CHAPTER 7

The Prediction of Unpredictability: Applications

of t h e N e w Paradigm o f Chaos i n Dynamical
Systems t o t h e Old Problem of t h e Stability
of a S y s t e m o f Hostile N a t i o n s

Alvin M. Saperstein

The social sciences have long tried to emulate the procedures and results o f
the physical sciences—the laws o f economic determinism o f Marx come
immediately to m i n d . To a large extent, the science o f international relations
has not been successful in this emulation; its capability o f predicting the
outcome o f international competition (war or peace?) has been very limited.
Perhaps this is because too many variables seem to be required, intuitively, to
describe the interactions between nations. Also, nations are composed o f
multitudes o f individually complex people, making it hard to believe that the
internation variables, representing their collective behaviors, can satisfy rela­
tively simple functional relationships. A n y easily expressible theory, relating
such variables, is bound to be incomplete and thus suspect, whether the theory
is expressed verbally or mathematically.
The physical sciences owe much o f their long-term successful hold on the
public imagination and pocketbook to their long record o f successful predic­
tions o f physical events and processes, predictions that have given rise to our
powerful modern technological society. In the process, the sciences have
learned that incomplete descriptions o f the systems being studied are still very
useful. O f the many variables evidently needed to describe a physical system,
some are major, more are minor, and often the latter can be ignored while still
producing very useful predictions about the system. For example, celestial
mechanics can give a very good predictive description o f planetary motion in
our solar system without including the effects o f the motions o f the various
moons about their respective planets. The minor variables only lead to impor­
tant effects in those regions o f the system where the incomplete model,
consisting only o f the major variables, is itself unstable. Successful detailed
predictions have usually been obtained by avoiding these unstable regions,

139
140 Chaos Theory in the Social Sciences

regions i n which small changes i n the (major) variables included in the model,
mirroring the effect o f the excluded (minor) variables, lead to major changes
in the system. Such instability, in a deterministic system, is now commonly
referred to as chaos.
Physics also deals w i t h multitudes o f complex entities, such as mole­
cules. When appropriate—that is, when the interest is in the collection, for
example, a gas, rather than in its individual constituent molecules—physics
successfully uses very few variables. These are often the result o f averaging
over the constituent multitudes, satisfying relatively simple functional rela­
tionships, such as the gas laws and the fluid flow equations (e.g., Kinetic
Theory, Fluid Physics; see Lerner and Trigg 1991). These laws are necessarily
incomplete and hence sometimes incapable o f leading to detailed predictions.
But we have learned how to predict their unpredictability and have found that
this incomplete knowledge is still useful as the model is extended to include
further variables, that is, to more completeness.
Following the lead o f physics, it w i l l be assumed in this chapter that
international relations may be productively modeled by relatively few vari­
ables, simply related; the choice o f variables depends upon the system o f
interest and the questions to be answered. It w i l l further be assumed that the
breakdown o f these incomplete models into chaos may be more useful than
the models themselves, that the qualitative prediction o f model instability may
be more meaningful than the quantitative predictions o f specific system varia­
tion.
Given these assumptions, a procedure is developed for exploring a num­
ber o f questions that are o f major concern to the understanding o f international
relations and the practical application o f this understanding to the creation o f
international security policy. The procedure is the creation o f simple mathe­
matical models for the interaction o f competing states and the examination
of the numerical output o f these models for regions o f stability and chaos
(Saperstein 1990). The general questions to be explored in this chapter are:
" W h i c h is more war-prone—a bipolar or a tripolar world?"; "Are democracies
more or less prone to war than autocracies?"; and " W h i c h is more war-
prone—a system o f shifting alliances or a collection o f go-it-alone states?"
(Further questions suitable for exploration with this procedure should become
apparent to the reader.) For example, it w i l l be shown that the region o f
instability in a tripolar w o r l d is larger than that o f the corresponding bipolar
w o r l d . From this it w i l l be concluded that a tripolar w o r l d is more prone to
war than a bipolar one.
Before applying these assumptions to the development o f models and an
analytical technique for getting useful information from these models, I w i l l
first try to justify them by attempting to formulate a general definition o f
science and prediction and then draw from this requirements for the more
 The Prediction of Unpredictability 141

specific science o f international relations. A n important part o f this definition

w i l l be the relation between science and technology, a relation that has been
fundamental to the growth o f the natural sciences and their related technolo­
gies. The technology corresponding to the science o f international relations
should be the making o f successful international security policy. Following
the general discussion o f prediction and nonpredictability, a tool for recogniz­
ing chaotic regions is developed. The chapter concludes with the development
of a simple model for each question and the application o f the chaos tool to the
answering o f that question.

Science ( U n d e r s t a n d i n g ) and Technology (Control)

Many students in introductory courses in the sciences have great difficulty

coming to grips w i t h the notion o f science as a system o f understanding, not a
compilation o f facts or ideas about the system to be understood. Exacerbating
the problem is the difficulty o f compactly setting down what is universally
meant by "understanding" in the physical or social sciences. B y "understand­
i n g , " I mean the development o f a relationship between the phenomenon to be
understood and a previous set o f concepts already understood (even i f only
innately). These prior concepts, axioms for this latest level o f understanding,
are similarly related, in turn, to previously understood axioms, which, in turn,
are related to still more primitive understandings. Thus, science is a tower o f
relationships constantly growing at the top and the bottom.
By "relationship" between two levels o f concepts or phenomena, I mean
that one level can be derived from (caused by) the other level acting as
axioms. " D e r i v e d , " in turn, means that the one follows logically from the
other, arising by necessity from the structure o f the language being used; there
is a "functional relationship" between the t w o levels. For example, i f the prior
level consists o f the two statements: " a l l men are mortal" and "Socrates is a
man," then the derived statement is "Socrates is a mortal." More formally, i f
the system in question can be described by a variable (or set o f variables), x,
and i f x and x
n represent t w o neighboring levels, then we might be able to
n + l

write

*n+\ =/(*„. A
)> (!)

where / designates the functional relation that may depend upon a set o f
parameters A. I n physical science this might be the relationship between the
present configuration o f planets in the solar system and a previous configura­
tion. I n biology, it might be the relation between the behavior o f a cell in a salt
solution to that o f the same or similar cell in plain water. I f there is a time
sequence between the two l e v e l s — i f one occurs at a later real time than the
142 Chaos Theory in the Social Sciences

other—then the implied later level is determined by the former. The resulting
system o f understanding—henceforth also referred to as "the theory"—is
then a deterministic system. ( A n illustration o f a deterministic system w o u l d
be the sequence o f positions along the orbit o f an artillery shell in the absence
of air resistance. A nondeterministic system would be the sequence o f posi­
tions o f an electron in an atom, though the probabilities o f these positions are
well determined by quantum physics.) For example, suppose the theory deter­
mines the value o f some variable, x, at all future times, t, given the starting
value o f the variable, x (this could be the result o f iterating the relationship o f
0

equation 1 for all positive values o f n starting w i t h n = 0 ) ; the result o f the

theory is the curve x(t) shown (see fig. 7 . 1 , part a); x{t) is determined by JC 0

and the theory.

