).
We now have in hand an answer to the question of what downward causation means in the context of complex systems theory. According to the line of thought just sketched, downward causation corresponds to the second-order contextual constraint of a system’s organization restricting or limiting the degrees of freedom of the system’s components. Downward causation corresponds to the influence the relatedness of the system’s components has on the behavior of those components. More precisely, it corresponds to the influence of the system’s topological organization on its constituent processes (Varela 1979; see also Deacon 2003). “Downward” is thus a metaphor for the formal or topological influence of a whole with respect to its parts.
It is questionable whether this metaphor is a good one. Although there are clearly empirical differences in scale and logical differences in order between the topology of a system and its constituent processes and elements, the two levels do not move in parallel, with one acting upward and the other acting downward, because the whole system moves at once. John Searle makes this point in a related discussion: “The right way to think of this is not so much ‘top down’ but as system causation. The system, as a system, has causal effects on each element, even though the system is made up of the elements” (Searle 2000b, p. 17). From this perspective, the term downward causation is symptomatic of a partial recognition of system causation together with an inability to shift completely to a system-causation perspective.
Some philosophers might wonder whether the topological influence of a system on its elements should be considered a causal influence. This issue is inseparable from broader conceptual issues about causation. Philosophical debates about emergence and downward causation tend to be structured not only by a strong ontology of part and whole (or basic features versus emergent features), but also by a strong ontology of “causal powers”—the causal powers of basic features versus the causal powers (or lack thereof) of emergent features. Rather than get caught up in the philosophical debates about causal-power views of causation, I am content to explicate what dynamic systems theorists have in mind when they talk about whole-to-part influence in complex systems. As we have seen, this influence corresponds to the organizational constraint of a system with respect to its components. Such influence is topological. It is therefore not an external force that acts on something, but an interconnectedness or relatedness among processes. This interrelatedness structures the context and background of local interactions, such that certain kinds of events can take place that otherwise would not occur. Some authors describe the constraint of a system’s organization as a standing or ongoing, “structuring” cause, by contrast with an episodic “triggering” cause (see Dretske 1995b for this distinction), and liken organizational constraints so understood to Aristotle’s formal cause (Emmeche, Køppe, and Stjernfelt 2000; Juarrero 1999, pp. 125–128). This recuperation of formal causation gains support from the phenomenological and morphodynamic analysis of form presented in Chapter 4.
According to the ontological thesis of part/whole reductionism (“mereological supervenience”), all the properties of a whole are determined by the intrinsic (nonrelational) properties of its most fundamental parts. (Hence the whole is said to supervene on the intrinsic properties of its parts.) In contrast, according to holism (“mereological emergence”) certain wholes possess emergent features that are not determined by the intrinsic properties of their most basic parts. Such emergent features are irreducibly relational. They are constituted by relations that are not exhaustively determined by or reducible to the intrinsic properties of the elements so related. These holistic relations do not simply influence the parts, but supersede or subsume their independent existence in an irreducibly relational structure.3
One might think that relational holism could be invoked to legitimize the notion of downward causation. Downward causation is supposed to be the determinative influence that the relatedness of the system’s components has on their behavior. If this relatedness were holistic, then this fact would presumably account for the determinative influence of the system as a whole on its parts. The problem with this line of thought is that relational holism implies that the relation is the most basic unit and therefore that the terms of the relation have no independent (nonrelational) status. Hence the components could not constitute an independent lower level subject to higher-level “downward” influence, as the term downward causation cannot help but suggest. In other words, given relational holism, downward causation seems a misnomer. The presence of relational holism would thus provide another reason for doubting that the concept of downward causation is appropriate for describing the influence of a whole on its parts.
