On the Very Idea of a System Dynamics Model of Kuhnian Science
Jason Wittenberg
Department of Political Science
MIT
April 15, 1990
Paper to be presented at the International System Dynamics
Conference, Boston, July 1990. The title is suggested by an article
by Donald Davidson. This paper has benefitted from a conversation
with Thomas Kuhn, as well as eritical readings from and
conversations with Hayward R. Alker, Jr., John Sterman, and John
M. Richardson, Jr. I am solely responsible for the views expressed.
1332
System Dynamics '90 . 1333
Introduction
The appearance of Thomas Kuhn's Structure of Scientific
Revolutions engendered considerable discussion about the nature of
scientific change. Kuhn challenges the prevailing view of science
as a continuous, logical enterprise by attempting to debunk
science's myth of rationalism. As an historian as well as
philosopher of science!, he attempts to explain science's
extraordinary success not by developing methodological canons
divorced from scientific practice, but by looking at how scientists
actually work. (Lakatos and Musgrave 1970, 236-237)
Acknowledging the philosophical importance of actual
scientific practice is controversial. Kuhn's critics question both
his characterization of science as mostly "puzzle-solving", as well
as his claim that such practice is necessary for scientific
development.” It will not be the task of this essay to rehearse
these still unresolved debates. That is better left to historians
and philosophers. Rather, I would like to recognize another
important contribution to the discussion, one that is orthogonal
to any other that I know of. In "The Growth of Knowledge: Testing
a Theory of Scientific Revolutions with a Formal Model,"*® John
Sterman has built a model of Kuhn's account of scientific change.
He asks not whether science does, could or should correspond to
Kuhn's view, but whether Kuhn's theory is dynamically consistent.
He is interested in whether the behavior Kuhn describes (i.e.,
paradigm emergence, normal science, crisis and revolution) actually
follows logically from the assumptions Kuhn makes. To do so he
constructs a system dynamics computer model.
Modeling
Modeling is a worthwhile enterprise. Unlike traditional
attempts to resolve disagreement, in which scholars attempt to
intuitively describe the behavior implied by a complex and often
unstated set of assumptions, the modeling process requires that
assumptions and logic be made explicit, and thus open to debate and
discussion. Simulation infallibly yields the behavior implied by
the assumptions. (Forrester 1971, 54; 1987, 147; Meadows 1980, 27)
A successful model of a theory can lend extra confidence that the
theory being modeled is a sensible one.
But what constitutes a successful model? Good system dynamics
models are designed with well-defined boundaries, so that changes
1 some of his views are foreshadowed in Thomas S. Kuhn, The
Copernican Revolution. (Cambridge, Mass.: Harvard University
Press, 1957).
? The best collection of critical views remains Lakatos and
Musgrave (1970). For a very brief review of the issues in these
critiques, see Barnes 1982, 58-63.
5 See Technological Forecasting and Social Change 28, 1985,
93-122.
1834 System Dynamics '90
in system behavior may be understood as arising endogenously.
Undesirable behavior is assumed to result from the system's loop
structure rather than from uncontrollable exogenous factors.‘
Establishing the system boundary requires simplification and
reinterpretation. A model wouldn't be useful if it replicated the
complexity of the system being modeled. Indeed, "[t]he art of model
building is knowing what to leave out." (Sterman 1988, 137) Within
system dynamics the exclusion criterion is dictated by the model
purpose. A good model contains only those features necessary to
accomplish the model's purpose. (ibid)
While simplification is universally recognized as necessary,
it is always controversial when put into practice. Sterman is well
aware of potential criticism: "In the course of formalizing Kuhn's
theory, certain changes in emphasis and interpretation are
necessarily introduced..." (Sterman 1985, fn 4, 94) "The model
should be viewed more as a rough translation of the theory into
formal terms than as a definitive rendering." (ibid, 100) I do not
question the need for reinterpretation and simplification; that is
part of the modeling process. I also recognize that particular
reinterpretations and simplifications will be value judgments of
the modeler. But there comes a point when these activities go
beyond the bounds of reasonableness, when they begin to do violence
to the crucial ideas underlying the system being modeled.
