Chaos, Self-Organization, and Psychology

Scott Barton

A variety of investigators in recent years have proposed Shaw, 1990), coordination (Turvey, 1990), and condi-
models of psychological systems based on the concepts of tioning in animals (Hoyert, 1992). In clinical psychology,
chaos, nonlinear dynamics, and self-organization. Unfor- the concept of self-organization and nonlinear systems
tunately, psychologists in general have little understanding has been applied to models of family systems and marital
of these important ideas. These terms are defined, and therapy (Elkaim, 1990; Gottman, 1993), psychotherapy
their relationships are discussed. The value of applying (Goudsmit, 1989; Reidbord & Redington, 1992), and the
these concepts to psychological systems is demonstrated role of cognitive development in psychopathology (Gui-
by exploring their utility in areas ranging from neurosci- dano, 1991; Putnam, 1988). In addition, a number of
ence to clinical psychology. Some of the difficulties in using authors have applied these concepts to core issues in the
nonlinear concepts and methodologies in empirical inves- philosophy of science and systems theory. Their work has
tigations are also discussed. been helpful in understanding the processes of growth,
change, and development among a broad range of bio-
logical and psychological systems (Jantsch, 1980; Levine
& Fitzgerald, 1992; Odum, 1988; Sabelli & Carlson-Sa-

I n recent years, a new paradigm for understanding sys-

tems has been gaining the attention of psychologists
from a wide variety of specialty areas. This paradigm
has no single name but has been described in terms of
chaos, nonlinear dynamics (sometimes called nonlinear
belli, 1989; Vandervert, 1991; Waldrop, 1992).
The level of technical understanding required to un-
derstand chaos, nonlinear dynamics, and self-organization
from the perspective of mathematics or physics is generally
not necessary for psychologists. However, a certain degree
dynamical systems theory), and self-organization. Al- of familiarity with the mathematical and physical un-
though these interrelated concepts have generated a great derpinnings is helpful. One of the most basic ideas is the
deal of interest in physics, chemistry, and biology (e.g., concept of dynamics itself. Some of the principles in-
Gleick, 1987; Kauffman, 1993; Stewart, 1989), the ma- volved in dynamics are described in the next section.
jority of psychologists know very little about them. The
purpose of this article is to define these concepts, clarify Dynamics in Nonlinear Systems
the types of issues they are applicable to, and discuss their
significance for research and clinical practice. At the most basic level, dynamics is the study of the way
in which systems change (Morrison, 1991). Dynamics
Understanding Growth, Change, and explores the effect of various forces on the behavior of
Development in Psychological Systems systems over time and the manner in which these systems
seek optimal stable states. Dynamics may be used to ex-
The dynamic behavior of complex psychological systems plore a variety of systems. Some of these systems are rel-
is often difficult to understand. Why, for instance, do atively simple (e.g., a study of the forces acting on an
groups of neurons often synchronize their firing patterns apple that cause it to fall to a stable rest on the ground),
in a unique spatial manner (Freeman, 1991)? How can whereas others are dauntingly complex (e.g., the forces
a person have two or more separate and distinct person- that act on the fertilized human egg that lead to the de-
alities (Putnam, 1988)? Why do various belief systems velopment of a full-term infant).
link up with one another to create family dysfunction
From a mathematical perspective, dynamics can be
(Elkaim, 1990)? The answers to these and many more
thought of as linear or nonlinear. The basic assumption
questions about dynamic psychological systems can be
underlying linear dynamics is that the way a system
explored using the concepts of chaos, nonlinear dynamics,
changes can be most effectively modeled with two or more
and self-organization.
In neuroscience and psychophysiology, these con-
cepts have been used to investigate the way memories are Larry R. Vandervert served as action editor for this article.
I wish to thank Larry Vandervert for his extraordinary perseverance,
formed (Freeman, 1990, 1991; Kohonen, 1988), the way patience, and enthusiasm while serving as action editor for this article.
attention affects the dynamics of human electroenceph- I also wish to thank Julian Libet, Charleston Veterans Administration
alograms (EEGs; Basar, 1990a), the dynamic nature of Medical Center, for providing me with his generous support and feedback
sleep (Roschke & Aldenhoff, 1992), and the way connec- throughout the writing process. The input and assistance of Walter Free-
man, William McCown, and Dana Redington were also invaluable, and
tionist models account for learning (Carpenter & Gross- I wish to acknowledge them as well.
berg, 1987; Hanson & Olson, 1990). In experimental Correspondence concerning this article should be addressed to Scott
psychology, nonlinear dynamics have been used to model Barton, Department of Veterans Affairs, Ralph H. Johnson Medical
approach-avoidance conflicts (Abraham, Abraham, & Center, Rt 116B, 109 Bee Street, Charleston, SC 29401-5799.