The concept of relational holism was introduced largely in connection with nonseparability or entanglement in quantum mechanics, in which the state of the system is not constituted by the states of its parts and only the whole system can be said to be in a definite state (Teller 1986; see also Belousek 2003). On the basis of nonseparability, a number of philosophers have argued that quantum mechanical systems are holistic or mereologically emergent.4 Silberstein and McGeever (1999) have proposed that nonseparability is a paradigm of “ontological emergence.” By this they mean (i) that it is a kind of emergence that belongs to a system or whole in itself, as opposed to being an artifact of our theories or models; and (ii) that it violates the metaphysical doctrine of (atomistic) mereological supervenience, which states that every property of the whole is exhaustively determined by the intrinsic properties of its most fundamental parts.5 Silberstein and McGeever’s question with regard to complex dynamic systems is whether they, too, exhibit ontological emergence or only epistemological emergence.6
The main reason usually given for thinking that nonlinear dynamic systems are not holistic or ontologically emergent is that they are classical deterministic systems. In a deterministic nonlinear dynamic system, at all times (i) every variable has a definite value, and (ii) any change in the value of any variable is nonstochastic. As Silberstein states: “It is hard to imagine how any system that is ‘deterministic’ in both these ways could exhibit mereological emergence” (Silberstein 2001, p. 83). Nevertheless, because it is impossible for us to deduce mathematically the behavior or global dynamic properties (attractors) of the system, these behaviors and properties can be described as epistemologically emergent. For instance, chaotic systems are deterministic, but they can appear random and are unpredictable in the long run. They are highly sensitive to initial conditions (small differences in perturbations produce exponentially divergent effects), and we are able to specify the values of their variables only to a finite degree of precision. Stephen Kellert expresses this notion of epistemological emergence: “chaos theory argues against the universal applicability of the method of micro-reductionism, but not against the philosophical doctrine of reductionism. That doctrine states that all properties of a system are reducible to properties of its parts. Chaos theory gives no examples of ‘holistic’ properties which could serve as counter-examples to such a claim” (Kellert 1993, p. 90).
More recent discussions may cast doubt on this statement. Frederick Kronz (1998) proposes that the key to chaos in classical systems is the nonseparability of the Hamiltonian energy function (see also Kronz and Tiehen 2002, pp. 332–333). Whereas the Newtonian formulation of classical mechanics takes forces as fundamental, the Hamiltonian formulation takes energies as fundamental. The Hamiltonian of a system corresponds to the total energy of the system (kinetic energy plus potential energy). In classical mechanics, the Hamiltonian describing a linear system is separable into a sum of Hamiltonians, with one element in the sum for each constituent of the system. If there are nonlinear terms in the equations of motion, however, then the Hamiltonian is nonseparable. Building on Kronz’s discussion, Robert Bishop (2002) proposes that the nonseparability of the Hamiltonian is crucial for understanding the emergent global coherence of a complex system:
[T]he properties of integrity, integration and stability exhibited by Bénard cells are global properties and involve the nonlocal relation of all fluid elements to each other. This global behavior differs from holistic entanglement in quantum mechanics in the sense that fluid elements may be distinguished from each other while they are simultaneously identified as members of particular Bénard cells and participate in interaction with fluid elements throughout the system. In this context focusing on the nonseparability of the Hamiltonian may be more appropriate because, in contrast to the quantum case, classical states are always separable even when the Hamiltonian is nonseparable. (Bishop 2002, p. 7)
According to this proposition, there is a kind of holism proper to complex systems that does not seem compatible with the philosophical doctrine of reductionism. At the same time, this form of holism (non-separability of the Hamiltonian) is not incompatible with determinism.
What determinism means, however, is a whole other matter (see Atmanspacher and Bishop 2002). I do not intend to enter this thicket here, but two basic points are worth making. First, it is important to distinguish between determinism as a feature of a scientific model and determinism as a metaphysical thesis about nature. According to the metaphysical thesis, all physical properties in nature are definite and determinate, and the evolution of the natural world is fixed uniquely. (The complete and instantaneous state of the world fixes its past and future with no alternatives.) This thesis hardly follows from the fact that we can construct nonstochastic dynamic-system models of observable phenomena. Second, it is also important to distinguish between nonlinear dynamic systems as abstract mathematical models and as observable biophysical systems. Any concrete empirical system will involve some degree of randomness in the form of stochastic fluctuations (Kelso 1995). For any given empirical system, the analytical techniques most appropriate for characterizing the system’s behavior depend on the hypotheses one makes about its degree of nonlinearity and its degree of stochasticness (Le Van Quyen 2003; Schreiber 1999). Metastable dynamic systems display different mixes of nonlinearity and stochasticness at different spatial and temporal scales (Friston 2000b; Kelso 1995; Le Van Quyen 2003).
