In the Theoretician's Laboratory: Thought Experimenting as Mental Modeling

Author(s): Nancy J. Nersessian

Source: PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association,
Vol. 1992, Volume Two: Symposia and Invited Papers (1992), pp. 291-301
Published by: The University of Chicago Press on behalf of the Philosophy of Science Association
Stable URL: http://www.jstor.org/stable/192843 .
Accessed: 09/05/2014 13:34

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

The University of Chicago Press and Philosophy of Science Association are collaborating with JSTOR to
digitize, preserve and extend access to PSA: Proceedings of the Biennial Meeting of the Philosophy of Science
Association.

http://www.jstor.org

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
 In the Theoretician's Laboratory:
Thought Experimenting as Mental Modeling1

Nancy J. Nersessian

Princeton University

1. Introduction

After a long period of neglect there has been a recent wave of interest in thought
experiments in science, in mathematics, and in philosophy (See, e.g., Horowitz &
Massey, 1991). I will restrict my analysis to thought experiments as they function in
science, although I believe the account has implications for thought experiments, gen-
erally. The two most influential views on the topic in philosophy and history of sci-
ence in this century represent the extremes of empiricism and rationalism. Pierre
Duhem dismissed all thought experiments as bogus precisely because they are "not
only not realized but incapable of being realized" (Duhem 1914, p.202). That is, either
they can be turnedinto real experiments - and, thus the "thought"dimension is incon-
sequential-or they are to be dismissed because they are not "experimental"at all.
Alexandre Koyr6 (1939, 1968), on the other hand, arguedthat the idealizing function
of thought experiments is essential to scientific thinking. Idealizationis requiredfor
the "mathematization"of nature and this can only be carriedout in the mind, not in the
laboratory. Thus, Koyre concluded thought experiment supplantsreal-worldexperi-
mentationand demonstrates the synthetic a priori natureof scientific knowledge.

Historians have argued against Koyre mainly on the basis of evidence that Galileo
actually performed many of what he presented as thought experiments. This does not,
however, undermine the point that extrapolation to the limit and other forms of ab-
straction can only be carried out in thought. Philosophers of science, under the influ-
ence of logical positivism, have found Duhem's view most sympathetic and, until
quite recently, it has been the predominant position. Thought experiments, while
perhaps of psychological value, make no significant contributionto scientific reason-
ing. This stance is based on a limited conception of what constitutes "reasoning".
"Reasoning"is customarily taken to comprise applying formal rules of inference to
systems of propositions. However, a fuller account needs to extend the notion of rea-
soning to include the types of non-algorithmic inferences employed in a "reasoned
change of view" (Harman 1986, Nersessian 1988, 1992). Thought experimenting is a
principal means through which scientists change their conceptual structures. I pro-
pose that thought experimenting is a form of"simulative model-based reasoning".
That is, thought experimenters reason by manipulatingmental models of the situation

PSA 1992, Volume 2, pp. 291-301

Copyright ? 1993 by the Philosophy of Science Association

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
292

depicted in the thought experimental narrative. As I have discussed elsewhere

(1991a, 1991b, 1992), the narrativeform of presentationplays a central role in com-
municating a thought experiment within a community of scientists. Although my hy-
pothesis derives from examining the role of thought-experimentalnarrativesin theo-
retical reasoning practices, it is reinforced by David Gooding's presentationin this
volume, which derives from examining their role in experimentalpractices.

Briefly, my hypothesis is that what distinguishes thought experiments from logical

argumentsand other forms of propositionalreasoning is that reasoning by means of a
thought experimentinvolves constructingand making inferences from a mental simu-
lation. This is what makes a thought experiment both "thought"and "experimental".
The original thought experiment is the construction of a dynamical model in the mind
by the scientist who imagines a sequence of events and processes and infers out-
comes. She then constructs a narrativeto describe the setting and sequence in order
to communicate the experiment to others, i.e., to get them to construct and run the
correspondingsimulation and presumably obtain the same outcomes. Although lan-
guage is used to construct that simulation, the operations thought experimentersper-
form in executing the experiment are not on linguistic representations,but are on the
model the narrativehas enabled them to construct. While thought experimenting is a
truly creative part of scientific practice, the basic ability to construct and execute a
thought experimentis not exceptional. The practice is highly refined extension of a
common form of reasoning. It is rooted in our abilities to anticipate, imagine, visual-
ize, and re-experience from memory. That is, it belongs to a species of thinking by
means of which we grasp alternatives,make predictions, and draw conclusions about
potential real-world situations we are not participatingin at that time.

