Commentary on Lewis

Abstract: 58 words
Main Text: 1289 words
References: 33 words
Total Text: 1394 words

The concept of circular causality should be discarded

Bram Bakker
Informatics Institute
University of Amsterdam
Kruislaan 403
1098 SJ Amsterdam
The Netherlands
+31 20 525 7524
bram@science.uva.nl
http://www.science.uva.nl/~bram/


Abstract

This commentary argues that one specific but central concept in Lewis's theory, circular causality, is fundamentally flawed and should be discarded – first, because it does not make theoretical sense, and, second, because it leads to problems in practice, such as confounding the interaction between different systems with the relationship between different levels of analysis of a single system.


In recent years, the dynamic systems framework has become increasingly popular with cognitive scientists and neuroscientists. Lewis' article is a valuable contribution to this area, and it extends the reach of the framework to the psychology and neurobiology of emotions. He makes a convincing case for the claim that in emotional interpretations (EIs), the strong interaction between emotions and appraisals may be understood best in terms of the dynamic systems concepts of positive and negative feedback, and the rapid unfolding of an EI may be understood as a phase transition ending in convergence to one of a limited set of attractors. Furthermore, Lewis argues convincingly that such explanations can be plausibly connected to the underlying neural machinery. Even if not all details have been worked out, his comprehensive account of emotions and appraisals provides significant insights as well as valuable guidance to further work.

However, this commentary focuses on one specific but central element in Lewis' paper, the concept of "circular causality". This concept is used in much dynamic systems theorizing and is defined in section 3.2.4 as follows:

Feedback is one form of nonlinear causation. A second form, termed circular causality (Haken 1977), describes bidirectional causation between different levels of a system. A coherent, higher-order form or function causes a particular pattern of coupling among lower-order elements, while this pattern simultaneously causes the higher-order form. The top-down flow of causation may be considered an emergent constraint (by the system as a whole) on the actions of the parts.
One typical example cited by theorists using this term is the phenomenon of emerging orderly patterns in boiling water, called convection rolls (e.g. Kelso 1995). In this example, circular causality refers to the relationship between the water molecules and the global convection roll pattern. As is apparent in Lewis' definition, and also in other literature (e.g. Haken 1977; Kelso 1995), circular causality is construed as a kind of "nonlinear" causality, to be contrasted with ordinary, "linear" causation, which is predominant in theorizing about emotions (sections 1, 2.2, 2.3, 2.4, 3.1, 7). Throughout the paper, Lewis develops the argument that emotion theories based on linear causal processes should be replaced by theories based on nonlinear, circular causality.

This commentary's main argument is that the concept of circular causality is fundamentally flawed and should be discarded. Circular causality, as it is construed by Lewis and other authors, is about the relationship between different levels of analysis of a single system, between parts and wholes: the behavior of the parts "causes" the whole form, and the whole form in turn "causes" the behavior of the individual parts. However, since we are considering a single system, the whole form simply is the collection of individual parts interacting in a particular way. The word "correspondence" seems more appropriate than "causality" to describe this relationship. The relationship between a single constituent part and the whole form is the relationship of being one of multiple components making up the whole; again it is not properly conceived as a causal relationship. In the example of self-organizing convection rolls in boiling water, the convection roll pattern corresponds to the combined behavior of all constituent water molecules; and a single water molecule is one component of the whole system. In fact, this relationship between different levels of analysis of the same system is not fundamentally different for self-organizing systems than for other types of systems. A car corresponds to the collection of its interacting parts, and a wheel is one component of the whole system.

One of the hallmarks of circular causality (see the definition above) is the "top-down flow of causation" in a self-organizing system, which is considered a special kind of constraint by the whole system on the actions of the parts. However, such "constraints" are also there in other kinds of systems. The behavior of a wheel of a car is "constrained", in the same sense, by the behavior of the whole car and vice versa: the wheel is moving when the whole car is moving and vice versa. Such a "constraint" is only natural as the wheel is part of the car; but one would not normally refer to it as "circular causality". In other words, these types of constraints are always there between the whole system and its parts, and no special form of causality needs to be invented to enforce them.

Obviously, something special is going on in self-organizing systems. Unlike cars, self-organizing systems can "spontaneously" go from unordered regimes to ordered regimes that have surprising complexity, without any obvious, external instruction or programming agency---and this type of behavior may be very relevant for understanding cognition. However, there are already many appropriate concepts to describe what is special about it: self-organization, emergence, phase transitions, attractors, nonlinearities, etc. There is no need or justification for invoking a special kind of causality here, "circular causality", which should be distinguished from normal, "linear" causality.

One might agree with this argument in principle, but not see the relevance beyond a philosophical discussion of fairly arbitrary definitions and semantics. However, unfortunately the concept of circular causality leads to identifiable problems in scientific practice, both in Lewis' article and in other work.

The main problem is that circular causality suggests an interaction between separable entities that does not exist. At worst, this leads to suggestions of cartesian dualism: "circular causality has been identified between superordinate mental states (e.g., attention, expectancy) and subordinate neural events" (section 4). At best, it becomes very difficult for the reader (and the author) not to confound circular causality, which according to the definition is about parts and wholes of a single system, with "bidirectional" or "reciprocal" causality, which refers to the mutual interaction (feedback) between different systems (e.g. sections 1, 2.4, 3.1, 3.2, 3.3, 4.1, 4.2, 5.1, 5.3, 6.1, 6.2, 7). Not surprisingly, this is particularly apparent when the situation is considered of interaction between systems located at different levels of the brain hierarchy or neuroaxis (section 5.3): "Vertical integration could instantiate circular causality at the whole-brain level, thus providing an additional mechanism of neural integration." Here it is hard not to interpret this as putting on a par, on the one hand, vertical integration, which refers to connecting systems at different levels of the brain hierarchy through bidirectional feedback loops, with, on the other hand, circular causality, which refers to different levels of analysis of a single system.

All this makes it difficult to determine what exactly the claims are that are put forward by Lewis, and what scientists should do to verify the claims. Should experimentalists start looking for mechanisms that realize circular causality in the brain? Should theorists start developing mathematical theories that describe formally how circular causality operates in the brain? This commentary argues that circular causality cannot and will not be found in the brain, no matter the amount or sophistication of experimentation. And mathematical formalization of the proposed self-organizing emotion/appraisal system (or any self-organizing system) should reveal immediately that the relationship between parts and the whole form is not properly understood as causal. For instance, the equations governing the parts of a self-organizing system do not need, as input, variables describing the whole form, but instead depend only on variables describing other parts. This is why the system is self-organizing in the first place. In general, this highlights the value of providing, together with a conceptual theory, corresponding equations that clarify the concepts and make them precise.

In summary, circular causality is a concept that does not make sense, that does not add anything to the theory, and that leads to problems in scientific practice. Therefore, it should be discarded. In Lewis' paper, the concept of circular causality unfortunately distracts and takes away from his otherwise comprehensive, novel, and important contributions.


References

Haken, H. (1977) Synergetics – An introduction: Nonequilibrium phase transitions and selforganization in physics, chemistry and biology. Springer Verlag.

Kelso, J. A. S. (1995) Dynamic patterns: The self-organization of brain and behavior. MIT Press/Bradford.