A check on the adequacy o f the understanding system (the theory) is
provided by observation o f the system being understood. Are the axioms
always observed to be accompanied by the derived concepts or phenomena
(logical equivalence)? Or are they always followed by the next level? In this
case, the consequences are predicted by the axioms. Such an empirical check
requires confirmation o f the validity o f the axioms in the system under consid­
eration, as w e l l as the observation o f their outcome. Hence, the prior l e v e l —
the initial state o f the system to be understood—must also be carefully
observed. Actual measurements o f the real world—either physical or social—
are always accompanied by noise, random fluctuations in the values that
would be obtained in a series o f identical measurements upon identical sys­
tems (e.g., "error analysis" in Lerner and Trigg 1991). Hence there is always
some uncertainty (also commonly called experimental error) i n the observed
system parameters or starting values, and hence some lack o f certainty in the
adequacy o f the system o f understanding. The result, therefore, is a range o f
possible starting values, 8x , and a consequent range in determined outcomes,
0

8x(t) (fig. 7 . 1 , part b).

The phenomena being investigated may not refer to definite events but to
the probability o f occurrence o f the events. For example, we may be investi­
gating the probability o f survival o f a bacterium as an antibiotic is added to its
growth medium. The events themselves may be random, nondeterministic,
even though their probability is determined by the relations o f the theory. I n
this case, single observations o f the system, no matter how precise, are not
sufficient. Repeated observations on an ensemble o f identical systems (theo­
retically an infinite number o f identical observations) are required to deter­
mine a probability (see, e.g., Feller 1950). For example, a single toss o f a
coin w i l l not tell you the probability o f a head for that coin. N o r w i l l the
observation o f a single human life span give any actuarial probability informa­
tion.
Thus a scientific theory relates some initial state (definite or a probability,
 The Prediction of Unpredictability 143

Fig. 7 . 1 . T i m e e v o l u t i o n o f m o d e l s o l u t i o n s under v a r y i n g d y n a m i c s :
(a) s i n g l e s o l u t i o n o f m a t h e m a t i c a l m o d e l ( d e t e r m i n i s m ) ; (b) f l o w o f
s o l u t i o n s ( p r e d i c t a b i l i t y ) ; (c) f l o w o f s o l u t i o n s (chaos).

but observed somehow) o f the system to be understood to some other state(s)

o f the same system. Since the initial state is uncertain, to the extent o f the
observational noise encountered in empirically determining i t , knowledge o f
the related states may also be uncertain. I n linear theories, small changes in
the parameters describing the initial state lead to correspondingly small
changes in the related states. For example, i f the theory is described by
equation 1, then the small changes are similarly related:

x n + l + 8x„ +l =f(x„ + 8x„, A ) . (2)

144 Chaos Theory in the Social Sciences

Hence, given the theory, the parameters o f the determined states are as
w e l l k n o w n as those o f the initially given states (see fig. 7 . 1 , part b). It
follows that such a deterministic theory is also a predictive theory. The future
state o f the system is as well known as its present initial state. The ability o f
the physical sciences to make predictions, subsequently verified, has given
them a strong hold on the human imagination. A popularly observed eclipse
bolsters the public support o f astronomy, as well as o f all the other sciences,
presumably applying the same methods. (Naturally enough, other intellectual
and pseudointellectual disciplines thus strive to at least appear to be using the
same scientific approaches.)
In nonlinear theories, differences between initial states may not evolve
into allowed differences o f final states; that is, equation 2 is not valid. Hence,
small initial uncertainties about the system may lead to large final uncertain­
ties; little or nothing may be k n o w n about the outcome, no matter how precise
the knowledge o f the initial state. For example, i n figure 7 . 1 , part c the range
in outcomes determined by the theory is equal to the full range o f values
possible to the system; prediction may be impossible. I f a small change i n the
initial configuration o f a deterministic theory leads to large indeterminate
changes in the output configurations, the theory is said to be chaotic (Schuster
1988); it precludes precise prediction. A given theory may be chaotic for
some ranges o f input or system parameters and nonchaotic and predictive for
other possible conditions.
For example, the theory obtained by applying the usual Newtonian dy­
namical laws to the motion o f physical fluids is often referred to as the Navier-
Stokes equations. Given the initial state of, say, a body o f water—the velocity
of every particle o f the water at some given starting time—the solutions o f the
equations w i l l give the velocity o f each o f these particles at every subsequent
time. O f course, noise and measurement errors are associated with the deter­
mination o f the initial and system parameters (e.g., the configurations o f the
rocks i n the creek bed over which the water flows; the random fluctuations o f
the molecules making up the fluid around their mean, fluid, motion). Hence,
there w i l l be fluctuations in the results produced by the equations. I f the
output fluctuations are small, the motion o f the fluid is well predicted: k n o w l ­
edge o f the behavior o f the water at one place, at one time, gives knowledge
of the fluid flow at neighboring places at subsequent times. Such a nonchaotic
flow is termed laminar flow and is characteristic o f liquids flowing slowly in
smoothly varying channels. The same theory w i l l describe the motion o f the
same water flowing rapidly in a boulder-strewn river, a flow that we usually
describe as turbulent (see, e.g., Lerner and Trigg 1991). I n this chaotic
regime, knowledge o f what the water is doing at one time and place provides
almost no insight into what it w i l l be doing later at the same and neighboring
positions.
 The Prediction of Unpredictability 145