Curiously,the most comprehensive view of the natureand function of thought ex-

periments-that of Ernst Mach-has had little influence in philosophy and history of
science. Mach (1898, 1905) held that thought experiments are on a continuum with
real-world experiments and saw them as providing empirical data with epistemologi-
cal status comparableto that of real-world experiments. His argumentfor why the
outcomes of thought experiments have empirical import rested on sensationalist psy-
chology and evolutionary theory. In contemporaryparlance, Mach's approachwas
"naturalistic".He sought to provide the psychological and biological basis for the
epistemic status of the outcomes of thought experiments as empirical knowledge.
While the specific psychological and biological theories he relied on are outmoded,
his naturalisticapproachto an explanation of why and how thought experiments work
has much in common with the one taken here (See also, Sorrenson 1992).

I have been arguing for some time now that if we treat conceptual change not as
something inherentin languages or ideas, but as something accomplished by human
agents, then how human cognitive capacities and limitations facilitate and constrain
the practices scientists employ in conceptual innovation and change become pertinent
to philosophical analysis. From the perspective of methodological practices of scien-
tists, thoughtexperimenting has proven highly effective in numerous instances of
conceptual change. One of my concerns in developing an account of conceptual
change in science is to develop what Ronald Giere (1992) has called "the cognitive
foundations"for such model-based reasoning, which, on my analysis, includes the use
of analogical and visual models as well as thought experiments. Providing this foun-
dation will establish that these heuristics are not ancillary, dispensable aids to think-
ing-while the "real"reasoning takes place by deductive or inductive arguments-but
are reasoning methods essential to the practice of science. Recent work in cognitive
psychology is pertinentfor developing a framework in which to analyze the practice
of thought experimenting. Examining the literatureon mental modeling duringnarra-

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
 293
tive comprehension led to my initial proposals that the narrativeform of presentation
plays a significant role in the process of thought experimenting and that in one of its
functions the real-world experimental narrativeplays a similar role (1991a; 1991b).

2. "Mental Modeling" in Narrative Comprehension

In comprehending a thought-experimentalnarrativewe make use of cognitive

structuresand operations in common use that have been investigated in some detail
by cognitive psychologists. Reading, comprehending, and thinking about stories
would seem to epitomize thinking with language. Yet, there is a significant body of
cognitive research that supports the hypothesis that the inferences subjects make are
derived from constructing and manipulatinga mental model of the situation depicted
by the narrative,ratherthan by applying rules of inference to a system of propositions
representing the content of the text.
The contemporarynotion that mental modeling plays a significantrole in human
reasoning was formulated,initially, by Kenneth Craikin 1943. Craikproposed that
people reason, in general, by carryingout thoughtexperimentson internalmodels.
Since it was the heyday of behavioristpsychology when he proposedit, nothing much
was done to furtherthe hypothesis. The development of a cognitive psychology in the
1960s created a more hospitable environmentfor its articulationand exploration.
Though not uncontroversial,the centralityof mental modeling to cognition is a hypoth-
esis under investigation by many domains. The main impetus for the resurgence of the
hypothesis is experimentalresults that demonstratethe effect of semantic information
on reasoning (See, Johnson-Laird1983, for an extensive discussion). Mental modeling
has been investigated in a wide range of phenomenafrom thinkingabout causality in
physical systems (See, e.g., deKleer & Brown 1983) to reasoning with representations
of domain knowledge (See, e.g., Gentner& Stevens 1983) to analogicalreasoning (See,
e.g. Gentner & Gentner 1983) to deductive inferencing (See, e.g., Johnson-Laird1983)
to comprehendingnarratives(referencesbelow). Because the potentialrange of appli-
cation is so extensive, some have arguedthat the hypothesis of mental models can pro-
vide a unifying frameworkfor the study of cognition (Gilhooly 1986). I, too, find the
hypothesis attractivebecause it provides the possibility of furnishinga unified analysis
of the widespread modeling practices implicated in conceptualchange.

