See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/232993588

Mental Models and Thought Experiments

Article  in  International Studies in the Philosophy of Science · January 1992

DOI: 10.1080/02698599208573432

CITATIONS READS

57 136

1 author:

Nenad Miscevic
University of Maribor
103 PUBLICATIONS   218 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Radicalisation and Violent Extremism (2018-2021) View project

curiosity View project

All content following this page was uploaded by Nenad Miscevic on 27 May 2019.

The user has requested enhancement of the downloaded file.

This article was downloaded by: [New York University]
On: 12 May 2015, At: 14:12
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Studies in the

Philosophy of Science
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/cisp20

Mental models and thought

experiments
a
Nenad Miščević
a
I.G. Kovacica 26, Rijeka, 5100, Croatia
Published online: 09 Jun 2008.

To cite this article: Nenad Miščević (1992) Mental models and thought
experiments, International Studies in the Philosophy of Science, 6:3, 215-226, DOI:
10.1080/02698599208573432

To link to this article: http://dx.doi.org/10.1080/02698599208573432

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information
(the “Content”) contained in the publications on our platform. However, Taylor
& Francis, our agents, and our licensors make no representations or warranties
whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and
views of the authors, and are not the views of or endorsed by Taylor & Francis. The
accuracy of the Content should not be relied upon and should be independently
verified with primary sources of information. Taylor and Francis shall not be liable
for any losses, actions, claims, proceedings, demands, costs, expenses, damages,
and other liabilities whatsoever or howsoever caused arising directly or indirectly in
connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.
Terms & Conditions of access and use can be found at http://www.tandfonline.com/
page/terms-and-conditions
 INTERNATIONAL STUDIES IN THE PHILOSOPHY OF SCIENCE, VOL. 6, NO. 3,1992 215

Mental models and thought experiments

NENAD MIŠČEVIĆ
I.G. Kovacica 26, 5100 Rijeka, Croatia

Introduction
Downloaded by [New York University] at 14:12 12 May 2015

People are generally good at reorganizing things 'in their heads'.1 A prospective buyer
perusing a Sears & Roebuck catalogue and contemplating a purchase of a new piece of
furniture, say a large sofa, might indulge on the spot in planning how to rearrange the living
room. She or he would effortlessly move the pieces of old furniture around, until enough
room is made for the large new sofa—taking into account myriads of important details of
what supports what, what can be placed near the fireplace, and what should be kept far
from fire, etc. Psychologists are quite interested in how people manage to do that. A
popular and plausible answer is that our buyer builds or has built a 'mental model' of
the room, and so to speak, moves furniture 'in the model'. This way of manipulating
representations is called 'reasoning in the mental model'.
A prospective buyer who is otherwise interested in science, might have come across
some of the famous thought experiments in physics where the subject is typically invited to
'imagine an experimental situation' which is being described, e.g. to 'picture yourself as a
freely falling body, or a chain placed over a triangular beam, and then asked to imagine
various things happening to these objects, or to 'do things' with them—link the chain,
rotate an imagined object mentally, etc. Eventually the subject reaches a conclusion and
'sees' that the body will fall very slowly, or that the chain will stay still, etc.
I believe that the thought processes involved in the two lines of reasoning—one
concerning the furniture, and the other concerning the 'experimental situation' in arm-
chair physics are essentially the same. As the saying goes, science is refined common
sense.
The view that thought experiments are sophisticated 'remodelling in the head'
belongs to the tradition which sees thought experiments as processes of reorganizing or
rearranging the old pieces of knowledge—old experiences, beliefs and the like. I concur
with the tradition in believing that the insight yielded by such experiments is a posteriori.
The thought experimenter is allowed to use deduction, inductive reasoning and math-
ematics (usually arithmetic and geometry), to freely imagine things and then rearrange
these imagined items. The material with which the subject works is experiential, and so his
or her conclusions will be, in the last resort derived from experience. His or her physics will
be of the usual, a posteriori sort.
There is, however, another tradition which sees thought experiments as yielding a
priori knowledge. The armchair physics is a priori, this Platonist tradition claims.
The best-known defender of the uncompromising Platonism about thought exper-
iments is Alexandre Koyre. Recently, the Canadian philosopher James Brown has pub-
lished a sophisticated defence of Platonism. He gives a detailed and inspiring description of
 216 N. MIŠČEVIČ