Purpose of this Essay
This essay has two purposes. First, I want to show that
Sterman has indeed exceeded a modeler's prerogative of
reinterpretation. The model excludes components of the theory which
for Kuhn play an important role in paradigm change. I will
illustrate this by comparing Sterman's and Kuhn's accounts of the
causes of scientific revolution, and then judging how faithful the
model is to Kuhn's theory. Given that some reinterpretation is
inevitable, this is not an easy task. It first requires that we be
able to point to some well-bounded body of discourse and call it
"Kuhn's interpretation." It is well known in political science that
Hobbes waS no "Hobbesian" and that Machiavelli was no
“Machiavellian.” It think it a safe bet that neither is Kuhn
"Kuhnian." Thus, applications and interpretations considered
somehow Kuhnian must, to the extent possible, be distinguished from
views Kuhn himself has expressed.* Indeed, it will be difficult
enough to establish a stable account of Kuhn's own thoughts,
evolving as they have over more than two decades. The second
requirement is that we be able to distinguish fundamental from
peripheral arguments. This. can be done by examining criticisms of
* For an excellent and brief discussion of system dynamics,
see Jay W. Forrester (1987), "Lessons from System Dynamics
Modeling," System Dynamics Review, Vol. 3, No. 2, Summer 1987,
Pp. 142-146. ,
® Gutting (1980) is a nice collection of interpretations and
applications. The bibliography lists 250 works about Kuhn, as
well as 48 books and articles by Kuhn himself.
System Dynamics '90 1335
Kuhn and his subsequent refinements. Those aspects of his
formulation which have received the most attention are deemed to
be the most important. My second purpose is to highlight a few
methodological problems which must be faced before a truly valid
model of Kuhnian science can be constructed. These will emerge in
the process of model evaluation.
If my purpose is to illustrate invalidity, it is not enough
merely to know that such a project is possible. There are degrees
of validity, and while an appropriate model boundary is necessary,
it is by no means sufficient. A more fully specified definition of
validity will permit a finer-grained evaluation.
Validity
The best discussion of validity criteria within system
dynamics remains Forrester (1961, 115-129), who maintains that
validity only has a useful meaning with respect to a model's
purpose. (ibid, 115) The purpose of most models is to understand
the behavior of complex systems, so that undesirable outcomes may
be minimized or avoided. Forrester identifies two requirements a
valid model must satisfy. I will call these reality conditions.®
First, the model must generate behavior that doesn't significantly
differ from that of the real system. (ibid, 119) Second, the
relationships in the model must represent the true causes of
action. (ibid, 122) The second criterion is added in recognition
that any number of models can be constructed to reproduce a given
set of behaviors, but a model can only be said to “explain" this
behavior if its equations reflect the real causal relationships in
the system.
While these criteria are meant to have general applicability,
they presuppose that a well-defined distinction can be made between
the model and reality. I admit that for most models such a
distinction can readily be made. In these, the behavior to be
explained is easily identifiable empirically. Inventory, profits,
sales, fear, pleasure and so forth, are data that are "given" in
the sense that they can be understood independently of any
comprehension of the forces which determine their behavior.’ Thus,
it isn't necessary to understand the myriad of forces causing sales
to fluctuate in order to recognize that sales do fluctuate. The
knowledge base used in constructing the model will normally be
quite heterogenous. Many different explanations and observations
will be taken into account. |
There are other mode and Sterman's is one, that do not
purport to represent real-world systems. These models are of
theories, and their purpose is not to solve problems, but to probe
an argument's internal consistency. Thus, the purpose of Sterman's
®° Forrester has much more to say about validity, but for the
present purposes these two criteria suffice.
7 Note that I list both quantitative and nonquantitative
variables. The distinction between the two, and any validity
problems associated with the latter, are not of direct interest
ere.
1836 System Dynamics '90
model is "to test the dynamic consistency of Kuhn's theory...by
formalizing it and then testing the formalized theory with a
computer simulation model." (1985, op. cit., 94) I understand
"testing" in this context to mean being able to account for the
behavior Kuhn postulates. That is, Sterman's model will "explain"
some body of data by reproducing its qualitative behavior. The
problem is that the model has no data in the same sense as the
real-world models discussed above. Here the behavior to be
explained must necessarily come from the same database from which
the model was built. They are both interpretations of the same
authoritative texts. This considerably blurs the distinction
between the model and the system being modeled, so that to maintain
that the model reflects "reality" is to border on the
tautological.® This would not be a problem if Sterman had chosen to
model an actual scientific revolution rather than a generic one.