There are several distinct theoretical accounts of mental models that tend to be
conflated in the literature. The most significant distinction for our purposes is be-
tween those investigations that treat mental models as structuresstored in long term
memory and then called upon in reasoning and those that treatthem as temporary
structuresconstructed in working memory for a specific reasoning task. I am con-
cerned with the latter in this analysis, where the mental model is constructed from the
thought experimental narrativeand used in reasoning. Since Philip Johnson-Laird's
account is the best articulatedof those analyses that focus on the temporaryreasoning
structure,it will inform my discussion. In general terms, a mental model is a struc-
tural analog of a real-world or imaginary situation, event, or process that the mind
constructs to reason with. What it means for a mental model to be a structuralanalog
is that it embodies a representationof the spatial and temporal relations among and
the causal structureconnecting the events and entities depicted.

Further,although it is a matter of some debate whatformat a mental model takes

and what are the generative processes in the brain for creating and operating on men-
tal models, these issues do not have to be resolved before we can make progress on an
account of thought experimenting. The essential points are that a mental model is
non-propositional in form and the mental mechanisms are assumed to be such that

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
294

they can satisfy the model-building and simulative constraintsnecessary for the activ-
ity of mental modeling. I cannot go deeply into the "format"issue here, but to ally
possible objections to the "image-like" nature of such models, I want to stress that
most researcherswould concur in the view that mental modeling, even if it does make
use of the mechanisms of the visual cortex, is not like constructinga picture in the
mind. That great thought experimenters, such as Bohr, have claimed not to be able to
visualize well does not undermine my claim that thought experimentingis mental
modeling. Mental modeling does not require introspective access to an image in the
"mind's eye". It only requires the ability to reason by means of an analog model.
The relationship between a mental model and what has been called "mentalimagery"
is something that still needs to be worked out.

Advocates of mental modeling argue that the original capacity developed as a way
of simulating possible ways of maneuvering within the physical environment. It
would be highly adaptive to possess the ability to anticipatethe environmentand pos-
sible outcomes of actions, so it likely that many organisms have the capacity for men-
tal modeling. It is also likely that humans have the possibility of creating models
from both perception and description. Additionally, there is mounting evidence from
neuropsychology that the perceptual system plays a significant role in imaginative
reasoning (See, e.g., Farah 1988). Again, this makes sense from an evolutionary per-
spective. The visual cortex is one of the oldest and most highly developed regions of
the brain. As Roger Shephard, a psychologist who has done extensive research on vi-
sual cognition, has put it, perceptual mechanisms "have, throughevolutionaryeons,
deeply internalized an intuitive wisdom about the way things transformin the world.
Because this wisdom is embodied in a perceptual system that antedates,by far, the
emergence of language and mathematics, imagination is more akin to visualizing than
to talking or to calculating to oneself' (1988, p.180). The point is that contraryto the
picture philosophers have constructed of the inferior status of visual modes of think-
ing, the visual cortex is a more highly evolved portion of the brain (Wimsatt 1990).
Although the original ability to envision by mental modeling may have developed as
a way of simulating possible courses of action in the world, it is highly plausible that,
as human brains have developed, this ability has been "bentto the service of creative
thought"(Shephard 1988, p.180).
In the domain of research into narrativecomprehension the mental models hypoth-
esis is that in understandingthe meaning of a narrative,the linguistic expressions as-
sist the reader/listenerin constructing a mental model and in reasoning about the situ-
ation depicted by the narrativethrough this model. Johnson-Lairdin psycholinguis-
tics and others in formal semantics, linguistics and AI have proposed a theory of "dis-
course models" for narratives. The central idea is that "discourse models make ex-
plicit the structurenot of sentences but of situations as we perceive or imagine them"
(Johnson-Laird1989, p.471). The principal tenets of the theory are: (1) the referent
of a discourse is the situation the discourse describes; (2) the meaning of a discourse,
i.e. the set of all possible situations it could describe, comprises both the linguistic
representationand the mental mechanisms for constructingand runningmental mod-
els; and (3) if a discourse has at least one model that can be embedded in a model of
the world it is judged to be true (p.475).