the task of explaining key characteristics of thought experiments, and I shall take his
description as the foil against which to develop the mental-model account of thought
experiments, so I shall treat him as 'the Platonist' tout court.
We are offered a set of alleged data or phenomena about thought experiments and the
Platonist is claiming that these data cannot be explained in non-apriorist fashion. Since I
accept all the alleged data except one, I shall henceforward simply call them 'data'. I shall
also accept the description and attribution of two paradigmatic thought experiments given
by J. Brown—I am no historian of science, and in any case I think it is the best policy to
secure the common bases for the debate about issues of principle.
I want to show that the view of thought experiment as 'rearranging things in the head',
or, to put it a bit more technically, reasoning in mental models, can accommodate all the
relevant data within non-apriorist framework. Without the theory of reasoning in mental
model, the non-apriorist did not have the necessary instruments at his or her disposal—
non-apriorists like Kuhn talk vaguely or metaphorically about seeing old data in new light,
Downloaded by [New York University] at 14:12 12 May 2015

and even propose the extremely implausible view that the thought experimenter must
already have had the experience described in the experiment, all this because of the lack of
adequate psychological theory.

Examples: the free fall and the linked chain

Let me recount two key examples of thought experiments, on which our Platonist, J.
Brown, builds his challenge to non-Platonists.
The first is Galileo's thought experiment, recounted by Salviati-Galileo to Simplicio
in Discorsi. Salviati wants to show that all bodies, regardless of their weights, fall at the
same speed. He begins by noting that on Simplicio's suppositions heavier bodies fall faster
than light ones. Then, he bids Simplicio to consider what happens when we tie the heavier
and the lighter body together: "If then we take two bodies whose natural speeds are
different, it is clear that on uniting the two, the more rapid one will be partly retarded by the
slower, and the slower will be somewhat hastened by the swifter" (Galileo, p. 158). It
seems then that the combined system will be slower than the heavy body alone. But, on the
other hand, the combined system is heavier than the heavy body alone, so it should fall
more quickly than the heavy body alone. The supposition that weight is relevant to speed
leads to absurdity. So, Salviati can conclude that the difference in weight is irrelevant to the
speed. Let us call this the intermediate conclusion. From this conclusion, Salviati jumps to
the final conclusion: all bodies fall at (approximately) the same speed.
I have closely followed J. Brown's reading of Galileo. In addition, I have explicitly
separated the intermediate from the final conclusion, in order to underscore at the very
start the point to which the Platonist will later want to draw our attention.
One might be tempted to see at least the first part of Galileo's story as a concealed
reductio argument. The peculiar flavour of'experiment' as opposed to deductive reasoning
is better suggested by the second example.
The second example concerns weight resting on a plane. Suppose a chain is draped
over a prism-like pair of inclined (frictionless) planes of unequal slope. Will the chain
move?
Simon Stevin says no, and proposes a thought experiment. Suppose you link the
chain, adding links at the bottom. The closed loop "would rotate if the force on the left
were not balanced by the force of the right. Thus, we would have made a perpetual motion
machine, which is presumably impossible" (Brown, 1991, p. 4).
 MENTAL MODELS AND THOUGHT EXPERIMENTS 217

The challenge: the virtues of armchair physics

Let us take as the starting point for our discussion of thought experiments the following
interesting, and in my opinion quite correct observation:
The burden of any thought experiment rests on the establishment (in the
imagination) of a phenomenon. Once the phenomenon is established, the infer-
ence to a theory is fairly unproblematic; that is, the jump from data to theory is
relatively small (Brown, 1986, p. 4)
The small jump can lead very far—in the most famous cases it is a small jump for the
experimenter but a big leap for science (this does not mean that the theory suggested by
successful thought experiments is accepted without testing—quite the contrary is the
case). The contrast between the importance of the jump and the ease and quickness of it is
crucial for the Platonist argument. Our Platonist analyses the features of the jump and
Downloaded by [New York University] at 14:12 12 May 2015