The difference between the laminar and turbulent regimes o f the theory is
determined by the value o f a parameter, the Reynolds number, calculated
from the fluid speed and the size o f the obstructions to the fluid flow. For
values o f the Reynolds number less than a critical value, determined by the
theory, the flow is laminar; exceeding the critical Reynolds number implies
turbulence. The turbulent regime can further be divided into regimes of soft
chaos and hard chaos. I n the soft regime, the output fluctuations, though
large w i t h respect to the input parameter fluctuations, are small compared to
the extent o f the system. For example, there may be turbulent ripples on the
surface o f a body o f water even though the motion o f the bulk o f the fluid is
well predicted. I n hard chaos the fluctuations dominate the entire system, as
does the flow uncertainty i n a fully turbulent, chaotically tumbling and s w i r l ­
ing river rapids.
Although the flow o f water in a turbulent stream may not be predictable,
the onset o f turbulence itself may be predicted. The transition from laminar
flow to turbulent flow depends upon the Reynolds number, a parameter that
changes as the stream is followed. K n o w i n g how it changes, from a k n o w l ­
edge o f the configuration o f the stream bed and its sources, allows a prediction
as to where and when the stream w i l l change from laminar to turbulent flow
and vice versa. Similarly, the transition to and from turbulent air flow over an
aircraft's wings can be predicted as the shape o f the aircraft and its aerial
maneuvers are varied. Or, since a large storm may be construed as a turbulent
part o f the earth's atmosphere, the possibility o f such a storm may be pre­
dicted even i f the air flow w i t h i n such a storm remains chaotic and unpredict­
able. Hence, a theory o f a system that manifests considerable chaotic, and
therefore unknowable, regimes, but that allows the prediction o f where and
when these chaotic regimes w i l l start, still represents considerable under­
standing o f that system. Such theories may not have the presumably a l l -
powerful predictability o f celestial mechanics, which has caught the public
eye. But all natural systems may not be predictable (as we have learned in this
century), and one o f the characteristics o f successful science is to do the best
you can. The study o f such unknowable systems has become very popular
recently in the physical sciences (see, e.g., chaos in Lerner and Trigg 1991).
The possibility o f prediction implies the possibility o f deliberate control.
Control o f a process means that known intended consequences w i l l follow
from deliberate acts: prior conditions are modified to ensure desired outcomes
or to avoid undesired outcomes. For example, i f the outcome x„ +l in equation
1 is not desired, either the prior condition, x„, or the system parameters, A,
can be changed appropriately. ( I f a car is going in a dangerous direction, the
parameter A—the steering wheel position—can be changed or the system can
be started w i t h the car on a different highway.) I f prediction is not possible,
there is no way o f k n o w i n g the outcome o f a given act or policy, which is
146 Chaos Theory in the Social Sciences

synonymous w i t h saying control doesn't exist. Hence, technology, which is

the controlled manipulation o f some part o f the human environment, requires
predictability. A successful technology, one that produces anticipated ends in
a system from given input policies and means, is thus a confirmation (though
not a proof!) o f the understanding o f that system. Conversely, an unsuccessful
technology, one i n which anticipation is thwarted, is a definitive proof o f the
failure o f the underlying theoretical understanding. The failure o f a bridge or
an agricultural policy is an indication that the appropriate understanding o f the
behavior o f physical materials and environments or o f farmers is lacking. The
success o f the bridge or policy, however, does not guarantee the correctness o f
the theoretical understanding. (There may be alternative theories to the one
that maintains that guardian angels hold up the bridge and protect the crops.)
It follows that there is a symbiotic relationship between the physical and
biological sciences and engineering and medical technologies, a relationship
that has existed ever since these sciences have existed. Scientific advances
have led to new technologies that, in turn, "proved" the correctness o f the
scientific understanding and provided the tools for further measurements and
theory. This symbiotic relationship has led to the success o f both natural
science and technology and has served as a model for all other sciences.

The Science o f I n t e r n a t i o n a l Relations

Turning now to international relations, whose understanding is part o f the

political and social sciences, the corresponding technology is international
policy m a k i n g , w i t h its subset, international security policies. I f we truly
understand the international system, we should be able to make successful
policy, that is, policy whose desired outcome, given the existing system and
starting configurations, actually occurs (see, e.g., Saperstein 1990). I n the
past, a successful international security policy meant that a nation obtained
what it wanted, either without having to resort to war or by winning its wars
without unreasonable costs to itself. That is, the international system, o f
which the nation is a part, either doesn't change to the detriment o f the nation
or, at the very least, the component o f the system encompassing the nation
thrives even i f the rest o f the system is fundamentally changed. In the present
nuclear age, a war between major nuclear powers is unlikely to leave either
the w o r l d system or any o f its subcomponents intact. Hence, a major goal o f
any international policy has to be the avoidance o f such a war. Thus, the
symbiotic science must be able to predict the outbreak o f war.
I f that science could also predict the outcome o f a war, there would be no
need for the war. Nations often strive for overwhelming military strength in
the hope o f ensuring (predicting) their success in the event o f war. They have
often been unsuccessful in this prediction because o f the large uncertainties in
 The Prediction of Unpredictability 147

the reactions o f their potential opponents. Considering war as a chaotic pro­

cess (see, e.g., Clausewitz 1982), like a turbulent flow over boulders, it may
be possible to predict its outbreak but not its outcome—that is, which side o f
the boulder gets a given portion o f churning water or which side wins the war.
A science o f international relations that only dealt w i t h probabilities
could lead to no policy technology and hence would be useless. Testing o f
probabilistic arguments requires an ensemble o f identical systems, whereas
we only have one such system—our real w o r l d . Thus, a scientific conclusion
that there is a likelihood o f nuclear war w i l l be treated as a prediction o f
certainty; the corresponding policy w i l l almost certainly be avoided. I n such a
situation, probabilistic predictions such as " l i k e l y " or "unlikely" (where is the
separation?) w i l l be treated by policymakers as "yes" or " n o , " which are the
results o f a definitive or deterministic science. Hence, we seek a science o f
international security with at least some deterministic aspects.
Can we postulate a system o f competing, hostile states to be determinis­
tic? Experience indicates that nations respond to one another's actions in
reasonably determined ways—at least for major actions and responses. I f
these patterns o f action and response didn't exist, there would be no political
science, and perhaps no history, o f international relations—just readings o f
tea leaves or goat's entrails! But why should we expect deterministic behavior
o f a collection o f nations, given that each one is composed o f many, many
individuals, each o f w h o m often acts in an unpredictable, chaotic manner? A n
answer—that deterministic behavior o f entities exists precisely because o f the
random nature o f large numbers o f their subentities—is obtained by analogy
to the physicist's description o f a gas. Such a gas is a collection o f an
enormous number o f randomly moving molecules whose precise description
requires a similarly enormous number o f stochastic variables. Yet its overall
physical behavior is deterministically described by a few gas variables, such
as pressure and temperature, each a complex average over the many molecular
variables. O n a gross scale o f observation these few gas variables usually
change smoothly, though on a very fine scale they exhibit random minor
fluctuations, mirroring the underlying random molecular variables. So we
might expect the behavior o f modern nations to be governed by a relatively
few deterministic variables, each a complex average over the behaviors o f
each o f its multitudinous population. The usual goal is to find rules governing
these variables, admitting that there w i l l be occasional minor fluctuations—a
love affair or assassination at the higher levels o f government or society—not
subject to these rules. M y additional goal is to see how a system, evolving
according to such rules, responds to such fluctuations.

