Lecture 1

2/1/94
Introduction
Philosophy of science is part of a range of sub-disciplines known as "philosophy of X" (where X may be filled in
with art, history, law, literature, or the various special sciences such as physics). Each of the activities for which
there is a "philosophy of X" is an investigation into a certain part of the world or a particular type of human activity.
What we will talk about today, in very general terms, is what distinguishes philosophy of X from sociology, history,
or psychology of X. These approaches to science cannot be sharply demarcated, though many people have
thought that they can be. However, there are clear differences of emphasis and methods of investigation between
these approaches that we can outline here in a preliminary way.
Let us attempt to characterize the particular emphases of several approaches to studying science:
Sociology of Science - Sociology of science studies how scientists interact as social groups to resolve

differences in opinion, how they engage in research, and how they determine which of many theories and
research programs are worth pursuing, among other things.
Psychology of Science - Psychology of science studies how scientists reason, i.e., the thought processes that

scientists follow when they are judging the merits of certain kinds of research or theoretical speculations; how
they reason about data, experiment, theories, and the relations between these; and how they come up with
new theories or experimental procedures.
History of Science - History of science studies how scientists have engaged in the various activities noted

above in the past, i.e., how social interactions among scientists (and between scientists and the greater
society) and styles of scientific reasoning have changed over the centuries; and how particular scientific
achievements came to be accepted, both by individual scientists and by the scientific community as a whole.
In each of these cases, the data on which the approach to studying science is based is empirical or observational.
What is at issue is how scientists in fact interact as social groups, reason, or how scientific reasoning styles and
scientific theories have in fact changed over time. To adjudicate disputes within these approaches thus requires
engaging in much like scientific activity itself: one must gather evidence for one's views (or in the case of certain
approaches to history, one's interpretation of the nature of scientific activity or the causes of scientific revolutions).
What is philosophy of science? How does it differ from these other approaches to studying science? Well, that's
not an easy question to answer. The divisions among philosophers of science are quite striking, even about
fundamentals, as will become apparent as the course proceeds. One reason for this is that philosophers of
science, on occasion, would find many of the things that sociologists, psychologists, and historians of science
study to be relevant to their own studies of science. Of course, the degree to which philosophers of science are
interested in and draw upon the achievements of these other disciplines varies greatly among individuals--e.g.,
some philosophers of science have been far more interested in the history of science, and have thought it more
relevant to their own endeavors, than others. However, there are some tendencies, none of them completely
universal, that would serve to mark a difference between philosophers of science on the one hand and
sociologists, historians, and psychologists of science on the other.
The first difference is that philosophy of science is not primarily an empirical study of science, although empirical
studies of science are of relevance to the philosopher of science. (Like everything else you might cite as a
peculiarity of philosophy of science, this point is a matter of dispute; some philosophers of science, for example,
claim that philosophy of science ought to be considered a special branch of epistemology, and epistemology
ought to be considered a special branch of empirical psychology.) Philosophers of science do not generally
engage in empirical research beyond learning something about a few branches of science and their history. This

type of study, however, is simply a prerequisite for talking knowledgeably about science at all. Philosophers
primarily engage in an activity they call "conceptual clarification," a type of critical, analytical "armchair"
investigation of science. For example, a philosopher of science may try to answer questions of the following sort.
What is scientific methodology, and how does it differ (if it does) from the procedures we use for acquiring
knowledge in everyday life?
How should we interpret the pronouncements of scientists that they have gained knowledge about the invisible,
underlying structure of the world through their investigations?
Part of what is open to philosophy of science, insofar as it is critical, is to question the methods that scientists use
to guide their investigations. In other words, philosophers of science often seek to answer the following question.
What reason is there to think that the procedures followed by the scientist are good ones?
In a sense, philosophy of science is normative in that it asks whether the methods that scientists use, and the
conclusions that they draw using those methods, are proper or justified. Normally, it is assumed that the methods
and conclusions are proper or justified, with it being the task of the philosopher of science to explain precisely
how they can be proper or justified. (In other words, the philosopher of science seeks to understand the practice
of science in such as way as to vindicate that practice.) This opens up the possibility of revision: that is, if a
philosopher of science concludes that it is impossible to justify a certain feature of scientific practice or
methodology, he or she might conclude that that feature must be abandoned. (This would be rare: most
philosophers would react to such a situation by rejecting the view that did not vindicate that feature of scientific
practice.)
Let us take another approach to the question of what distinguishes philosophy of science from other approaches
to studying science. Philosophy has, since Plato, been concerned with the question of what a particular kind of
thing essentially is. In other words, philosophers seek a certain sort of answer to questions of the following form.
What is X?
In asking a question of this form, philosophers seek to understand the nature of X, where by "nature" they mean
something like X's essence or meaning.
We will start the course by considering the question, "What is scientific explanation?" We will also seek to answer
the question, "What makes a scientific explanation a good one?" Most people take the notion of explanation for
granted; but as you will soon find out, philosophers take a special interest in the concepts others take for granted.
Philosophers emphasize the difference between being able to identify something as an explanation and being
able to state in precise terms what an explanation is, i.e., what makes something an explanation. Philosophers
seek to do the latter, assuming that they are able (like everyone else) to do the former.
None of this, of course, will mean very much until we have examined the philosophy of science itself, i.e., until we
start doing philosophy of science. To a large degree, each of you will have to become a philosopher of science to
understand what philosophy of science is.

Lecture 2
2/3/94
The Inferential View Of Scientific Explanation
Last time we discussed philosophy of science very abstractly. I said that it was difficult to separate completely
from other studies of science, since it draws on these other disciplines (sociology, history, psychology, as well as
the sciences themselves) for its "data." What distinguishes philosophy of science from other studies of science is
that it (1) takes a critical, evaluative approach--e.g., it aims at explaining why certain methods of analyzing data,
or as we shall see today, the notion of explanation--are good ones. There is also an emphasis on conceptual
analysis--e.g., explaining what explanation is, or, in other words, what it means when we say that one thing
"explains" another. (Philosophers often discuss the meaning of many terms whose meaning other people take for
granted.) I also noted that the best way to see how philosophy of science is differentiated from other studies of
science is by examining special cases, i.e., by doing philosophy of science. We will start our investigation by
examining the notion of explanation.
The Difference Between Explanation And Description
It is commonplace, a truism, to say that science aims at not only describing regularities in the things that we
observe around us--what is often called observable or empirical phenomena--but also at explaining those
phenomena. For example, there is the phenomenon of redshift in the spectra of stars and distant galaxies. The
physical principles behind redshift are sometimes explained by analogy with the Doppler effect, which pertains to
sound: observed wavelengths are increased if the object is moving away from us, shortened if the object is
moving towards us. (Also, there is a derivation in general relativity theory of the redshift phenomenon, due to
gravitation.) In 1917, de Sitter predicted that there would be a relationship between distance and redshift, though
this prediction was not generally recognized until Hubble (1929), who was inspired by de Sitter's analysis. Another
example: There's a periodic reversal in the apparent paths of the outer planets in the sky. This can be predicted
purely mathematically, based on past observations, but prediction does not explain why the reversal occurs. What
is needed is a theory of the solar system, which details how the planets' real motions produce the apparent
motions that we observe. Final example: Hempel's thermometer example (an initial dip in the mercury level
precedes a larger increase when a thermometer is placed into a hot liquid).
Three Approaches To Explanation
A philosopher of science asks: What is the difference between describing a phenomenon and explaining it? In
addition, what makes something an adequate explanation? Philosophers have defended three basic answers to
this question.
Inferential View (Hempel, Oppenheim) - An explanation is a type of argument, with sentences expressing laws of
nature occurring essentially in the premises, and the phenomenon to be explained as the conclusion. Also
included in the premises can be sentences describing antecedent conditions.
Causal View (Salmon, Lewis) - An explanation is a description of the various causes of the phenomenon: to
explain is to give information about the causal history that led to the phenomenon.
Pragmatic View (van Fraassen) - An explanation is a body of information that implies that the phenomenon is
more likely than its alternatives, where the information is of the sort deemed "relevant" in that context, and the
class of alternatives to the phenomenon are also fixed by the context.
In the next few weeks, we will examine each of these accounts in turn, detailing their strengths and weaknesses.
Today we will start with Hempel's inferential view, which goes by several other names that you might encounter.

The "received" view of explanation (reflecting the fact that philosophers generally agreed with the inferential view
until the early 1960s)
The "deductive-nomological" model of explanation (along with its probabilistic siblings, the "inductive-statistical"
and "deductive-statistical" models of explanation)
The Inferential Theory of Explanation
The original Hempel-Oppenheim paper, which appeared in 1948, analyzed what has come to be known as
"deductive-nomological" (D-N) explanation. Hempel and Oppenheim consider patterns of explanation of a
particular sort and try to generalize that pattern. As already noted, they view explanation as a certain sort of
argument, i.e., a set of premises (sentences) which collectively imply a conclusion. The arguments they
considered were deductive, i.e., arguments such that if the premises are true the conclusion has to be true as
well.
All human beings are mortal.
Socrates is a human being.
Socrates is mortal.
Not every argument that is deductive is an explanation, however. How can we separate those that are from those
that are not?
To accomplish this task, Hempel and Oppenheim describe their "General Conditions of Adequacy" that define
when a deductive argument counts as an adequate explanation.
An explanation must:
(a) be a valid deductive argument (hence "deductive")
(b) contain essentially at least one general law of nature as a premise (hence "nomological")
(c) have empirical content (i.e., it must be logically possible for an observation-statement to contradict it)
The first three conditions are "logical" conditions, i.e., formal, structural features that a deductive argument must
have to count as an explanation. To complete the conditions of adequacy, Hempel and Oppenheim add a fourth,
"empirical" condition.
The premises (statements in the explanans) must all be true.
On the inferential view, explanations all have the following structure (where the antecedent conditions and laws of
nature make up the "explanans").
C1, ... , Cn [antecedent conditions (optional)]
L1, ... , Ln [laws of nature]
Therefore, E [explanandum]
Later, we will be looking at a statistical variant of this pattern, which allows the laws of nature to be statistical, and
the resulting inference to be inductive.
Laws Of Nature
Since Hempel and Oppenheim's analysis is stated in terms of laws of nature, it is important to state what a law of
nature is on their view. Hempel and Oppenheim take a law to be a true "law-like" sentence. This means that a law
is a linguistic entity, to be distinguished by its peculiar linguistic features. According to Hempel and Oppenheim,
laws are to be distinguished from other sentences in a language in that they are (1) universal, (2) have unlimited
scope, (3) contain no designation of particular objects, and (4) contain only "purely qualitative" predicates.

The problem Hempel and Oppenheim face is distinguishing laws from accidental generalizations, i.e., general
truths that happen to be true, though they are not true as a matter of physical law. For example, suppose that all
the apples that ever get into my refrigerator are yellow. Then the following is a true generalization: "All the apples
in my refrigerator are yellow." However, we do not deem this sentence to be a law of nature. One reason that this
might be so is that this statement only applies to one object in the universe, i.e., my refrigerator. Laws of nature,
by contrast, refer to whole classes of objects (or phenomena). (Consider the sentence "All gases that are heated
under constant pressure expand.") It is for this reason that Hempel and Oppenheim include the requirement that
to be a law of nature a sentence must not designate any particular objects.
Discuss: Can we get around this requirement by a bit of linguistic trickery? Consider if we replace the designation
"in my refrigerator" by "in any Lyle Zynda-owned refrigerator," or, even better, by some physical description in
terms of "purely qualitative" predicates that single out my refrigerator from the class of all refrigerators in the
universe. Would the existence of such a sentence convince you that you have discovered a new law of nature?
Moreover, consider the following two sentences.
G: No gold sphere has a mass greater than 100,000 kilograms.
U: No enriched uranium sphere has a mass greater than 100,000 kilograms.
The former is not a law of nature, though the latter is (though it is of a relatively low-level variety of law).
One reason that the statements about apples and gold might not be laws of nature, a reason not adequately
captured by the Hempel-Oppenheim analysis of laws, is that they do not support inferences to counterfactual
statements. For example, it cannot be inferred from the fact that all the apples that ever get into my refrigerator
are yellow that if a red apple were to be placed into my refrigerator, it would turn yellow. Laws of nature, by
contrast, support counterfactual inferences. From the fact that all gases that are heated under constant pressure
expand we can infer that if a sample of gas in a particular container were heated under constant pressure, it
would expand. (We can reasonably infer this even if we never heat that sample of gas under constant pressure.)
Similarly, we can infer that if were to successfully gather 100,000 kilograms of uranium and try to fashion it into a
sphere, we would fail. (We could not infer a similar statement about gold from the true generalization that no gold
sphere has a mass greater than 100,000 kilograms.)
The difference between statements G and U has never been captured purely syntactically. Thus, Hempel and
Oppenheim's view that laws of nature are sentences of a certain sort must be fundamentally mistaken. What laws
of nature are is still a matter of dispute.
Counterexamples To The Inferential View Of Scientific Explanation: Asymmetry and Irrelevance
The Hempel-Oppenheim analysis of scientific explanation has the following overlapping features. (Review and
explain each.)
(a) Inferential - Explanations are arguments: to explain why E occurred is to provide information that would have
been sufficient to predict beforehand that E would occur.
(b) Covering Law - Explanations explain by showing that E could have been predicted from the laws of nature,
along with a complete specification of the initial conditions.
(c) Explanation-Prediction Symmetry - The information (i.e., laws, antecedent conditions) appearing in an
adequate explanation of E could have been used to predict E; conversely, any information that can be used
to predict E can be used after the fact to explain why E occurred.
(d) No Essential Role For Causality - Laws of nature do not have to describe causal processes to be used
legitimately in scientific explanations.
Many counterexamples have been given to the Hempel-Oppenheim analysis of scientific explanation that target
one or more of these features. The first group of counterexamples reveals that the Hempel-Oppenheim analysis
faces the Problem of Asymmetry: Hempel-Oppenheim assert that explanation and prediction are symmetric,
whereas that does not seem to be the case, as the following examples show.