As a form of mental model, a discourse model embodies a representationof the

spatial, temporal, and causal relationships among the events and entities of the narra-
tive. In constructing and updating a model, the readercalls upon a combination of
pre-existing conceptual and real-world knowledge and employs the tacit and recursive
inferencing mechanisms of her cognitive apparatusto integrate this with the informa-
tion contained in the narrative. A number of experiments have been conducted to in-

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
 295

vestigate the hypothesis that in understandinga narrativereaders spontaneously con-

struct mental models to represent and reason about the situations depicted by the text
(Franklin& Tversky 1990; Johnson-Laird 1983; Mani & Johnson-Laird 1982;
McNamara & Sternberg 1983; Morrow et al. 1989; Perrig & Kintsch 1985).
Although no instructions are given to imagine or picture the situations, when queried
about how they had made inferences in response to the experimenter's questioning,
most subjects reportit was by means of "seeing" or "being in the situation"depicted.
That is, the reader sees herself as an "observer". Whether the view of the situation is
"spatial",i.e., a global perspective, or "perspectival",i.e., from a specific point of
view, is currentlyunder investigation.
Some of the experimental evidence for the hypothesis of mental modeling during
narrativecomprehension comes from chronometric studies which claim to show that
model-based reasoning is faster than reasoning with propositions. A situation that is
representedby a mental model should allow the reasoner to generate conclusions
without having to carry out the extensive operations needed to process the same
amount of backgroundinformation to make inferences from an argumentin proposi-
tional form. The situational constraints are built into the model, making many conse-
quences implicit in it that would requireconsiderable inferential work in propositional
form. For example, moving an object changes, immediately, its spatial relationships
to all the other objects. The reasoner would grasp this simply by means of the
changes in the model and not need to make additional inferences. Further,reasoning
through a model should restrict the scope of the conclusions drawn. For example,
moving an object in specified manner both limits and makes immediately evident the
consequences of that move to those directly relevant to the situation depicted by the
narrative. Thus other support comes from demonstrations that inferences subjects
make are much more difficult or not made at all when they are required to reason with
the situation reformulatedpropositionally.

3. Characteristicsof Thought Experiments

Specific features of thought experiments have led me to propose that thought-ex-

perimental narrativesfunction in much the same way as other narratives. It is not
possible within the confines of this paperto present specific thought experiments.
There is great variety among them and it would be a difficult task to construct a list of
all their salient features. Nor need any one thought experiment exemplify all possible
features. James Brown's taxonomy of thought experiments provides a useful classifi-
catory schema (Brown 1991; See also, Anapolitanos 1991). The features I list are
ones that are pertinentto understandingthem as a species of simulative model-based
reasoning and cut across these categories.
Feature 1: By the time a thought experiment is public it is presented in the form
of a narrative. Occasionally, presentations of thought experiments also include
some form of visual illustration, so my use of the term "narrative"is to be taken as
broad enough to encompass these. The narrativehas the characterof a simulation.
It calls upon the reader/listenerto imagine a dynamic scene: one that unfolds in
time and follows a specific causal sequence.

Feature 2: The readeris invited to follow througha sequence of events or process-

es as one would in the real world. Thatis, even if the situation may seem bizarreor
fantastic, such as being in a chest in outer space, there is nothing bizarrein the un-
folding. Objects behave as they would in the real world in the presence or absence
of gravity. The assumptionis that if the experiment could be performed,the chain
of events would unfold according to the way things usually take place in the world.

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
296
Feature 3: By the time a thought experiment is communicated it is in a polished
form. That is, we are not shown the tinkering that went into setting it up and refin-
ing it. The reader/listenerrarely, if ever, gets a glimpse of failed thought experi-
ments or avenues explored in constructing the one presented to them. When con-
veyed, a thought experiment is as packaged and polished as a real-world experi-
ment is when it is published. This does not mean that it cannot fail (See, Janis
1991). Nor does it mean that it will be effective in establishing agreement on the
part of the reader. The reader may execute it incorrectly or lack the competence to
perform it at all. It may take some time to grasp the consequences of a subtle ex-
periment and its interpretation. Further,the significance of a thought experiment
is sometimes a matterof prolonged debate within the community.

Feature 4: The thought-experimentalnarrativedepicts abstractions. For exam-

ple, certain features of objects that would be present in a real-world experiment are
not included, such as the color of rocks and the physical characteristicsof ob-
servers. That is, there has been a prior selection of the pertinentdimensions on
which to focus that evidently derives from experience in the world. From experi-
ence we believe, e.g., that the color of a rock does not effect its rate of fall. Such
information is customarily excluded from real-world experimental narrativesas
well. This facilitates the reader's recognition of the situation as prototypical, i.e.
as representing a class of experimental situations.