In this case behavior would consist of historical data, culled from
sources separable from Kuhn and others' interpretations of him. The
model would become that of a real-world system.
I do not want to make too much of the distinction between
these two model types. The differences are a matter of degree, not
of kind. It is not so much that models of theories are easier to
validate, but that they are easier to invalidate. The first reality.
condition presents a difficult hurdle for modelers of real-world
systems because of the sheer complexity of exhibited behavior. This
behavior is the result of both stochastic and structural factors.
Reproducing the behavior predicted by a theory is less difficult
for two reasons. One has already been discussed, the very close
dependence of model and data on the same database. Another is that
this behavior is simply less complex than real-world behavior.
Theories, after all, simplify.
The differences in the complexity of the behavior these models
generate reflect the relative transparency of the corresponding
systems being modeled. Discerning the real causes of behavior is
considerably more difficult for a real-world system than it is for
a theory-constructed system. Theories simplify behavior because the
causal connections they postulate simplify real-world causes.
Ceteris paribus, the simpler the causes, the easier they are to
distinguish.
It follows that the easier the causes are to identify, the
easier particular claims to have recognized these causes are to
critique. There can be no appeal to particular authoritative texts
in evaluating models of real-world systems, while such appeals
necessarily occur in the case of theoretical systems. Thus, an
assessment of a model of why scientific revolutions actually occur
would be legitimized by reference to a wide variety of historical,
sociological, psychological and philosophical texts. No particular
text would, prima facie, be privileged over any other. On the other
hand, a model of Kuhnian science may only be challenged through
® I do not mean to imply that Sterman intentionally "cooked"
his reference mode to match his base run. In all models of this
type there is necessarily an iterative process of model
development and reference mode refinement.
System Dynamics ‘90 : 1337
appeal to Kuhn and perhaps a few of his interpreters. While the
existence of authoritative texts by no means guarantees an
authoritative interpretation of those texts’, the space of
interpretive possibilities must nonetheless be smaller.
Assessing the Model
Of the many points of contact between Sterman and Kuhn, the
most important issue is why scientists reject paradigms. The core
of Sterman's model is scientists' confidence in the paradigm (CP).
(1985, op. cit., 104) A confidence level of one represents total
commitment, while a level of zero indicates total rejection.
Confidence is a function of the relative number of accumulated
anomalies (RA) and the rate of progress of the paradigm (RSP,
defined roughly as the ratio of the number of puzzles solved ina
year to the total number solved). If the number of anomalies
increases above some acceptable number, or the rate of progress
falls below some expected level, then confidence will decline.
Practitioners join or leave a paradigm based on the confidence
of those in the paradigm relative to the confidence of outsiders
in other paradigms. (ibid, 105) The higher the practitioners'
confidence, relative to other paradigms, the higher the recruitment
rate. The lower the relative confidence, the higher the defection
rate. Recruitment and defection are modeled as the same process,
though arithmetically inverse. Membership changes are determined
through the difference between recruitment and defection.
The model's validity problems center on the role of
alternative paradigms. For Kuhn the distinguishing feature of
normal science is that few scientists engage in inventing novel
theories; they are busily solving puzzles the dominant paradigm
supplies. (1970, 24) Novel theories emerge only when the old
paradigm is in crisis. These issues are not peripheral. His critics
have focused precisely on the distinction between normal and
revolutionary science. Note some of the chapter titles in Lakatos
and Musgrave (1970): "Against Normal Science", "Does the
Distinction Between Normal and Revolutionary Science Hold Water?",
and "Normal Science and its Dangers." The postulation of normal
science is one of Kuhn's most controversial claims.
Sterman is aware of Kuhn's position, but chooses not to model
it, due to a combination of a system dynamicist's need to preserve
endogeneity and any modeler's natural wish to create as
parsimonious a model as possible. Thus, he correctly notes that
competing paradigms "tend to be born in the crisis phase of an
existing paradigm," and are "part and parcel of the dynamic
process," but then avers that credible models must also generate
the predicted behavior "without relying on external driving forces
such as the emergence, as if by magic, of a new and better theory."