The core o f international relations can be described as follows: Nation A

responds to the actions o f nation B in the context o f the other nations o f the
w o r l d system and its o w n present and past internal situations. Nation B then
148 Chaos Theory in the Social Sciences

responds to nation A's response in a similar context. Each response is deter­

mined by the system's previous responses (Richardson 1960a,b). Hence, the
situation is recursively deterministic. I f A and B designate the appropriate sets
o f variables describing the two nations, the situation might be represented in a
manner similar to that o f equation 1 :

n+l = a(A„, B „ , A)
(3)
B, = b(A , B„, A),
n

where a and b are functional relationships representing the responses o f the

two nations; the parameters À may also represent some average over the other
nations in the w o r l d system. For example, A„ might represent the hostility (or
the exports) o f A to B in year n, w i t h B similarly defined. Then equation 3
n

describes how next year's hostility (or exports) o f A to B w i l l be determined

by this year's hostility (or exports) o f A to B and o f B to A . The role o f the
political scientists (and the need o f the policymakers) is to attempt to predict
the outcome o f the intertwining chain o f responses on the basis o f their
understanding o f the system and its history. Usually this understanding is
verbal, expressed in the words o f historical and policy treatises. ( I n equation
3, the relationships a and b and the parameters À would be verbally ex­
pressed.)
Verbal understanding possesses the vividness, vagueness, redundancy,
and contradictions o f ordinary, daily, spoken and written language. It usually
contains many implicit assumptions and innate biases. The logic taking it from
premise to conclusions is often incomplete and/or fuzzy. It is often difficult to
know what the premises are and how well they are verified, what the conclu­
sions are, and how we got from one to the other (e.g., the functions a and b in
equation 3 are not explicit). Hence, how do you test the model against reality?
In contrast, mathematical formalisms tend to be explicit in their premises,
transparent i n their logic, and concrete in their conclusions. There is thus an
often-expressed desire to describe all aspects o f the macroscopic interactive
behavior o f nations via explicit mathematical recursion relations between
numerical variables representing those behaviors. Equation 3 becomes a set o f
mathematical relations between numerically defined variables: A, B, A, and so
on. I f the resultant equations could be solved, given all the necessary input
representing system parameters and starting conditions, they would produce
concrete predictions that could be tested against observed behavior to confirm
or negate the mathematical model.
However, given the complexity o f the real-world system, we are unlikely
to ever have all the necessary input information, or sufficient understanding,
to create a model system o f equations complete enough to really represent the
w o r l d in all o f its manifestations. A n y practical model would have fewer
 The Prediction of Unpredictability 149

variables than are really required. The resultant equations, though far short
of the complexity o f the w o r l d , w o u l d still be complicated and nonlinear i f
they are to represent some o f the essential attributes o f our strongly interacting
real w o r l d , ruling out the possibility o f simple, explicit solutions (see, e.g.,
Prigogine and Stenger 1984). These obvious shortcomings o f any mathemati­
cal model are no different than the difficulties o f any reasonable verbal under­
standing o f the w o r l d . The difference is that the incompleteness o f the mathe­
matical model is usually immediately obvious, whereas that o f verbal models
is concealed in their fuzzy verbosity. In both cases, the resultant predictions
are necessarily incomplete, making theoretical testing and practical policy
formulation difficult.
The more relevant factors left out o f a model or incompletely described
w i t h i n the model, the more gross and incomplete its consequent predictions
w i l l be. Overall system variations may be correctly predicted though the
detailed outcomes may not be trustworthy. Such incomplete predictions may
or may not be useful, depending upon the circumstance o f the policy making
or theory building. It becomes important to determine what kind o f gross
predictions, from necessarily incomplete theoretical models, can be useful in
a given arena o f human experience. I t is the contention o f this chapter that the
prediction o f unpredictability can be very useful i n attempts to understand and
control the advent o f war in the w o r l d system o f nations (Saperstein 1984).

The Prediction o f Chaos a n d t h e O u t b r e a k of War

Crisis instability in the international system usually implies a configuration i n

which small insults can lead to major changes—the loss o f a nail leads to the
loss o f a shoe . . . to the loss o f a kingdom; an assassination o f a minor duke
can lead to the deaths o f millions and the profound transformation o f the
w o r l d system. Such instability represents the loss o f control and the great
potential o f war. This parallels the definition o f chaos (Schuster 1988): small
disturbances o f a deterministic mathematical system lead to dispropor­
tionately large changes in the system and the consequent loss o f control.
Prediction o f unpredictability i n a system is a prediction o f the onset o f
chaos—soft or hard. The range between soft and hard chaos among nations is
the range between minor and major loss o f control i n international relations. I t
is the range between the possibility o f loss o f an unspecified soldier i n a given
battle, the w i n n i n g or losing o f that battle, o f the war, and o f the existent
w o r l d system. We postulate that the presence o f hard chaos, in a theoretical
system modeling the international system o f competing nations, is a represen­
tation o f major crisis instability and o f the extreme likelihood o f the outbreak
of major war in the real w o r l d being modeled. I f the prediction o f hard chaos
is believable, then the international security policies associated with that part
150 Chaos Theory in the Social Sciences

o f the model leading to the onset o f chaos should be avoided. A useful

analogy is to the testing o f a new aircraft to determine its behavior under a
wide variety o f desirable and undesirable circumstances. I f the theoretical
model (either purely mathematical or based upon w i n d tunnel modeling)
indicates that certain maneuvers are likely to lead to loss o f control and
possible loss o f plane and crew, the pilots w i l l be instructed to avoid those
maneuvers—even i f they w o u l d lead to otherwise desirable results. Even i f
the model is not sufficiently complete to indicate how the aircraft w i l l behave
in the danger zone but is believable in predicting the existence o f the loss-of-
control zone, that w o u l d be sufficient to mandate changing policy so as to stay
away from that zone. Thus, i f the prediction o f unpredictability i n an inter­
national system is believable, the ability to make such predictions, even i f
incomplete i n details, can be used to answer political science questions as w e l l
as to help determine practical national security policy.
Thus we must face the question: is the prediction o f unpredictability
believable in an incomplete model? I f the onset o f chaos portends the outbreak
of war, can we believe a model's prediction o f war even though the extent and
outcome o f the war are not predicted and may not even be describable by the
model? Is the prediction o f instability (or stability) o f a model itself stable i n
the face o f expansions o f the model to include more relevant aspects o f the
system? W i l l a model's prediction o f chaos be softened or disappear as more
variables and/or more complicated interactions are added to the model?
A formal answer to the above question requires a working knowledge o f
the complete model, knowledge that doesn't exist. ( I f it d i d , the question
would be irrelevant!) In its place, we must rely on analogy with other
systems—usually from mathematics or physics—where complete models,
and their incomplete component models, exist and are used. Mathematical
experience w i t h dynamical systems indicates that "chaos first appears in the
neighborhood o f non-linear resonances" (Reichl 1992, 14), where new vari­
ables first make their impact upon the system. There is no experience o f such
chaotic regions disappearing as new variables are introduced. Empirical expe­
rience with fluids indicates that chaos (turbulence) appears earlier and stron­
ger when new variables, such as temperature differences and heat flows,
become important in the system. Theoretical experience with specific mathe­
matical models o f real phenomena (the Navier-Stokes equation for fluid flows;
the recursive equations' modeling o f the evolution o f tripolar systems from
bipolar ones [Saperstein 1991], developed later i n this chapter) suggests that
the regions o f stability (areas o f absence o f chaos) decrease in extent when
additional variables come into play.
We thus presume that qualitative (gross) predictions o f the loss o f model
stability are much more believable (more stable) than are the quantitative
(detailed) predictions from the same model. Hence, the prediction o f hard
 The Prediction of Unpredictability 151