Science has barely begun to chart this vast sea of nonlinearity and stochasticness. Within this context, “deterministic” seems best understood as describing certain nonlinear analysis techniques (those in which there are no noise terms), not as an ontological characteristic of nature (in a classical observer-transcendent sense).
The discussion up to this point provides further support for a conception of emergence as dynamic co-emergence. Dynamic co-emergence means that part and whole co-emerge and mutually specify each other. Kronz and Tiehen (2002), in their discussion of emergence and quantum mechanics (with reference also to nonlinear dynamic systems), advocate the same idea, which they call dynamic emergence:
Emergent wholes have contemporaneous parts, but these parts cannot be characterized independently from their respective wholes. Emergent wholes are produced by an essential, ongoing, interaction of its [sic] parts. These are the central features of the new view sketched here; the nonseparable Hamiltonian constitutes an essential ongoing interaction . . . By adopting [this] view, we can say that it does not make sense to talk about reducing an emergent whole to its parts, since the parts are in some sense constructs of our characterization of the whole. (Kronz and Tiehen 2002, p. 345)
My last task in this appendix is to examine some recent and well-known objections to emergence and downward causation raised by the philosopher Jaegwon Kim (1993, 1998, 1999). Kim argues that “reflexive downward causation”—the causal influence of a whole on its own microconstituents—is incoherent when understood to happen simultaneously or synchronically, and is either otiose or violates the “causal closure of the physical” when understood to happen diachronically (Kim 1999, pp. 28–33). If Kim is right, then emergence and downward causation turn out to be to merely epistemological in the ontological reductionist’s sense. For this reason, it will be instructive to consider his arguments.
Here is Kim’s description of the first case of synchronic reflexive downward causation:
At a certain time t, a whole W, has emergent property M, where M emerges from the following configuration of conditions: W has a complete decomposition into parts a1, . . . , an; each ai has property Pi; and relation R holds for the sequence a1, . . . , an. For some aj, W’s having M at t causes aj to have Pj at t. (Kim 1999, p. 28)
What Kim finds troubling and ultimately incoherent about this case is the synchronic way in which it combines upward determination and downward causation. The upward determination of W’s having M at t as an effect of aj’s having Pj at t is simultaneous with the downward causation of aj’s having Pj at t as an effect of W’s having M at t.
Nevertheless, there is no obvious incoherence in this case yet. Simultaneous causation, though counterintuitive to commonsense, is not obviously incoherent. For instance, certain theories in physics postulate instantaneous action at a distance; whether such influences exist, and whether they should qualify as causal, are difficult empirical and conceptual issues. Furthermore, simply to pronounce causal circularity in general “unacceptable,” as Kim does (p. 28), is too facile, given that the notion of circular causality plays a central role in complex systems theory. Whether the circular causality of complex systems theory is best seen as synchronic and/or diachronic, or as simply a shorthand for reciprocal causal explanation, is an open and difficult question, which cannot be decided simply by armchair pronouncements.
Hence it comes as no surprise that Kim has to derive the incoherence he discerns by introducing an additional metaphysical principle about causation—the “causal-power actuality principle”—to which he thinks “we tacitly subscribe” (though he does not tell us whether “we” means metaphysicians, scientists, or the ordinary person in the street):
For an object x, to exercise, at time t, the causal/determinative powers it has in virtue of having property P, x must already possess P at t. When x is caused to acquire P at t, it does not already possess P at t and is not capable of exercising the causal/determinative powers inherent in P. (Kim 1999, p. 29)
With this principle applied to the case, synchronic reflexive downward causation is incoherent. W’s having M at t is supposed to be an effect of aj’s having Pj at t, which implies (by the causal-power actuality principle) that aj must already possess Pj at t, but aj’s having Pj at t is supposed to be an effect of W’s having M at t, which implies that aj does not already have Pj at t.