(Sterman, 105; 96) If having an alternative paradigm is desirable,
and it can not be introduced exogenously, then the only remaining
choices are either to model the emergence of the new paradigm, or
® The voluminous literature concerned with interpreting Karl
Marx is proof of this.
1838 . System Dynamics ‘90
posit a continuously existing alternative. Preferring parsimony to
accuracy, Sterman chooses the latter strategy.
In principle such a tactic is not impermissible. The effect
of this alternative paradigm can be neutralized during normal
science and then switched on right before a crisis by adjusting the
values of confidence in alternative paradigms (CAP) and effect of
confidence on recruitment and defection (ECR and ECD). But note
that unless the switch conditions are determined within the model,
the net result would be to introduce the switch exogenously,
something Sterman explicitly declares to be unsatisfactory.
A bigger problem with positing a continual alternative
paradigm, but negating its effects, is that it reifies structure.
There is a big difference between novel theories emerging in
crisis, and novel theories always existing but only gaining
salience during crisis. The first view is of the emergence of a
new structure, the second of an ever-existing structure that
suddenly gains importance. The difference is ontological, and may
not matter in terms of model results, but it surely matters if one
is concerned with how accurately the model represents the theory.
This relatively minor problem of system boundary is made much
worse by the way in which alternative paradigms are actually
implemented in the code. Disregarding the differential effects of
alternative paradigms in normal and revolutionary science, he sets
the confidence in these paradigms to a constant.” This represents
a fundamental error in the model. Because the confidence in
alternative paradigms never changes, it can be removed without
changing the qualitative or quantitative behavior of the model.
Alternative paradigms in Sterman's model are superfluous."
For Kuhn alternative paradigms are not only necessary but
crucial to the dynamics of the fall of a paradigm. He is quite
explicit on this:
{T]he act of judgment that leads scientists to
reject a previously accepted theory is always
based upon more than a comparison of that
theory with the world. The decision to reject
one paradigm is always simultaneously the
decision to accept another, and the judgment
leading. to that decision involves the
comparison of both paradigms with nature and
with each other. (1970, op. cit., 77)
Once again, the validity of the model depends on how critical this
aspect of Kuhn's theory is. I take it to be decisive. The "and" in
the last sentence quoted is italicized in the original, indicating
that Kuhn knew that the most provocative part of the thesis was
. 10 The value of CAP, 0.5, corresponds to maximum uncertainty
in the competing paradign.
1 to remove the effect of other ‘paradigms, I Set the
confidence in alternative paradigms, CAP, equal to one, and then
rescaled the effect of confidence on recruitment and defection,
ECR and ECD, to range from zero to one.
System Dynamics '90 ‘ 1389
contained in the last clause. Indeed, Kuhn's elaborate and very
interesting comparisons of Lavoisier's and Priestley's views of
chemistry, and Copernican and Ptolemaic astronomy, testify to the
importance of alternative paradigms in the revolutionary process.
It is here that Sterman has sacrificed too much in the name of
parsimony. One can't reject the role of competing paradigms and
still claim to be providing a valid model of Kuhnian scientific
revolutions.
It is illuminating-- and intriguing, that despite the
injudicious choice of model boundary, the model has satisfied the
first reality condition. Sterman does manage to reproduce behavior
characteristic of Kuhnian paradigm change. This can mean one of
two things. Either Kuhn's theory contains propositions (i.e., the
whole business of alternative paradigms) that are not necessary to
produce the forces of change he postulates, or the model has failed
the second reality criterion. That is, the forces modeled are not
the same forces Kuhn postulates. Postponing discussion of the first
possibility for the time being, let me consider the second. Does
the model validly portray even those aspects of Kuhn's theory that
it claims to represent?