January 1994 • American Psychologist

In the public domain
Vol. 49, No. 1, 5-14
 equations whose solutions can be combined to obtain
another solution (Morrison, 1991; Stewart, 1989). Linear Figure 1
equations work quite well for a number of problems in Bifurcation Diagram Arising From Iterations of the Equation
the physical sciences. For instance, they are very useful x' = x + rx (1 — x)
if one wants to predict the orbit of the planets or under- 1.34 - ,
stand the effects of wind resistance and gravity on the
trajectory of a missile. Because they are additive, they
also work well for a number of problems in psychology.
They are, for instance, the cornerstone of statistics. When
we perform an analysis of variance or enter data into a
multiple regression equation, we are using linear equa-
tions to describe the relationships among variables.
The problem with linear equations is that they can-
not always describe what happens in natural systems. This
failure is especially noticeable when continuous changes i
1.5 r 3.0
in certain control parameters lead to sudden jumps in
Note. This diagram illustrates the way in which solutions to a very simple nonlinear
behavior. Although linear equations are helpful in de- equation can result in sudden jumps from one set of solutions to the next. Ultimately,
scribing the smooth flow patterns of a liquid within a they can jump to very complex chaotic behavior.
certain range of flow rates, for instance, nonlinear equa-
tions are necessary to describe these same patterns when
that range is exceeded and the sudden jump to turbulence
sets in (Gleick, 1987). This occurs in human psycho- to result in one of four typical patterns (Abraham et al.,
physiological systems as well. The speed of alternately 1990; Tufillaro et al., 1992). These patterns, when graphed
tapping one's index fingers (antiphase tapping), for in- in diagrams that show periodic changes in behavior, are
stance, can be adjusted in a linear manner within a certain called attractors. The trajectories of these attractors typ-
range of tapping speeds. However, when the high limit of ically converge on a discrete point, a simple oscillating
that range is exceeded, a sudden nonlinear jump to in- cycle, a quasiperiodic cycle, or a chaotic cycle (see Figure
phase finger tapping occurs (Kelso & Schoner, 1988). 2). The attractors in Figure 2 are portrayed in two di-
To explore systemic change, nonlinear dynamics uses mensions.
nonlinear equations (Abraham et al., 1990; Morrison, The chaotic attractor requires a special explanation.
1991). Nonlinear equations are not additive; therefore, The pattern in such an attractor is bounded, but after a
they are often difficult to solve. Sometimes a single so- certain number of repetitions within the system, it be-
lution can be obtained, but often the answer involves a comes very irregular. This irregularity results in unpre-
pattern of solutions. To find such an pattern, the data are dictability, despite the fact that it derives from a com-
generally run through a system of equations so that the pletely deterministic system. This unpredictability is as-
results ultimately feed back into the system itself. If, for sociated with a property of chaotic systems known as
instance, one takes the equation x1 = x + rx (1 —x) and sensitive dependence on initial conditions. This means that
feeds the results back into the equation (so that what was if two sets of initial conditions differ by any arbitrarily
x1 for the first solution becomes x for the second), one small amount at the outset, their specific solutions will
can explore the behavior of a classic nonlinear equation diverge dramatically from one another over the long range.
(Tufillaro, Abbott, & Reilly, 1992). This process is called In the case of the pattern in Figure 1, for instance, if x =
iteration. When x is set at 0.5000, the values of r are 0.5000 had been replaced with x = 0.5001, the solutions
allowed to range between 1.5 and 3.0, and the results are in the chaotic portion of the attractor would have been
plotted on the y axis; the peculiar pattern characteristic completely different, whereas the general chaotic pattern
of bifurcations (the geometrical splitting of the solution) would be the same. Given that no measurement system
to chaos emerges (see Figure 1). These splits occur quite is without some error, it becomes clear that if a system
suddenly at certain values of r. Actually, the points in is chaotic, general patterns of future behavior may be
Figure 1 represent the solutions after they have had time predictable but specific behaviors over the long range
to "settle down," in this case after 150 iterations for each will not.
value of r. Readers who have trouble making sense of this It is possible to construct a chaotic attractor from a
diagram might want to think of Figure 1 as a two-di- set of equations, as I have demonstrated here, or it can
mensional map of a time series. The only difference is be reconstructed from a time series of observable repet-
that just enough numbers are plotted to show the repet- itive behaviors. The latter solution is often the easiest way
itive pattern for each value of r. For those who are inter- to map out the attractor associated with a complex psy-
ested in exploring this pattern on their own, an extremely chophysiological system (e.g., Hoyert, 1992; Reidbord &
simple computer program in BASIC is available for Redington, 1992; Roschke & Aldenhoff, 1992).
studying a similar equation (Stewart, 1989). Chaotic behavior is more than just a mathematical
As I have noted, nonlinear systems tend to settle anomaly. It occurs in the real world as well. A wide range
down over time. This settling down, or convergence, tends of physical, chemical, and biological systems are now

January 1994 • American Psychologist

 (BZ) reaction. My goal is to help build an understanding
of the fundamental characteristics of such systems, char-
Figure 2 acteristics that are relevant to chemistry, biology, and
Types of Attractors Typically Found in Nonlinear Systems-.
(A) Point Attractor, (B) Cyclical or Oscillating Attractor, (C)
psychology (Schore, 1981). The BZ reaction, discovered
Quasi periodic Attractor, (D) Chaofic Attractor in Russia in the late 1950s, is often used to illustrate the
concept of self-organization (Gleick, 1987; Prigogine &
Stengers, 1984; Schore, 1981). Its behavior can be defined
in terms of a cyclical or sometimes even chaotic attractor
and can be modeled by a system of nonlinear differential
equations (Epstein, Kustin, De Kepper, & Orban, 1983).
In the BZ reaction, the system has the potential to
exist in two different states. Assuming that an iron catalyst
A B
is used, one state appears red and the other blue (Winfree,
1974). If the reaction is allowed to run in a continuously
stirred beaker and the concentration of the reactants
crosses a critical threshold, it will oscillate between the
red state and the blue state at intervals of about 30 sec-
onds. It first appears red and then blue. Unlike a "typical"
C D chemical reaction, it does not move toward a static equi-
librium point in a linear manner. Instead, it cycles in an
obvious and dramatic fashion. The process of moving
from red to blue or vice versa is sudden and discontinuous
(see Figure 3A). This occurs because the chemical pro-
known to exhibit deterministic chaos (Prigogine & Sten- cesses that result in the red state coming into existence
gers, 1984; Stewart, 1989). The presence of chaos suggests become linked to the processes resulting in the blue state.
that even if we are able to characterize all the variables When this happens, the two states codetermine one an-
in a nonlinear system completely, general patterns of fu- other in a cyclical, nonlinear fashion.
ture behaviors may be the best we can hope to predict.
In an insightful treatment of this problem as it relates to
behavior analysis, Hoyert (1992) explored the behavior
of a hypothetical system designed to predict within-in-
terval variability in a fixed-interval reinforcement sched-
Figure 3
ule. He demonstrated that chaotic behavior can arise even (A) Simplified Illustration of Oscillating States in the
when the variables in such a system are completely de- Beluzhov-Zhabotinsky Reaction; (B) Propagation of Spiral
termined. Hoyert went on to note that the interdepen- Waves in the Beluzhov-Zhabotinsky Reaction
dence of variables in a nonlinear system, along with sen-
sitivity to initial conditions, lead to the implication that
studying each factor in isolation may not lead to useful
knowledge about the behavior of the system as a whole.
This concept, long a tenet of general systems theory, has
now been unequivocally demonstrated in complex non-
linear systems.