challenges the opponent to account for them in empiricist fashion, implying, of course, that
such an empiricist account will not be forthcoming. The Platonist is then expected to win
by default—he or she was the only game in town.
I shall try to summarize the main points of the Platonist argument (as presented in the
section entitled "Why a priori?" of Brown, 1986), regrouping them a bit so as to stress
those elements which present the challenge to the empiricist.
The first point is that in the thought experiment "there has been no new observational
data" (Brown, 1986, p. 11). The experimenter has obviously learned something through
the experiment—if there were no new data, how could anything new be learned?
At this point, the empiricist might be tempted to present learning by thought exper-
iments as deductive learning. Let us assume (but only for the sake of argument) that there
are purely propositional ways of recounting the reasoning performed in the experiment,
which do not involve the use of imagining, or 'picturing in front of one's inner eye', but
consist of sentences or formulas only. (I am not sure that such an assumption makes sense
at all, but the deductivist is committed to claiming that the reasoning in thought exper-
iment can be captured by such ways.) We need to imagine the propositional alternative as
streamlined as possible, with reasoning made explicit and formalizable, if not outright
formalized. The 'deductivist' must claim that learning by thought experiment in fact
reduces the grasping dimly and maybe unconsciously something like the regimented
alternative. This is clearly implausible. Such purely propositional alternatives might not
be available for the enquirer who is inventing or performing a thought experiment. (For
example, Stevin might not have known any other way to prove that the system he is
describing is in equilibrium, than the way of his thought experiment, but latter day
physicists have such other ways at their disposal.) Furthermore, a regimented alternative
is heuristically of much less value than the 'experimental' alternative. (J. Brown gives in
The Laboratory of the Mind a nice example of geometrical thought experiment which is
unmanageable in any regimented alternative.) So, we may agree with our opponent that the
purely deductivist escape is no solution for the anti-Platonist.
So here is the second point: we face the challenge—to explain what makes thought
experiments superior to a deductive alternative (if there is one) in respect of heuristic power,
obviousness and ease.
The third point is that the learning by thought experiments is 'not a case of seeing old
empirical data in a new way'.
The point cuts two ways. First, there are cases of successful thought experiments in
which there was no old way of seeing data, and J. Brown seems specially to insist upon this
 218 N. MIŠČEVIČ

(since the proposal criticized is Kuhn's, Brown also spends some time on showing that the
more general picture of paradigm change does not cover what is essential for thought
experiments). Second, most data offered in thought experiments are not just 'old data'—
the contraptions of Stevin or Einstein are not things we have been seeing all our lives,
nor are quasi-experiences they are recommending the experiences we have already had,
waiting only for a fresh look, to teach us new things.
The fourth point needs some elaboration. (It is not stated among the main arguments
in Brown's list, but seems to be pervasive although implicit in the whole discussion.) To
put it bluntly, learning by thought experiments does not look like normal inductive learn-
ing ('inductive' being taken in the widest sense). Commenting on the jump from data to
theory, Brown writes: "There seemed to be no problem of an inductive leap; if we got the
phenomenon right then the theory followed more or less automatically" (Brown, 1986,
p. 13). We are not making 'an inference based on the sense perception of a finite number of
instances of the universal and then having to generalize from that'.
Downloaded by [New York University] at 14:12 12 May 2015

I would group Brown's question about Galileo's thought experiment together with
these considerations. Remember that the experiment yields the intermediate conclusion,
i.e. that weight is irrelevant for speed of free-falling bodies. But how does one reach the
final conclusion, i.e. that all bodies fall at the same speed? " . . . all bodies need not fall with
the same speed. They might fall at a different rate due to their color, or as has been recently
argued, due to their chemical composition" (Brown, 1986, p. 12).
What I see as a real challenge is to explain the ease and the unproblematic character of the
jump from the intermediary to the final conclusion. Do people unconsciously weight and
eliminate alternatives? This, if properly done, seems to be an enormous task, certainly not
to be performed in a few seconds and with no apparent effort.
One can put the same problem in a slightly different way which brings it more into line
with the traditional approach to methodology of science. If the experimenter wants to get a
specific answer to the experimental question, he or she has to isolate potentially relevant
factors, keeping the rest fixed, and then see how the experimental set-up behaves. Put in
this way, the precept amounts to the requirement of maximum specificity of background
conditions in thought experiments. Here is a good formulation of the requirement, from
the chapter on thought experiments in the book by Kathy Wilkes:

One vital constraint is this. The experimenter—any experimenter, in thought or

actuality—needs to give us the background conditions against which he sets his
experiment. If he does not, the results of his experiment will be inconclusive
(Wilkes, 1988, p. 7)
It is here that we face a problem—a variant of the famous 'frame problem': how does
the enquirer select what is pertinent? How are (what are considered to be) the irrele-
vant factors kept at distance, so that they don't overburden the argument? How is the
information about all the relevant factors assembled?
The four points listed present in my opinion the major epistemological challenge to
the non-Platonist (J. Brown also adduces some ontological considerations about laws, but
we shall stick here to epistemology).
There is a further alleged datum which J. Brown lists. He is very much impressed with
Galileo's discovery of contradiction in Aristotle:
Here we have a transition from one theory to another which is quite remarkable.
There has been no new empirical evidence. The old theory was rationally
believed before the thought experiment, but was shown to be absurd by it. The
 MENTAL MODELS AND THOUGHT EXPERIMENTS 219

thought experiment established rational belief in a new theory (Brown, 1986,

p. 10)
and further
The transition from Aristotle to Galileo's theory is not just a case of making the
simplest overall adjustment in the old theory, (ibid., p. 12)
The comment that the revision occasioned by successful thought experiments is not
a case of simplest (or minimal) belief revision seems to overestimate the revolutionary
potential of thought experiments. It is exaggerated to claim that deeply entrenched
patterns of beliefs just tumble down upon the sole impact of thought experiments.
Take first the case of Galileo. Koyre's Platonism about thought experiments in
Galileo was plausible within his total picture of Galileo's work. After Stillman Drake the
picture of Galileo looks different—with stress on real experiment, exact measurement, etc.
Downloaded by [New York University] at 14:12 12 May 2015