chaos in a model reasonably representing the international system o f compet­

ing nations is a fair warning to policymakers: embarkation upon the modeled
course o f action is to be done with extreme dread and care. However, a
contrary prediction o f model stability is not an absolute assurance o f system
stability; complacent acceptance o f the corresponding policy is not warranted
since a more faithful, more complete model o f the system in which the policy
is being applied may very well allow chaotic breakdown. Similarly, when
exploring theoretical questions about the international system, the appearance
o f chaos in an appropriate model portends the breakdown o f stability in the
system; however, stability o f the model does not necessarily imply a corre­
sponding stability i n the system.
The stability o f a mathematical system can be tested directly i f sufficient
computing power is available. For example, a large number o f possible differ­
ent but neighboring initial configurations o f the system o f interest may be
specified. Each one o f these starting configurations gives rise, via the mathe­
matical model, to a final configuration (the system at some specified future
time). I f these predicted final states are as similar to each other as the initial
states, the system is stable. I f they are w i l d l y divergent, as compared to the
starting values, the system is chaotic. As an illustration, the S D I system
proposed by President Reagan can be described by variables representing the
number o f offensive missiles available to the opponents, the number o f defen­
sive antimissile weapons, and the number o f anti-anti-missile weapons (also
offensive) (Saperstein and Mayer-Kress 1988). The model consists o f rela­
tions between these variables specifying how each o f the opponent parties
procures weapons in response to the analogous procurements o f the other
nations in the system—a typical arms race. The initial configuration is spe­
cified by the initial numbers o f the different weapons stocks and some numeri­
cal estimates o f their capabilities: neighboring starting configurations imply
slightly different values for these initial numbers. (In the real w o r l d , these
numbers and parameters would not be precisely known to any o f the compet­
ing parties.) The computer then tracks the evolution with time o f the system
emanating from each starting configuration and displays the variations in
the possible outcomes—the numbers o f the different types o f weapons i n the
stocks o f each o f the opponents. I f the final variations are comparable to the
corresponding initial variations, the system is stable. I f the final variations are
large compared to the initial ones, but still small enough that system d o m i ­
nance (superior and inferior weapons stocks) can be distinguished, the system
displays soft chaos. I f the final outcomes vary so widely that they encompass
the entire variation possible to the system and obscure all differences between
the variables corresponding to the different nations, then no prediction o f the
outcome o f the arms race is possible—we have hard chaos.

A more modest approach, suitable for systems with fewer variables, is

152 Chaos Theory in the Social Sciences

the calculation o f Lyapunov coefficients for the model in question. These

coefficients are direct measures o f the rate at which initially neighboring
configurations drift apart as the model system evolves. Each possible starting
point traces out a path toward the future by means o f the model mechanism, as
is illustrated in figure 7 . 1 . I f the paths from closely neighboring starting points
remain close (fig. 7.1b), prediction is possible; i f the paths separate exponen­
tially (fig. 7.1c), so that the final outcomes cover all possibilities allowed in
the system, everything is possible in the future, meaning nothing is k n o w n
about the future, and prediction is impossible—the situation is chaotic. The
Lyapunov coefficient (see, e.g., Schuster 1988) is a useful measure o f such
path separation that is easily calculated given a computer able to follow the
evolution o f neighboring paths. Let X and XQ be any two possible neighbor­
0

ing starting configurations, separated by a small distance 8 . I n the course o f

0

time n these evolve respectively (e.g., via equation 1) to the t w o configura­

tions X and X ' , separated by the distance 8„, where
N N

= 8 0 *"«*o>. (4)

This defines the Lyapunov coefficient Ç(X ), which depends upon the starting
0

configuration as w e l l as the dynamics o f the system. Since the interest is in the

long-time evolution o f the system, it follows that

limit limit 1
log x: (5)
S&o) =
°° fir, 0 n
I f £ < 0, two configurations starting close to each other w i l l remain close to
each other. Thus, predictability is possible (fig. 7.1b), so such a system w i l l
be defined as stable. I f £ > 0, the initially close configurations w i l l drift ever
further apart, making prediction impossible. Thus £ > 0 is the signature o f
chaos or instability.
In this chapter, several aspects o f international arms competitions, nor­
mally modeled qualitatively and verbally, w i l l be put into equivalent algebraic
forms. Only algebra is used! The variable expressing the arms level or pro­
curement level o f one country in one year is expressed in terms o f similar
variables describing the status o f that country and o f its competitor nations in
the prior year (e.g., equation 3). These expressions only involve a small
number o f additions, subtractions, multiplications, and divisions. Hence, they
can be put into standard spreadsheets on a desktop computer and iterated from
year to year to generate the time evolution o f the armaments levels o f the
competing nations. Thus, the future o f these levels, and o f the arms competi­
tion among these states, is determined. Similarly, the corresponding
Lyapunov coefficients, describing how the uncertainty associated w i t h these
levels, given that there are initial uncertainties, evolves with time, are also
 The Prediction of Unpredictability 153

computed. O f course, one doesn't reach limits o f infinity or zero, as is

required by the definition in equation 5, on a desktop computer. Instead, the
iterations are run up to large values o f n ( i . e . , 100) for varying small values
of 8 .0

The prediction by a mathematical model is meaningful only i f the results

are stable, that is, i f small changes in the input and/or model parameters lead
to corresponding small changes in the predicted time evolutions. Otherwise,
the model's output is chaotic and indicative that the system being modeled is
crisis-unstable and likely to undergo a phase transition from a competitive,
but nonshooting, cold war to a hot war. Since the models are necessarily
incomplete, their detailed predictions are not really believable, even i f mean­
ingful within the context o f the model itself. But, as we have presumed
before, the prediction o f chaos is itself meaningful and believable. Using the
desktop computer, the models are explored over different ranges o f parame­
ters, or different algebraic forms, to see which regions lead to stable solutions,
which to chaos. Since different parameter ranges or algebraic forms represent
different policy choices, we are thus able to predict which policies w i l l be
dangerous, in that they may lead to crisis-unstable situations.
We have briefly described one approach that can lead to specific policy
choices (e.g., S . D . I , should not be developed i f it w i l l lead to a decrease i n
the range o f stability o f the international offensive nuclear warhead missile
system). I n the remainder o f this chapter the chaos approach w i l l be applied to
three political theory questions: Are bipolar international systems more or less
stable than corresponding tripolar systems? Is a system o f democratic nations
more or less stable than corresponding systems o f autocratic states? Is a
system o f nations that strives for a balance o f power via shifting coalitions o f
states more or less stable than one in which each nation individually seeks to
balance the power o f the others?
There is a great body o f literature exploring the answers to these ques­
tions via qualitative verbal analysis, and no claims are made for new or
surprising answers. However, since all o f these theoretical approaches, quali­
tative and quantitative, are incomplete and hence not completely convincing,
a juxtaposition o f different approaches that lead to similar answers should add
to our understanding o f these questions and their applicability. A disagree­
ment between the results o f the different approaches should be a signal that
understanding is still lacking.