Self-Organization
Fundamentals of Self-Organizing Systems
For any discussion of chaos to be linked meaningfully to
psychological systems, it must be linked to the broader
concept known as self-organization (Abraham et al., 1990; BLUE
Kauffman, 1993; Prigogine & Stengers, 1984). Self-or-
ganization denotes a process by which a structure or pat-
Note. Substances X and Y are used generically to refer to any two substances
tern emerges in an open system without specifications that will support an oscillating chemical reaction. Note that when the solution is in
from the outside environment. When a system of this the red state, the concentrations of Substances X and Y change slowly within a
type receives a sufficient amount of energy, it may become certain range. Beyond this range, however, the entire system switches suddenly
to the blue state. After this, the concentration- of reactants once again change
unstable. As a result of this instability, an originally uni- slowly, although now in the opposite direction, until they reach a certain point.
form state can give rise to a variety of complicated tem- After this point, the whole system jumps suddenly into the red state. These linked
poral, spatial, and behavioral patterns (Prigogine & Sten- processes occur repeatedly in a cyclical fashion.
Bl -B4 illustrate spatiotemporal self-organization in a chemical system. Note
gers, 1984). the entrainment of the slower wave by the faster wave in the lower right-hand
In this section, I focus primarily on self-organization corner as the reaction progresses.
in a chemical system known as the Beluzhov-Zhabotinsky

January 1994 • American Psychologist

 When the BZ solution is constantly stirred, only tion, I discuss the work of an investigator who has begun
temporal oscillations will occur. However, if it is poured to explore such structures to understand the neural or-
into a petri dish and allowed to sit quietly, a whole new ganization underlying perception, memory, and behavior.
type of organization emerges. Under these conditions, any
small perturbation in the system, such as a piece of dust Self-Organization and Memory
or a hot needle, will create a local region of instability. A variety of exciting approaches to exploring cognition
This instability then triggers the formation of spiral or and memory are arising from the perspective of chaos
circular waves that slowly propagate throughout the sys- and self-organization. Some of the most fascinating studies
tem (see Figure 3B). Thus, the system develops not only have emerged from the laboratory of Walter Freeman
temporal self-organization but spatial self-organization (1990, 1991), one of the primary proponents of nonlinear
as well. The waves propagate with a diverse set of fre- models in the study of brain function. Freeman, a neu-
quencies, and a number of different waves can be created rophysiologist at the University of California at Berkeley,
simultaneously. If a faster wave meets a slower one, it will has extensively explored the manner in which odors are
overtake it, causing the slower one to disappear. This phe- remembered and represented in the olfactory system of
nomenon is called entrainment. It can be observed in rats. Using EEG measurements, the results he has ob-
Figure 3B in the lower right-hand corner of the petri dish. tained have given us new insight, not only into the mech-
In the stirred solution, the role of diffusion is nullified by anism of olfaction but into the general role of memory
constant stirring, so that only reaction-based temporal and nonlinear systems in the brain.
oscillations can occur. In the still of the petri dish, however, Understanding how odors are represented is a chal-
a linkage is established between reaction and diffusion, lenging task. One of the first approaches taken was to
leading to both temporal and spatial oscillations (Winfree, investigate the global EEG of the olfactory lobe itself.
1974). When a rat inhales an odor, it sets off a burst of electrical
In discussing this reaction, Schore (1981) asked, activity in the lobe. This burst is characterized by a wave-
"Where else do we find systems possessing a high degree form known as the carrier wave. The carrier wave is an
of naturally generated organization that are highly sen- aspect of the EEG that reflects the integrative actions of
sitive to perturbations, switch rapidly from one state to local pools of neurons in the brain. Its elicitation following
another, and operate spontaneously (in response to minor exposure to an odor is a very reliable finding. However,
fluctuations) as well as with external triggering?" (p. 454). the frequency and amplitude of the wave varies in an
His answer was, "in living systems." Specifically, he com- unreliable (functionally random) manner both between
mented on how neural thresholds, states of consciousness, and within odorant exposures. Attempts at signal detec-
and various biological oscillators exhibit this type of be- tion (e.g., signal averaging and detection of evoked po-
havior. The remainder of this article expands on this idea tentials) have proved incapable of delineating a reliable
by focusing on the way in which these properties manifest differential response from one odor to the next when
in animal and human behavior. temporal patterns alone are investigated.
The BZ reaction illustrates the general characteristics Freeman, however, knew that the information re-
of self-organization that apply to psychological systems layed to the brain by odorant exposure occurred in the
as surely as to chemical and biological systems. These form of a spatial pattern of pulse activity across neurons.
include the readiness to exhibit (a) multiple stable states He suspected that odors might therefore be represented
that can change suddenly from one to another when a on the lobe by some sort of spatial map. To explore this
parameter value crosses a critical threshold, (b) cyclical idea, he took a unique approach. Instead of placing a
state changes, (c) the structural coupling of component single electrode on the olfactory lobe, he placed an array
processes, (d) temporal, spatial, and behavioral organi- of 60 electrodes on a representative portion of the bulb.
zation, (e) localized instabilities that can lead one part of The electrodes, arranged in a 6 X 10 array, allowed him
the system to organize itself differently from another part to obtain EEGs at a far higher level of spatial resolution
of the system, (f) the ability of one unit to cause other than he could previously obtain. When he did this, he
units to oscillate at a harmonically related frequency (en- found that each odor was indeed distinguishable by the
trainment), and (g) behavior that can sometimes be mod- spatial pattern of the amplitude of the wave (see Figure
eled by a system of nonlinear equations. 4). In short, the message (e.g., "this is peppermint") was
Although chemical systems are valuable as meta- not in the waveform at all; it was in the spatial pattern
phors, most biological and psychological systems are of the amplitudes of the waveform. When animals learned
considerably more complex. Psychological systems lack a response to an odor, each odor was shown to have a
the precise temporal or spatial symmetry seen in physical specific spatial amplitude pattern. In the process of ex-
systems and instead involve complex neurological struc- ploring this area, Freeman found clear-cut evidence of
tures and behaviors. To understand these types of systems, biologically significant self-organization: stable spatio-
it is important to realize that some self-organizing prop- temporal structures in olfactory EEGs triggered by a small
erties can only be found in living things. One of the most perturbation (the odor).
general of these properties involves the ability to develop If the spatial structure of odor representations had
stable yet flexible structures that serve important biolog- been all that Freeman had found, it would have been an
ical needs (Prigogine & Stengers, 1984). In the next sec- important contribution in and of itself. The full signifi-