Drake has shown that the material from Dialogues and Discourses is just the tip of the
iceberg. Galileo has not revised his opinions just on the strength of a few thought exper-
iments, but thanks to the impressive amount of his actual, carefully performed exper-
iments. (Generally, to make Galileo into a champion of Platonic or aprioristic physics is
now completely implausible.)
Similarly, Einstein's thought experiments with light beams in a running train did not
by themselves lead to the massive belief revision—the abandonment of classical mech-
anics. These experiments show what the world is like if the speed of light is constant, and if
the principle of relativity of inertial movement is upheld in full generality. That the speed
of light is constant was not impressed on the scientific community by thought experiments,
but by the overall impact of Maxwell's electrodynamics. The thought experiments only
clearly showed the consequences of accepting certain principles, they did not make the
scientific community accept them. (For similar caveats derived from fallibility of thought
experiments, see A. Irvine's essay 'Thought experiments in scientific reasoning'.)
It seems then that the impression of revolutionary impact of thought experiments in
isolation from the real ones is mistaken. It is no phenomenon, so it does not need explaining
and presents no problem for the anti-Platonist.
Here then is the challenge for non-apriorist posed by the four points of Platonic
argument. The following questions have to be answered:
1. How can one learn new things without new observational data?
2. Why are thought experiments superior to deduction in terms of heuristic power,
obviousness and ease?
3. Where does the 'experiential' element in thought experiments come from? Are
there any new experiences or quasi-experiences present in thought experiment,
and of what nature are they?
4. If the reasoning in thought experiment is broadly inductive, how can it eliminate
alternatives and reach its conclusion so quickly and effortlessly, and assert it with
such force?

Reasoning in mental models

I have suggested at the beginning of this paper that performing thought experiments
consists in a particular way of manipulating a certain kind of representations. The rep-
resentations and operations on them have a quasi-spatial quality, which distinguishes them
 220 N. MIŠČEVIČ

from their counterparts couched in purely verbal or propositional medium. Although the
basic mechanism enabling the construction and manipulation of the representations might
be computational, the most important feature of representations and operations is pre-
cisely their concrete and quasi-spatial character. This has earned them the name of 'mental
models'.
Typical examples of mental models are involved in understanding stories, or in ordin-
ary planning of activities. When a reader encounters a description of a situation, she builds
a model, a quasi-spatial 'picture' of it. As new details are supplied by the story-teller, the
model becomes updated. The background conditions are dictated by the reader's general
knowledge about the world.
A number of researchers have investigated reasoning in mental models. I shall follow
the lead of the best-known author in the field, P. N. Johnson-Laird, supplementing his
characterization with the more detailed description of the knowledge structures from the
joint work of Holland, Holyoak, Nisbett and Thagard. I shall limit myself to the non-
Downloaded by [New York University] at 14:12 12 May 2015

problematic aspects of the theory (e.g. Johnson-Laird claims that people use reasoning in
mental models in all reasoning tasks. I do not need, and I am not endorsing, this very strong
claim—all I shall need is the uncontroversial claim that people sometimes use it and are
generally good at it).
Mental models are highly specific (Johnson-Laird, 1983, p. 157). They purport to
represent concrete situations, with determinate objects and relations (precisely what is
demanded in thought experiments). Their structure is not arbitrary, but 'plays a direct
representational role since it is analogous to the corresponding state of affairs in the world'
(ibid.). One can distinguish simple static 'frames' representing relations between a set of
objects, temporal models consisting of sequences of such frames, kinematic models which
is the temporal model with continuous time, and finally dynamic models which model causal
relations. Galileo's and Stevin's experiments apparently demand at least a kinematic model.
The experimenter's 'view' of the model is his 'image' of it, a perspectival representation of
the model.
Reasoning in mental models demands rules for manipulation. Some constraints on
manipulation come directly from the geometry of the model. It is essential that manipu-
lation of elements mobilizes the spatial skills of the subject, her 'knowledge how' which is
generally not verbalizable. Other constraints on manipulation concern the consistency and
coherence of the model. Johnson-Laird hypothesizes the existence of general procedures
which add new elements to the model, and 'a procedure that integrates two or more
hitherto separated models if an assertion interrelates entities in them'. The integration of
models is subject to consistency requirements—if the joint model is logically impossible,
some change has to be made.
The point of building a mental model and working within it is to provide a medium
which adequately represents the task and is convenient in being easily manipulable and
'mobilizing'. Let me quote:

The notion of mental model is central to our analysis of problem solving and
induction. In common with many recent theoretical treatments, we believe that
cognitive systems construct models of the problem space that are then mentally
'run' or manipulated to produce expectations about the environment...
Although mental models are based in part on static prior knowledge, they are
themselves transient, dynamic representations of particular unique situations...
mental models are the major source of inactive change in long-term knowledge
structures. The reason is simple. Because mental models are built by integrating
 MENTAL MODELS AND THOUGHT EXPERIMENTS 221

knowledge in novel ways in order to achieve the system's goals, model construc-
tion provides the opportunity for new ideas to arise by recombination and as a
consequence of inductive model based prediction. (Holland et al., 1986, pp.
12-14)

I hope that even this sketchy and meagre description makes it obvious that mental
models are ideal medium for thought experiments. The recommendations of Galileo or
Einstein really sound like precepts to build some such models—conversely, the theory of
mental models seems to offer a description of mechanisms by which a thought experiment
is performed.
The most important points about the use of mental models for our subsequent
discussion are:

1. Many problems, particularly those involving spatial relations, are much easier to
Downloaded by [New York University] at 14:12 12 May 2015

solve in mental models, then verbally.

2. Manipulating mental models gives the thinker a kind of 'view' of changes of situ-
ation. It is close to experience, not only in respect of its 'feel', but, much more
importantly, in respect of its concreteness, specificity and spatial character.
3. The solution of the problem is 'imaged' (or 'seen'), and this is just due to the
peculiarity of the medium.

We now pass the characteristics of knowledge used in constructing models. Although

much is controversial, there seems to be agreement that the construction and use of models
mobilizes the everyday knowledge of the user. (When you read a story beginning with
'Mary entered the room' you suppose that the room has a door, that people normally enter
rooms by their doors, that they walk on the floor and not on the walls, etc., and you use this
knowledge in constructing your model of the situation.) Now, this knowledge is the object
of intense research in psychology and artificial intelligence. It is clear that everyday
knowledge is stratified and quite organized—researchers talk about 'folk physics', 'folk
pyschology', about represented social knowledge, etc.
One of the most popular proposals about the micro-structure of everyday knowledge
is that it is stratified in so-called default hierarchy. Here is an example and a sketch of
explanation:

When you hear someone saying 'Johnny ran towards the bird, but i t . . . ' you probably
automatically expect him to end saying ' . . . flew away'. You do not expect birds to attack
people savagely, to evaporate or to turn into pigs.
The explanation for the speed and ease with which you evoked the 'flew away' alterna-
tive is that your knowledge about birds is stratified. If we represent what you know by
conditional sentences grouped together into larger chunks—sometimes called schemata by
psychologists—we might put at the top of the 'bird-schema' the sentence '1. If something
is a bird, it flies'. This will be your default alternative. If you know nothing else about a
situation involving a bird, you will simply reach for (1) as the natural—most accessible,
most plausible—candidate.
Of course, you might have at your disposal a list of 'defeaters', conditions which
suspend (1)—for instance, you know that in the context of a Hitchcock movie, the right
thing to expect is that the bird will attack Johnny. The default conditionals (rules) together
with defeaters and alternatives, stratified into more and less readily accessible solutions
constitute the default hierarchies.
 222 N. MIŠČEVIČ

The building of default hierarchies is usually presented in terms of inductive learning,

in the very broad sense of 'induction'. Alternative are tried and rejected, rules (con-
ditionals) are recombined and transformed, their 'credibility' being subject to empirical
checking.
In this way, the finished default hierarchy embodies results of a long learning process.
When a reasoner builds her model, she can use such stratified knowledge in order to reach
swift and effortless cognitive decisions.
To sum up, mental models allow for reorganization of knowledge—particularly of
past experience. The medium they provide can be heuristically superior to the purely
verbal medium, especially to the aseptic medium of presumed regimented deductive
alternative.
The characteristics of mental models essential for us are:
1. Flexibility allowing for ad hoc combinations and recombination of items.
Downloaded by [New York University] at 14:12 12 May 2015

2. Rules for integrating given models into larger ones.

3. Non-propositional component, particularly suitable for geometrical and visual
geometry-based reasoning.
4. Integration of chunked material, like Schemas and scripts allowing for the use of
stored knowledge organized in default hierarchies.
These characteristics ground precisely the pragmatic virtues of thought experiments—
manipulability and 'mobilizing force'. So, let us go back to the Platonist challenge!