(1) Eclipse - You can predict when and where a solar eclipse will occur using the laws governing the orbit of the
Earth around the Sun, and the orbit of the Moon around the Earth, as well as the initial configuration these three
bodies were in at an earlier time. You can also make the same prediction by extrapolating backwards in time from
the subsequent positions of these three bodies. However, only the first would count as an explanation of why the
eclipse occurred when and where it did.
(2) Flagpole - Using the laws of trigonometry and the law that light travels in straight lines, you can predict the
length of the shadow that a flagpole of a certain height will cast when the Sun is at a certain elevation. You can
also predict what the height of the flagpole is by measuring the length of its shadow and the elevation of the Sun.
However, only the first derivation would count as an explanation.
(3) Barometer - Using the laws governing weather patterns, storm formation, and the effect of air pressure on the
behavior of barometers, you can predict that when a barometer falls that a storm will soon follow. You can also
predict that when a storm is approaching, the barometer will fall. However, neither of these are explanatory, since
both are explained by antecedent atmospheric conditions.
The second group of counterexamples reveals that the Hempel-Oppenheim analysis of explanation faces the
Problem of Irrelevance: the Hempel-Oppenheim analysis sometimes endorses information as explanatory when it
is irrelevant to the explanandum.
(4) Birth Control Pills - All men who take birth control pills never get pregnant. Thus, from the fact that John is
taking birth control pills we can infer logically that he won't get pregnant. However, this would hardly be an
explanation of John's failing to get pregnant since he couldn't have gotten pregnant whether or not he took birth
control pills.
(5) Hexed Salt - All salt that has had a hex placed on it by a witch will dissolve in water. Hence, we can logically
infer from the fact that a sample of salt had a hex placed on it by a witch that it will dissolve in water. However,
this wouldn't give us an explanation of the dissolution since the salt would have dissolved even if it hadn't been
hexed.

Lecture 3
2/8/94
The Causal Theory Of Explanation, Part I
Last time, we saw that the inferential view of explanation faced the asymmetry and irrelevance problems. There is
another problem, however, that comes out most clearly when we consider the inductive-statistical (I-S) component of
the inferential view. This problem strikes most clearly at the thesis at the heart of the inferential view, namely, that to
explain a phenomenon is to provide information sufficient to predict that it will occur.
I-S explanation differs from D-N explanation only in that the laws that are cited in the explanation can be statistical.
For example, it is a law of nature that 90% of electrons in a 90-10 superposition of spin-up and spin-down will go up if
passed through a vertically oriented Stern-Gerlach magnet. This information provides us with the materials for stating
an argument that mirrors a D-N explanation.
Ninety percent of electrons in a 90-10 superposition of spin-up and spin-down will go up if passed through a vertically
oriented Stern-Gerlach magnet. (Law of Nature)
This electron is in a 90-10 superposition of spin-up and spin-down and is passed through a vertically oriented SternGerlach magnet. (Statement of Initial Conditions)
Therefore, this electron goes up. (Explanandum) [90%]
This argument pattern is obviously similar to that exhibited by D-N explanation, the only difference being that the law
in the inductive argument stated above is statistical rather than a universal generalization. On the inferential view, this
argument constitutes an explanation since the initial conditions and laws confer a high probability on the
explanandum. If you knew that these laws and initial conditions held of a particular electron, you could predict with
high confidence that the electron would go up.
The problem with the inferential view is that you can't always use explanatory information as the basis for a
prediction. That is because we frequently offer explanations of events with low probability. (Numbers in the examples
below are for purposes of illustration only.)
Atomic Blasts & Leukemia. We can explain why a person contracted leukemia by pointing out the person was once
only two miles away from an atomic blast, and that exposure to an atomic blast from that distance increases one's
chances of contracting leukemia in later life. Only 1 in 1,000 persons exposed to an atomic blast eventually contract
leukemia. Nevertheless, exposure to an atomic blast explains the leukemia since people who haven't been exposed
to an atomic blast have a much lower probability (say, 1 in 10,000) of contracting leukemia.
Smoking & Lung Cancer. We can explain why someone contracted lung cancer by pointing out that the person had
smoked two packs of cigarettes a day for forty years. This is an explanation since people who smoke that much have
a much higher probability (say, 1 in 100) of contracting lung cancer than non-smokers (say, 1 in 10,000). Still, the
vast majority of smokers (99 percent) will never contract lung cancer.
Syphilis & Paresis. We can explain why someone contracted paresis by pointing out that the person had untreated
latent syphilis. This is an explanation since the probability of getting paresis is much higher (e.g., 1 in 100) if a person
has untreated latent syphilis than if he does not (e.g., 0). Still, the vast majority of people with untreated latent
syphilis will never contract paresis.
In each of these cases, you can't predict that the result will occur since the information does not confer a high
probability on the result. Nevertheless, the information offered constitutes an explanation of that result, since it
increases the probability that that result will occur.

In the 1960s and 1970s, Wesley Salmon developed a view of statistical explanation that postulated that, contrary to
what Hempel claimed earlier, high probability was not necessary for an explanation, but only positive statistical
relevance.
Definition. A hypothesis h is positively relevant (correlated) to e if h makes e more likely, i.e., pr(h|e) > pr(h).
The problem Salmon faced was distinguishing cases where the information could provide a substantive explanation
from cases where the information reported a mere correlation and so could not. (For example, having nicotine stains
on one's fingers is positively correlated with lung cancer, but you could not explain why a person contracted lung
cancer by pointing out that the person had nicotine stains on their fingers.) Distinguishing these cases proved to be
impossible using purely formal (statistical) relations. Obviously, some other type of information was needed to make
the distinction. Rejecting the received view of explanation, Salmon came to believe that to explain a phenomenon is
not to offer information sufficient for a person to predict that the phenomenon will occur, but to give information about
the causes of that phenomenon. On this view, an explanation is not a type of argument containing laws of nature as
premises but an assembly of statistically relevant information about an event's causal history.
Salmon points out two reasons for thinking that causal information is what is needed to mark off explanations. First,
the initial conditions given in the explanatory information have to precede the explanandum temporally to constitute
an explanation of the explanandum. Hempel's theory has no restriction of this sort. The eclipse example illustrates
this fact: you can just as well use information about the subsequent positions of the Sun and Moon to derive that an
eclipse occurred at an earlier time as use information about the current positions of the Sun and Moon to derive that
an eclipse will occur later. The latter is a case of retrodiction, whereas the latter is a (familiar) case of prediction. This
is an example of the prediction-explanation symmetry postulated by Hempel. However, as we saw earlier when
discussing the problem of asymmetry, only the forward-looking derivation counts as an explanation. Interestingly,
Salmon points out that the temporal direction of explanation matches the temporal direction of causation, which is
forward-looking (i.e., causes must precede their effects in time).
Second, not all derivations from laws count as explanations. Salmon argues that some D-N "explanations" (e.g., a
derivation from the ideal gas law PV = nRT and a description of initial conditions) are not explanations at all. The
ideal gas law simply describes a set of constraints on how various parameters (pressure, volume, and temperature)
are related; it does not explain why these parameters are related in that way. Why these constraints exist is a
substantive question that is answered by the kinetic theory of gases. (Another example: People knew for centuries
how the phases of the Moon were related to the height of tides, but simply describing how these two things are
related did not constitute an explanation. An explanation was not provided until Newton developed his theory of
gravitation.) Salmon argues that the difference between explanatory and non-explanatory laws is that the former
describe causal processes, whereas non-explanatory laws (such as the ideal gas law) only describe empirical
regularities.

Lecture 4
(2/10/94)
The Causal Theory Of Explanation, Part II
As we discussed last time, Salmon sought to replace the inferential view of explanation, which faces the asymmetry
and irrelevance problems, with a causal theory, which postulates that an explanation is a body of information about
the causes of a particular event. Today we will discuss Salmon's view in detail, as well as the related view of David
Lewis.
Salmon's theory of causal explanation has three elements.
(1) Statistical Relevance - the explanans (C) increases the probability of the explanandum (E), i.e., pr(E|C) > pr(E).
(2) Causal Processes - the explanans and the explanandum are both parts of different causal processes
(3) Causal Interaction - these causal processes interact in such a way as to bring about the event (E) in question
This leaves us with the task of saying what a causal process is. Basically, Salmon's view is that causal processes are
characterized by two features. First, a causal process is a sequence of events in a continuous region of spacetime.
Second, a causal process can transmit information (a "mark").
Let us discuss each of these in turn. There are various sequences of events that are continuous in the required
sense--e.g., a light beam, a projectile flying through space, a shadow, or a moving spot of light projected on a wall.
An object that is sitting still, e.g., a billiard ball, is also deemed a causal process. Each of these is a continuous
process in some sense but not all of them are causal processes--e.g., the shadow and light spot. Let's look at an
example that makes this clearer. As some of you may know, relativity theory says that nothing can travel faster than
light. But what is the "thing" in nothing? Consider a large circular room with a radius of 1 light year. If we have an
incredibly focused laser beam mounted on a swivel in the center of the room, we can rotate the laser beam so that it
rotates completely once per second. If the laser beam is on, it will project a spot on the wall. This spot too will rotate
around the wall completely once per second, which means that it will travel at 2p light-years per second! Strangely
enough, this is not prohibited by relativity theory, since a spot of this sort cannot "transmit information." Only things of
that sort are limited in speed.
Salmon gives this notion an informal explication in "Why ask 'Why'?" He argues that the difference between the two
cases is that a process like a light beam is a causal process: interfering with it at one point alters the process not only
for that moment: the change brought about by the interference is "transmitted" to the later parts of the process. If a
light beam consists of white light (or a suitably restricted set of frequencies), we can put a filter in the beam's path,
e.g., separating out only the red frequencies. The light beam after it passes through the filter will bear the "mark" of
having done so: it will now be red in color. Contrast this with the case of the light spot on the wall: if we put a red filter
at one point in the process, the spot will turn red for just that moment and then carry on as if nothing had happened.
Interfering with the process will leave no "mark." (For another example, consider an arrow on which a spot of paint is
placed, as opposed to the shadow of the arrow. The first mark is transmitted, but the second is not.)
Thus, Salmon concludes that a causal process is a spatiotemporal continuous process that can transmit information
(a "mark"). He emphasizes that the "transmission" referred to here is not an extra, mysterious event that connects
two parts of the process. In this regard, he propounds his "at-at" theory of mark transmission: all that transmission of
a mark consists in is that the mark occurs "at" one point in the process and then remains in place "at" all subsequent
points unless another causal interaction occurs that erases the mark. (Here he compares the theory with Zeno's
famous arrow paradox. Explain. The motion consists entirely of the arrow being at a certain point at a certain time; in
other words, the motion is a function from times to spatial points. This is necessary when we are considering a
continuous process. To treat as the conjunction of the discrete events of moving from A halfway to C, moving halfway
from there, and so on, leads to the paradoxical conclusion that the arrow will never reach its destination.

Transmission is not a "link" between discrete stages of a process, but a feature of that process itself, which is
continuous.
Now we turn to explanation. According to Salmon, a powerful explanatory principle is that whenever there is a
coincidence (correlation) between the features of two processes, the explanation is an event common to the two
processes that accounts for the correlation. This is a "common cause." To cite an example discussed earlier, there is
a correlation between lung cancer (C) and nicotine stains on a person's fingers (N). That is,
pr(C|N) > pr(C).
The common cause of these two events is a lifetime habit of smoking two packs of cigarettes each day (S). Relative
to S, C and N are independent, i.e.,
pr(C|N&S) = pr(C|S).
You'll sometimes see the phrase that S "screens C off from N" (i.e., once S is brought into the picture N becomes
irrelevant). This is part of a precise definition of "common cause," which is constrained by the formal probabilistic
conditions. We start out with pr(A|B) > pr(A). C is a common cause of A and B if the following hold.
pr(A&B|C) = pr(A|C)pr(B|C)
pr(A&B|C) = pr(A|C)pr(B|C)
pr(A|C) > pr(A|C)
pr(B|C) > pr(B|C)
(The first condition is equivalent to the screening off condition given earlier.) These conditions are also constrained:
A, B, and C have to be linked suitably as parts of a causal process known as a "conjunctive" fork.

(Consider the relation between this and the smoking-lung cancer, and leukemia-atomic blast cases given earlier.)
This does not complete the concept of causal explanation, however, since some common causes do not "screen off"
correlated by (causally) independent events. Salmon gives Compton scattering as an example. Given that an
electron e- absorbs a photon of a certain energy E and is given a bit of kinetic energy E* in a certain direction as a
result, a second photon will be emitted with E** = E - E*. The energy levels of the emitted photon and of the electron
will be correlated, even given that the absorption occurred. That is,
pr(A&B|C) > pr(A|C)pr(B|C).
This is a causal interaction of a certain sort, between two processes (the electron and the photon). We can use the
probabilistic conditions here to analyze the concept: "when two processes intersect, and both are modified in such
ways that the changes in one are correlated with changes in the other--in the manner of an interactive fork--we have
causal interaction." Thus, a second type of "common cause" is provided by the C in the interactive fork.
Salmon's attempt here is to analyze a type of explanation that is commonly used in science, but the notion of causal
explanation can be considered more broadly than he does. For example, Lewis points out that the notion of a causal
explanation is quite fluid. In his essay on causal explanation, he points out that there is an extremely rich causal

history behind every event. (Consider the drunken driving accident case.) Like Salmon, Lewis too argues that to
explain an event is to provide some information about its causal history. The question arises, what kind of
information? Well, one might be to describe in detail a common cause of the type discussed by Salmon. However,
there might be many situations in which we might only want a partial description of the causal history (e.g., we are
trying to assign blame according to the law, or we already know a fair chunk of the causal history and are trying to
find out something new about it, or we just want to know something about the type of causal history that leads to
events of that sort, and so on). Lewis allow negative information about the causal history to count as an explanation
(there was nothing to prevent it from happening, there was no state for the collapsing star to get into, there was no
connection between the CIA agent being in the room and the Shah's death, it just being a coincidence, and so on). To
explain is to give information about a causal history, but giving information about a causal history is not limited to
citing one or more causes of the event in question.
(Mention here that Lewis has his own analysis of causation, in terms of non-backtracking counterfactuals.)
Now given this general picture of explanation, there should be no explanations that do not cite formation about the
causal history of a particular event. Let us consider whether this is so. Remember the pattern of D-N explanation that
we looked at earlier, such as a deduction of the volume of a gas that has been heated from the description of its
initial state, how certain things such as temperature changed (and others, such as pressure, did not), and an
application of the ideal gas law PV = nRT. On Hempel's view, this could count as an explanation, even though it is
non-causal. Salmon argues that (1) non-causal laws allow for "backwards" explanations, and (2) cry out to be
explained themselves. Regarding the latter point, he says that non-causal laws of this sort are simply descriptions of
empirical regularities that need to be explained. The same might occur in the redshift case, if the law connecting the
redshift with the velocity was simply an empirical generalization. (Also, consider Newton's explanation of the tides.)
Let's consider another, harder example. A star collapses, and then stops. Why did it stop? Well, we might cite the
Pauli Exclusion Principle (PEP), and say that if it had collapsed further, there would have been electrons sharing the
same overall state, which can't be according the PEP. Here PEP is not causing the collapse to stop; it just predicts
that it will stop. Lewis claims that the reason this is explanatory is that it falls into the "negative information" category.
The reason that the star stopped collapsing is that there was no physically permissible state for it to get into. This is
information about its causal history, in that it describes the terminal point of that history.
There are other examples, from quantum physics, especially, that seem to give the most serious problems for the
causal view of explanation, especially Salmon's view that explanations in science are typically "common cause"
explanations. One basic problem is that Salmon's view relies on spatiotemporal continuity, which we cannot assume
at the particulate level. (Consider tunneling.) Also, consider the Bell-type phenomenon, where we have correlated
spin-states, and one particle is measured later than the other. Why did the one have spin-up when it was measured?
Because they started out in the correlated state and the other one that was measured had spin-down. You can't
always postulate a conjunctive fork. This should be an interactive fork of the type cited by Salmon, since given the
set-up C there is a correlation between the two events. However, it is odd to say the setup "caused" the two events
when they were both space-like separated! We will consider these problems in greater detail next week.