It is true that, as John Norton (1991) has pointed out, extremely colorful narratives
may include highly specific details. Ratherthan being "irrelevant",as he main-
tains, though, these details usually serve to reinforce crucial aspects of the experi-
ment. For example, in one version of the chest, or "elevator",experiment,
Einstein depicts the physicist as being drugged and then waking up in a box. This
colorful detail serves to reinforce the point that the observer could not have known
before entering the chest if he were falling in outer space or sitting in a gravitation-
al field. It also reinforces the condition that the observer cannot know whether or
not there are gravitational sources around.

Feature 5: A thoughtexperimentis usually so compelling thateven in those cases

where it is possible to carryit out, the readerfeels no need to do so. The constructed
situation,itself, is apprehendedas pertinentto the real world in several ways. It can
reveal something in our experience thatwe did not see the importof before, such as
that the measurablecurrentin a stationaryand in a moving conductorcannot support
the distinctionmade in the theoreticalexplanationof them as differentphenomena. It
can generatenew datafrom the limiting case, for example, thatin no medium lead
and wood would fall at the same speed. It can make us see the empiricalconse-
quences of something in our existing conceptions, such as thatthe attributescalled
"gravitationalmass" and "inertialmass"are the same propertyof bodies.

4. Thought Experimenting as Mental Modeling

We can only speculate about what goes on in the mind of the scientist who devises
the original thought experiment. Scientists have rarely been asked to discuss the de-
tails of how they formulated these experiments. However, reports of thought experi-
ments are presented, customarily, in the form of narratives. Because the thought-ex-
perimental narrativesare what we have access to and because they are a central form
of effecting conceptual change within a scientific community, I propose to examine
how the narrativesfunction. From that analysis we can infer that the original experi-
ment involves a similar form of reasoning. To explicate the notion that thought exper-
imenting is simulative model-based reasoning, we need to determine:(1) how a narra-

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
 297
tive facilitatesthe constructionof a modelof an experimental
situationin thoughtand
(2) how one canreachconceptualandempiricalconclusionsby mentallysimulating
the experimentalprocesses.

Accordingto the mentalmodelshypothesisdiscussedabove,thefunctionof the

narrativeformof presentationof a thoughtexperiment(Feature1) is to guidetheread-
er in constructinga structural
analogof the situationdescribedby it andto makeinfer-
encesthroughsimulatingtheeventsandprocessesdepictedin it. So, as withother
formsof discoursemodels,the operationscarriedoutin executingthethoughtexperi-
mentareperformednot on propositionsbuton theconstructed model. Unlikethe fic-
tionalnarrative,however,thecontextof the scientificthoughtexperimentmakesthe
intentionclearto thereaderthatthe situationis one thatis to representa potentialreal-
worldsituation(Feature2). Thatthoughtexperimentis presentedin a polishedform
(Feature3) shouldmakeit an effectivemeansof gettingcomparable mentalmodels
amongthemembersof a communityof scientists.Thenarrative has alreadymadesig-
nificantabstractions (Feature4), whichaidsin focusingattentionon the salientdimen-
sionsof themodelandin recognizingthe situationas prototypical.Thus,theexperi-
mentalconsequencesgo beyondthe specificsituationof thethoughtexperiment.
Whilesomekindsof mentalmodelingmayemploystaticrepresentations, those
derivedfromthought-experimental narrativesareinherentlydynamic.Thesituations
undergodevelopment.The narrativedelimitswhicharethespecifictransitionsthat
governwhattakesplace. In constructingandconductingtheexperiment,we use in-
ferencingmechanisms,existingrepresentations, andscientificandgeneralworld
knowledgeto makerealistictransformations fromonepossiblephysicalstateto the
next. Muchof whatwe employin thesetransformations is tacit. Thus,expertiseand
learningplaya crucialrole in thepractice;as does whatGoodingcalls "embodiment"
(1990, 1992). Theconstructedsituationinheritsempiricalforceby beingabstracted
frombothourexperiencesandactivitiesin, andourknowledge,conceptualizations,
andassumptionsof, the world(Features5). In thisway,thedatathatderivefrom
thoughtexperimentinghaveempiricalconsequencesandat the sametimepinpointthe
locus of theneededrepresentationalchange. Thisunderstanding formsthebasisof
problem-solving effortsto constructan empiricallyadequateconceptualization.
WhileI agreewith Norton(1991) thatthoughtexperimentscan oftenbe recon-
structedas arguments,the modelingfunctioncannotbe supplantedby an argument.
As Nortonacknowledges,the argumentcan be constructedonly afterthe fact. That
is, on my account,the argumentis notevidentuntilafterthe thoughtexperimenthas
beenconstructedandexecuted. Exhibitingthe soundnessof a thoughtexperimentby
reconstructingit as an argumentcanperforman important rhetoricalfunction.
However,real-worldexperimentaloutcomescanbe recastin argumentformas well,
butno one wouldarguethatthe experimentcan be replacedby theargument.In simi-
larfashion,we need to differentiatebetweenthereasoningthatis donewiththe
thoughtexperimentandthatwhichis donewith thereconstruction of it. Onmy view,
thoughtexperimentingis a complexformof reasoningthatintegratesvariousformsof
information-propositions,models,andequations-into dynamicmentalmodels. By
linkingtheconceptualandthe experientialdimensionsof humancognitiveprocess-
ing, thoughtexperimentingdemonstrates the undesirable
real-worldconsequencesof
a representation,therebycompellingrepresentationalchange.
4. ConcludingRemarks
In concludingI wantto raisesomeissuesthatwill requirefurtheranalysisin lightof
theinterpretation
of thoughtexperimentingas simulativemodel-based reasoning.