W h i c h Is M o r e W a r - P r o n e — A Bipolar
or a Tripolar W o r l d ?

Is a bipolar w o r l d more or less stable than a tripolar world? As we leave a

world configuration dangerously dominated by two nuclear superpowers for a
w o r l d perhaps dominated by none, one, or many, it is important to gain
154 Chaos Theory in the Social Sciences

insight into whether w e must be even more cautious or can be a little less
t i m i d in addressing the many nonmilitary security problems o f a twenty-first-
century w o r l d . One approach to this question is to compare the range o f stability
o f a two-power system w i t h that o f a corresponding three-power w o r l d .
The variable chosen to describe national behavior in the interactive
model o f equation 3 is the devotion o f a given nation to war preparation—the
ratio o f arms procurement ( i n the most general sense) i n a given year to
the gross national product o f that nation i n the same year. O f necessity, the
numerical value o f this variable must lie between zero and one. The bipolar
w o r l d is then heuristically modeled by the simple nonlinear Richardson-type
model represented by the coupled equations below

X
n+l = 4ar„(l - Y ) n (6)
Y n+l = 4bX (\ - X„).
n

X is the devotion o f nation X to war in year n. The procurement o f arms

n

by X in the year n + 1 is in response to, and assumed to be proportional to, the

previous year's devotion o f its opponent, nation Y. The nonlinear term (1 —
Y„) expresses the assumption that i f the opponent's resources are stretched to
the breaking point ( i . e . , his previous year's procurements were so great that
they almost preempted his entire G N P , making i t impossible for h i m to
procure any more), there is no need to stretch any further this year, so that this
year's procurements may be correspondingly diminished. This model is so
simple that its region o f stability may be analytically computed (Saperstein
1984) as a function o f the model's proportionality parameters, a and b, which
must also lie between zero and one so that the X , Y„ remain between these
n

bounds. The resultant curve represents the critical relation between a and b;
the region above the curve, in which the two Lyapunov coefficients, £(X ) and 0

£(Y ), each defined as in equation 5, are positive, is the model's chaotic

0

region.
In equation 7 the model is extended to three nations (when e = 0,
equation 7 reduces to the previous two-nation model, equation 6):

X n+1 = 4aY„(l - Y ) + 4 e Z „ ( l - Z„)

n

Y n+l = 4bX„(l - X„) + 4eCZ„(l - Z„) (7)

Z„ + 1 = 4e[X„(l - X„) + CY„(l - /„)].

As e increases from zero, the coupling between X and Y and the third
nation, Z , becomes increasingly significant. Numerical computations o f the
three Lyapunov coefficients pertinent to this model, £(X ), £(Y ), and £(Z ),
0 0 0

made using simple spreadsheets on a small desktop computer (Saperstein

 The Prediction of Unpredictability 155

1991), indicate that the stability region decreases in area as e increases, that
is, as the third nation becomes more significant in the w o r l d system. Hence,
the model suggests, in accord w i t h a great deal o f other scholarly evidence,
that a tripolar w o r l d is more dangerous than a bipolar one.

A r e Democracies M o r e or Less Prone t o War?

Given the propensity o f many o f the world's major democracies to aid and
abet less-than-democratic governments elsewhere, it would be useful to know
i f such policies hinder or further the w o r l d peace that these democracies
presumably seek and guard. Required first is a definition o f democracy that is
easy to quantify and fit into a model o f competitive arms procurement, such as
that o f equation 6, suitably generalized.
It seems easiest to define democracy in terms o f diffuseness o f decision
making. The larger the fraction o f the population that has significant input into
matters o f peace and war, the more democratic the nation w i l l be considered
to be. W i t h this definition, the numerical difference between democracy and
autocracy w i l l be based upon population-sampling theory. I assume that every
citizen o f nation 1 has some intrinsic feeling for the totality o f its opponent
nation 2, and conversely, ranging from complete confidence to extreme fear
and loathing. Let a, be the "fear and loathing" coefficient for any one citizen
of nation 1. Then the policy o f nation 1 toward its opponent w i l l be deter­
mined by (a,), the mean o f a, over the class o f citizens o f 1 who are significant
decision makers; (a ) is similarly defined. The fraction o f the citizens belong­
2

ing to this class w i l l be small in an autocracy, large in a democracy.

Assume that there is a natural spread o f values o f the fear and loathing
coefficient whose distribution is characteristic o f humanity, not nationality.
Each large nation w i l l contain many people manifesting small values o f the
coefficient, intermediate values, and so on. Thus the mean, a, o f this coeffi­
cient, taken over an entire large population, is a human characteristic, inde­
pendent o f nation: d , = a = a. (Note that the known examples o f extreme
2

interethnic hostility usually occur in comparatively small groups.)

The decision makers o f a nation are a sample o f the entire nation. I f it is a
very large sample, means over this sample w i l l be close in value to means
over the entire population. Hence, for large democracies, (a) = d. Means over
small samples may vary significantly from the means over the entire popula­
tion. Thus, given a random set o f autocratic nations, there w i l l be some for
which (a) greatly exceeds a and some for which (a) is much less than d . I n a
corresponding set o f democratic nations, all w i l l have intermediate values o f
{a) = d. Therefore, i f is the effective fear and loathing coefficient o f nation
X for nation Y , i f a l l nations i n the set containing X and Y are large democra­
cies, = d , whereas i f the set contains autocracies, can range from very
small values to very large values.
156 Chaos Theory in the Social Sciences

The effective fear and loathing coefficient is taken to be the propor­

tionality constant i n a nonlinear Richardson-type arms procurement recursion
relation. I n a competitive situation between nations X and Y, the annual
procurement o f arms by X , A X , w i l l be proportional to the amount by which
X's arms stocks were exceeded by Y in the previous year: Y — X . (The n n

proportionality constant w i l l be taken to be a^; the similar coefficient for Y's

annual arms buildup is a .) Procurement w i l l also increase proportionally to
yx

the total strength o f Y : Y„; i f Y is very weak, i t w o n ' t matter i f Y is stronger

than X ; the more powerful Y is, the more damaging any discrepancy in power
w i l l be i n any potential conflict. However, arms buildups cannot continue
indefinitely. There w i l l be economic constraints; nations can't procure more
than the total economy w i l l allow. I f C is the m a x i m u m annual expenditure x

for arms possible by X , assume a smoothed economic cutoff function

0 ( 1 - AX/C ) X = (1 - AX/C ) X 0 ( 1 - AX/CJ, (8)

where 0 is the unit step function (0(TJ) = 1 for 17 > 1, 0 = 0 otherwise).