Lecture 5
2/15/93
The Causal Theory Of Explanation, Part III
Last time, we discussed Salmon's analysis of causal explanation. To review, Salmon said that explanation involved
(1) statistical relevance, (2) connection via a causal process, and (3) change after a causal interaction. The change is
the event to be explained. The notion of a causal process was filled out in terms of (a) spatiotemporal continuity and
(b) the ability to transmit information (a "mark"). While we can sometimes speak simply of cause and effect being
parts of a single causal process, the final analysis will typically be given in terms of more complex (and sometimes
indirect) causal connections, of which Salmon identifies two basic types: conjunctive and interactive forks.
Today, we are going to look at these notions a bit more critically. To anticipate the discussion, we will discuss in turn
(1) problems with formulating the relevant notion of statistical relevance (as well as the problems associated with a
purely S-R view of explanation), (2) the range of causal information that can be offered in an explanation, and (3)
whether some explanations might be non-causal. If we have time, we will also discuss whether (probabilistic or nonprobabilistic) causal laws have to be true to play a role in good explanations (Cartwright).
Statistical Relevance
The basic problem with statistical relevance is specifying what is relevant to what. Hempel first noticed in his analysis
of statistical explanation that I-S explanations, unlike D-N explanations, could be weakened by adding further
information. For example, given that John has received penicillin, he is likely to recover from pneumonia; however,
given the further information that he has a penicillin-resistant strain of pneumonia, he is unlikely to recover. Thus, we
can use true information and statistical laws to explain mutually contradictory things (recovery, non-recovery). On the
other hand, note that we can also strengthen a statistical explanation by adding more information (in the sense that
the amount of inductive support the explanans gives the explanandum can be increased). This "ambiguity" of I-S
explanation--relative to one thing, c explains e, relative to another, it does not--distinguishes it in a fundamental way
from D-N explanation.
As you know, the inferential view said that the explanans must confer a high probability on the explanandum for the
explanation to work; Salmon and other causal theorists relaxed that requirement and only required that the
explanans increase the probability of the explanandum, i.e., that it be statistically relevant to the explanandum. Still,
the ambiguity remains. The probability that Jones will get leukemia is higher given that he was located two miles
away from an atomic blast when it occurred; but it is lowered again when it is added that he had on lead shielding
that completely blocked the effects of any radiation that might be in the area. This led Hempel, and Salmon, too, to
add that the explanation in question must refer to statistical laws stated in terms of a "maximally" specific reference
class (i.e., the class named in the "given" clause) to be an explanation. In other words, it is required that dividing the
class C further into C1, C2, and so on would not affect the statistics, in that pr(E|Ci) = pr(E|Cj). This can be
understood in two ways, either "epistemically," in terms of the information we have at our disposal, of "objectively," in
terms of the actual "objective" probabilities in the world. (Hempel only recognized the former.) If our reference class
can't be divided ("partitioned") into cells that give different statistics for E, then we say that the class is
"homogeneous" with respect to E. The homogeneity in question can be either epistemic or objective: it must be the
latter if we are really talking about causes rather than what we know about causes.
The problem with this is that dividing up the class can result in trivialization. For example, a subclass of the class of
people who receive penicillin to treat their pneumonia (P) is the class of those people who recover (R). Obviously, it is
always the cause that pr(|P&R) = 1. However, this type of statistical law would not be very illuminating to use in a
statistical explanation of why the person recovered from pneumonia.
There are various ways around this problem. For example, you can require that the statistical law in question not be
true simply because it is a theorem of the probability calculus (which was the case with pr(R|P&R) = 1). Hempel used

this clause in his analysis of I-S explanation. Salmon adds that we should further restrict the clause by noting that the
statistical law in question not refer to a class of events that either (1) follow the explanandum temporally, or (2)
cannot be ascertained as true or false in principle independently of ascertaining the truth or falsity of the
explanandum. The first requirement is used to block explanations of John's recovery that refer to the class of people
who are reported on the 6:00 news to have recovered of pneumonia (supposing John is famous enough to merit
such a report). This is the requirement of maximal specificity (Hempel) or that the reference class be statistically
homogeneous (Salmon).
Of course, as we mentioned earlier, there might be many correlations that exist in the world between accidental
events, such as that someone in Laos sneezes (S) whenever a person here recovers from pneumonia (R), so that
we have pr(R|P&S) > pr(R|P). (Here the probabilities might simply be a matter of the actual, empirical frequencies.) If
this were the case, however, we would not want to allow the statistical law just cited to occur in a causal explanation,
since it may be true simply by "accident." We might also demand that there be causal processes linking the two
events. That's why Salmon was concerned to add that the causal processes that link the two events must be
specified in a causal explanation. The moral of this story is two-fold. Statistical relevance is not enough, even when
you divide up the world in every way possible. Also, some ways of dividing up the world to get statistical relevance
are not permissible.
What Kinds Of Causal Information Can Be Cited In An Explanation?
Salmon said that there were two types of causal information that could be cited in a causal explanation, which we
described as conjunctive and interactive forks. Salmon's purpose here is to analyze a type of explanation that is
commonly used in science, but the notion of causal explanation can be considered more broadly than he does. For
example, Lewis points out that the notion of a causal explanation is quite fluid. In his essay on causal explanation, he
points out that there is an extremely rich causal history behind every event. (Consider the drunken driving accident
case.) Like Salmon, Lewis too argues that to explain an event is to provide some information about its causal history.
The question arises, what kind of information? Well, one might be to describe in detail a common cause of the type
discussed by Salmon. However, there might be many situations in which we might only want a partial description of
the causal history (e.g., we are trying to assign blame according to the law, or we already know a fair chunk of the
causal history and are trying to find out something new about it, or we just want to know something about the type of
causal history that leads to events of that sort, and so on).
Question: How far back in time can we go to statistically relevant events? Consider the probability that the accident
will occur. Relevant to this is whether gas was available for him to drive his car, whether he received a driver's license
when he was young, or even whether he was lived to the age that he did. All of these are part of the "causal history"
leading up to the person having an accident while drunk, but we would not want to cite any of these as causing the
accident. (See section, "Explaining Well vs. Badly.")
Something to avoid is trying to make the distinction by saying that the availability of gas was not "the" cause of the
person's accident. We can't really single out a given chunk of the causal history leading up to the event to separate
"the cause." Lewis separates the causal history--any portion of which can in principle be cited in a given explanation-from the portion of that causal history that we are interested in or find most salient at a given time. We might not
interested in information about any portion of the causal history, Lewis says, but it remains the case that to explain
and event is to give information about the causal history leading up to that event.
In addition, Lewis points out that the range of ways of giving information about causal histories is quite broad. For
example, Lewis allow negative information about the causal history to count as an explanation (there was nothing to
prevent it from happening, there was no state for the collapsing star to get into, there was no connection between the
CIA agent being in the room and the Shah's death, it just being a coincidence, and so on). To explain is to give
information about a causal history, but giving information about a causal history is not limited to citing one or more
causes of the event in question.

Lecture 6
2/17/94
Problems With The Causal Theory Of Explanation
Last time we finished our exploration of the content of causal theories of explanation, including the kinds of caveats
that have to be added to make the theory workable. Today, we will examine whether the causal approach as a whole
is plausible, and examine an alternative view, namely van Fraassen's pragmatic theory of explanation. (Note
Kourany's use of the term "erotetic.") As I stated at the end of the period last time, there are two basic challenges that
can be given to the causal view, namely that sometimes non-causal generalizations can explain, and that laws can
be explained by other laws (a relationship that does not seem to be causal, since laws don't cause other laws, since
neither is an event).
(1) Non-causal generalizations - Suppose that someone was ignorant of the various gas laws, or once learned the
laws and has now forgotten them.
General Law: PV = nRT
Boyle's Law: At constant temperature, pressure is inversely proportional to volume, i.e., PV = constant.
Charles' Law: At constant pressure, volume is directly proportional to temperature, i.e., V/T = constant.
Pressure Law: At constant volume, pressure is directly proportional to temperature, i.e., P/T = constant.
They wonder why it is that a certain container containing gas expands when heated. You could then give various
answers, such as that the pressure was constant, and then cite Charles's law to finish the explanation. Alternately,
you might say that there was a more complex relationship, where the pressure increased along with the temperature,
but that the increase in pressure was not enough to compensate for the increase in temperature to that the volume
had to rise too, according to the general gas law. Q: Is this an explanation? We have to distinguish whether it's the
"ultimate" or "fundamental" explanation of the phenomenon in question or whether it's an explanation of the
phenomenon.
Example from statics: an awkwardly posed body is held in place by a transparent rod. Why is it suspended in that
odd way? Well, there's a rod here, and that compensates for the force of gravity.
Example from Lewis, the collapsing star: Why did it stop? Well, there is no causal story that we can tell, other than by
giving "negative" information: there was no state for it to get into if it collapsed further, because of Pauli Exclusion
Principle. (Here identical fermions cannot have the same quantum numbers, n, l, m, and ms.) Here PEP is not
causing the collapse to stop; it just predicts that it will stop. Lewis claims that the reason this is explanatory is that it
falls into the "negative information" category. The reason that the star stopped collapsing is that there was no
physically permissible state for it to get into. This is information about its causal history, in that it describes the
terminal point of that history.
(2) Explanation of Laws by Laws. Newton explained Kepler's laws (ellipses, equal areas in equal times, p2 = d3) by
deriving them from his laws of motion and gravitation (inertia, F = ma, action-reaction, and F = Gm1m2/r2). This is
the kind of explanation to which the inferential view is especially well suited (as well as the pragmatic view that we
will consider next time), but it does not fit immediately into the causal view of explanation (Newton's laws don't cause
Kepler's laws to be true.) That is because the causal view of explanation seems best-suited for explaining particular
events, rather than general regularities. Lewis responds to several objections to his theory by saying that his theory
does not intend to cover anything more than explanations of events. (Not a good answer as is, since then we would
not have a general theory of explanation, but only a description of certain kinds of explanation.) However, he does
have an answer at his disposal: one way to give information about causal histories is to consider what is common to
all causal histories for events of a given type (e.g., a planet orbiting around a star according to Kepler's laws). Q: Is
this enough?

There are of course the other problems mentioned earlier: Salmon's view relies on the notion of spatiotemporal
continuous processes to explain the notion of causal processes. (Lewis is not subject to this problem, since he has
an alternative theory of causation: linkage by chains of non-backtracking counterfactual dependence.)

Lecture 7
2/22/94
Van Fraassen's Pragmatic View Of Explanation
Van Fraassen's pragmatic view of explanation is that an explanation is a particular type of answer to a why-question,
i.e., an answer that provides relevant information that "favors" the event to be explained over its alternatives. For van
Fraassen, these features are determined by the context in which the why-question is asked.
The Basic Elements Of The Pragmatic View Of Explanation
According to van Fraassen, a why-question consists of (1) a presupposition (Why X), (2) a contrast class (Why X
rather than Y, Z, and so on), and (3) an implicitly understood criterion of relevance. Information given in response to a
particular why-question constitutes an explanation of the presupposition if the information is relevant and "favors" the
presupposition over the alternatives in its contrast class. (Explain and give examples.)
Both the contrast class and the criterion of relevance are contextually determined, based on interests of those
involved. Subjective interests define what would count as an explanation in that context, but then it's an objective
matter whether that information really favors the presupposition over the alternatives in its contrast class. (Explain
and give examples.)
Contrasts Between The Pragmatic And Causal Views Of Explanation
1. Any type of information can be counted as relevant (of course, it's a scientific explanation if only information
provided by science counts; however, there might be different kinds of scientific explanation; not any old
information will do).
2. Context (interests) determines when something counts as an explanation, vs. when we would find an
explanation interesting or salient. (According to Lewis, what makes it an explanation is that it gives information
about the causal history leading up to a given event; whether we find that explanatory information interesting or
salient is another matter.)
3. Distinction: Pragmatic theory of explanation vs. theory of the pragmatics of explanation. On the pragmatic view,
God could never have a "complete" explanation of an event, unless he had interests. (A mere description of the
causal history leading up to an event--even a complete one--is not an explanation of any sort according to the
pragmatic view.)
On the pragmatic view, asymmetries only exist because of the context; thus, they can be reversed with a change in
context. That is what van Fraassen's Tower example is supposed to illustrate. (Recount the Tower example.) Lewis'
Objection to the Tower example: What is really doing the explaining is the intention of the Prince, and that's a cause
of the flagpole being that particular height. Discuss: Can you think of a story in which the redshift would explain the
galaxies moving away? Where human intention is not possible, it seems difficult; this would seem to confirm Lewis'
diagnosis of the Tower story.