The notion of downward causation as global-to-local structuring influence casts doubt on the generality of Kim’s principle. The problem with the principle can be seen in the case of Rayleigh-Bénard convection rolls. The orderly rolling motions (which in a closed container form parallel cylinders called “Bénard cells”) emerge from the local dynamics of the fluid molecules and at the same time constrain the states of motion available to those molecules. As Robert Bishop states:
When the cells are established at t, this governing property is established at t; it did not exist prior to t. Likewise, prior to t, the trajectories of fluid elements had the property of accessing various states of motion, a property they lose at t due to the global governing effects of Bénard cells. The causal constraints/modifications on the motion of the fluid elements in this case are synchronic: The emergence of the self-regulating global pattern is simultaneous with the modifications of the accessible states of the system. If it were not synchronic, the pattern would not arise. (Bishop 2002, p. 9)
Contrary to Kim, therefore, synchronic reflexive downward causation seems perfectly coherent, as long as it is not conceptualized as efficient causation (in which cause and effect are external to one another), but rather as global-to-local structuring influence.
Let us turn to Kim’s argument against diachronic downward causation. Here matters require still more disentangling. Kim distinguishes between two kinds of downward causation—one in his view ordinary and unproblematic, the other highly problematic—but neither kind fits the kind of downward causation found in complex systems theory.
The first kind is simply causation from the properties of a whole to properties of its parts—where Kim leaves the notions of “whole” and “part” unanalyzed. A whole causes one of its microconstituents to change in a certain way, and there is a time-delay between cause and effect. Kim interprets Roger Sperry’s (1969) analogy of a rolling wheel constraining the motion of its own molecules as a case of this kind of diachronic downward causation. The structural properties of the wheel constrain the movements of the molecules composing it, so that the molecules are caused to move in a particular way as the wheel rolls downhill. Here are Kim’s other examples and his verdict about this kind of diachronic downward causation:
I fall from the ladder and break my arm. I walk to the kitchen for a drink of water and ten seconds later, all my limbs and organs have been displaced from my study to the kitchen. Sperry’s bird flies into the blue yonder, and all of the bird’s cells and molecules, too, have gone yonder. It doesn’t seem to me that these cases present us with any special mysteries rooted in self-reflexivity, or that they show emergent causation to be something special and unique. (1999, p. 30)
Kim concludes that “emergent downward causation should not simply be identified with causation from properties of the whole to properties of its own parts, that is, reflexive downward causation” (1999, p. 31). What emergent downward causation requires is not simply whole-to-part influence, but a difference in kind between the macrolevel and microlevel events. In other words, the causal relations between the two levels must implicate different properties—unlike the rolling wheel or the flying bird, in which the property of motion is common to the whole and its constituent parts.
Kim seems right about these cases. But there is a shortcoming to his analysis: none of the examples he gives in the passage above contains the kind of global-to-local influence found in complex systems. Nowhere in these examples is there the sort of nonlinear coupling of system components typical of a self-organizing system. This absence is particularly evident in the case of Sperry’s wheel, which is an aggregative system, not a self-organizing one. Nowhere is there any dynamic global pattern formation (as in convection rolls and Bénard cells). Nor is there any self-reflexivity because there is no complex system with operational closure and second-order contextual constraints (as in autopoiesis). Kim neglects the most important and relevant cases for emergence and downward causation. He draws the right general conclusion, but not for the right reason. It is true that emergent downward causation should not be identified simply with whole-to-part causation, but that is because emergent downward causation (at least in the complex systems context) is not meant to describe any kind of whole-to-part influence. Rather, it should describe the specific kind of reflexive global-to-local influence that happens in a system that has dynamic global coherence in and through collective self-organization.