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
298
4.1 Thought Experimenting and Idealization

Idealization is a significant dimension of thought experimenting, but focusing on

this one dimension has led to misconstruals of its nature and function. Thought-world
models are abstractionsbut idealization is only one form of abstraction(Nersessian,
in press). Limiting case analysis is a form of idealization employed frequently in
thought experimenting. In this species of thought experimenting, the simulation con-
sists in abstractingspecific physical dimensions to create an idealized representation.
Isolating the physical system in thought allows us to manipulatevariables beyond
what is physically possible and this creates data we did not possess before.

Idealization is not, however, the most salient dimension of the reasoning done with
thought experiments. It is more significant that the thought experiment is understood
to representa prototypical occurrence of a situation. This is what gives the outcomes
their generality and contributesto the impact of the experiment. Although the thought
experimenterconstructs a single model, its significance for a whole class of phenome-
na and situations is grasped in its execution. The thought-experimentalmodel has as-
pects that are generic and others that are highly specific. For example, in Galileo's
thought experiments with falling bodies any objects will do. The color and shape of
the objects are not significant. The specific weights are also not salient, but that one
object weighs more than the other needs to be specified. Most importantly,in thought
experimenting the causal sequences are usually highly specific.
4.2 Thought Experiments and Real-world Experiments

Reformulatingboth kinds of experiments in argument form can help in persuading

others and in justifying conclusions about the experimental outcomes. But, the mod-
eling function of a thought experiment cannot be eliminated in favor of an argument,
any more than the real-world experiment can be replaced by an argument.
Further,there is a significantconnection between the two types of experimentalnar-
rative (Nersessian 1991b). Recently, sociologists of science have pointed out thatearly
experimentalnarratives,i.e. those that helped to establish the practiceof communicating
experimentalresults, were much more richly detailed than is now customary(Dear 1985,
Shapin 1984). In discussing Boyle's "literarytechnology", Steven Shapin (1984) claims
thatthe style of the experimentalnarrativereflects the circumstancesthat Boyle was at-
temptingto gain authorityfor his results by creating a way of "witnessing"an experi-
ment while not being present To achieve this purpose, "virtual"witnessing needed to
create an "impressionof verisimilitude",which Shapin interpretsas conveying authority
and compelling assent. From the mental models perspective developed here, thatthe
narrativesfunctionedto assist his readersin constructingtheir own mental simulation,
therebycreatingan understandingof what they did not actually witness themselves, is a
significantfactorin why Boyle's narrativeswere effective as rhetoricaldevices.
In contrast,Larry Holmes (1990) has pointed out that the experimental narratives
produced by the members of the Acadermiedes Sciences during the same historical pe-
riod are much more succinct and quite similar to the modem experimental narrative.
He argues that this stems, in part, from the practice of French Academicians of carry-
ing out communal investigations, so authoritywas not in question. We have seen that
tacit and explicit community knowledge and practices figure significantly in the men-
tal modeling process. Thus, where there is a highly developed community of experi-
mental practitioners,a more succinct experimental narrativewill be effective.
Knowledge of procedures and apparatuscan be presumed and are part of what the
community of readers will interpolate into their mental models of the experiment.