Lecture 8
2/24/94
Carnap vs. Popper
The idea is to say when we "test" scientific theories, and then form opinions regarding those theories based on such
tests. What does scientific investigation consist in, and what are the rules governing it?
Induction - Scientific investigation is a type of inductive process, where we increase the evidential basis for or against
a particular theory without the evidence conclusively establishing the theory. "Induction" has had various meanings in
the past: in the Renaissance, it was thought that the way to develop scientific theories was to examine all the
evidence you could and then extrapolate from that to form theories. (This was a method of developing theories as
well as a method of justifying them.) This was in contrast to positing "hypotheses" about unobservable entities to
explain the phenomena. (Indeed, Newton was criticized for formulating such "hypotheses.") This view did not survive,
however, since it became apparent that you can't form theories in this way. Thus, we have to understand "induction"
differently (supposing that it is a useful concept at all).
Carnap is an inductivist, and in this respect he differs from Popper. However, both agree (taking inspiration from
Hume) that there is a serious problem with the justification of "inductive inference." Carnap discusses it in terms of a
puzzle about how we arrive at and form opinions regarding laws. (Note that notion of a law that Carnap assumes is
similar to Hempel's.) Laws are universal statements (at least), hence apply to an at least potentially infinite domain.
However, our empirical data is always finite. (Consider ideal gas law.) What does deductive logic give us to evaluate
theories?
Suppose that h -> e1, e2, e3,.... If we show the truth any finite number of the ei, we haven't established that h;
however, if we show that one of these is false, h must also be false.
Thus deductive logic cannot give us the tools to establish a scientific theory; we cannot "infer" from evidence to
theory. There are different ways in which this can be so. Carnap distinguishes between empirical and theoretical
laws. The latter refer to unobservable entities or properties; theoretical laws are absolutely necessary in science, but
they cannot simply be derived from large bodies of research. Science postulates new categories of things not just
generalizations about regularities of things we can observe. (Consider the kinetic theory of gases.) The upshot is that
theories can always be wrong, no matter how much evidence we've found that is consistent with those theories.
Scientific theories are not "proven" in the sense that given that a certain body of empirical data they are immune from
all refutation. Q: How can we form an opinion regarding theories? Moreover, what kind of opinion should we form?
Both of these authors assume that the answer to this question will be in the form of a "logic of scientific discovery."
(Logic = formal system of implication, concerned with relations of implication between statements, but not relations of
statements to the world, i.e., whether they are true or false). Indeed, this is the title of Popper's famous book. The
point on which they differ is the following: Is deductive logic all that we're limited to in science? (Popper - yes, no
inductive logic; Carnap - No, there's an "inductive logic" too, which is relevant to scientific investigation.)
Popper and Carnap also agree on is that there is a distinction between the contexts of justification and discovery.
This is a contentious view, as we'll see when we start looking at Kuhn and Laudan. The traditional approach to
induction assumed that gathering evidence would enable one to formulate and justify a theory. (Consider Popper's
example of a person who gave all his "observations" to the Royal Society. Also, consider the injunction to "Observe!")
Both Carnap and Popper distinguish the two: what they were concerned with is not how to come up with scientific
hypotheses--this is a creative process (e.g., the formulation of a theoretical framework or language), not bound by
rules of logic but only (perhaps) laws of psychology--but how to justify our hypotheses once we come up with them.
Carnap's answer is that we can't "prove" our hypotheses but we can increase (or decrease) their probability by
gathering evidence. (This is inductive since you proceed to partial belief in a theory by examining evidence.)
Let's make this a bit more precise: is induction a kind of argument? That is, is there such a thing as "inductive
inference?" This is one view of inductive logic: a formal theory about relations of (partial) implication between
statements. Can be thought of in two ways. (1) You accept a conclusion (all-or-nothing) based on evidence that
confirms a theory to a certain degree (e.g., if sufficiently high). (2) You accept a conclusion to a certain degree (i.e.,

as more or less probable) based on certain evidence. The latter is what Carnap's approach involves. The basic
notion there is degree of confirmation. What science does when it "justifies" a theory (or tests it) is to provide
evidence that confirms or disconfirms it to a certain degree, i.e., makes it more or less probable than it was before
the evidence was considered.
We've already made informal use of the notion of probability. However, the precise sense of the term turns out to be
of importance when thinking about induction in Carnap's sense.
Frequentist (Statistical)
Inductive (Logical)
Objective (Physical Propensity)
Subjective (Personal Degree of Belief)
Carnap thought that the Logical notion was the one operative in scientific reasoning. Analogy with deductive logic:
formal, relation between statements because of their formal properties (i.e., irrespective of what facts obtain).
Disanalogy: No acceptance or belief: pr(h|e) = x means that e partially implies h, to degree x. Only partial belief,
guided by the logical probabilities. The latter is a matter of the logical relationship between the two statements; it
should guide our opinion (degrees of belief), but it does not on Carnap's view reduce to it. Scientific reasoning is then
the formulation of a broad framework (language) in which theories can be expressed; inherent in that framework will
be relations of partial implication (conditional logical probabilities) between evidence and hypotheses, i.e., pr(h|e).
Evidence confirms the theory if pr(h|e) > pr(h) (and if pr(h|e) is sufficiently high). Then making an observation
involves determining whether e is true; we seek to develop and confirm theories in this way.
Popper rejected this framework altogether. Popper thought that Carnap's inductive logic--indeed, the very idea of
inductive logic--was fundamentally mistaken. (How do we establish our starting point in Carnap's theory, i.e., the
logical probabilities? It didn't work out very well.) There is no such thing: the only logic available to science is
deductive logic. (Note that this doesn't mean that you can't use statistics. Indeed, you'll use it frequently, though the
relationship between statistical laws and statistical data will be deductive.) What is available then? Well, we can't
verify a theory (in the sense of justifying belief or partial belief in a hypothesis by verifying its predictions), but we can
certainly falsify a theory using purely deductive logic. If h e1, e2, e3, ..., then if even one of e1, e2, e3, ... turn out to
be false, the theory as a whole is falsified, and must be rejected (unless it can be amended to account for the falsity
of the falsifying evidence). Thus, Popper argues that science proceeds ("advances") by a process of conjecture and
refutation. He summarizes several important features of this process as follows (page 141).

It's too easy to get "verifying" evidence; thus, verifying evidence is of no intrinsic value.
To be of any use, predictions should be risky.
Theories are better (have more content) the more they restrict what can happen.
Theories that are not refutable by some possible observation are not scientific (criterion of demarcation).
To "test" a theory in a serious sense is to attempt to falsify it.
Evidence on "corroborates" a theory if it is the result of a serious ("genuine") test.

The process is then to start with a conjecture and try to falsify it; it that succeeds, move on to the next conjecture,
and so on, until you find a conjecture that you do not falsify. Keep on trying, though. If you have trouble falsifying it,
you say that it has been "corroborated." This does not mean that it has a high degree of probability, however. It still
may be improbable, given the evidence at hand. (Indeed, it should be improbable if it says anything of interest.) We
only tentatively accept scientific theories, while continuing to try to refute them. (Here "tentative acceptance" does not
mean to believe that they are true, or even to be highly confident in their truth.)

Lecture 9
3/1/94
An Overview Of Kuhn's The Structure Of Scientific Revolutions
Today we will be looking at Kuhn's The Structure of Scientific Revolutions (SSR) very broadly, with the aim of
understanding its essentials. As you can gather from the title of Kuhn's book, he is concerned primarily with those
episodes in history known as "scientific revolutions." During periods of this sort, our scientific understanding of the
way the universe works is overthrown and replaced by another, quite different understanding.
According to Kuhn, after a scientific discipline matures, its history consists of long periods of stasis punctuated by
occasional revolutions of this sort. Thus, a scientific discipline goes through several distinct types of stages as it
develops.
I. The Pre-Paradigmatic Stage
Before a scientific discipline develops, there is normally a long period of somewhat inchoate, directionless research
into a given subject matter (e.g., the physical world). There are various competing schools, each of which has a
fundamentally different conception of what the basic problems of the discipline are and what criteria should be used
to evaluate theories about that subject matter.
II. The Emergence Of Normal Science
Out of the many competing schools that clutter the scientific landscape during a discipline's pre-paradigmatic period,
one may emerge that subsequently dominates the discipline. The practitioners of the scientific discipline rally around
a school that proves itself able to solve many of the problems it poses for itself and that holds great promise for future
research. There is typically a particular outstanding achievement that causes the discipline to rally around the
approach of one school. Kuhn calls such an achievement a "paradigm."
A. Two Different Senses Of "Paradigm"--Exemplar And Disciplinary Matrix
Normal science is characterized by unanimous assent by the members of a scientific discipline to a particular
paradigm. In SSR, Kuhn uses the term paradigm to refer to two very different kinds of things.
1. Paradigms As Exemplars
Kuhn at first uses the term "paradigm" to refer to the particular, concrete achievement that defines by example the
course of all subsequent research in a scientific discipline. In his 1969 postscript to SSR, Kuhn refers to an
achievement of this sort as an "exemplar." Among the numerous examples of paradigms Kuhn gives are Newton's
mechanics and theory of gravitation, Franklin's theory of electricity, and Copernicus' treatise on his heliocentric theory
of the solar system. These works outlined a unified and comprehensive approach to a wide-ranging set of problems
in their respective disciplines. As such, they were definitive in those disciplines. The problems, methods, theoretical
principles, metaphysical assumptions, concepts, and evaluative standards that appear in such works constitute a set
of examples after which all subsequent research was patterned. (Note, however, that Kuhn's use of the term
"paradigm" is somewhat inconsistent. For example, sometimes Kuhn will refer to particular parts of a concrete
scientific achievement as paradigms.)
2. Paradigms As Disciplinary Matrices
Later in SSR, Kuhn begins to use the term "paradigm" to refer not only to the concrete scientific achievement as
described above, but to the entire cluster of problems, methods, theoretical principles, metaphysical assumptions,
concepts, and evaluative standards that are present to some degree or other in an exemplar (i.e., the concrete,
definitive scientific achievement). In his 1969 postscript to SSR, Kuhn refers to such a cluster as a "disciplinary
matrix." A disciplinary matrix is an entire theoretical, methodological, and evaluative framework within which scientists
conduct their research. This framework constitutes the basic assumptions of the discipline about how research in that
discipline should be conducted as well as what constitutes a good scientific explanation. According to Kuhn, the
sense of "paradigm" as a disciplinary matrix is less fundamental that the sense of "paradigm" as an exemplar. The
reason for this is that the exemplar essentially defines by example the elements in the framework that constitutes the
disciplinary matrix.

B. Remarks On The Nature Of Normal Science

1. The Scientific Community
According to Kuhn, a scientific discipline is defined sociologically: it is a particular scientific community, united by
education (e.g., texts, methods of accreditation), professional interaction and communication (e.g., journals,
conventions), as well as similar interests in problems of a certain sort and acceptance of a particular range of
possible solutions to such problems. The scientific community, like other communities, defines what is required for
membership in the group. (Kuhn never completed his sociological definition of a scientific community, instead leaving
the task to others.)
2. The Role Of Exemplars
Exemplars are solutions to problems that serve as the basis for generalization and development. The goal of
studying an exemplar during one's scientific education is to learn to see new problems as similar to the exemplar,
and to apply the principles applicable to the exemplar to the new problems. A beginning scientist learns to abstract
from the many features of a problem to determine which features must be known to derive a solution within the
theoretical framework of the exemplar. Thus, textbooks often contain a standard set of problems (e.g., pendulums,
harmonic oscillators, inclined plane problems). You can't learn a theory by merely memorizing mathematical formulas
and definitions; you must also learn to apply these formulas and definitions properly to solve the standard problems.
This means that learning a theory involves acquiring a new way of seeing, i.e., acquiring the ability to group problems
according to the theoretical principles that are relevant to those problems. The "similarity groupings" of the mature
scientist distinguish her from the scientific neophyte.
3. Normal Science As "Puzzle-Solving"
According to Kuhn, once a paradigm has been accepted by a scientific community, subsequent research consists of
applying the shared methods of the disciplinary matrix to solve the types of problems defined by the exemplar. Since
the type of solution that must be found is well defined and the paradigm "guarantees" that such a solution exists
(though the precise nature of the solution and the path that will get you to a solution is often not known in advance),
Kuhn characterizes scientific research during normal or paradigmatic science as "puzzle-solving."
III. The Emergence Of Anomaly And Crisis
Though the paradigm "guarantees" that a solution exists for every problem that it poses, it occasionally happens that
a solution is not found. If the problem continues to persist after repeated attempts to solve it within the framework
defined by the paradigm, scientists may become acutely distressed and a sense of crisis may develop within the
scientific community. This sense of desperation may lead some scientists to question some of the fundamental
assumptions of the disciplinary matrix. Typically, competing groups will develop strategies for solving the problem,
which at this point has become an "anomaly," that congeal into differing conceptual "schools" of thought much like
the competing schools that characterize pre-paradigmatic science. The fundamental assumptions of the paradigm
will become subject to widespread doubt, and there may be general agreement that a replacement must be found
(though often many scientists continue to persist in their view that the old paradigm will eventually produce a solution
to the apparent anomaly).
IV. The Birth And Assimilation Of A New Paradigm
Eventually, one of the competing approaches for solving the anomaly will produce a solution that, because of its
generality and promise for future research, gains a large and loyal following in the scientific community. This solution
comes to be regarded by its proponents as a concrete, definitive scientific achievement that defines by example how
research in that discipline should subsequently be conducted. In short, this solution plays the role of an exemplar for
the group--thus, a new paradigm is born. Not all members of the scientific community immediately rally to the new
paradigm, however. Some resist adopting the new problems, methods, theoretical principles, metaphysical
assumptions, concepts, and evaluative standards implicit in the solution, confident in their belief that a solution to the
anomaly will eventually emerge that preserves the theoretical, methodological, and evaluative framework of the old
paradigm. Eventually, however, most scientists are persuaded by the new paradigm's growing success to switch their
loyalties to the new paradigm. Those who do not may find themselves ignored by members of the scientific
community or even forced out of that community's power structure (e.g., journals, university positions). Those who
hold out eventually die. The transition to the new paradigm is complete.

Lecture 10
3/3/94
Paradigms and Normal Science
Last time we examined in a very general way Kuhn's account of the historical development of particular scientific
disciplines. To review, Kuhn argued that a scientific discipline goes through various stages: the pre-paradigmatic,
paradigmatic ("normal"), and revolutionary (transitional, from one paradigm to another). Each stage is characterized
in terms of the notion of a paradigm, so it is highly important that we discuss this notion in detail. Today, we will limit
ourselves primarily to the context of the transition from pre-paradigmatic science to "normal" (paradigm-governed)
science.
Paradigms
Let's look a bit a Kuhn's characterization of the notion of paradigm. He introduces paradigms first as "universally
recognized scientific achievements that for a time provide model problems and solutions to a community of
practitioners" (page x). A paradigm is "at the start largely a promise of success discoverable in selected and still
incomplete examples" (pages 23-24), and it is "an object for further articulation and specification under new or more
stringent conditions" (page 23); hence from paradigms "spring particular coherent traditions of scientific research"
(page 10) that Kuhn calls "normal science." Normal science consists primarily of developing the initial paradigm "by
extending the knowledge of those facts that the paradigm displays as particularly revealing, by increasing the extent
of the match between those facts and the paradigm's predictions, and by further articulation of the paradigm itself"
(page 24). The paradigm provides "a criterion for choosing problems that, while the paradigm is taken for granted,
can be assumed to have solutions" (page 27). Those phenomena "that will not fit the box are often not seen at all"
(page 24). Normal science "suppresses fundamental novelties because they are necessarily subversive of its basic
commitments." Nevertheless, not all problems will receive solutions within the paradigm, even after repeated
attempts, and so anomalies develop that produce "the tradition-shattering complements to the tradition-bound activity
of normal science" (page 6).
At the outset of SSR, we are told that a paradigm is a concrete achievement that provides a model for subsequent
research. Such achievements are referred to in textbooks, lectures, and laboratory exercises; typically, they are the
standard problems that a student is required to solve in learning the discipline. The task given the student provides
some of the content of the mathematical equations (or more generally, theoretical descriptions) that comprise the
main body of the text. In other parts of SSR, however, we are told that a paradigm is much more, e.g., that it includes
law, theory, application, and instrumentation together (page 10); or that it is a set of commitments of various sorts,
including conceptual, theoretical, instrumental, methodological, and quasi-metaphysical commitments (pages 41-42).
Paradigms sometimes are characterized as definitive, concrete patterns or models for subsequent research, but at
other times seem to be characterized as vague theories or theory schemas to be subsequently articulated. In its
broadest sense, the paradigm is taken to included theories, laws, models, concrete applications
(exemplars--"paradigms" in the narrower sense), explicit or implicit metaphysical beliefs, standard for judging
theories, and particular sets of theoretical values. In short, anything that is accepted or presupposed by a particular
scientific community can seemingly be part of a "paradigm."
There is no doubt that all these elements are present in science. The question is whether it is informative to
characterize science in this way. A basic problem is as follows: is "paradigm" is defined in its broadest sense, where
anything that is accepted or presupposed by a scientific community is part of the "paradigm" that defines that
community, then it is a relatively trivial matter to say that there are paradigms. (That is, it's not really a substantive
historical thesis to say that scientific eras are defined by the universal acceptance of a paradigm if a paradigm is
simply defined as whatever is universally accepted.)
In his 1969 Postscript to SSR, Kuhn recognizes these problems and distinguishes between two senses of "paradigm"
as used in SSR: a "disciplinary matrix" (framework) and an "exemplar." The former (disciplinary matrix) is the entire
framework--conceptual, methodological, metaphysical, theoretical, and instrumental--assumed by a scientific
tradition. The latter (exemplars) are the concrete, definitive achievements upon which all subsequent research is
patterned. Kuhn's thesis about paradigms is not empty, since he argues that the definitive, concrete achievement