January 1994 • American Psychologist

 associated with positive reinforcement, for instance, is
completely different from the structure associated with
Figure 4 punishment.
Left Side-. 60 Electroencephalograms Recorded
Simultaneously From the Olfactory Cortex of a Rabbit as It These findings throw light on what, historically, has
Recognizes a Scent; Right Side: Contour Plot been one of the core problems in understanding memory:
Corresponding to the Amplitudes its extraordinary flexibility as a function of changes in
mental state, emotional needs, more recent learnings, and
external cues. Bartlett (1932) was keenly aware of this
problem. He noted that "some widely held views [of
memory] have to be completely discarded, and none more
completely than that which treats recall as the reexcite-
ment in some way of fixed and changeless traces" (Bartlett,
1932, p. vi). Bartlett went on to note that
Note. The electrodes are arranged in a 6 X 10 rectangular array. The carrier
wave is nearly the same in each recording, except that the amplitude varies. The A new incoming impulse must become not merely a cue setting
shape of the wave, however, does not indicate the identity of the scent. Information up a series of reactions in afixedtemporal order, but a stimulus
about identity is contained in the spatial pattern of the amplitudes. The differences which enables us to go directly to that portion of the organized
between amplitudes are represented in a manner analogous to the differences in setting of past responses which is most relevant to the needs of
elevation on a topographic map. From "Spatial Properties of an EEG Event in the
Olfactory Bulb and Cortex" by W. Freeman, 1978, Eledroencepha/ography and the moment. . .. There is one way in which an organism could
Clinical Neurophysio/ogy, 44, p. 589-590. Copyright 1978 by Elsevier Scientific learn to do this. . . . It may be the only way. . . . An organism
Publishers, Ireland Ltd. Reprinted by permission. has to somehow acquire the capacity to turn round upon its
own "schemata" and construct them afresh, (p. 206)
Freeman's work, along with that of others, is now allowing
us to confirm that this "reconstruction" of memory ac-
cance of his discovery, however, depended on more details tually occurs at an objective physiological level.
about the process of learning. He continued to investigate Although the findings showing changes in spatial
this phenomenon and found a multitude of fascinating patterns as new odors were learned were exciting, Free-
qualities. man (1990) also wondered about dynamic processes in
After making this initial discovery, Freeman asked the more classical sense of the word. He was especially
an interesting question. "What happens to the original interested in the rate at which neural information spread
spatial pattern associated with the first odor when a second from the sensory receptors to the bulb itself and the shape
odor is learned?" If the original representation is truly of that spread. When he explored this, he discovered that
stable (like the stimulus response connection it is asso- once an odor had been learned, its recognition could begin
ciated with), it should not be changed by new learning. anywhere among the sensory receptors on the bulb and
To illustrate this metaphorically, assume that one had spread rapidly and coherently throughout the entire bul-
learned to associate peppermint with some behavior. After bar structure in the manner of a two-dimensional spread-
the learning was completed, one's spatial EEG pattern ing wave. He noted that this type of spreading response
for peppermint was determined. Subsequently, one was necessary to account for the bulb's ability to rapidly
learned to associate cinnamon with another behavior. Af- create the same psychological meaning from a variety of
ter this, one's pattern for cinnamon was assessed as well. spatial points on the sensory receptors. As noted in the
If one once again smelled peppermint and were tested last section, the property by which information necessary
again, one might expect the pattern associated with pep- for creating a spatiotemporal structure is "stored" in a
permint to look the same as it did before. When Freeman nascent form throughout the system and in which a per-
ran an analogous experiment with animals, however, he turbation can generate this structure through a wavelike
found that just the opposite occurred. The pattern for response is a hallmark of self-organization in chemical
peppermint changed when it was tested again. He was and biological systems (Goldbeter & Segel, 1977; Prigo-
forced to conclude that the neural representation of an gine & Stengers, 1984; Winfree & Strogatz, 1984).
odor is not fixed like a photograph. Instead, the structure With every new finding adding further evidence that
of old learnings reform in the context of more recent he was dealing with a complex nonlinear system, Freeman
learnings. This reorganization of the nervous system is finally attempted to model the behavior of the bulb using
not consistent with the view that discrete categories of a system of nonlinear differential equations. He wanted
experience are stored away in fixed physiological patterns a model that allowed him to mimic many of the properties
in the brain. It is also additional evidence of self-orga- he observed naturally in the bulb, including the ability
nization: spatiotemporal structures that have some degree to suddenly "turn on" a state of odor recognition asso-
of stability but that can reconstruct themselves when de- ciated with an appropriate activity pattern and the ability
stabilized by new information. to learn new information about an odor without losing
Freeman (personal communication, July 8, 1993) all information about previous odors. Freeman was able
noted that it is important to be aware that the structures to create such a model. Two examples of chaotic attractors
he discovered reflect the meaning of the stimulus, not his equations formed are illustrated in the phase portraits
merely its presence. The spatial structure of peppermint appearing in Figure 5.