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
 299
4.3 Thought Experimenting and Conceptual Change

In his influential 1964 essay, Thomas Kuhn characterizedthought experimenting as

"one of the essential analytical tools which are deployed duringcrises and which then
help to promote basic conceptual reform"(Kuhn 1977, p.263). The historical record
does indeed show the preponderanceof thought experiments in periods of conceptual
change in science. But, to understandwhy requires a fundamentalrevision in how we
conceive of conceptual change, which I can only sketch here. The basic ingredients of
that revision are to represent a concept by a system of constraintsfor generating the
members of a class of models and to view a conceptual structureas an agglomeration
of such constraint systems (Nersessian, in process). Thought experimentingplays a
crucial role in conceptual change by showing that existing systems of constraintscan-
not be integrated into consistent models of the physical world. This process involves
integratingconstraints derived from existing conceptual structures,mathematicalrep-
resentations, and the world. The thought experimentpinpoints the locus of the needed
conceptual reform, often providing the basis from which to constructa new representa-
tion. Thought experimenting may facilitate recognizing the undesirableconsequences
of our conceptualizations in much the way that experimenting by computer simulation
exposes undesirable consequences of the constraintsof a scientific representation. By
creating a simulative model that attempts to integrate specific systems of constraints,
thought experimenting enables the scientist to grasp essential points of conflict and
infer their consequences more readily than would reasoning throughthe logical conse-
quences of a representation. Once the initial experimenterunderstandsthe implica-
tions of a thought experiment, she can guide others in the community to see them as
well by crafting a description of the experiment into a narrative.

Notes

1I acknowledge and appreciate the supportof NSF Scholars Awards DIR 8821422
and DIR 9111779 in conducting the research discussed here.

References

Anapolitanos, D.A. (1991), "ThoughtExperiments and Conceivability Conditions." in

Horowitz & Massey, pp. 87-98.

Brown, J.R. (1991), The Laboratory of the Mind. London: Routledge.

Craik, K. (1943), The Nature of Explanation. Cambridge:Cambridge University

Press.

Dear, P. (1985), "Totisin verba: Rhetoric in the Early Royal Society." Isis 76: 145-
161.

De Kleer, J. and Brown, J.S. (1983), "Assumptions and Ambiguities in Mechanistic

Mental Models." in Gentner & Stevens, pp. 155-190.

Duhem, P. (1914), The Aim and Structureof Physical Theory. (Trans.)P. Wiener.
Princeton: Princeton University Press, 1954.

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
300
Farah,M.J. (1988), "Is Visual Imagery Really Visual? Overlooked Evidence from
Neuropsychology." Psychological Review 95: 307-317.
Franklin,N. and Tversky, B. (1990), "Searching Imagined Environments."Journal
of ExperimentalPsychology 119: 63-76.

Gentener,D. and Gentner, D.R. (1983), "Flowing Watersor Teaming Crowds: Mental
Models of Electricity." in Gentner & Stevens, pp. 99-130.

Gentner,D. and Stevens, A.L. (1983), Mental Models. Hillsdale: Lawrence Erlbaum.

Giere, R. (1992), "Concepts, Categories, and Scientific Theories." unpublished

manuscript.

Gilhooly, K.J. (1986), "Mental Modelling: A Frameworkfor the Study of Thinking."

in Thinking:Progress in Research and Teaching, (eds.) J. Bishop, J.
Lochhead, and D. Perkins. Hillsdale, NJ: Lawrence Erlbaum,pp. 19-32.

Gooding, D. (1990), Experiment and the Making of Meaning. Dordrecht:Kluwer

Academic Publishers.

_ _ _ _ _ __. (1992), "The ProceduralTurn; or Why do Thought Experiments Work?"

in Cognitive Models of Science, (ed.) R. Giere. Minnesota Studies in the
Philosophy of Science 15. Minneapolis: U. of Minnesota Press, pp. 45-76.
Harman,G. (1986), Change in View. Cambridge: MIT Press.