The virtues explained

Let us first look at the example.
As you recall, the first part of Galileo's experiment concerning free fall (as presented
by J. Brown) involves reasoning about the role of weight and culminates in the intermedi-
ate conclusion that weight plays no role in determining speed. Simplicio is asked first to
consider a light body L and a heavy body H falling freely, and then to consider a package P,
made of L and H tied together. He is supposed to 'become aware' of the contradiction
present in the initial assumption that speed is proportional to weight. How does he achieve
this?
In terms of mental models this is how he does it:
He first builds two smaller models for L and H; first an 1-model, with L falling slowly;
and an h-model, with H falling rapidly. Now, he is asked to consider how H and L would
fall together. This demands an integration of h-model and l-model. When he tries to perform
the integration he discovers that an integrated lh-model, which would be at the same time
model for P, is impossible.
Instead of manipulating formulae (the way a logician would do) Simplicio manipu-
lates more concrete representations enabling him to use his everyday skills and will cer-
tainly reach with ease the required conclusion (the interim conclusion that it is incoherent
to maintain that speed is proportional to weight). The purely propositional way might have
been, on the contrary, closed to Simplicio.
Now we can address the phenomena which pose a challenge for the non-Platonist.

1 Making data available

The first challenge was introduced by the claim that no new data are supplied by thought
experiments. But building mental models enriches the enquirers data basis by deploying
 MENTAL MODELS AND THOUGHT EXPERIMENTS 223

pieces of knowledge which were not seen to be relevant before, and items of practical
know-how which can be manifested only in concrete manipulation. So, against the claim
that no new data are introduced through thought experiments it should be pointed out that
the mobilization of the previously unavailable knowledge does constitute an important
widening of the data base and a vital part of process of discovery. But, we can be more
specific:

You know that a circular chain retains its topological properties while sliding. But
unless you are a trained scientist or philosopher of science, you have no way to express this
knowledge in a simple propositional way. Not even Stevin could have done it easily. So, he
and his public do not have an adequate propositional representation at their disposal, and
the implicit knowledge they have stays dormant and idle. However, with a quasi-spatial
model the implicit knowledge becomes usable.
This brings us to the most conspicuous contribution of quasi-spatial models to
Downloaded by [New York University] at 14:12 12 May 2015

reasoning in physics, namely their use in eliciting and mobilizing geometric intuitions.
Geometry is paramount in physics reasoning. (Of course, if geometry is a priori this
does not make physics a priori. Suppose Euclidean geometry is a priori. Suppose my table
is rectangular. It follows, a priori, that the sum of its interior angles equals four right
angles. This does not make the sentence 'The sum of interior angles of the top of my table
is four right angles' a piece of a priori science of carpentry. I have a sneaking suspicion
that certain prima facie plausibility of apriorist's claims stems from confusion between
mathematics and physics.)
Now, geometrical reasoning is notoriously easier on drawings than on propositions,
and geometrical intuitions are more readily available when you evoke spatial models. For
most people they might be available only through models (mental or graphic). Everybody
knows in practice which properties a circular chain retains while sliding. But unless one has
been doing some topology, one has no way to express this knowledge systematically in a
simple propositional way. Presumably, not even Stevin could have done it easily. So, he
and his public do not have an adequate RV at their disposal. However, when you build your
model or Stevin builds his, you use your spatial skills—the same ones you use when
orienting yourself in space (or playing chess or tetris). The know-how involved in these
skills certainly cannot be routinely verbalized (nor a fortiori 'regimented'), so they
would not be available for Stevin or Galileo (or high school students learning elements of
mechanics) outside the context of model building in thought experiment.
We have just mentioned one set of relevant properties of Stevin's chain. Here is the
other, which relates to forces acting on the chain. When a student learns to use vectors she
starts with physical representation (cart pulled by two horses in different directions),
passes to the geometrical one, and ends on an algebraic one. Why? Because geometrical
representation is vastly more easy to manipulate. (Constrast this with the Platonist claim
that we have direct access to physical laws as alleged abstract objects.) Now, this is pre-
cisely the factor which makes it easier to understand Stevin's thought experiment, than to
understand a description of the same situation couched in terms of vector algebra. We are
given a representation on which we can work—we find it easy to imagine the links of the
chain sliding down the inclined plane, setting the lower part into motion, and we find it
easy to recognize that the overall picture will not change by any amount of sliding. Try to
express this in a more abstract medium, and to reason in that medium, and you will be able
to feel the difference.
Now, thought experiments from Galileo to Einstein typically bring geometrical
intuitions to bear upon issues in physics (either exclusively or in conjunction with others).
 224 N. MIŠČEVIČ