January 1994 • American Psychologist

 Despite the success of Freeman's approach in un-
derstanding the dynamics of the olfactory system, it has Figure 5
its limitations. One limitation is that his model, although Phase Portrait of Two Chaotic Attractors Modeled by
it accounts quite well for a number of observed properties, Nonlinear Differential Equations to Represent Odorant
does not correspond with the actual EEG patterns in the Responses
olfactory lobe. By comparison, the real patterns seem
impossibly complicated and noisy. This discontinuity be-
tween model and reality is a common problem in inves-
tigating natural nonlinear systems. Not only are inves-
tigators rarely able to completely characterize all the
variables that affect a complex system, but they must
isolate a system well enough to cut through what Mor-
rison (1991) called a "sea of noise" (p. 271). Achieving
the necessary degree of isolation is a difficult task, even
in very simple physical systems (e.g., see Gleick, 1987).
In neural systems, the problem is compounded by an Note. Observe the geometric similarities to the chaotic attractor in Figure 2D. From
"Simulation of Chaotic EEG Patterns With a Dynamic Model of the Olfactory
even greater degree of complexity and interconnectedness. System" by W. Freeman, 1987, Biological Cybernetics, 56, p. 143. Copyright
Freeman (1990) noted that many of the better known 1987 by Springer-Verlag. Reprinted by permission.
tests for nonlinearity in mathematics, physics, and chem-
istry are inadequate in neuroscience because of this com-
plexity. The initial conditions of the brain shift irreversibly
every time something new is learned. and unreliability. They cannot be manufactured in the
It is because of this irreversible process of growth laboratory for the convenience of the investigator. There-
and pattern development in psychological systems that I fore, they are fundamentally more difficult to quantita-
stress the characteristics and dynamics of self-organizing tively model than the systems described in the section on
systems rather than their mathematics. In Freeman's self-organization in chemical systems. Many investigators
work, these properties include sudden state changes, spa- are working on the development of quantitative methods,
tiotemporal organization of odor representations, a but their efforts are in the early stages at the present time
wavelike spread of information, and stable yet flexible (e.g., Levine & Fitzgerald, 1992; Reidbord & Redington,
structures associated with learning. What is truly ex- 1992).
traordinary is that the processes by which the brain creates In the past, a number of prominent psychotherapists
memories can be similar in so many ways to the processes and researchers speculated briefly about the role of chaos
driving an oscillating chemical system. It is the common and nonlinear systems in psychotherapy (e.g., Meehl,
base of nonlinear dynamic processes that connect the 1978; Minuchin & Fishman, 1981). Although they sus-
two systems. pected it might have some utility, they did not develop
Through nonlinear dynamics and a growing under- their ideas to any significant degree. In recent years, the
standing of self-organization, a whole new way of thinking number of interested model makers has increased dra-
about brain function is beginning to emerge. Basar matically. Various aspects of nonlinear dynamics have
(1990b) has edited an entire book of contributions in this been applied to Jungian therapy (Abraham et al., 1990;
area. Edelman (1992) has devised an ambitious theory Biitz, 1992;Eenwyk, 1991), psychoanalysis (Langs, 1992),
using the concept of self-organization and evolution to posttraumatic stress disorder (Glover, 1992), psychic de-
explain brain function and development. Milton, Longtin, velopment and individual psychopathology (Guidano,
Beuter, Mackey, and Glass (1989) have reviewed the role 1991), family systems (Elkaim, 1990), the genesis and
of nonlinear dynamics in clinical neurology. There is little treatment of multiple personality disorder (Putnam, 1988,
doubt that the growth and development of nonlinear dy- 1989), schizophrenia (Schmid, 1991), and psychiatric
namic models in neuroscience will continue into the disorders in general (Sabelli & Carlson-Sabelli, 1989).
foreseeable future. These models are notable for their general reference to
topics and features derived from nonlinear dynamics, but
Self-Organization and Clinical Psychology otherwise are remarkably heterogeneous. In general, these
There is perhaps no other area in which chaos theory, authors have used these concepts to model process,
nonlinear dynamics, and self-organizing systems are so change, and development in psychological systems. A
intuitively appealing yet so analytically difficult as in great deal of the impetus for this work originated in Eu-
clinical psychology. With a few notable exceptions, their rope, driven by the perspectives of Maturana and Varela
application to clinical issues are metaphorical and qual- (1980) in neuroscience, and Prigogine (Prigogine & Sten-
itative in nature. This is due to the inherent complexity gers, 1984) in nonequilibrium thermodynamics.
of the clinical realm and the difficulty measuring behavior The concept of nonlinear transitions in mental states
with the reliability necessary for mathematical modeling is one of the most common themes among clinicians
(Maturana & Varela, 1987). Individuals and family sys- writing about nonlinear dynamics. In keeping with Emde,
tems must be taken as they come, with all their complexity Gaensbaur, and Harmon (1976), a mental state is denned