I f X ' s arms stocks exceed those o f Y , then X can decrease his arma­
ments. Assume that the arms build-down in any one year (AX negative) w i l l
be proportional to the amount by which X exceeds Y ' s strength in the previous
year and also proportional to the total power o f X : the more confident a nation
is in its strength, the more it can afford to build d o w n . The proportionality
parameter for build-down is the confidence coefficient, w h i c h , for simplicity,
is taken to be the inverse o f the build-up fear and loathing coefficient. Finally,
there are no economic constraints for build-down. (It is assumed that recon­
version o f the military-industrial complex works!)
The result o f combining the coupled arms buildups and build-downs is
the following set o f recursion relations, relating the total arms stocks o f X and
Y i n any year to the corresponding stocks in the previous year:

x„ +i =x„ + Y„(Y„
axy - x ) 0 (1 - [x
n n+l - x yc ) 0
n x (Y n

~^-X (X n n - Y„)@(X -Y ) n n
u
xy (9)

Y n+l =Y n + a X (X yx n n - Y„) 0 (1 - [Y„+l - Y ]IC )

n y 0 (X n

- Y {Y n n - X„) 0 (Y„ - X ).n

"xy

These relations may be iterated, given arbitrary starting values X and Y 0 0

at n = 0, again using standard spreadsheet routines on a desktop computer.

Going on to large n (n = 100), the t w o Lyapunov coefficients can be calcu­
lated from equation 5. Typical results (Saperstein 1992a) are given in figure
 The Prediction of Unpredictability 157

x<0) = y(0) = 0.3

epsilon = 0.001

20

V
0 10 20 30 40
a

Fig. 7.2. T r a n s i t i o n t o e s c a l a t i n g a r m s race

7.2, where the region above the curve is the region o f instability. Note that
small values for both a = a^ and b = a imply stability whereas instability
yx

results when both a^ and a get large or when either gets very large.
yx

Since large values o f the effective fear and loathing coefficient imply
chaos and war, and since any collection o f autocratic states (or large mixed
collection o f autocratic and democratic states) is more likely to have some
pairs o f nations w i t h large values o f this coefficient than a corresponding
collection o f democratic states, it follows that an outbreak o f war is more
likely i n a collection o f autocratic states than in a similar collection o f demo­
cratic states. Even though it is often easier, in foreign affairs, to deal with
autocratic governments, it is evident that, in addition to the usual human
desire to nurture similar systems elsewhere, democracies are more likely to
ensure w o r l d peace by supporting the democratization o f other members o f
the international system.

W h i c h is M o r e War-Prone: A S y s t e m of S h i f t i n g
Alliances o r a Collection o f G o - l t - A l o n e States?

As a last example, consider the question: which is a more stable system o f

competing states: one in which each nation individually procures arms so as to
158 Chaos Theory in the Social Sciences

match any other state in the system or one i n which weaker nations j o i n
coalitions to collectively procure arms against the strong and in which they
shift alliances when a coalition member becomes too strong—independence
security or balance-of-power security? To model the t w o cases simply, con­
sider three competing states, either independent o f each other or in which any
two w i l l be allied against the stronger third.
Considering any pair o f the three, arms buildup or build-down is pre­
sumed to occur just as in the previous case (equation 9), w i t h the effective
fear and loathing coefficient o f X for Y and l/a^ the corresponding confi­
dence coefficient. However, now X may pair against Y or Z , i f all three are
independent, so the resultant arms procurement or build-down is the sum o f
the corresponding pair terms:

AX = X N+L -X„ = [ Y„(YAXY N - X„) 0 (Y N - X„) + a Z (Z XZ N N - X) N

l
e (z„ - x )] © ( - = ^ ) n

- ~- x (x n n - Y ) 0 (x -y„)~
N n ~ x (x n n - Z„) 0 (X - Z„) N

AY = y„ + 1 -Y„ = [a X {XYX N N - Y ) 0 (X
N N - Y„) + a Z (Z YZ N N - Y) N

0 (Z„ - Y )} 0 ( ^ - ^ )
N (10)

- - L Y (Y N N - x„) © (Y -x„)-^- N Y„(Y„ - Z ) 0 (Y - Z„) N N

u u
yx yz

AZ = Z„ + 1 - Z„ = [a X„(X„ ZX - Z„) 0 (X„ - Z„) + a Y (Y ZY N N - Z„)

0 (Y N - Z„)] 0 ( X
-=^ )

- -L z (z n n - x ) © (z„ - x ) - -L z (z
n n n n - Y„) © (z„ - Y„).

In the alliance model, X w i l l j o i n its procurement w i t h Y to match a

superior Z , or j o i n w i t h Z to match a superior Y , and similarly for build-down
(here the appropriate coefficients are written b^, b , and so on): x z

AX = [b^Y,, - X „ - Z„) 0 (Y N ~ X „ - Z„) + b Z (Z XZ N N - X N - Y)

N

0 (Z„ - X n - Y )} 0
N ( )
 The Prediction of Unpredictability 159

TT~ V« + z )(X n n + Z n - Y ) 0 (X
n n + Z„ - Y) n

xy

(X n + Y )(X n n + Y - Z„) 0 (X„ + Y -

n n Z„)

AY = [b X {X„ yx n - Y „ - Z„) 0 (X n - Y n - Z„) + b Z (Z„ yz n - Y „ - X)

n

1 - AY
0 (z„ - y„ - x„>] ©

- r - (Y + z„)(Y n n + z„ - © (r„ + z„ - x) n

(ID
- ^ - (y„ + x„)(Y n + x„~ z„) 0 (K„ + x„ - z„)

AZ = [b X (X ZX N N - Z „ - Y„) 0 (X„ - Z„ - K„) + fo /„(K„ - Z„ - X„)

z

0 (K„ - Z„ - X„)] 0 ( )