They help the enquirer and the reader mobilize the vast resources of practical geometrical
knowledge—the data which are old in the sense of being 'implicit' in the resources, but new
(or 'feeling' new) in the sense of not being at the enquirers conscious disposal, or not even
being in principle verbalizable for her.
This point can be generalized to cover the whole non-verbal practical knowledge which
might be mobilized by thought experiments, together with the implicit (but verbalizable)
everyday knowledge usable in building and manipulating mental models. The use of such
knowledge goes a long way towards accounting for the impression that thought exper-
iments draw upon a rich array of data, richer than usually allowed for in philosophical
reconstruction of them.
Still the data do not amount to a theory. The pieces of old information have to be put
together in novel ways in order to yield substantial theoretical insights. This brings us to
the second and third challenge, which we shall address together.
Downloaded by [New York University] at 14:12 12 May 2015

2 Constructing new experiences

Against the claim that armchair physics is 'not a case of seeing old empirical data in new
ways', the obvious reply is that some thought experiments do involve seeing old data in the
appropriate way, and some involve combining old data in new way.
If no one had looked at the given set of data before, or if no one has had any idea how to
'see' them in any systematic way, then our armchair innovator has just seen them in the
right way, without seeing them in a 'new' way, because there was no old way of seeing.
Much more important is the possibility of producing new data by manipulating the old
ones. Thought experiments must provide first, an adequate representation of the problem,
i.e. representation giving the opportunity for the subject to manipulate the problem situ-
ation (in his head, of course) in a particularly easy fashion, and so that it makes it easy to
mobilize subjects' cognitive resources—skills, implicit background knowledge, perceptual
beliefs, etc., in a way superior to regimented reasoning.
There is no special mystery about either component.
It is well-known that some forms of representation are easier to read and to manipulate
than others, and that the heuristic power of a representation depends crucially on these
qualities. Johnson-Laird and his followers have shown that at many tasks people perform
dramatically better when allowed to use quasi-spatial representation. Stevin's linked chain
is a good example of such easily manipulable representation. (Why exactly are some rep-
resentations better than others is an interesting problem, but it is a technical problem
which certainly does not threaten a general empiricist line about cognition.)
The manipulability and the mobilizing force should account for the heuristic value
of thought experiments. The theory of mental models makes correct prediction that a
visualizable thought experiment will have more heuristic value that a corresponding
propositional alternative, and explains in principle why.
It is plausible that manipulating quasi-spatial models can be equivalent to actually
seeing the movement of pieces manipulated. The manipulation is quasi-experiential in that
it yields non-inferential beliefs akin to perceptual ones. This might explain the felt
difference between thought experiment and a piece of regimented propositional reasoning.
I shall rely upon the generally acknowledged fact that differences in the represen-
tational medium and in the representation available or chosen are greatly responsible for
differences in the performance in problem solving (including scientific discovery), and
shall not expand this point, given that it is the least problematic issue of all.
 MENTAL MODELS AND THOUGHT EXPERIMENTS 225

3 The ease of the jump

Let us look at the issue of the inductive character of some key steps in thought experiments.
The demand of maximum specificity (exclusion of irrelevant alternatives) seems incom-
patible with quickness and ease of the final step towards the conclusion of thought exper-
iments in which irrelevant alternatives might crop up. If the experimenter must identify
and weigh alternatives before reaching his conclusion, as canons of inductive procedure
demand of him, he will never make his step so gracefully.
The answer lies in the fact that reasoning in mental models mobilizes the resources of
highly organized background knowledge—chunked, with undetachable 'parts', stratified
in default hierarchy and medium-specific. Look at the example. You all know that colour
does riot affect speed. How do you know it? Your knowledge is organized (and stored) in a
default way and it might be present implicitly in various scenarios concerning situations of ^
falling. It is difficult to handle outside a concrete context (try to think of all the things that
Downloaded by [New York University] at 14:12 12 May 2015