10 January 1994 • American Psychologist

as "a constellation of certain patterns of physiological states from heart rate data (see Figure 6). Although the
variables and/or behaviors which seem to repeat them- utility of this approach to the solution of clinical problems
selves and appear to be relatively stable" (p. 29). This has yet to be determined, it is an important advance in
general concept of mental states can be applied to a wide basic research. Roschke and Aldenhoff (1992) reported
range of problems in clinical and experimental work. an analogous investigation of the correspondence between
Wolff (1987) was one of the first to apply concepts chaotic attractors reconstructed from EEGs and various
derived from nonlinear dynamics to the development of stages of sleep.
mental states in infants. He explored a wide range of phe-
nomena, including waking, smiling, sleeping, and crying
Difficulties With This Paradigm
behaviors. Wolff demonstrated how some states, such as Despite the many advantages of nonlinear dynamics and
waking, are very unstable in newborn infants. As infants self-organization in expanding the description and anal-
mature, however, they develop longer and longer unin- ysis of psychological systems, these concepts have nu-
terrupted periods of wakefulness, indicating increased merous problems in their application to both basic and
stability. Wolff's concern with understanding the stability applied research. The problem of not knowing the factors
of states and the change in stability with the passage of involved and separating the signal from the noise was
time is very much in the tradition of nonlinear dynamics. discussed in the section on memory, but a number of
Building on the work of Wolff (1987), Putnam (1988, other problems deserve discussion as well.
1989) began to explore change in a very different domain:
state transitions in adults with multiple personality dis- Confusion of Concepts and Techniques Among
order. He proposed that children naturally develop various Different Fields
mental states, each imbued with a different sense of self. Varela (1989) issued an important warning to those who
Normally, with the passage of time and the presence of would compare neuroscience to family therapy in the
emotional support, these states are consolidated into a context of self-organization: "Circulation of concepts be-
more or less coherent self. However, when a child is se- tween diverse approaches is reciprocal, but we cannot
verely and repetitively traumatized, he or she may enter simply and directly export or import such notions.
these states defensively to avoid emotional or physical
pain. When this occurs over a long period, the sense of
self fails to consolidate. The various states become elab-
orated and develop a different set of memories, affective
Figure 6
qualities, and identities. They also become unstable and Four Phase Plots Constructed From a Time Series of Heart
discontinuous, predisposing the individual to sudden Rate Data
jumps between one state and another. Recall from the
discussion of the BZ reaction that one part of a self-or-
ganizing system can organize itself separately from an-
other. In this case, memories and perceptions become
organized so differently in each state that communication
between one state and another is blocked. Unfortunately,
this communication is the very thing that is needed to
develop a coherent sense of self. Putnam (1989) described
how the process of developing coherence involves opening
channels of communication between states. This occurs
through discouraging pathological dissociation and en-
couraging the integration of dissociated states, memories,
and affects.
State oscillations following trauma have been de-
scribed in areas other than identity. Horowitz (1986), for
instance, noted that the oscillation of intrusive memories
Note. All correspond with different clinical states observed during a single session
(e.g., nightmares and flashbacks) with the avoidance of of psychotherapy. The three axes used to construct these plots are listed in the
situations associated with the trauma (e.g., phobias and lower left-hand corner. The darker portions of the plots show the paths that the
withdrawal) is typical of the pathology following a trau- trajectories were most likely to settle into. For the sake of clarity, a varying number
of phase plots were superimposed on one another in Types a, c, and d. In Type
matic event. He explained how an appreciation of the a trajectories, the patient's behavior tended to be avoidant and overcontrolled.
oscillatory nature of these symptoms can help prevent In Type b trajectories, the characteristic behavioral pattern had a narrative quality
but was somewhat less avoidant. In Type c trajectories, the patient's pattern
their misinterpretation and assist in treatment. showed the highest level of affect. These trajectories were the most clinically
In a very different approach to exploring nonlinear variable overall. In Type d trajectories, the patient was more likely to discuss the
dynamics and mental states, Reidbord and Redington focal topic in therapy and their conversation flowed more easily between topics
and emotions. From "Psychophysiological Processes During Insight Oriented Psy-
(1992) looked at the relationship between the phase por- chotherapy" by S. Reidbord and D. Redington, 1992, The Journal of Nervous and
traits of a patient's heart rate and their behavioral state Mental Disease, 180, p. 652. Copyright 1992 by Williams & Wilkins. Reprinted
by permission.
during psychotherapy. They reconstructed a variety of
chaotic attractors associated with the patient's mental

January 1994 • American Psychologist 11

 .. . stable patterns in natural systems . .. have a clear nored, their meaning becomes too hard to discern. Al-
resonance with the establishment of human institutions, though all modelers must ignore some features of a com-
but the differences between them are profound" (p. 24). plex system, little is gained by ignoring them to the point
The wisdom of this statement should be obvious. The of making the model impractical.
dynamics of a propagating wave are a poor model for
therapists wishing to understand dysfunctional family re- Confusion Over How to Test Hypotheses
lationships. On the other hand, therapists might be very Testing hypotheses may seem problematic when the no-
interested in the linkage of attributions, beliefs, and coping tion of linear causality no longer applies and correlation
skills that underlie and maintain a family's pathology. A seems irrelevant. Although much work remains to be done
major problem in the psychological literature on chaos, to develop good nonlinear methodologies (see Basar,
nonlinear dynamics, and self-organizing systems is that 1990a; Levine & Fitzgerald, 1992), this realization need
this important distinction is often ignored. not pose an insurmountable problem for investigators.
One way that the distinction between fields is set The primary thing to remember is that the fundamental
aside is when authors use rigorous terminology from goal of modeling or analyzing a self-organizing system is
nonlinear dynamics to refer to psychological variables to understand a pattern of dynamic behavior. Hypotheses
that are multidimensional and difficult to quantify. Biitz must be built around such a pattern. Freeman (1990), for
(1992), for instance, defined chaos as "overwhelming instance, explored the hypotheses that the spatial pattern
anxiety," whereas Sabelli, Carlson-Sabelli, and Javaid of olfactory EEGs was related to specific odors. Reidbord
(1990) conceptualized creativity and destructiveness as and Redington (1992) explored the hypotheses that dy-
"chaotic attractors." These definitions, although clearly namic flow patterns deriving from heart rate data were
metaphorical, bear little resemblance to the definition of related to mental states. In both cases, the unit of analysis
chaos in the physical sciences. Terms that refer to specific was a dynamic pattern of observed behavior. Testing hy-
and limited ideas in mathematics and physics should not potheses about the difference between mean values of
be confused with the broader characteristics of self-or- individual variables before identifying the dynamic pat-
ganizing psychological systems. Using these terms as tern of interest may obscure a focus on the patterns that
metaphors may be acceptable as a heuristic device, but are naturally present in the system (Prigogine & Stengers,
the two are not the same. Although useful at times, all 1984).
metaphors eventually break down or lose their validity In regard to data analysis, the following three ap-
when more and more exacting parallels are drawn be- proaches are worth considering:
tween them and reality (Chubb, 1990). 1. If it is feasible to model a self-organizing system
Another common and related source of confusion with various nonlinear equations, then modelers should
involves taking a perfectly good hypotheses about a psy- by all means attempt such strategies. Nonlinear differ-
chological process and pitching it as if we could measure ential equations are especially useful in modeling neural
that process precisely. Jung (1946/1969), for instance, sytems (e.g., Carpenter & Grossberg, 1987; Freeman,
posited that the more one represses a particular feeling 1990). In modeling social systems, a system dynamics
or belief, the more likely it is to get converted to its op- approach has often proved productive (Levine & Fitz-
posite. The more an individual represses the belief "I am gerald, 1992).
worthless," for instance, the more likely he or she will be 2. If it is possible to reconstruct a nonlinear attractor
to express it as "You are worthless." Jung's is a perfectly from a time series, then modelers should try this as well.
reasonable hypothesis. However, to model such a process This approach has proved helpful in behavior analysis
with a phase diagram based on a system of differential (Hoyert, 1992), and in studying psychophysiological sys-
equations (e.g., Abraham et al., 1990) is to imply a level tems (e.g., Reidbord & Redington, 1992; Roschke & Al-
of measurement precision we don't have in clinical psy- denhoff, 1992). Taking this approach, however, is a func-
chology. This type of analogy may create a sort of "halo tion of one's ability to get reliable and meaningful data
effect" that makes the targeted construct seem more easily out of an experiment in which stable, clearly delineated
and accurately assessed than it really is. cycles of behavior are apparent.
The confusion of techniques appropriate at one level 3. If it seems unlikely that mathematical techniques
of analysis to those appropriate at another can also be that focus on the analysis of repetitive cycles alone will
seen when the wide variety of variables observed in the prove helpful or practical (e.g., when exploring develop-
real world are collapsed into a few simple dimensions. mental changes with maturity), the properties of self-or-
Callahan and Sashin (1990), for instance, developed a ganizing systems can be subsumed in one's model and
dynamic model based on the factors of feelings, thoughts, various aspects of the model tested using standard statis-
and actions to predict What they call affect-response. Us- tical techniques. Wolff (1987), for instance, was able to
ing an ordinal scale, they divided each dimension into demonstrate increasing stability in the sleep-wake cycle
low, medium, and high and plotted them on a three-di- of infants using statistical methods alone. His focus, how-
mensional state plot. However, to describe feelings, ac- ever, remained on dynamic phenomena.
tions, and thoughts only as low, medium, and high would, The various methodologies, both linear and nonlin-
for most therapists, reduce them beyond recognition. ear, are mutually compatible, not contradictory. They can
When the qualitative features of these dimensions are ig- be used to study different aspects of a system, depending