("paradigm" in the narrow sense) provides the foundation of the disciplinary matrix ("paradigm" in the broader sense).
In other words, a scientific tradition is defined not by explicitly stated theories derived by explicit methodological rules,
but by intuitive abstraction from a particular, concrete achievement. Let's look at this in more detail.
The function of textbook examples - Textbook examples provide much of the content of the mathematical or
theoretical principles that precede them. Here's a common phenomenon: you read the chapter, think you understand,
but can't do the problems. This doesn't evince any failure on your part to understand the text, simply shows that
understanding the theory cannot be achieved simply by reading a set of propositions. Learning the theory consists in
applying it (it's not that you learn the theory first and then learn to apply it). In other words, knowledge of a theory is
not always propositional knowledge (knowledge that--accumulation of facts); it is sometimes procedural or
judgmental knowledge (knowledge how--acquiring judgmental skills). Scientists agree on the identification of a
paradigm (exemplar), but not necessarily on the full interpretation or rationalization of the paradigm (page 44).
Scientific training is a process of incorporation into a particular community - The goal of training is to get the student
to see new problems as like the standard problems in a certain respect: to group the problems that he will be faced
with into certain similarity classes, based on the extent to which they resemble the concrete, standard exemplars.
Being able to group things in the right way, and to attack similar problems using methods appropriate for solving that
particular type of problem evinces understanding and so incorporation into the community. Kuhn's thesis is that it
does not evince the internalization of methodological rules explicit or implicit in scientific procedure. (Here a "rule" is
understood as a kind of statement about how to proceed in certain circumstances.) That is not to say that none, or
even most, of scientific practice can be codified into rules; it is just that rules of how to proceed are not required, nor
does scientific competence (in the sense of judgmental skills) consist in the acquisition of rules (statements about
how to proceed) and facts (statements about the way the world is). This distinguishes him from Carnap and Popper,
who see at least the "justification" of scientific theories are a rule-governed procedure.
Thus, to adopt a paradigm is to adopt a concrete achievement as definitive for the discipline. (Example: Newton's
Principia, with its application to particular problems such as the tides, the orbits of the planets, terrestrial motion such
as occurs with projectiles, pendulums, and springs, and so on.) It's definitive in the sense that the methods that were
used, the result that was obtained, and the assumptions behind the methods (mathematical generalizations,
techniques for relating the formalism to concrete situations, etc.). The discipline then grows by extending these
procedures to new areas--the growth is not simply one of "mopping up" however (despite Kuhn's own
characterization of it as such), but rather an extension of the disciplinary matrix (framework) by organic growth. (No
sufficient & necessary conditions; instead, a family resemblance.)
Normal Science
Pre-Paradigmatic Science - Kuhn's model is the physical sciences; almost all of his examples are taken from
Astronomy, Physics, or Chemistry. This affects his view of science. Consider, on the other hand, Psychology,
Sociology, Anthropology, are they "pre-paradigmatic"? Is there any universally shared framework? If so, are they
sciences at all?
Some properties - each person has to start anew from the foundations (or at least, each subgroup); disagreement
over fundamentals (what counts as an interesting problem, what methods should be used to solve the problem, what
problems have been solved); in this context, all facts seem equally relevant:
In the absence of a paradigm or some candidate for a paradigm, all of the facts that could possibly pertain to the
development of a given science are likely to seem equally relevant. As a result, early fact-gathering is a far more
nearly random activity than the one that subsequent scientific research makes familiar ... [and] is usually restricted to
the wealth of data that lie ready to hand (page 15).
Thus, some important facts are missed (e.g., electrical repulsion).
Why accept a paradigm?

It solves a class of important though unsolved problems.

The solution has great scope and so holds much promise for generating further research. That is, it must be
both broad enough and incomplete enough to provide the basis for further research: "it is an object for
further articulation and specification under new or more stringent conditions" (page 23).

It gives direction and focus to research, i.e., selects out a feasible subset of facts, possible experiments as
promising or relevant. It says what kinds of articulations are permissible, what kinds of problems have
solutions, what kind of research is likely to lead to a solution, and the form an adequate solution can take. A
puzzle: the outcome is known, the path that one takes to it is not (also, it requires ingenuity to get there).

Important to recognize that the form of the solution is not exactly specified, but only the categories in which it will be
framed and the set of allowable paths that will get one there. However, this again is not a rule-bound activity, but
imitation (and extension) of a pattern. "Normal science can proceed without rules only so long as the relevant
scientific community accepts without question the particular problems-solutions already achieved. Rules should
therefore become important and the characteristic unconcern about them should vanish whenever paradigms or
models are felt to be insecure" (page 47).
Discuss: What level of unanimity exists in present sciences? Aren't there disagreements even in physics?

Lecture 11
3/8/94
Anomaly, Crisis, and the Non-Cumulativity of Paradigm Shifts
Last time we discussed the transition from pre-paradigmatic science to paradigmatic or "normal" science, and what
that involves. Specifically, we discussed the nature and role of a paradigm, distinguishing between the primary,
narrow sense of the term (an exemplar, i.e., a definitive, concrete achievement) and a broader sense of the term (a
disciplinary matrix or framework, which includes conceptual, methodological, metaphysical, theoretical, and
instrumental components). We also noted that though Kuhn usually talks about each scientific discipline (as
distinguished roughly by academic department) as having its own paradigm, the notion is more flexible than that and
can apply to sub-disciplines (such as Freudian psychoanalysis) within a broader discipline (psychology). The
question of individuating paradigms is a difficult one (though Kuhn sometimes speaks as if it's a trivial matter to
identify a paradigm), but we will not investigate it any further. Now we want to turn to how paradigms lead to their
own destruction, by providing the framework necessary to discover anomalies, some of which lead to the paradigm's
downfall.
As you recall, normal science is an enterprise of puzzle-solving according to Kuhn. Though the paradigm
"guarantees" that the puzzles it defines have solutions, this is not always the case. Sometimes puzzles cannot admit
of solution within the framework (disciplinary matrix) provided by the paradigm. For example, the phlogiston theory of
combustion found it difficult to explain the notion of weight gain when certain substances were burned or heated.
Since combustion was the loss of a substance on that view, there should be weight loss. This phenomenon was not
taken to falsify the theory, however (contrary to what Popper might claim); instead, phlogiston theorists attempted to
account for the difference by postulating that phlogiston had "negative mass," or that "fire particles" sometimes
entered an object upon burning. The paradigm eventually collapsed for several reasons:

None of the proposed solutions found general acceptance; i.e., there was a multiplicity of competing
solutions to the puzzle presented by the anomaly (weight gain).
The proposed solutions tended to create as many problems as they solved; one method of explaining
weight gain tended to imply that there would be weight gain in cases in which there was none.
This led to a sense of crisis among many practitioners in the field; the field begins to resemble the preparadigmatic stage (i.e., nearly random attempts at observation, experimentation, theory formation),
signaling the demise of the old paradigm. Nevertheless, it is not abandoned: to abandon it would be to
abandon science itself, on Kuhn's view.
Eventually, a competitor arose that seemed more promising than any of the other alternatives, but which
involved a substantial conceptual shift. This was the oxygen theory of combustion.

Paradigm Change as Non-Cumulative

There is a common but oversimplified picture of science that sees it as a strictly cumulative enterprise (science
progresses by finding out more and more about the way the world works). The "more and more" suggests that
nothing is lost. Kuhn argues that on the contrary there are substantial losses as well as gains when paradigm shifts
occur. Let's look at some examples.

Some problems no longer require solution, either because they make no sense in the new paradigm, or they
are simply rejected.
Standards for evaluating scientific theories alter along with the problems that a theory must according to the
paradigm solve.

Example: Newtonian physics introduced an "occult" element (forces), against the prevailing view of the corpuscular
view that all physical explanation had to be in terms of collisions and other physical interactions between particles.
Newton's theory did not accord with that standard, but it solved many outstanding problems. (Corpuscular view could
not explain other than in a rough qualitative way why planets would move in orbits; Kepler's Laws were separate
descriptions of why this was so. Thus it was a great achievement when postulating forces led to the derivation of
Kepler's Laws.) "Forces" were perceived by many as odd, "magical" entities. Newton himself tried to develop a
corpuscular theory of gravitation, without success, as did many Newtonian scientists who followed him. Eventually,
when it became apparent that the effort was futile, the standard that corpuscular-mechanistic explanation was
required was simply disregarded, and gravitational attraction was accepted as an intrinsic, unexplainable property of
matter.
Another example: Chemistry before Dalton aimed at explaining the sensory qualities of compounds (colors, smells,
sounds, etc.). Dalton's atomic paradigm was only suited for explaining why compounds went together in certain
proportions; so when it was generally accepted the demand that a theory explain sensory qualities was dropped.
Other examples: (a) Phlogiston. The problems of explaining how phlogiston combined with calxes to form particular
metals were abandoned when phlogiston itself was abandoned. The ether (medium for electromagnetic radiation);
explaining why movement through the ether wasn't detectable vanished along with the ether. There simply was no
problem to be solved. (b) Michelson-Morley experiment. This experiment was first "explained" by Lorenz based on
his theory of the electron, which implied that since forces holding together matter were electromagnetic and hence
influenced by movement through the ether, parts of bodies contract in the direction of motion when moving through
the ether. Relativity explains the contraction, but in a new conceptual framework that does not include the ether.

Some new entities are introduced along with the paradigm; indeed, only make sense (i.e., can be
conceptualized) when introduced as such. (Oxygen wasn't "discovered" until the oxygen theory of
combustion was developed.)
Elements that are preserved may have an entirely different status. (The constant speed of light is a
postulate in special relativity theory, a consequence of Maxwell's theory.) Substantive conceptual theses
might be treated as "tautologies" (e.g., Newton's Second Law).

The fact that new standards, concepts, and metaphysical pictures are introduced makes the paradigms not only
incompatible, but also "incommensurable." Paradigm shift is a shift in worldview.

Lecture 12
3/10/94
Incommensurability
At the end of the lecture last time, I mentioned Kuhn's view that different paradigms are incommensurable, i.e., that
there is no neutral standpoint from which to evaluate two different paradigms in a given discipline. To put the matter
succinctly, Kuhn argues that different paradigms are incommensurable (1) because they involve different scientific
language, which express quite different sorts of conceptual frameworks (even when the words used are the same),
(2) because they do not acknowledge, address, or perceive the same observational data, (3) because they are not
concerned to answer the same questions, or resolve the same problems, and (4) they do not agree on what counts
as an adequate, or even legitimate, explanation.
Many authors took the first sense of incommensurability (linguistic, conceptual) to be the primary one: the reason
scientists differed with regard to the paradigms, and there was no neutral standpoint to decide the issues between
the two paradigms, is that there is no language (conceptual scheme) in which the two paradigms can be stated. That
is why the two sides "inevitably talk past each other" during revolutionary periods. Kuhn seems to assume that
because two theories differ in what they say mass is like (it is conserved vs. it is not, and exchangeable with energy)
that the term "mass" means something different to the two sides. Thus, there is an assumption implicit in his
argument about how theoretical terms acquire meaning, something like the following.
The two sides make very different, incompatible claims about mass.
The theoretical context as a whole (the word
determines the meaning of a theoretical or observational term.

and

its

role

within

paradigm)

The two sides mean something different by "mass."

On this interpretation of Kuhn, the two sides of the debate during a revolution talk past each other because they're
simply speaking different (but very similar sounding) languages. This includes not only the abstract terms like "planet"
or "electron," but observational terms like "mass," "weight," "volume" and so on. This contrasts with the older
("positivist") view espoused by Carnap, for example, which held that there was a neutral observational language by
which experimental results could be stated when debating the merits (or deficiencies) of different candidates for
paradigm status. The two scientists might disagree on whether mass is conserved, but agree on whether a pointer on
a measuring apparatus is in a certain position. If one theory predicts that the pointer will be in one location, whereas
other predicts it will be in a different location, the two cannot both be correct and so we then need only check to see
which is right. On Kuhn's view (as we are interpreting him here), this is a naive description, since it assumes a sharp
dichotomy between theoretical and observational language. (On the other hand, if his view is simply holistic, so that
any change in belief counts as a conceptual change, it is implausible, possibly vacuous.)
This interpretation of Kuhn makes him out to be highly problematic: if the two groups are talking about different
things, how can they really conflict or disagree with one another? If one group talks about "mass" and another about
"mass*" it's as if the two are simply discussing apples and oranges. Theories are on this interpretation strictly
incomparable. Newtonian and relativistic mechanics could not be rivals. This is important since Kuhn himself speaks
as though the anomalies are the deciding point between the two theories: one paradigm cannot solve it, the other
can. If they are dealing with different empirical data, then they're not even trying to solve the same thing.
The problem becomes more acute when you consider remarks that Kuhn makes that seem to mean that the
conceptual scheme (paradigm) is self-justifying, so that any debate expressed in the different languages of the two
groups will necessarily be circular at some point. That is, there is no compelling reason to accept one paradigm
unless you already accept that paradigm: reasons to accept a paradigm are conclusive only within that paradigm
itself. If that is so, however, what reason could anyone have to give up the old paradigm? (In addition, it cannot be
literally correct that the paradigm is self-justifying, since otherwise there would be no anomalies.)