Kim’s next step is to examine the sort of downward causation in which the macrolevel and microlevel events have different characteristic properties. His model is mental-to-physical causation, but the form of his argument has nothing to do with the mental and the physical as such; it could just as easily apply to the relationship between biological and physical events. The argument he advances rests on three principles (see Kim 1993, pp. 336–357):
1. The Physical Realization Principle: Every emergent event or property M must be realized by (or determined by, or supervenient on) some physical event or property P (its “emergence base”).
2. The Causal Inheritance Principle: If M is instantiated on a given occasion by being realized by P, then the causal powers of this instance of M are identical with (or a subset of) the causal powers of P.
3. The Principle of the Causal Closure of the Physical Domain: Any physical event that has a cause at time t has a physical cause at t. Hence, “if we trace the causal ancestry of a physical event, we need never go outside the physical domain” (1993, p. 280).
The critical question Kim then asks is why the physical emergence base of the emergent cannot displace the emergent as a cause of any of its putative effects. Here is the basic form of his argument. Suppose emergent M causes M* (which may or may not be an emergent). M* has its own physical emergence base P*. Therefore M’s causation of M* presupposes M’s causation of P*. In other words, the only way for M to bring about M* is by bringing about its physical emergence base P* (a case of downward causation). Now, M has its own physical emergence base P. The presence of P is sufficient for the presence of M. (Note that P is supposed not to cause M, but synchronically to realize M.) But M is causally sufficient for P* and thereby M*. It follows that P is causally sufficient for both P* and M*. But then M’s status as a cause is preempted by P, thus making M “otiose and dispensable” as a cause of P* and M*. The upshot of this argument is a dilemma for the emergentist who believes in downward causation: either downward causation is otiose, because the putative causal power of the emergent is preempted by the causal power of the physical elements on which the emergent is based, or downward causation violates the principle that the physical domain is causally closed.
This argument, when applied to the case of emergent downward causation in self-organizing systems, is tantamount to denying that the system’s organization exerts any influence on the system’s components. The thought seems to be that all of what we call the macrolevel and emergent causation really happens at the microlevel (by the Causal Inheritance Principle). Talk of the system’s organization having an emergent causal influence is simply a macrolevel description of the microlevel causal events. Such a description can be epistemologically useful for us, even though ontologically speaking all the real causal action goes on at the microlevel of the emergence base (Kim 1999, p. 33). In this view, the system’s organization is in fact epiphenomenal—an effect of lower level causal transactions with no significant causal status of its own—although it can still be useful for us to talk about the system as if its organization did play a causal role in bringing about certain events (see Kim 1993, pp. 92–108).
The problem with this way of thinking is its refusal to countenance the causal importance of a system’s organization (the relations that define it as a system). It does not acknowledge that the microlevel interactions happen as they do because of the way the local processes are organized, which is a macrolevel, relational characteristic of the system. That the macrolevel organization of a complex system is not simply epiphenomenal is evident in at least two respects.
First, it is multiply realizable: the very same organization can obtain, even though the constituents that realize it can vary. Indeed, in the case of any dissipative (energetically open) self-organizing system, not only can the particular constituents vary, but they also must vary, because they have to be incessantly renewed as the system dissipates energy into its environment. This kind of material renewal is especially striking in autopoiesis. There is and must be a constant turnover of material constituents in the cell while the autopoietic organization remains invariant (otherwise the cell dies). This aspect of a complex system is part of its robustness or dynamic stability, which means that the organization of the system is resilient to small changes or perturbations at the microlevel, and adaptive to large changes.
Second, the dynamic stability of a complex system is reflected in the fact that different counterfactual statements are true for the macrolevel organization and for the microlevel processes. The organization is necessary for certain subsequent events, but the particular constituents are not. If the organization had been different and those constituents the same, the events would not have occurred; but if the constituents had been different (within certain limits) and the organization the same, the events would still have occurred (because of the way the constituents were organized).7 Given that different counterfactual statements are true of the microlevel constituents and the macrolevel organization, the two cannot be identified.