Consider again the question of why scientists reject
paradigms. In Sterman's formalism this question is equivalent to |
asking why recruitment falls and defection rises. Falling
confidence plays the key role. Confidence rises or falls due to the
combination of two factors: relative number of anomalies and
progress in puzzle solving. Note the character of the equations
that determine the effects of anomalies and progress on confidence:
CC. KL=NCC*ICC.K*RCC.K
ICC. K=EAC.K + EPC.K
EAC. K=TABHL (TEAC, RA.K,0,6,.5)
TEAC=5/2.15/0/-1.2/-2.15/-2.9/-3.4/-3.9/-4.4
EPC. K=TABLE (TEPC, RSP.K,0,5,.5)
TEPC=-5/-2.15/0/1.2/2.15/2.9/3.4/3.9/4.2 ...1
cc is total change in confidence. ICC is indicated change in
confidence, and does most of the work. NCC is normal change in
confidence and RCC is receptiveness to change in confidence. They
are multipliers which reduce or magnify the effect of ICC. EAC is
the effect of anomalies on confidence, and EPC is the effect of
progress on confidence. For reasons of space not all the values
of TEAC (or TEPC) are listed. RA is relative anomalies and measures
how many anomalies there are relative to the acceptable number. RSP
stands for relative solved puzzles, and compares the current rate
of puzzle solving with the total number of puzzles the paradigm has
solved. The coefficients for anomalies and solved puzzles are left
oer of the Icc equation because in the base run they have a value
of one.
” Sterman, 1985, op. cit., 120.
1340 System Dynamics '90
Consider the first equation. Since NCC and RCC are positive
and constant, any change in the sign of CC must be due to the
change in sign of ICC in the second equation. This will occur when
the sum of EAC and EPC is less than zero. Thus, confidence declines
when the effects of anomalies and progress on confidence are
negative. This will only happen when the number of anomalies (RA)
rises too high, and not enough puzzles (RSP) are being solved. From
our perspective the important point is that confidence falls
because there are too many anomalies and too little progress.
Kuhn would not agree that paradigms are abandoned, at least
in the beginning, because there are too many anomalies, or too
little progress. As he says, "paradigm debates are not really about
relative problem-solving ability, though for good reasons they are
usually couched in those terms." (1970, 157) Neither Lavoisier's
oxygen theory nor the phlogiston theory could account for all the
facts that the other could account for. Each could explain
phenomena the other couldn't account for. Similarly, Copernican
astronomy did not surpass Ptolemaic in accuracy until over a half
century after Copernicus had died, yet Copernican theory prevailed.
(Kuhn 1977, 323) Lavoisier's oxygen theory and Copernican astronomy
were initially accepted not because of problem-solving ability, but
despite it.
The point is extremely important, because it is precisely
Kuhn's unwillingness to privilege problem-solving ability that
prompts his detractors to accuse him of abandoning science to "mob
rule" and “irrationality." Bell and Bell (1980), for example,
identify two views of Kuhn, one of which places him fairly close
to Popperian refutationism, and another "dogmatic" interpretation,
which insists there is no intellectual criterion for paradigm
comparison. (ibid, 18-20)
Kuhn recoils from such accusations.’ (See 1970b.) It is not
that paradigm choice is made irrationally, but that accuracy is not
the only criterion of choice. Paradigms are compared not only with
reality, but with each other. Evaluation will be based as well on
consistency, scope, simplicity and prospects for future progress.
Together with accuracy, these criteria form the shared basis for
choice. (1977, 321-322)
In any given historical situation these criteria may conflict
with one another. While both Copernican and Ptolemaic astronomy
were internally consistent, only Ptolemaic was also consistent with
other physical theories. Thus, consistency spoke in favor of
Ptolemy. Simplicity, on the other hand, favored Copernicus. At
least in terms of mathematical apparatus, Copernican astronomy
required only one circle per planet, while Ptolemaic required two.
(1977, 323-324) Resolution of these conflicts requires ranking
these criteria in order of importance.
Note, however, that "importance" is a value judgement. There
is no a priori reasoning for favoring simplicity over consistency,
or any one criterion over any other. Science is silent on the
issue, so that scientists will differ over which is the more
important. Copernican astronomy triumphed because Copernicus had
faith that his simpler, more elegant theory, once fully
articulated, would surpass the Ptolemaic system in accuracy. But
during Copernicus' life such success was but a dream. Ultimate
System Dynamics '90 1841
triumph was not achieved by Kepler until long after Copernicus had
died.