An alternative view would reject the thesis that the incommensurability of scientific concepts or language is the
primary one; rather it is the incommensurability between scientific problems. That is, if the two paradigms view
different problems as demanding quite different solutions, and accept different standards for evaluating proposed
solutions to those problems, they may overlap conceptually to a large degree, enough to disagree and be rivals, but
still reach a point at which the disagreement cannot be settled by appeal to experimental data or logic. "When
paradigms change," he says, "there are usually significant shifts in the criteria determining both the legitimacy of
problems and of proposed solutions...." "To the extent ... that two scientific schools disagree about what is a problem
and what is a solution, they will inevitably talk through each other when debating the relative merits of their respective
paradigms." The resulting arguments will be "partially circular." "Since no paradigm ever solves all the problems it
defines and since no two paradigms leave all the same problems unsolved, paradigm debates always involve the
question: which problems is it more significant to have solved?"
On this view, what makes theories "incommensurable" with each other is that they differ on their standards of
evaluation; this difference is the result of their accepting different exemplars as definitive of how work in that
discipline should proceed. They are, indeed, making different value judgments about research in their discipline.
How are different value judgments resolved? One focus of many critics has been Kuhn's insistence to compare
scientific revolutions with political or religious revolutions, and with paradigm change as a kind of "conversion." Since
conversion is not a rational process, it is argued, then this comparison suggests that neither is scientific revolution,
and so science is an irrational enterprise, where persuasion--by force, if necessary--is the only way for proponents of
the new paradigm to gain ascendancy. Reasoned debate has no place during scientific revolutions. Whether this is
an apt characterization of Kuhn's point depends on whether conversion to a religious or political viewpoint is an
irrational enterprise.
Kuhn himself does not endorse the radical conclusions just outlined; he does not view science as irrational. In
deciding between different paradigms, people can give good reasons for favoring one paradigm over another, he
says; it is just that those reasons cannot be codified into an algorithmic "scientific method," that would decide the
point "objectively" and conclusively. There are different standards of evaluation for what counts as the important
problems to solve, and what counts as an admissible solution. For pre-Daltonian chemistry (with the phlogiston
theory of combustion), explaining weight gains and losses was not viewed to be as important as explaining why
metals resembled each other more than they did their ores. Quantitative comparisons were secondary to qualitative
ones. Thus the weight gain problem was not viewed as a central difficulty for the theory, though an anomalous one.
Things were different in the new theory, in which how elements combined and in what proportions became the
primary topic of research. Here there was a common phenomenon on which they could agree--weight gain during
combustion--but it was not accorded the same importance by both schools.
Under this interpretation, much of what Kuhn says is misleading, e.g., his highly metaphorical discussion about
scientists who accept different paradigms living in different worlds. Kuhn sometimes seems to be making an
argument, based on Gestalt psychology, of the following form.
Scientists who accept different paradigms experience the world in different ways; they notice some things the others
do not, and vice versa.
The world consists of the sum of your experiences.
Scientists who accept different paradigms experience different worlds.
Some of his arguments depend on the assumption that to re-conceptualize something, to view it is a different way, is
to see a different thing. Thus, he speaks as if one scientist sees a planet where another saw a moving star, or that
Lavoisier saw oxygen whereas Priestley saw "dephlogisticated air."
This is an incommensurability of experience; it is dubious, but this does not detract from the very real
incommensurability of standards that Kuhn brought to the attention of the philosophical, historical, and scientific
communities.

Lecture 13
3/22/94
Laudan on Kuhn's Theory of Incommensurable Theories
Before the break, we finished our discussion of Kuhn's theory of scientific revolutions by examining his notion of
incommensurability between scientific theories. To review, Kuhn claimed that rival paradigms are always
incommensurable. Roughly, that means that there is no completely neutral standpoint from which one can judge the
relative worth of the two paradigms. As we discussed, incommensurability comes in three basic varieties in Kuhn's
SSR:

Incommensurability of Standards or Cognitive Values - Scientists espousing different paradigms may agree
on certain broadly defined desiderata for theories (that a theory should be simple, explanatory, consistent
with the empirical data, of large scope and generality, and so on), but they typically disagree on their
application. For example, they might disagree about what needs to be explained (consider transition to
Daltonian chemistry) or what constitutes an acceptable explanation (consider transition to Newtonian
mechanics).
Incommensurability of Language - Scientists typically speak different languages before and after the
change; the same words may take on new meanings (consider mass, simultaneous, length, space, time,
planet, etc.). The two sides inevitably "talk past one another."
Incommensurability of Experience - Scientists see the world in different, incompatible ways before and after
a paradigm shift. Kuhn describes paradigm change as involving a kind of Gestalt shift in scientists'
perceptions of the world. Sometimes Kuhn speaks as though the world itself has changed (as opposed to
the conceptual framework in which the world is perceived, which is quite different), though such talk is best
construed as metaphorical. (Since otherwise you're committed to the view that the world is constituted by
how we conceive of it.)

As I noted then, the first sense of incommensurability is the fundamental one for Kuhn. That is because the paradigm
(in the sense of exemplar, a concrete, definitive achievement) defines by example what problems are worth solving
and how one should go about solving them. Since this defines the particular sense in which theories are deemed
"simple," "explanatory," "accurate," and so one, the paradigm one adopts as definitive determines which standards
one uses to judge theories as adequate.
Thus, according to Kuhn one's particular standards or cognitive values are determined by the paradigm one accepts.
There is on his view no higher authority to which a scientist can appeal. There are no "deeper" standards to which
one can appeal to adjudicate between two paradigms that say that different problems are important; thus, there is no
neutral standpoint from which one can decide between the two theories. It is primarily in this sense that theories are
"incommensurable" according to Kuhn.
In the next two weeks, we will examine the notion of cognitive values in science in detail, particularly as discussed in
Laudan's book Science and Values. (Note that Laudan's book does not deal with ethical values, but with cognitive
ones, and particularly with the notion that there is no paradigm-neutral algorithm for adjudicating between different
sets of cognitive values.) Laudan's aim: to find a middle ground between the rule-bound view of Carnap and Popper,
and the apparent relativism of Kuhn (i.e., his view that standards of theoretical worth are paradigm-relative, and that
there is no higher authority to adjudicate between these standards).
Laudan's claim is that both approaches fail to explain some aspect or another of science, and go too far in predicting
the degree of disagreement or consensus in science.
On the rule-bound view, there should normally be consensus as long as scientists are rational. To decide between
two competing hypotheses, one has only to examine the evidence. Evidence can be inconclusive, but it can never be
conclusive in one way for one person and conclusive in another way for another person. (It is this way according to
Kuhn, since to a large extent paradigms are "self-justifying.") In any case, it is always apparent how one could
proceed in principle to adjudicate between the two hypotheses, even if it is impossible or impractical for us to do so. If
there is a neutral algorithm or decision procedure inherent in "the scientific method," then you can see how the

degree of consensus that typically exists in science would be easy to explain. On the other hand, it is difficult to
explain how there could be disagreement over fundamentals when people have the same body of evidence before
them. Historical study does seem to suggest that indeed science is non-cumulative in important respects, i.e., that in
scientific revolutions some standards and achievements are lost at the same time that new ones are gained.
(Examples: Dalton, Newton).
On the other hand, it is hard to see how Kuhn could explain how consensus arises as quickly as it does in science,
given his incommensurability thesis. Indeed, it is difficult to see how Kuhn can explain why consensus arises at all.
Some of his explanations leave one cold, e.g., that all the young people adopt the new theory, and that the older
people, who remain loyal to the older framework, simply die off. Why shouldn't the younger scientists be as divided
as the older scientists? Similarly, arguing that some groups get control of the universities and journals does not
explain why the others don't go off and found their own journals. To see this, consider Kuhn's own analogies between
scientific revolutions and political revolutions of religious conversions. In these areas of human discourse, there is
little consensus and little prospect for consensus. (Indeed, in the fields of philosophy and sociology of science there
is little consensus or prospect for consensus, either.) Here we suspect that the groups differ with regard to their basic
(political or religious) values, and that since there is no way of adjudicating between these values, the rifts end up
persisting. If science is like that, and there is no "proof" but only "conversion" or "persuasion," then why should there
ever be the unanimity that arises during periods of normal science, as Kuhn describes it?
To complicate matters, Kuhn often notes that it usually becomes clear to the vast majority of the scientific community
that one paradigm is "better" than another. One important achievement leading to the adopting of the new theory is
that it solves the anomaly that created a sense of crisis within the old paradigm. Additionally, the new paradigm may
include very precise, quantitative methods that yield more accurate predictions, or they may simply be easier to apply
or conceptualize. Often, Kuhn claims not that these are not good considerations in favor of adopting the new
paradigm, but that they are "insufficient" to force the choice between scientists. In any case, they are not often
present in actual historical cases (e.g., he notes that at first Copernican astronomy was not sufficiently more accurate
than Ptolemaic astronomy).
When Kuhn talks like this, he sounds very little like the person who propounds radical incommensurability between
theories. Instead, the issue at hand seems to be that the empirical evidence simply does not determine logically
which theory is correct. This thesis is often called the "underdetermination" thesis. This is much less radical than
saying that scientists "live in different worlds" (incommensurability of experience). Instead, we simply have it that
empirical evidence does not determine which theory is correct, and that to fill in the gap scientists have to import their
own cognitive values, about which they differ. (These may not be supplied by the paradigm, but rather the paradigm
is favored because cognitive values differ.)
Laudan thinks that Kuhn shares many assumptions with Popper and Carnap, in particular the view that science (and
pursuit of knowledge in general) is hierarchically structured when it comes to justification. That is, we have the
following levels of disagreement and resolution.
Level of Disagreement
Factual
Methodological
Axiological

Level of Resolution
Methodological
Axiological
None

According to Laudan, Kuhn disagrees with Carnap and Popper about whether scientists share the same cognitive
values insofar as they are acting professionally (i.e., as scientists rather than individuals). If so, this would provide a
way of resolving any dispute. Kuhn says No; Carnap and Popper (in different ways) says Yes. They both seem to
agree that differences in cognitive values cannot be resolved. Thus, the reason Kuhn sees paradigms as
incommensurable is simply that on his view there is no higher level to appeal to in deciding between the different
values inherent in competing paradigms. Next time, we will examine the implications of the hierarchical picture in
detail.

Lecture 14
3/24/94
Laudan on the Hierarchical Model of Justification
At the end of the last lecture, we briefly discussed Laudan's view that Popper, Carnap, and Kuhn all shared an
assumption, i.e., that scientific justification is hierarchically structured. To review, Laudan thinks that Kuhn shares
many assumptions with Popper and Carnap, in particular the view that science (and pursuit of knowledge in general)
is hierarchically structured when it comes to justification. That is, we have the following levels of disagreement and
resolution.
Level of Disagreement
Factual
Methodological
Axiological

Level of Resolution
Methodological
Axiological
None

According to Laudan, Kuhn disagrees with Carnap and Popper about whether scientists share the same cognitive
values insofar as they are acting professionally (i.e., as scientists rather than individuals). If so, this would provide a
way of resolving any dispute. Kuhn says No; Carnap and Popper (in different ways) says Yes. They both seem to
agree that differences in cognitive values cannot be resolved. Thus, the reason Kuhn sees paradigms as
incommensurable is simply that on his view there is no higher level to appeal to in deciding between the different
values inherent in competing paradigms. Let us look at Laudan's description of the hierarchical model in detail.

Factual Disputes - disagreement about "matters of fact," i.e., any claim about what is true of the world,
including both what we observe and the unobservable structure of the world.

Factual disputes, on the hierarchical view, are to be adjudicated by appealing to the methodological rules governing
scientific inquiry.

Methodological Disputes - disagreement about "methodology," i.e., both high- and low-levels rules about
how scientific inquiry should be conducted. These include very specific, relatively low-level rules such as
"always prefer double-blind to single-blind tests when testing a new drug" to high-level rules such as "avoid
ad hocness," "only formulate independently testable theories," "assign subjects to the test and control
groups randomly," and so on. These rules would also include instructions regarding statistical analysis
(when to perform a t-test or c-squared test, when to "reject" or "accept" hypotheses at a given level of
significance, and so on). As Laudan remarks, settling a factual dispute on the hierarchical model is
somewhat similar to deciding a case in court; the rules are fairly well established for deciding cases;
evidence is presented on either side of the case, and the rules, when properly applied, result in an impartial,
fair, justified resolution of the dispute.

Note that it is not a part of the hierarchical view that any factual dispute can be immediately adjudicated by
application of methodological rules. For starters, the evidence may simply be inconclusive, or of poor quality. In that
case, the rules would simply tell you to go find more evidence; they would also tell you what type of evidence would
be needed to resolve the dispute. This does not mean, of course, that the evidence will be found. It may be
impractical, or maybe even immoral, to go out and get evidence of the required sort. That, however, only means that
some factual disputes cannot be settled as a practical matter; all disputes can be settled "in principle."
Methodological disputes, on the hierarchical view, are to be adjudicated by appealing to the goals and aims of
scientific inquiry. The assumption is that rules of testing, experiment, statistical analysis, and so on, are not ends in
themselves but means to achieving a higher goal.

Axiological Disputes - disagreements about the aims and goals of scientific inquiry. For example, do
scientists seek truth, or simply empirical adequacy? Must theories be "explanatory" in particular sense? As
remarked last time, Carnap and Popper seem to assume that scientists, insofar as they are acting rationally

and as scientists, do not disagree about the fundamental aims and goals of scientific inquiry. (Carnap and
Popper might differ on what those aims and goals are that all scientists share; but that is another story.)
Kuhn, on the other hand, assumes that scientists who commit themselves to different paradigms will also
differ about what they consider to be the aims and goals of scientific inquiry (in a particular discipline).
On the hierarchical view of justification, axiological disputes cannot be adjudicated; there is no higher level to appeal
to.
[Note: GOAL in subsequent discussion of hierarchical view is what it involves, and what it would take to refute it.
Contrast with "Leibnizian Ideal."]
Factual Disagreement (Consensus)
Can all factual disputes be settled by appeal to methodological rules? There is a basic problem with supposing that
they can. Although the rules and available evidence will exclude some hypotheses from consideration, they will never
single out one hypothesis out of all possible hypotheses as the "correct" one given that evidence. In other words,
methodological rules plus the available evidence always underdetermine factual claims. This may occur if the two
hypotheses are different but "empirically equivalent," i.e., they have the same observational consequences. In this
case, it is questionable whether the theories are even different theories at all (e.g., wave vs. matrix mechanics). In
many cases we might think that the theories really are different, but observation could never settle the issue between
them (e.g., Bohmian mechanics vs. orthodox quantum mechanics).
As noted earlier, this fact does not undercut the hierarchical model. The reason for this is that the hierarchical model
only says that when factual disputes can be adjudicated, they are adjudicated at the methodological level (by
applying the rules of good scientific inquiry). However, it is not committed to conceiving of methodology as singling
out one hypothesis out of all possible hypotheses, but simply as capable, often enough, of settling a dispute between
the hypotheses that we happen to have thought of, and giving us direction on how to gather additional evidence
should available evidence be insufficient. That is, on the hierarchical model are rules simply answer the question:
Which hypothesis out of those available to us is
best supported by the available evidence?
The rules, then, do not tell us what hypothesis to believe ("the truth is h"), but simply which of two hypotheses to
prefer. In other words, they provide criteria that partition or divide the class of hypotheses into those that are
permissible, given the evidence, and those that are not. Thus it may turn out in particular cases that given the
available evidence more than one hypothesis is acceptable.
Consider the following question: would it be rational for a scientist now, given our present state of empirical
knowledge, to believe in Cartesian physics, or phlogiston theory, and so on? The point here is that though there are
periods (perhaps long ones) during which the available rules underdetermine the choice, so that it is rationally
permissible for scientists to disagree, there comes a time when rational disagreement becomes impermissible. That
is, though whether a person is justified in holding a position is relative to the paradigm in the short term, in the long
term it isn't true that just "anything goes." (Feyerabend, some sociologists conclude from the fact that "reasonable"
scientists can and do differ, sometimes violently so, when it comes to revolutionary periods that there is no rational
justification of one paradigm over another, that there is no reason to think that science is progressing towards the
truth, or so on.)