you in fact know to be irrelevant for the speed of a falling body). If one had to write down a
list, and to explain why the particular characteristics are on the list, it would be a never
ending job.
On the contrary, the corresponding task of concentrating upon what is thought to be
relevant will be easy once you have imagined a concrete situation (once you have started
building a model). The default hierarchy becomes mobilized, and it controls and con-
strains the model building smoothly and effortlessly. This is an additional ground for
heuristic superiority of thought experiments over explicit propositional representations—
thought experiments are more economical about implicit assumptions.
Let us now look at Galileo's experiment, the second and final part—the jump from
irrelevance of weight to the positive definite conclusion that all bodies fall with roughly the
same speed.
The speed of free-falling objects is independent of its colour. It is also independent of
its taste. It is independent of the look of the object. These truths are built into our back-
ground knowledge. They might figure within knowledge representation in a default way (a
consequence would be, for example, that if an object is falling unusually slowly, its colour
would be the last property one should look at in search of explanation). If we had to make all
these truths explicit in order to arrive at a conclusion (i.e. that speed does not depend on
weight), we would never arrive at it, certainly not within a few seconds. But the simplest
(and therefore proper) explanation of our speed and efficiency is not that we see the right
explanation written down in the Platonic realm, but that truths about what is irrelevant are
given as implicit constraints on our model building. They determine the architecture of the
model, while remaining rather inconspicuous in the background.
All the irrelevant possibilities (irrelevant from the viewpoint of the enquirer) are
excluded thanks to the filtering power of default hierarchies. There is of course no special
reason to think that the background knowledge itself is apriori. The default hierarchies are
presumably built up in some kind of inductive procedure, embodying life-long experience
and learning from others. As we said, they contribute to specificity by narrowing down the
range of possibilities which have to be explicitly taken into account. Of course, together
with this stratified knowledge, the use of models also mobilizes knowledge embodied in
enquirers cognitive skills, most notably spatial ones, all contributing to the specificity of
the model, narrowing the range of the alternative possible world to make it as close as
possible to (what the experimenter takes to be) the actual world. This contribution is unique
to model building. It could not be normally achieved by a corresponding propositional
version.
 226 N. MIŠČEVIČ

This might go a long way towards explaining why the jump to conclusion is so
unproblematic. It is unproblematic because the irrelevant (or seemingly irrelevant) factors
have been neutralized and taken care of by our background knowledge, and its default
hierarchies. The significant amount of background knowledge is being brought to bear on
the issue, and this is done imperceptibly with no manifest effort.

Conclusion and a bonus

Let us resume. The assumption that performing a thought experiment consists in construct-
ing a mental model and reasoning in it, throws light on the most important characteristics of
thought experiments. It explains the nature of data, points to the skills employed in
manipulating the items in the model, and accounts for heuristic fruitfulness and relative
ease of the work in the laboratory of the mind. It postulates no mysterious capacities, unlike
the Platonic account, and is therefore preferable to it.
Downloaded by [New York University] at 14:12 12 May 2015

The non-Platonist account has an additional explanatory advantage over the Platonic
one—it holds promise of explaining not only the virtues but also the weaknesses and
liabilities of thought experiments in a systematic fashion. No such systematic explanation
is forthcoming if thought experiments are seen as return tickets for brief trips in the
Platonic realm of ideas. The Platonist must probably regard the unsuccessful thought
experiments as anomalies, random failures of sight on the part of the mind's eye. The non-
Platonist can point to the role of background knowledge in construction of mental models
and in attendant reasoning. This background knowledge (or rather 'knowledge'), never
made explicit in the model, contains a lot of false folk theories, which when mobilized are
certain to mislead the experimenter. Some rules for model building themselves might be
misleading, if the model is to picture situations radically different from the ordinary ones
(perhaps the EPR thought experiment is a case in point, with its assumptions of locality
and separability which may be a part of our folk-metaphysics, presupposed in every model
of a situation we might come to build). Our basic capacities are probably better tailored for
moving furniture around than for doing quantum mechanics.

Note
1. The earlier versions of this paper were presented at the meetings of the section for analytical philosophy of the
Croatian philosophical society in Rijeka and Zadar, and at the course on philosophy of science in Dubrovnik.
Thanks are due to the participants, above all to James Brown, who supported my efforts to write the final
versions, to my critics—Elvio Baccarini, Boran Berčić, Slavko Brkić, Andy Irvine, Wladislaw Krajewski,
Snježana Prijić, Nikica Simić, Zvonimir Šikie, and to Kathy Wilkes to whose book Real People I owe my
first interest in thought experiments and who has invited me to the course in the Inter-University Centre,
Dubrovnik.

References
BROWN, J. (1991) The Laboratory of the Mind (London, Routledge).
BROWN, J. (1986) Thought experiments since the Scientific Revolution, International Studies in the Philosophy of
Science,1,No. 1.
GALILEO, G. Dialogues Concerning the Two New Sciences, (tr. CREW & DE SALVIO) Encyclopaedia Britanica.
HOLLAND, J. et al. (1986) Induction (Cambridge, MA, MIT Press).
IRVINE, A. (1991) Thought Experiments in Scientific Reasoning, in: HOROWITZ & MASSEY (Eds) Thought
Experiments in Science and Philosophy (Totowa, Rowan & Littlefield).
JOHNSON-LAIRD, P.N. (1983) Mental Models Cambridge, University of Cambridge Press).
WILKES, K. (1988) Real People (Oxford, Oxford University Press).

View publication stats