Yet even if scientists could agree on how to rank the
criteria, they would still disagree on how to apply them in
particular situations. (1977, op. cit., 331) Thus, while Copernican
astronomy was simpler, it was so only in terms of mathematical
apparatus. In was not simpler in terms of the computational work
required to make predictions. Simplicity here has two different
meanings. Science, once again, does not recommend one or the other.
(ibid, 324) Similarly, scientists weighing the relative accuracy
of the oxygen and phlogiston theories would almost certainly
disagree on which was the more accurate, since the theories did not
account for the same phenomena. Thus, even if accuracy were the
agreed upon value, choosing between the two would require deciding
which phenomena were more important, surely a decision that would
vary with the individual. (ibid, 323)
Kuhn is reserving a role in the paradigm debate for the
scientist as a unique individual. Personality and education will
influence choice. (1970b, 241) Thus, one can not explain what
animated the early Copernicans without recourse to the "ear for
mathematical harmonies" provided by the rise of neoplatonism in
Renaissance Europe. (1957, 181) It is because of these individually
varying factors that Kuhn insists that an algorithm able to dictate
rational, unanimous choice is unattainable. (1977, op. cit., 326)
Although the model fails to represent these criteria of
decision, it is not necessarily completely invalid. Kuhn makes a
distinction, though not very explicitly, between the initial
acceptance and the ultimate acceptance of a paradigm. (1970, 156)
He argues that only a relatively few scientists need be converted
through these individual criteria. They will then develop the new
paradigm to a point where other scientists can adopt it purely for
reasons of predictive accuracy and puzzle-solving ability.
Ultimately, a proposed paradigm does not become the new paradigm
of normal science unless it surpasses the old one in its ability
to solve puzzles.
There are thus two general methods of paradigm change. The
first utilizes the idiosyncratic value systems of certain key
individuals. They adhere to the new paradigm despite the fact that
it may not be as accurate as the old paradigm. The new paradigm may
simply have aesthetic appeal. The second method involves collective
behavior, and is based on the new paradigm's ability to solve
puzzles. Practitioners convert. because the new paradigm solves more
puzzles. Given the way paradigm change is portrayed in Sterman's
model, it appears to be applicable only to the second group. The
individual behavior characteristic of the early stages of
revolution remains unmodeled.
rd_a Vali del_of Kuhnian Science
There are two areas in which the model is invalid. The first
is its neglect of alternative paradigms. In principle such
paradigms are not difficult to incorporate. Since they arise
endogenously, they are well-suited to being modeled within system
dynamics. It was an injudicious choice of system boundary to
1342 , System Dynamics '90
exclude them, but an understandable one. Done properly, the model
_would have at least doubled in size.”
The second aspect of: invalidity concerns the incomplete
representation of paradigm change. I submit that if we take Kuhn
literally about the lack of an algorithm dictating rational,
unanimous choice, no valid system dynamics model of Kuhnian science
is possible. To see this, let us begin by imagining the very best
possible model, a valid and accurate representation of Kuhn's
theory. Belief in this model implies belief in the dynamic
consistency of Kuhn's argument, since the purpose of the model is
to illustrate this consistency. But belief in the model is more
than just belief in dynamic consistency. Since model behavior is
just the logical consequence of simulating the model's assumptions,
belief in the model is equivalent to belief in the model's
assumptions. These assumptions are the interrelationships of the
variables as represented in the code. The American Heritage
Dictionary (2nd College Edition, 1982) defines "algorithm" as "a
mechanical or recursive computational procedure." (93) A system
dynamics model is precisely such a procedure. The code of our
imaginary model is nothing more than a mechanical method for
comparing paradigms. Thus, if we continue to believe in the
validity of this model, then we have accepted what Kuhn describes
as unattainable, an algorithm dictating rational, unanimous choice.
There is no prima facie contradiction here. I can grant Kuhn
consistency without having to agree with him if I only disagree
with his assumptions. This is so because by disagreeing with an
argument I am disagreeing either with the behavior predicted by
that argument, or with that argument's assumptions. By granting
dynamic consistency I am assenting to the behavior, given the
assumptions. Thus, any disagreement must be over the assumptions.