12 January 1994 • American Psychologist

 Edelman, G. M. (1992). Bright air, brilliant fire: On the matter of the
on which is most appropriate for addressing the specific mind. New York: Basic Books.
question at hand. The coherence of alpha waves in atten- Eenwyk, J. R. (1991). Archetypes: The strange attractors of the psyche.
tional tasks, for instance, has been explored both with Journal of Analytical Psychology, 36, 1-25.
linear correlation coefficients and nonlinear correlation Elkaim, M. (1990). If you love me, don't love me: Constructions of reality
dimensions (Basar, 1990a). Similarly, the mechanism un- and change in family therapy. New York: Basic Books.
derlying the wavelike spread of olfactory information has Emde, R. N., Gaensbaur, T. J., & Harmon, R. J. (1976). Emotional
expression in infancy: A biobehavioral study. Psychological Issues,
been studied by fitting the sums of cosine waves (a linear 10 (Monograph 37). New York: International Universities Press.
approach), whereas the pattern of olfactory EEGs has Epstein, I. R., Kustin, K., De Kepper, P., & Orban, M. (1983, March).
been modeled by a system of nonlinear differential equa- Oscillating chemical reactions. Scientific American, 248, 112-123.
tions (Freeman, 1990). Freeman, W. (1978). Spatial properties of an EEG event in the olfactory
bulb and cortex. Electroencephalography and Clinical Neurophysi-
In regard to research and design issues in general, ology, 44, 586-605.
it is important to note that this paradigm, although new Freeman, W. (1987). Simulation of chaotic EEG patterns with a dynamic
and exciting, offers no cure for the profound difficulties model of the olfactory system. Biological Cybernetics, 56, 139-150.
psychologists face in establishing reliability and validity Freeman, W. (1990). Searching for signal and noise in the chaos of brain
in all of their research. Instead, it provides a new way of waves. In S. Krasner (Ed.), The ubiquity of chaos (pp. 47-55). Wash-
ington, DC: American Association for the Advancement of Science.
thinking about psychological systems. Ultimately, its Freeman, W. (1991, February). The physiology of perception. Scientific
value to psychology will be a function of its ability to American, 264, 78-85.
solve problems and understand phenomena more effec- Gleick, J. (1987). Chaos: Making a new science. New York: Viking Pen-
tively than competing paradigms. As with all new para- guin.
digms, investigators need the latitude to be speculative at Glover, H. (1992). Emotional numbing: A possible endorphin-mediated
phenomenon associated with post-traumatic stress disorders and other
first. Following the generation of new ideas and models, allied psychopathological states. Journal of Traumatic Stress, 5, 643-
they must then subject their speculations to empirical 675.
tests. When the problems, generic principles, and research Goldbeter, A., & Segel, L. A. (1977). Unified mechanism for relay and
methodologies are all chosen carefully, however, the con- oscillation of cyclic AMP in Dicytostelium Discoideum. Proceedings
cepts of chaos, nonlinear dynamics, and self-organizing of the National Academy of Sciences, 74, 1543-1547.
systems can allow investigators to explore a variety of Gottman, J. M. (1993). The roles of conflict engagement, escalation,
and avoidance in marital interaction: A longitudinal view of five types
areas from new and promising angles, ones that many of couples. Journal of Consulting and Clinical Psychology, 61, 6-15.
may have never before considered. Goudsmit, A. L. (Ed). (1989). Self-organization in psychotherapy. New
York: Springer-Verlag.
Conclusion Guidano, V. F. (1991). The self in process. New York: Guilford Press.
Hanson, S. J., & Olson, C. R. (1990). Conneclionist modeling and brain
In this article, I have presented some of the basics of chaos, function: The developing interface. Cambridge, MA: MIT Press.
nonlinear dynamics, and self-organizing systems. I pro- Horowitz, M. J. (1986). Stress response syndromes. New York: Jason
pose that these revolutionary ideas, which are beginning Aronson.
Hoyert, M. S. (1992). Order and chaos in fixed interval schedules of
to prove productive for investigators in the physical sci- reinforcement. Journal of the Experimental Analysis of Behavior, 57,
ences, deserve more attention among psychologists. By 339-363.
applying them to specific fields of research, our under- Jantsch, E. (1980). The self-organizing universe. Elmsford, NY: Pergamon
standing of complex systems may be broadened and new Press.
ways found to view old problems. I hope that more in- Jung, C. G. (1969). The structure and dynamics of the psyche: Vol 8.
The collected works ofC G. Jung (2nd ed.). (R. F. C. Hull, Trans.).
vestigators and clinicians will take them up and use them Princeton, NJ: Princeton University Press. (Original work published
to explore systems in their own areas of interest. 1946)
Kauffman, S. A. (1993). The origins of order. New York: Oxford Uni-
REFERENCES versity Press.
Kelso, J. A. S., & Schoner, G. (1988). Self-organization of coordinative
movement patterns. Human Movement Science, 7, 27-46.
Abraham, F. D., Abraham, R. H., & Shaw, C. D. (1990). A visual in-
troduction to dynamical systems theory for psychology. Santa Cruz, Kohonen, T. (1988). Self-organization and associative memory. New
CA: Aerial Press. York: Springer-Verlag.
Bartlett, F. C. (1932). Remembering: An experimental and social study. Langs, R. (1992). Towards building psychoanalytically based mathe-
Cambridge, England: Cambridge University Press. matical models of psychotherapeutic paradigms. In R. L. Levine &
Basar, E. (1990a). Chaotic dynamics and resonance phenomena in brain H. E. Fitzgerald (Eds.), Analysis of dynamic psychological systems
function: Progress, perspectives, and thoughts. In E. Basar (Ed.), Chaos (Vol. 2, pp. 371-393). New York: Plenum Press.
in brain function (pp. 1-27). New York: Springer-Verlag. Levine, R. L., & Fitzgerald, H. E. (1992). Analysis of dynamic psycho-
Basar, E. (Ed.). (1990b). Chaos in brain function. New York: Springer- logical systems (Vols. 1, 2). New York: Plenum Press.
Verlag. Maturana, H., & Varela, F. (1980). Autopoiesis and cognition: The re-
Biitz, M. R. (1992). The fractal nature of the development of the self. alization of the living. Boston: D. Reidel
Psychological Reports, 71, 1043-1063. Maturana, H., & Varela, F. (1987). The tree of knowledge. Boston:
Callahan, J., & Sashin, J. I. (1990). Predictive models in psychoanalysis. Shambhala.
Behavioral Science, 35, 60-76. Meehl. P. (1978). Theoretical risks and tabular asterisks: Sir Karl, Sir
Carpenter, G. A., & Grossberg, S. (1987). Discovering order in chaos: Ronald, and the slow progress of soft psychology. Journal of Consulting
Stable self-organization in neural recognition codes. Annals of the and Clinical Psychology, 46, 806-834.
New York Academy of Sciences, 504, 33-51. Milton, J. G. Longtin, A., Beuter, A., Mackey, M. C , & Glass L. (1989).
Chubb, H. (1990). Looking at systems as process. Family Process, 29, Complex dynamics and bifurcations in neurology. Journal of Theo-
1669-1675. retical Biology, 138, 129-147.