Moral (Laudan): Underdetermination does not imply epistemic relativism with regard to scientific theories.

How is this relevant to Kuhn? Well, Laudan claims that Kuhn is implicitly assuming that the fact that there is no
neutral algorithm (methodological rule) that always tells us "This hypothesis is correct," and concludes from this that
the choice between them must be, at least in part, non-rational. Then he concludes at the end of the book that this
shows that science is not "progressing" towards the truth, considered as a whole. Progress can only be determined
when the relevant and admissible problems (goals and aims of the discipline) are fixed by a paradigm. (Analogy with
Darwinian evolution: no "goal" towards which evolution is aiming.) However, this may be true only in the short term.

Laudan says (pages 31-32) that though "observational accuracy" might be vaguely defined, there comes a point at
which it's apparent that one theoretical framework is more accurate observationally than another. Kuhn's problem is
emphasizing too strongly the disagreement that can occur because of the ambiguity of shared standards.
Methodological Disagreement (Consensus)
What are some examples of methodological disagreement? Predictions must be surprising or of wide variety. Another
example: disagreement over applications of statistics; statistical inference is not a monolithic area of investigation,
free of disagreement.
Laudan says that goals and aims cannot completely resolve disputes over many methodological claims; there is
underdetermination between goals and justification of methods just as there is between methods and justification of
factual claims. For example, simply accepting that our goal is that scientific theories be true, explanatory, coherent,
and of wide generality does not by itself determine which methodological principles we should seek.
This no more implies that methodological disputes cannot be resolved by appeals to shared goals than that factual
claims can never be settled by appeals to shared methods. We can often show that a certain rule is one way of
reaching our goal, or that it is better than other rules under consideration. Consider the rule that double-blind tests
are better than single-blind tests. NOTE: This example already shows that the hierarchical model is too simple, since
the lower-level facts influence what methods we think will most likely reach our goals of the truth about a subject
matter.

Lecture 15
3/29/94
Laudan's Reticulated Theory of Scientific Justification
Last time we examined Laudan's criticisms of the hierarchical model of scientific justification. As you recall, his point
was not that scientific debate is never resolved as the hierarchical model would predict; it is just that it does not
always do so. Laudan thinks that the hierarchical model is often plausible, so long as it is loosened up a bit. In
particular, the hierarchical model has to allow that not all disputes can be resolved by moving up to a higher level;
also, as I mentioned at the end of the last session, it has to allow for elements from a lower level to affect what goes
on at a higher level. (For example, it was mentioned that the methodological rule that double-blind tests be preferred
to single-blind tests was based on the factual discovery that researchers sometimes unintentionally cause patients
who have received a medication, but don't know whether they have, to be more optimistic about their prospects for
recovery and thereby affect how quickly people recover from an illness.) If these adjustments are made, the
hierarchical model becomes less and less "hierarchical."
In the end, however, the what makes the hierarchical view essentially "hierarchical" is that there is a "top" level (the
axiological) for which no higher authority is possible. To put it less abstractly, a theory of scientific justification is
hierarchical if it says that there is no way to adjudicate between disputes at the (axiological) level of cognitive values,
aims, or goals; disagreement at this level is always rationally irresolvable.
Laudan wants to dispute the claim that disagreement at the axiological level is always irresolvable. Instead, he
argues that there are several mechanisms that can and are used to resolve disagreements at the axiological level. To
see that these mechanisms exist, however, we have to drop all vestiges of the view that scientific justification is "topdown." Instead, scientific justification is a matter of coherence between the various levels; scientific disputes can be
rationally resolved so long as one or more of the levels is held fixed.
Central to this model of scientific justification is the view that the different levels constrain each other, so that holding
some of the levels fixed, there are limits to how far you can go is modifying the other level(s). This means that it must
be possible for some of the levels to change without there being change at all the other levels. Before Laudan
describes this model, which he the "reticulated" model of scientific justification, in detail, he first discusses a common
but flawed pattern of reasoning that leads many people to think that there must be "covariation" between all three
levels.
The Covariation Fallacy

Disagreement at one level (e.g., theory) are always accompanied by disagreement at all higher levels (e.g.,
method and goals).
Agreement at one level (e.g., aims) is always accompanied by agreement at all lower levels (e.g., method
and theory).

If these theses are correct, then theoretical disagreements between scientists would indeed have to be accompanied
by disagreements over aims, e.g., what counts as an acceptable scientific explanation. Laudan, relying on the fact
that there is underdetermination between every level, argues that this is not necessarily so. People can agree over
what counts as a good scientific explanation while differing over whether a specific theory meets whatever criteria are
necessary for a good scientific explanation, or over what methods would best help promote the acquisition of good
scientific explanations. (Kuhn's view, of course, is that there must be a difference at the higher level if disagreement
occurs at the lower level; thus, he argues that the scientists agree only at a shallow level--theories ought to be
"explanatory"--while disagreeing about what those criteria specifically amount to.) What is perhaps more important,
people can disagree about the aims of their discipline (e.g., truth vs. empirical adequacy, or consistency with the
evidence vs. conceptual elegance, simplicity, and beauty) while agreeing about methodology and theory. (TEST: Get
a group of scientists who agree on theory and method and then ask them what the ultimate aims of their discipline
are; you might find surprising differences.) This would occur if the methods in question would promote both sets of
aims. Similarly, the same theory can be deemed preferable by two different and deeply conflicting methodologies.
That is, a theory can win out if it looks superior no matter what perspective you take. This is how things have often
occurred in the history of science, according to Laudan.
Kuhn commits the covariance fallacy, Laudan argues, in his arguments for the view that theory, methods, and values,
which together make up a paradigm, form an inseparable whole. As Laudan construes him, Kuhn thinks that a
paradigm is a package deal: you can't modify the theory without affecting the methodological rules, or how the aims
of that discipline are conceived. On the contrary, Laudan argues, change can occur piecemeal, one or more level at a
time (with adjustments to the other levels coming later).
How Can Goals Be Rationally Evaluated?
Laudan describes two mechanisms that can be used to adjudicate between axiological disputes: (1) you can show
that, if our best theories are true, the goals could not be realized (the goal is "utopian"); and (2) the explicitly
espoused goals of a discipline are not (or even cannot be) reflected in the actual practice of that discipline (as
evinced in its methods). Mechanism (1) tries to show a lack of fit between theories and aims, keeping the former
fixed; mechanism (2) tries to show a lack of fit between methods and aims, keeping the former fixed.
Method (1) - Different Kinds of "Utopian" Strategies:
(a) Demonstrable utopianism (goals demonstrably cannot be achieved, e.g., absolute proof of general theories by
finite observational evidence);
(b) Semantic utopianism (goals are so vaguely that it is unclear what would count as achieving them, e.g., beauty or
elegance);
(c) Epistemic utopianism (it's impossible to provide a criterion that would enable us to determine if we've reached our
goal, e.g., truth).
Method (2) - Reconciling Aims and Practice
(a) Actual theories, methods cannot achieve those aims. Examples: Theories must be capable of proof by Baconian
induction from the observable evidence; Explanatory theories must not speculate about unobservables. Both were
rejected because the practice of science necessitated the postulation of unobservables; so Baconian induction was
rejected as an ideal and replaced with the Method of Hypotheses (hypothetical-deductivism). Here agreement over
theories and methods--which didn't make sense if the explicitly espoused aims were really the aims of science-provided the rational basis for adjusting the explicitly espoused aims of science.
(b) All attempts at achieving those aims have failed (e.g., certainty, explanatory, predictive theory that appeals only to
kinematic properties of matter).

Three Important Things to Note about Laudan's Reticulated Theory of Scientific Rationality: on Laudan's view,
(1) because the levels constrain but do not determine the other levels, it is sometimes the case that disagreements
over aims are rationally irresolvable--but this is not generally the case; (2) the levels are to a large degree
independent of one another, allowing for paradigm change to be piecemeal or gradual rather than a sudden
"conversion" or "Gestalt shift;" and (3) scientific "progress" can only be judged relative to a particular set of goals.
Thus, Laudan's view, like Kuhn's, is relativistic.
(Important: Like Kuhn, Laudan denies the radical relativistic view that progress does not exist in science; he simply
thinks that whether progress occurs in science can only be judged relative to certain shared goals, just like whether
certain aims are reasonable can only be judged if either theory or method is held fixed. Judgments that progress has
occurred in science, relative to fixed goals, can therefore be rationally assessed as true or false.)

Lecture 16
3/31/94
Dissecting the Holist Picture of Scientific Change
Last time, we discussed Laudan's reticulationist model of scientific justification. In this session, we will examine
Laudan's arguments for thinking that his model does better than Kuhn's quasi-hierarchical, "holist" model at
explaining both how agreement and disagreement emerges during scientific revolutions. As you recall, Laudan
argues that what change in the aims or goals of a scientific discipline can result from reasoned argument if there is
agreement at the methodological and/or theoretical (factual) level. In other words, if any of the three elements in the
triad of theories, methods, and aims is held fixed, this is sufficient to provide reasonable grounds for criticizing the
other elements. Laudan's view is "coherentist," in that he claims that scientific rationality consists in maintaining
coherence or harmony between the elements of the triad.
This picture of reasoned argument in science requires that the aims-methods-theories triad be separable, i.e., that
these elements do not combine to form an "inextricable" whole, or Gestalt, as Kuhn sometimes claimed. If some of
these elements can change while the others are held fixed, and reasoned debate is possible as long one or more of
the elements are held fixed, then this leaves open this possibility that scientific debates during what Kuhn calls
"paradigm change" can be rational, allowing (at least sometimes) for relatively quick consensus in the scientific
community. This could occur if scientific change were "piecemeal," i.e., if change occurred in only some elements of
the aims-methods-theories triad at a time. Indeed, Laudan wants to argue that when examined closely scientific
revolutions are typically piecemeal and gradual rather than sudden, all-or-nothing Gestalt switches. The fact that it
often looks sudden in retrospect is an illusion accounted for by the fact that looking back often telescopes the finegrained structure of the changes.
Kuhn on the Units of Scientific Change
For Kuhn, the elements of the aims-methods-theories triad typically change simultaneously rather than sequentially
during scientific revolutions. For example, he says: "In learning a paradigm the scientist acquires theory, methods,
and standards together, usually in an inextricable mix." In later chapters of SSR, Kuhn likens paradigm change to allor-nothing Gestalt switches and religious conversion. If this were so, it would not be surprising if paradigm debate
were always inconclusive and could never completely be brought to closure by rational means. Closure must then
always be ultimately explained by non-rational factors, such as the (contingent) power dynamics within a scientific
community. As Laudan argued in Chapter 1 of SV, factors such as these cannot fully explain why it is that closure is
usually achieved in science whereas it is typically not in religions or other ideologies, or why it is that closure is
normally achieved relatively quickly.
Laudan's solution to the problem of explaining both agreement and disagreement during scientific revolutions is not
to reject Kuhn's view entirely, but to modify it in two ways.

by dropping the hierarchical picture of scientific rationality, and replacing it with the "reticulated" picture, in
which scientific aims as well as methods and theories are rationally negotiable

by dropping the notion that all elements in the aims-methods-theories triad change simultaneously; change
is typically "piecemeal" during scientific revolutions

Question: How could piecemeal change occur? Laudan first sketches an idealized account of such change, and
then attempts to argue that this idealized account approximates what often happens historically. (Go through some
examples of the former in terms of a hypothetical "unitraditional" paradigm shift.) The fact that it does not look like
that in retrospect is normally due to the fact that history "telescopes" change, so that a decade-long period of
piecemeal change is characterized only in terms of its beginning and end-points, which exhibits a complete
replacement of one triad by another. "...a sequence of belief changes which, described at the microlevel, appears to
be a perfectly reasonable and rational sequence of events may appear, when represented in broad brushstrokes that
drastically compress the temporal dimension, as a fundamental and unintelligible change of world view" (78).
(Now go through what might happen if there are different, competing paradigms.) When there is more than one
paradigm, agreement can also occur in the following kinds of cases.

The theory in one complex looks better according to the divergent methodologies in both paradigms.
The theory in one complex looks like it better meets the aims of both theories than its rival (e.g., predictive
accuracy, simplicity).

Because the criteria (aims, methods) are different in the two theories, there may be no neutral, algorithmic proof;
nevertheless, it often turns out that as the theory develops it begins to look better from both perspectives. (Again, this
only makes sense if we deny three theses espoused by Kuhn--namely, that paradigms are self-justifying, that the
aims-methods-theories mix that comprises a paradigm is an "inextricable" whole, and that paradigm change cannot
be piecemeal.) Thus, adherents of the old paradigm might drop adopt the new methods and theories because they
enable them to do things that they recognize as valuable even from their own perspective; then they might modify
their aims as they find out that those aims don't cohere with the new theory. (Examples of Piecemeal Change:
Transition from Cartesian to Newtonian mechanics (theoretical change led later to axiological change); Transition
from Ptolemaic to Copernican astronomy (methodological change--i.e., it eventually because easier to calculate
using the methods of Copernican astronomy, though this wasn't true at first--led to eventual adoption of the
Copernican theory itself.)
Prediction of the Holist Approach (committed to covariance):

Change at one level (factual, methodological, axiological) will always be simultaneous with change at the
other levels (follows from the fact that theory, methods, and standards form an "inextricable" whole).