Now, if our imagined model is the best possible, then disagreeing
with Kuhn's assumptions is equivalent to disagreeing with the
model's assumptions, since the model's equations embody Kuhn's
assumptions, by assumption. But questioning a model's assumptions
means questioning its validity. Since by construction the model is
the best possible, this is the same as asserting the invalidity of
the most valid model. If the most valid model possible is invalid,
then no such model is possible.
Let me repeat, this is true only if Kuhn is taken literally.
As discussed previously, what Kuhn means is that in the crisis
stage of a paradigm, an individual's historical and cultural
context plays a key role. For the modeler of Kuhnian science the
problem then becomes incorporating these exogenous and contingent
elements into what is supposed to be a model of generic processes.
If Sterman's model were of an historic revolution rather than
a generic one, then the latter problem disappears. Thus, if the
model were of the Copernican revolution, then Copernicus’
neoplatonism, the necessity of calendar reform, and all the other
reasons that animated Copernicus could be validly incorporated into
18 a11 the structures of the model-- recruitment, defection,
confidence, etc., would have to be duplicated for the competitor.
Extra code would also have to be added to compare the paradigms.
System Dynamics '90 : 1343
the model. They become, if you will, the facts of the matter. But
generic models must identify forces common to the behavior of all
the revolutions to which Kuhn's theory applies. Causal factors
unique to any particular revolution may not be included.
One possible solution is to make the contingent forces
generic. The process is rather simple. One takes all the reasons
that these key individuals were committed to their paradigms, and
one constructs a category that includes all and only these reasons.
Thus, "simplicity", "accuracy", "scope", etc., all become lumped
into a category called, say, “aesthetic advantages." Then one
constructs a function representing the relationship between the
number of aesthetic advantages and confidence in the paradigm. Such
a function might indicate that as the degree to which the old
paradigm is aesthetically pleasing goes down, so does confidence.
The uniqueness of a Copernicus or an Einstein becomes only one
instantiation of a generic process of paradigm disillusionment. The
difficulty with this formulation is not that it is reductionist,
but that it implies that the practitioner somehow "chooses" the new
paradigm when the aesthetic problems of the old one become too
great. Yet Kuhn suggests that "choice" may not be the best way to
describe what is going on.
The most radical of Kuhn's theses, and one I have not touched
on in this paper, is the concept of paradigm incommensurability.
Incommensurability means that there can never be full translation
from one theory to another. (See Kuhn 1970, 148-159, 198-204;
1970b, 266-278). The idea here is that a theory's concepts change
their meanings and applicability in moving to the successor theory.
"Planet" did not mean the same thing to Copernicus as it did to
Ptolemy. Einsteinian "mass" is radically different from the
Newtonian version. Full translation would require conversion of
both theories into some neutral sense-datum language, and for Kuhn
no such language exists. This expresses his belief that "facts,"
which would normally be appealed to in paradigm debate, are not
independent of theory. People in different paradigms speak
different languages, so that point by point comparison of two
different paradigms is no more possible than such a comparison of
two languages. Yet it is just such a comparison process that one
must perform if one "chooses." The process of paradigm change is
better understood as a gestalt switch or conversion. The scientist
simply begins to practice in the new paradigm. (Kuhn 1977, 338-339;
1970, 204) |
Sylvan and Glassner (1985, 103) suggest that a model ought to
be judged based on the fit between the assumptions of the theory
being modeled and the mathematics used to model the theory. Kuhn's
turn toward semantics to illustrate incommensurability suggests
that more interpretive modeling methodologies, such as artificial
intelligence, might be more suitable for simulating the deep
structure of Kuhnian paradigm change.**
™ Kuhn's role as a thesis advisor for Kenneth Haase, who is
doing a dissertation on automated discovery systems, indicates
that he is aware of such possibilities.
1844 System.Dynamics '90
Conclusion
Sterman concludes his article by noting that "(i]t is not
necessary to invoke either competition between theories or 'great
men' hypotheses to account for scientific revolutions." (1985, op.
cit, 118) He is correct. His reproduction of the behavior
characteristic of paradigm change indicates as much. But this
quotation is marked by both a presence and an absence. The presence
is of the word "necessary," the absence of the word "Kuhnian." If
Sterman's theory doesn't require great men and competitor theories,
can the same be said for Kuhn's? There is no clear answer to this.