Kim recognizes some of these points, but he goes through contortions to deal with them. In a footnote at the very end of his article, “Making Sense of Emergence,” he admits that “complex systems obviously bring new causal powers into the world, powers that cannot be identified with causal powers of the more basic simpler systems. Among them are the causal powers of microstructural, or micro-based, properties of a complex system” (1999, p. 36). By a microstructural or micro-based property, he means a macrolevel property that is completely decomposable into a collection of microconstituents plus the microlevel properties and relations that characterize them (Kim 1998, p. 84). (Given this characterization, it would seem that Kim simply assumes that the microlevel does not include holistic relations, such as quantum entanglement or nonseparability of the Hamiltonian in nonlinear dynamic systems, but this is left unclear.) In this footnote (and the corresponding lengthier treatment in his book Mind in a Physical World), Kim gives away many of the issues at stake in the debate about emergence without ever acknowledging that he has done so (Campbell and Bickhard 2002). Thus he admits that “macroproperties can, and in general do, have their own causal powers, powers that go beyond the causal powers of their microconstituents” (1998, p. 85, emphasis omitted). Nevertheless, he declares, “these properties are not themselves emergent properties; rather, they form the basal conditions from which further properties emerge (for example . . . consciousness is not itself a microstructural property of an organism, though it may emerge from one)” (1999, p. 36).
The reasons Kim denies emergence to these macroproperties are complicated and problematic. First, his conception of emergent properties is derived from the British emergentists, who held that it is simply an inexplicable brute fact of nature—to be accepted in an “attitude of natural piety” in Samuel Alexander’s phrase—that a certain organization of lower-level elements gives rise to some emergent quality. According to this way of thinking, an emergent property is not to be identified with the macrolevel organization itself, but with a qualitatively distinct property that supervenes on and has as its emergence base that organization (see Kim 1998, pp. 85–86). There is no compelling reason today to endorse this peculiar conception of emergence. The British emergentists rightly drew attention to the significance of organization and relational structure, but did not have the scientific tools for studying them that we now do. It is not classical British emergentism but contemporary science that should guide our thinking about emergence.8
Second, Kim apparently accepts part/whole reductionism. Hence for him all macrolevel, micro-based properties are decomposable into the intrinsic properties of microlevel entities (although, as I mentioned earlier, this is left unclear). But part/whole reductionism as an ontological thesis about nature is unacceptable because (among other things) it presupposes a conceptual framework (of preexisting particulars with intrinsic properties) that apparently has no coherent formulation in the language of microphysical theory (which is, after all, supposed to be the ground floor on which everything else rests, according to the reductionist). Macrolevel characteristics can be micro-dependent and micro-involving, without being micro-based in Kim’s sense (analytically decomposable into preexisting microlevel entities and their intrinsic properties).
Third, Kim’s approach to emergence is entirely dominated by the Cartesian mind-body problem and its procrustean framework of the mental versus the physical. Given part/whole reductionism about the physical domain (a basically Cartesian conception of nature), and a very wide and problematic conception of “physical” that includes all of biology and psychology save phenomenal consciousness (which is excluded not as something immaterial but as something that resists physicalistic reductive analysis), the only candidates left over for emergence are “qualia,” the qualitative or phenomenal properties of conscious experience. By contrast, the approach to emergence presented in this book is oriented not by the Cartesian mind-body problem but by the Kantian problem of self-organization (see Chapter 6) and its relation to the threefold order of matter, life, and mind.9
Let me close this discussion by making a few comments about the three principles at the base of Kim’s argument. His argument is as strong as these principles, and although they have force against classical British emergentism, each one is dubious when applied to emergence in the context of contemporary science.
According to the first principle, every emergent is determined by its physical realization base or emergence base; according to the second, every emergent inherits its causal powers from its emergence base. These principles are simply an expression of physicalistic ontological reductionism. This position claims science for its support, but it is metaphysical in the sense of going beyond anything science itself tells us. I see no good reason to believe in such a thing as an “emergence base,” where this means a configuration of preexisting microphysical entities with intrinsic properties and causal powers that ground the macrophysical level. This image of nature as a mereologically ordered hierarchy grounded on a base level of particulars is a metaphysical picture projected onto science, whereas the image science projects is of networks of processes at various spatiotemporal scales, with no base-level particulars that “upwardly” determine everything else (Bickhard and Campbell 2002; Hattiangadi 2005).