Holmes, F.L. (1990), "Argumentand Narrative in Scientific Writing."in The Paper

Laboratory, (ed.) P. Dear. Philadelphia: University of Pennsylvania Press, pp.
164-181.

Horowitz, T. and Massey, G.J. (1991), Thought Experimentsin Science and

Philosophy. Savage, MD.: Rowman and Littlefield.

Janis,A. (1991), "CanThought ExperimentsFail?"in Horowitz& Massey, pp. 113-118.

Johnson-Laird,P.N. (1983), Mental Models. Cambridge:HarvardUniversity Press.

__ ?_____ _ _ . (1989), "Mental Models." in Foundations of Cognitive Science,

(ed.) M. Posner. Cambridge: MIT Press., pp. 469-500.

Koyre, A. (1939), Galileo Studies. Atlantic Highlands: Humanities Press, 1979.

?_____. (1968), Metaphysics and Measurement. Cambridge:HarvardUniversity
Press.

Kuhn, T.S. (1964), "A Function for Thought Experiments."in The Essential Tension:
Selected Studies in Scientific Tradition and Change. Chicago: University of
Chicago Press, 1977, pp. 240-265.
Mach, E. (1898), "On Mental Adaptation."in Popular Scientific Lectures, fifth edi-
tion, (trans.) T. McCormack. LaSalle, II.: Open Court, 1943, pp. 214-235.

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions
 301
_ _ _ __ . (1905), "On Thought Experiments." in Knowledge and Error. Dordrecht:
D. Reidel, 1975, pp. 134-147.

Mani, K. and Johnson-Laird,P.N. (1982), "The Mental Representation of Spatial

Descriptions." Memory and Cognition 10: 181-187.
McNamara, T.P. and Sternberg,R.J. (1983), "Mental Models of Word Meaning."
Journal of VerbalLearning and VerbalBehavior 22: 449-474.

Morrow, D.G., Bower, G.H., and Greenspan, S.L. (1989), "Updating Situation Models
during Narrative Comprehension." Journal of Memory and Language 28:
292-312.

Nersessian, N.J. (1988), "Reasoning from imagery and analogy in scientific concept
formation."in PSA 1988, (eds.) A. Fine & J. Leplin. East Lansing, MI.,
Philosophy of Science Association.

_____-. ? (1991a), "Why do thoughtexperiments work?"Proceedings of the

___
Cognitive Science Society 13. Hillsdale, NJ: Lawrence Erlbaum,pp. 430-38.

___?______._ (1991b), "The Cognitive Sciences and the History of Science," in

Conference on Critical Problems and Research Frontiers in the History of
Science and Technology.Madison, WI: HSS and SHOT,pp. 92-116.

_________. (1992), "How Do Scientists Think? Capturingthe Dynamics of

Conceptual Change in Science." in Cognitive Models of Science, (ed.) R.
Giere. Minnesota Studies in the Philosophy of Science 15. Minneapolis: U. of
Minnesota Press, pp. 3-44.

__ _ _ __. in press. "Abstractionvia Generic Modeling in Concept Formation

in Science." in Idealization in Science, (eds.) N. Cartwright& M.R. Jones.
Amsterdam:Editions Rodophi.

__ _ _ __ in process. Creativity and Conceptual Change: A Cognitive

Constructivist View. (to be published by MIT/BradfordBooks).

Norton, J. (1991), "ThoughtExperiments in Einstein's Work."in Horowitz and

Massey, pp. 129-148.

Perrig, W. and Kintsch, W. (1985), "Propositionaland Situational Representations of

Text."Journal of memory and Language 24:503-518.

Shapin, S. (1984), "Pumpand Circumstance:Robert Boyle's Literary Technology."

Social Studies of Science 14: 481-520.

Shephard,R. (1988), "The Imagination of the Scientist." in Imagination and the

Scientist, (eds.) K. Egan and D. Nadaner.NY: Teachers College Press, pp. 153-
185.

Sorenson, R. (1992), ThoughtExperiments. Oxford: Oxford University Press.

Wimsatt, W. (1990), "Tamingthe Dimensions-Visualization in Science." in PSA

1990, vol. 2, (eds.) A. Fine, M. Forbes, and L. Wessels. East Lansing, MI.:
PSA, pp. 111-135.

This content downloaded from 194.29.185.252 on Fri, 9 May 2014 13:34:17 PM

All use subject to JSTOR Terms and Conditions