While Kuhn never explicitly declares these forces to be necessary,
neither does he say they are unnecessary. They are simply there."
In some ways this is what makes him an historian rather then an
philosopher. But whether these forces are necessary or not is a
question of the validity of Kuhn's theory, not of Sterman's model.
Regardless of any necessity, great individuals and competitor
theories are indispensable, indeed constitutive of Kuhnian science.
Sterman most assuredly has a theory of scientific revolutions, but
not a model of Kuhnian paradigm change.
I am not suggesting that the model has no value; it does.
Indeed, it is a bold experiment, and as it now stands it is
splendidly representative of the feedback processes at work in
normal science as well as the dynamics of collective behavior
during paradigm change. It thus lends us extra confidence that
these parts of Kuhn's theory are sensible. Furthermore, its very
faults raise some fascinating questions about model construction
in general. The need to account for the role of individuals raises
the question of how it is possible to incorporate individual
behavior into a model representing collective behavior. There are
composition problems associated with having two very different foci
of analysis in the same model.’® And if the behavior can not be
understood in feedback terms, there is the further question of how
to create more "intelligent" system dynamics models. Merten (1988)
suggests the use of "intelligent logical loops" to model the
structure-transforming behavior characteristic of social evolution.
Insofar as this can generate qualitative behavioral change, it is
a step in the right direction.
BIBLIOGRAPHY
Barnes, Barry (1982), IT. S. Kuhn and Social Science. (N¥: Columbia
University Press).
Bell, James A. and Bell, James F. (1980), "System Dynamics and
Scientific Method." in Randers (1980).
18 He does assert that the functions performed by the
behavior he postulates are necessary if science is to flourish,
but leaves open the possibility that other behaviors might serve
similar functions. See Kuhn, 1970b, 237.
16 t am indebted to John M. Richardson, Jr. for pointing out
this implication of my argument.
System Dynamics '90 1345
Forrester, Jay W. (1961), Industrial Dynamics. (cambridge, Mass.:
MIT Press). z,
Forrester, Jay W. (1971), “Counterintuitive Behavior of Social
Systems." Technology Review, Volume 73, Number 3, January, 1971.
Forrester, Jay W. (1987), "Lessons From System Dynamics Modeling."
System Dynamics Review, Vol. 3, No. 2, Su. 1987.
Grant, Lindsey (1988), resi d oO cisions.
(University Press of America).
Gutting, Gary, ed. (1980), Paradigms and Revolutions. (Notre Dame:
University of Notre Dame Press).
Kuhn, Thomas S. (1957), The Copernican Revolution. (Cambridge,
Mass.: Harvard University Press).
Kuhn, Thomas S. (1970), Structure of Scientific Revolutions. 2nd
Edition. (Chicago: The University of Chicago Press).
Kuhn, Thomas S. (1970b), "Reflections on my Critics." in Lakatos
and Musgrave (1970).
Kuhn, Thomas S. (1977), The Essential Tension. (Chicago: The
University of Chicago Press).
Lakatos, Imre and Musgrave, Alan, eds. (1970), t.
Growth of Knowledge. (Cambridge: Cambridge University Press).
Meadows, Donella H. (1980), "The Unavoidable A Priori." in Randers
(1980).
Merten, Peter P. (1988), "Systems Simulation: The Simulation of
Social System Evolution with Spiral Loops." Behavioral Science,
Vol. 33, 1988, 131-157.
Randers, Jorgen, ed. (1980), ments of t! ni: cS:
Method. (Cambridge, Mass.: MIT Press).
Sterman, John D. (1985), "The Growth of Knowledge: Testing a Theory
of Scientific Revolutions with a Formal Model." Technological
Forecasting and Social Change 28, 93-122.
Sterman, John D. (1988), "A Skeptic's Guide to Computer Models."
in Grant (1988). 4 °
Sylvan, David and Glassner, Barry (1985), A_Rationalist Methodology
for the Social Sciences. (Oxford: Basil Blackwell).
The American Heritage Dictionary, 2nd College Edition, (Boston:
Houghton Mifflin, 1982). ' q r ¢