What about Kim’s third principle, the causal closure of the physical? The precise meaning and status of this principle are not clear. The main problem is that it is far from clear what “physical” means—what it includes and excludes—and it is hard to see how one could go about answering this question short of having a complete and true physics (whatever that means and assuming it even makes sense to suppose such a thing). Suppose at some point in the future physicists felt compelled to include mental properties (qua mental) as fundamental properties of physical theory (Montero 1999, 2001). Given that we cannot accurately predict the future course of physics, we have to at least allow for this possibility. In fact, some physicists and philosophers already believe such inclusion to be necessary to account for both mental and physical phenomena (Shimony 1997). But in that case, the closure of physics would include the mental qua mental (as opposed to the mental qua reduced to the physical). This possibility illustrates that the Cartesian mental/physical distinction has become useless.
As for Kim, he embraces a very wide sense of “physical” that includes every classical example of an emergent property from physics, chemistry, biology, and psychology, save qualia. Nowhere in my discussion of emergence and downward causation have I appealed to anything outside the physical domain so broadly conceived. If the principle of the causal closure of the physical is instead construed more narrowly to mean the causal closure of the microphysical, then the principle is not obviously true and may even be false or incoherent (Dupré 1993; Hattiangadi 2005). For instance, according to Bohr’s interpretation of quantum mechanics, macrophysical concepts are as indispensable to characterize microphysical phenomena as microphysical phenomena are to explain certain macrophenomena (Hattiangadi 2005). Moreover, the causal closure of physics has traditionally been linked to the ideal of the unity of science via intertheoretic reduction—that biology is in principle reducible to chemistry, and chemistry to physics. But such unity via intertheoretic reduction seems nowhere in sight (Dupré 1993; Garfinkel 1981). As Nancy Cartwright says about the classic case of the supposed intertheoretic reduction of physical chemistry to quantum mechanics: “Notoriously we have nothing like a real reduction of the relevant bits of physical chemistry to physics—whether quantum or classical. Quantum mechanics is important for explaining aspects of chemical phenomena but always quantum concepts are used alongside of sui generis—that is, unreduced—concepts from other fields. They don’t explain the phenomena on their own” (Cartwright 1997, p. 163).
One final remark about Kim’s conception of the “physical” is in order here. Part/whole reductionism goes hand in hand with an atomistic metaphysics of basic physical particulars and their mereological configurations, a metaphysics to which Kim apparently subscribes (1993, pp. 77, 96–97, 337). At the same time, he also apparently believes that nothing in the philosophical dispute about emergence depends on precise general definitions of “physical” (1993, p. 340). But this seems wrong on both counts. In the context of contemporary science, as we have seen, “nature” does not consist of basic particulars, but fields and processes, and this difference between a process-viewpoint and an elementary-particle-version of Cartesian substance metaphysics does make a difference to the philosophical issues about emergence (Campbell and Bickhard 2002; Hattiangadi 2005). In the former view, there is no bottom level of basic particulars with intrinsic properties that upwardly determines everything else. Everything is process all the way “down” and all the way “up,” and processes are irreducibly relational—they exist only in patterns, networks, organizations, configurations, or webs. For the part/whole reductionist, “down” and “up” describe more and less fundamental levels of reality. Higher levels are realized by and determined by lower levels (the “layered model of reality;” see Kim 1993, pp. 337–339). In the process view, “up” and “down” are context-relative terms used to describe phenomena of various scale and complexity. There is no base level of elementary entities to serve as the ultimate “emergence base” on which to ground everything. Phenomena at all scales are not entities or substances but relatively stable processes, and since processes achieve stability at different levels of complexity, while still interacting with processes at other levels, all are equally real and none has absolute ontological primacy.