- j - (z„ + r„)(z„ + Y„ - x ) 0 (z„ + Y„- n x)

n
zx

-T~(Z n + X )(Z„
n + X „ - Y ) 0 (Z„ + X
N N - Y ).
N

The t w o models each consist o f three coupled recursion relations (eqs. 10

and 11). The answer to the question posed is obtained by seeing which o f the
t w o models is more stable.
The analysis is done by assuming that the initial values, X , Y , Z , in 0 0 0

each o f the three nations are almost the same and asking whether these small
differences grow as the recursion progresses. A spreadsheet iteration o f the
recursion relations is again carried out, starting off w i t h XQ, Y' , Z' , represent­ 0 0

ing small random differences about an arbitrarily chosen starting point, X . 0

Lyapunov coefficients are computed (again by iterating out to n = 100 w i t h

small 5 ) , varying the starting value X and the single model parameter a
0 0

(representing a common value for a^, b^, a , and so on). xz

Four possibilities become evident, as the parameters a and X are varied, 0

for the independent nation model as a result o f the numerical computations

(Saperstein 1992b). I n the first case, strong stability, all sequence limits exist
and all Lyapunov coefficients are negative, which implies that all sequence
limits exist and are the same. There is complete predictability and hence no
war. I n the second case, weak stability, all sequences have limits that are
160 Chaos Theory in the Social Sciences

equal to each other, but the Lyapunov coefficients are positive. This occurs
because all limits are less than the starting value, X , except for the singular
0

case when all start off exactly equal to X , in which case the limiting value is
0

X . Hence, the numerator in the definition o f the Lyapunov coefficient doesn't

0

become less than the denominator, implying a positive coefficient. This situa­
tion is an illustration o f bifurcation, in that one point leads to results com­
pletely different from all other points. Since the limits are still well defined,
the sequences are predictable, not chaotic, and so again there is no war. I n the
third case, weak chaos, none o f the sequence limits exist and so there are no
Lyapunov coefficients. However, the variation within any one o f the se­
quences, X' , Y' , Z' , and the differences between the sequences (X' — X )
n n n n n

remain small compared to the arbitrary j o i n t starting value, X . The situation

0

resembles case b i n figure 7 . 1 , rather than case c; hence, the unpredictability

is minor and so, i n the paradigm o f this chapter, does not represent war. The
last possibility is one in which the sequences do not have limits and the
variations w i t h i n each sequence, as w e l l as the differences between them, are
large compared to the starting value. In this case the range o f unpredicta­
bility is the entire range possible, there is no way o f knowing—even
approximately—the outcome o f any policy or action, and hence major fluctu­
ations may result from minor perturbations. This is the criterion for crisis
instability and war.

The numerical results for the independent nation model are displayed in
figure 7.3. N o sharp boundary was evident between the regimes o f weak
stability and weak chaos and so the two regimes are considered to be a single
realm i n w h i c h crisis stability is still valid. Note that there is no upper l i m i t to
the weak chaos regime, no breakdown o f crisis instability, i f the fear and
loathing coefficient is small enough (a ^ 1). I f there is enough confidence in
the system, perturbations to the independent nation international system may
lead to continued fluctuations and uncertainty, but not to war, no matter how
large the initial, equal, stocks o f arms are. On the other hand, with sufficient
fear and loathing (a > I), war w i l l always break out i f the initial arms stocks
are large enough.
The numerical predictions o f the alliance model are quite different. For
large enough symmetric starting values, X = Y = Z , the system imme­
0 0 0

diately and completely disarms, no matter what value is chosen for the fear
and loathing coefficient. This is region I o f figure 7.4; since all sequences tend
immediately to a zero l i m i t i n g value, it is a strong stability region. I n region I I
o f the figure, a l i m i t has not yet been reached numerically at n = 100;
nonetheless, the fluctuations in the monotonically decreasing sequences are
small and hence this region is appropriately characterized as weak chaos: the
course o f events is clear and the eventual complete disarmament is completely
predictable.
 a

Fig. 7.4. N u m e r i c a l p r e d i c t i o n f o r t h e alliance m o d e l

162 Chaos Theory in the Social Sciences

Hence, using the possibility o f strong chaos in the model as an indicator

o f the possibility o f war in the corresponding international system, it follows
that a policy o f building (perhaps shifting) alliances is guaranteed to keep an
initially symmetric system peaceful. A policy o f seeking national security by
"going i t alone" allows the possibility o f war in such a system i f the contend­
ing parties start out w i t h large enough levels o f armed force.
I know o f no real-world situation in which a system o f comparable
competing states have balanced themselves d o w n to a complete disarmament
as is predicted by the alliance model o f this chapter. Perhaps this is because a
shifting o f alliances has costs i n the real w o r l d , costs not included in this
model. Also possible is that there are no real balance-of-power situations; real
nations may resort to balance-of-threat instead, a much harder situation to
model (Walt 1987). Or, the presumed anarchy may not have always been
present.

Conclusion

Humans have always developed theories (often mythologies) to explain their

w o r l d . W i t h the advent o f science, theories had to have practical as well as
literary or philosophical implications. A successful theory o f international
relations should be an important component o f competent policy making in
the arena o f international security in a competitive world. Conversely, a
careful analysis o f practical policy—successful or unsuccessful—must be the
foundation for any scientific theory.
Given the complexity o f the real international system, any theory o f it is
bound to be incomplete, describing some aspects o f it more thoroughly than
others. A truly massive modeling enterprise—such as is done i n economics
w i t h national input/output models or in meteorology w i t h large-scale
computer models o f the atmosphere—may at best lead to very rough, short-
term, quantitative agreements between predicted and observed experience.
The quality o f the results is often vastly disproportionate to the effort required
to secure them. It thus seems useful to look for simple models and methods o f
model analysis that can lead to qualitative robust predictions.
One such approach is the prediction o f unpredictability in simple non­
linear models o f the competitive interactions between rival states. It can be an
important approach to policy and theoretical questions, given the fundamental
assumption that such unpredictability may represent crisis instability and war
in the international system. Given that policy questions are usually more
specific and complex than more theoretical questions, we expect to require
more complex models for the former, although they would still be consider­
ably simpler than any intended to address quantitative questions. For example,
very simple theoretical models, easily analyzed w i t h spread sheets on a desk-
 The Prediction of Unpredictability 163

top computer, have allowed us to conclude that a tripolar world is less stable
than a corresponding bipolar w o r l d , that a set o f democratic nations is more
likely to be stable than a similar set that includes autocracies, and that a group
o f nations attempting to guard their security via balance-of-power alliance
formation is more stable than a group in which each member makes individual
attempts to cope w i t h other potentially hostile members. A bigger and more
complex computer model indicates that a policy o f introducing S D I into a
situation similar to the recent cold war nuclear confrontation between the
United States and the Soviet Union would likely dangerously destabilize such
a world.
None o f these theoretical or policy results is new. Similar conclusions
have been reached before by many others using conventional verbal analyses.
But the alternative route presented in this chapter—the qualitative analysis o f
simple quantitative models—leads very quickly to the answers without any
hidden implicit assumptions, biases, or faulty logic. Finally, agreement o f
results obtained via traditional means and the methods developed here brings
added credence to both approaches. Science is a fabric: its ability to cover the
w o r l d depends upon the existence o f many different fibers acting together to
give it structure and strength.