January 1994 • American Psychologist 13

Minuchin, S., & Fishrnan, H. C. (1981). Family therapy techniques. Schmid, G. B. (1991). Chaos theory and schizophrenia: Elementary
Cambridge MA: Harvard University Press. aspects. Psychopathology, 24, 185-198.
Morrison, F. (1991). The art of modeling dynamic systems. New York: Schore, N. E. (1981). Chemistry and human awareness: Natural and
Wiley. scientific connections. In R. S. Valle & R. von Eckartsburg (Eds.),
Odum, H. T. (1988). Self-organization, transformity, & information. The metaphors of consciousness (pp. 437-460). New York: Plenum
Science, 242, 1132-1139. Press.
Prigogine, I., & Stengers, I. (1984). Order out of chaos: Man's new dia- Stewart, I. (1989). Does God play dice? The mathematics of chaos. Cam-
logue with nature. New York: Bantam Books. bridge, MA: Basil Blackwell.
Putnam, F. (1988). The switch process in multiple personality disorder Tufillaro, N. B., Abbott, X, & Reilly, J. (1992). An experimental approach
and other state-change disorders. Dissociation, 1, 24-32. to nonlinear dynamics and chaos. Reading, MA: Addison Wesley.
Putnam, F. (1989). Diagnosis and treatment of multiple personality dis- Turvey, M. T. (1990). Coordination. American Psychologist, 45, 938-
order. New York: Guilford Press. 953.
Vandervert, L. (1991). A measurable and testable brain-based emergent
Reidbord, S. P., & Redington, D. J. (1992). Psychophysiological processes
interactionism: An alternative to Sperry's mentalist emergent inter-
during insight oriented therapy: Further investigations into nonlinear
actionism. The Journal of Mind and Behavior, 12, 201-220.
psychodynamics. Journal of Nervous and Mental Disease, 180, 649-
Varela, F. J. (1989). Reflections on the circulation of concepts between
657. a biology of cognition and systemic family therapy. Family Therapy,
Roschke, J., & Aldenhoff, J. B. (1992). A nonlinear approach to brain 28, 15-24.
function: Deterministic chaos and sleep E E C Sleep, 15, 95-101. Waldrop, M. (1992). Complexity: The emerging science at the edge of
Sabelli, H. C , & Carlson-Sabelli, L. (1989). Biological priority and psy- order and chaos. New York: Simon & Schuster.
chological supremacy: A new integrative paradigm derived from pro- Winfree, A. T. (1974, June). Rotating chemical reactions. Scientific
cess theory. American Journal of Psychiatry, 146, 1541-1551. American, 230, 82-95.
Sabelli, H. C , Carlson-Sabelli, L., & Javaid, J. I. (1990). The thermo- Winfree, A. X, & Strogatz, S. H. (1984). Organizing centers for three
dynamics of bipolarity: A bifurcation model of bipolar illness and dimensional chemical waves. Nature, 311, 611-614.
bipolar character and its psychotherapeutic applications. Psychiatry, Wolff, P. H. (1987). The development of behavioral states and the expres-
53, 346-368. sion of emotions in early infancy. Chicago: University of Chicago Press.

14 January 1994 • American Psychologist