Counterexamples: piecemeal change given above; cross-discipline changes not tied to any particular paradigm
(e.g., the acceptance of unobservables in theories, the rejection of certainty or provability as a standard for
acceptance of theories)
Because of the counterexamples, it is possible for there to be "fixed points" from which to rationally assess the other
levels. "Since theories, methodologies, and axiologies stand together in a kind of justificatory triad, we can use those
doctrines about which there is agreement to resolve the remaining areas about which we disagree" (84).
Can Kuhn respond?
(1) The "ambiguity of shared standards" argument - those standards that scientists agree on (simplicity, scope,
accuracy, explanatory value), they often interpret or "apply" differently. Laudan's criticism: not all standards are
ambiguous (e.g., logical consistency) - A response on behalf of Kuhn: it's enough that some are, and that they play a
crucial role in scientists' decisions
(2) The "collective inconsistency of rules" argument - rules can be differently weighted, so that they lead to
inconsistent conclusions. Laudan's criticism: only a handful of cases, not obviously normal. No one has ever shown
that Mill's Logic, Newton's Principia, Bacon or Descartes' methodologies were internally inconsistent. A response of

behalf of Kuhn: Again, it's enough if it happens, and it often happens when it matters most, i.e., during scientific
revolutions.
(3) The shifting standards argument - different standards applied, so theoretical disagreements cannot be
conclusively adjudicated. Laudan's criticism: It doesn't follow (see earlier discussion of underdetermination and the
reticulationist model of scientific change).
(4) The problem weighting argument - Laudan's response: one can give reasons why these problems are more
important than others, and these reasons can be (and usually are) rationally critiqued. "...the rational assignment of
any particular degree of probative significance to a problem must rest on one's being able to show that there are
viable methodological and epistemic grounds for assigning that degree of importance rather than another" (99). Also,
Laudan notes that the most "important" problems are not the most probative ones (the ones that most stringently test
the theory). For example, explaining the anomalous advance in Mercury's perihelion, Brownian motion, diffraction
around a circular disk. These problems did not become probative because they were important, but became
important because they were probative.

Lecture 17
4/5/94
Scientific Realism Vs. Constructive Empiricism
Today we begin to discuss a brand new topic, i.e., the debate between scientific realism and constructive empiricism.
This debate was provoked primarily by the work of Bas van Fraassen, whose critique of scientific realism and
defense of a viable alternative, which he called constructive empiricism, first reached a wide audience among
philosophers with the publication of his 1980 book The Scientific Image. Today we will discuss (1) what scientific
realism is, (2) what alternatives are available to scientific realism, specifically van Fraassen's constructive empiricism.
What Is Scientific Realism?
Scientific realism offers a certain characterization of what a scientific theory is, and what it means to "accept" a
scientific theory. A scientific realist holds that (1) science aims to give us, in its theories, a literally true story of what
the world is like, and that (2) acceptance of a scientific theory involves the belief that it is true.
Let us clarify these two points. With regard to the first point, the "aims of science" are to be distinguished from the
motives that individual scientists have for developing scientific theories. Individual scientists are motivated by many
diverse things when they develop theories, such as fame or respect, getting a government grant, and so on. The
aims of the scientific enterprise are determined by what counts as success among members of the scientific
community, taken as a whole. (Van Fraassen's analogy: The motives an individual may have for playing chess can
differ from what counts as success in the game, i.e., putting your opponent's king in checkmate.) In other words, to
count as fully successful a scientific theory must provide us with a literally true description of what the world is like.
Turning to the second point, realists are not so naive as to think that scientists' attitudes towards even the best of the
current crop of scientific theories should be characterized as simple belief in their truth. After all, even the most
cursory examination of the history of science would reveal that scientific theories come and go; moreover, scientists
often have positive reason to think that current theories will be superseded, since they themselves are actively
working towards that end. (Example: The current pursuit of a unified field theory, or "theory of everything.") Since
acceptance of our current theories is tentative, realists, who identify acceptance of a theory with belief in its truth,
would readily admit that scientists at most tentatively believe that our best theories are true. To say that a scientist's
belief in a theory is "tentative" is of course ambiguous: it could mean either that the scientist is somewhat confident,
but not fully confident, that the theory is true; or it could mean that the scientist is fully confident that the theory is
approximately true. To make things definite, we will understand "tentative" belief in the former way, as less-than-full
confidence in the truth of the theory.
Constructive Empiricism: An Alternative To Scientific Realism

There are two basic alternatives to scientific realism, i.e., two different types of scientific anti-realism. That is because
scientific realism as just described asserts two things, that scientific theories (1) should be understood as literal
descriptions of the what the world is like, and (2) so construed, a successful scientific theory is one that is true. Thus,
a scientific anti-realist could deny either that theories ought to be construed literally, or that theories construed literally
have to be true to be successful. A "literal" understanding of a scientific theory is to be contrasted with understanding
it as a metaphor, or as having a different meaning from what its surface appearance would indicate. (For example,
some people have held that statements about unobservable entities can be understood as nothing more than veiled
references to what we would observe under various conditions: e.g., the meaning of a theoretical term such as
"electron" is exhausted by its "operational definition.") Van Fraassen is an anti-realist of the second sort: he agrees
with the realist that scientific theories ought to be construed literally, but disagrees with them when he asserts that a
scientific theory does not have to be true to be successful.
Van Fraassen espouses a version of anti-realism that he calls "constructive empiricism." This view holds that (1)
science aims to give us theories that are empirically adequate, and (2) acceptance of a scientific theory involves the
belief that it is empirically adequate. (As was the case above, one can tentatively accept a scientific theory by
tentatively believing that the theory is empirically adequate.) A scientific theory is "empirically adequate" if it gets
things right about the observable phenomena in nature. Phenomena are "observable" if they could be observed by
appropriately placed beings with sensory abilities similar to those characteristic of human beings. On this construal,
many things that human beings never have observed or ever will observe count as "observable." On this
understanding of "observable," to accept a scientific theory is to believe that it gets things right not only about the
empirical observations that scientists have already made, but also about any observations that human scientists
could possibly make (past, present, and future) and any observations that could be made by appropriately placed
beings with sensory abilities similar to those characteristic of human scientists.
The Notion Of Observability
Constructive empiricism requires a notion of "observability." Thus, it is important that we be as clear as possible
about what this notion involves for van Fraassen. Van Fraassen holds two things about the notion of observability:
(1) Entities that exist in the world are the kinds of things that are observable or unobservable. There is no reason to
think that language can be divided into theoretical and observational vocabularies, however. We may describe
observable entities using highly theoretical language (e.g., VHF receiver," "mass," "element," and so on); this does
not, however, mean that whether the things themselves (as opposed to how we describe or conceptualize them) are
unobservable or not depends on what theories we accept. Thus, we must carefully distinguish between observing an
entity from observing that an entity exists meeting such-and-such a description. The latter can be dependent upon
theory, since descriptions of observable phenomena are often "theory-laden." However, it would be a confusion to
conclude from this that the entity observed is a theoretical construct.
(2) The boundary between observable and unobservable entities is vague. There is a continuum from viewing
something with glasses, to viewing it with a magnifying lens, with a low-power optical microscope, with a high-power
optical microscope, to viewing it with an electron microscope. At what point should the smallest things visible using a
particular instrument count as "observable?" Van Fraassen's answer is that "observable" is a vague predicate like
"bald" or "tall." There are clear cases when a person is bald or not bald, tall or not tall, but there are also many cases
in between where it is not clear on which side of the line the person falls. Similarly, though we are not able to draw a
precise line that separates the observable from the unobservable, this doesn't mean that the notion has no content,
since there are entities that clearly fall on one side or the other of the distinction (consider sub-atomic particles vs.
chairs, elephants, planets, and galaxies). The content of the predicate "observable" is to be fixed relative to certain
sensory abilities. What counts as "observable" for us is what could be observed by a suitably placed being with
sensory abilities similar to those characteristic of human beings (or rather, the epistemic community to which we
consider ourselves belonging). Thus, beings with electron microscopes in place of eyes do not count.
Arguments In Favor Of Scientific Realism: Inference To The Best Explanation
Now that we have set out in a preliminary way the two rival positions that we will consider during the next few weeks,
let us examine the arguments that could be given in favor of scientific realism. An important argument that can be

given for scientific realism is that we ought rationally to infer that the best explanation of what we observe is true. This
is called "inference to the best explanation." The argument for this view is that in everyday life we reason according
to the principle of inference to the best explanation, and so we should also reason this way in science. The best
explanation, for example, for the fact that measuring Avogadro's number (a constant specifying the number of
molecules in a mole of any given substance) using such diverse phenomena as Brownian motion, alpha decay, x-ray
diffraction, electrolysis, and blackbody radiation gives the same result is that matter really is composed of the
unobservable entities we call molecules. If it were not, wouldn't it be an utterly surprising coincidence that things
behaved in very different circumstances exactly as if they were composed of molecules? This is the same kind of
reasoning that justifies belief that an apartment has mice. If all the phenomena that have been observed are just as
would be expected if a mouse were inhabiting the apartment, isn't it then reasonable to believe that there's a mouse,
even though you've never actually seen it? If so, why should it be any different when you are reasoning about
unobservable entities such as molecules?
Van Fraassen's response is that the scientific realist is assuming that we follow a rule that says we should infer the
truth of the best explanation of what we have observed. This is what makes it look inconsistent for a person to insist
that we ought not to infer that molecules exist while at the same time insisting that we ought to infer that there is a
mouse in the apartment. Why not characterize the rule we are following differently--i.e., that we infer that the best
explanation of what we observe is empirically adequate? If that were the case, we should believe in the existence of
the mouse, but we should not believe anything more of the theory that matter is composed of molecules than that it
adequately accounts for all observable phenomena. In other words, van Fraassen is arguing that, unless you already
assume that scientists follow the rule of inference to the (truth of the) best explanation, you cannot provide any
evidence that they follow that rule as opposed to following the rule of inference to the empirical adequacy of the best
explanation.
We will continue the discussion of inference to the best explanation, as well as other arguments for scientific realism,
next time.
Lecture 18
4/7/94
Inference To The Best Explanation As An Argument For Scientific Realism
Last time, I ended the lecture by alluding to an argument for scientific realism that proceeded from the premise that
the reasoning rule of inference to the best explanation exists. Today, we will examine in greater detail how this
argument would go, and we'll also discuss the notion of inference to the best explanation in greater detail.
The Reality Of Molecules: Converging Evidence
As I noted last time, what convinced many scientists at the beginning of this century of the atomic thesis (i.e., that
matter is composed of atoms that combine into molecules, and so on) was that there are many independent
experimental procedures all of which lead to the same determination of Avogadro's number. Let me mention some of
the ways in which the number was determined.
(1) Brownian Motion. Jean Perrin studied the Brownian motion of small, microscopic particles known as colloids.
(Brownian motion was first noted by Robert Brown, in the early 19th century.) Though visible only through a
microscope, the particles were much larger than molecules. Perrin determined Avogadro's number from looking at
how particles were distributed vertically when placed in colloidal suspensions. He prepared tiny spheres of gamboge,
a resin, all of uniform size and density. He measured how the particles were distributed vertically when placed in
water; calculating what forces would have to be in place to account for this keeping the particles suspended, he could
calculate their average kinetic energy. If we know the mass and velocities, we can then determine the mass of a
molecule of the fluid, and hence Avogadro's number, which is the molecular weight divided by the mass of a single
molecule.

(2) Alpha Decay. Rutherford recognized that alpha particles were helium nuclei. Alpha particles can be detected by
scintillation techniques. By counting the number of helium atoms that were required to make up a certain mass of
helium, Rutherford calculated Avogadro's number.
(3) X-ray diffraction - A crystal will diffract x-rays, the matrix of atoms acting like a diffraction grating. From the
wavelength of the x-rays and the diffraction pattern, you can calculate the spacing of the atoms. Since that is regular
in a crystal, you could then determine how many atoms it takes to make up the crystal, and so Avogadro's number.
(Friedrich & Knipping)
(4) Blackbody Radiation - Planck derived a formula for the law of blackbody radiation, which made use of Planck's
constant (which was obtainable using Einstein's theory of the photoelectric effect) and macroscopically measurable
variables such as the speed of light to derive Boltzmann's constant. Then you use the ideal gas law PV = nRT; n is
the number of moles of an ideal gas, and R (the universal gas constant) is the gas constant per mole. Boltzmann's
constant k is the gas constant per molecule. Hence R/k = Avogadro's number (number of molecules per mole).
(5) Electrochemistry - A Faraday F is the charge required to deposit a mole of monovalent metal during electrolysis.
This means that you can calculate the number of molecules per mole if you know the charge of the electron, F/e = N.
Millikan's experimental measurements of the charge of the electron can then be used to derive Avogadro's number.
Now, the scientific realist wants to claim that the fact that all of these different measurement techniques (along with
many others not mentioned here) all lead to the same value for Avogadro's number. The argument, then, is how you
can explain this remarkable convergence on the same result if it were not for the fact that there are atoms &
molecules that are behaving as the theories say they do? Otherwise, it would be a miracle.
Inference To The Best Explanation (Revisited)
We have to now state carefully what is being claimed here. The scientific realist argues that the fact that the reality of
molecules explains the convergence on Avogadro's number better than its rival, that the world is not really molecular
but that everything behaves as if it were. That is because given the molecular hypothesis, convergence would be
strongly favored, whereas if the underlying world were not molecular we wouldn't expect any stability in such a result.
Now it is claimed that being the best explanation of something is a mark of truth. Thus, we have an inference pattern.
A explains X better than its rivals, B, C, and so on.
The ability of a hypothesis to explain something better
than all its rivals is a mark of its truth.
A is true.
Now why should we think that this is a good reasoning pattern? (That is, why should we think that the second
premise is true?) The scientific realist argues that it is a reasoning pattern that we depend on in everyday life; we
must assume the truth of the second premise if we are to act reasonably in everyday life. To refer to the example we
discussed the last session, the scientific realist argues that if this reasoning pattern good for the detective work that
infers the presence of an unseen mouse, it is good enough for the detective work that infers the presence of unseen
constituents of matter.
Van Fraassen has argued that the hypothesis that we infer to the truth of our best explanation can be replaced by the
hypothesis that we infer to the empirical adequacy of our best explanation without loss in the case of the mouse,
since it is observable. How then, can you determine whether we ought to follow the former rule rather than the latter?
The only reason we have for thinking that we follow IBE is everyday examples such as the one about mice; but the
revised rule that van Fraassen suggests could account for this inferential behavior just as well as the IBE hypothesis.
Thus, there is no real reason to think that we follow the rule of IBE in our reasoning.
Van Fraassen's Criticisms Of Inference To The Best Explanation
Van Fraassen also has positive criticisms of IBE. First, he argues that it is not what it claims to be. In science, you
don't really choose the best overall explanation of the observable phenomena, but the best overall explanation that

you have available to you. However, why should we think that the kinds of explanations that we happen to have
thought of are the best hypotheses that could possibly be thought up by any intelligent being? Thus, the IBE rule has
to be understood as inferring the truth of the best explanation that we have thought of. However, if that's the case our
"best" explanation might very well be the best of a bad lot. To be committed to IBE, you have to hold that the
hypotheses that we think of are more likely to be true than those that we do not, for that reason. This seems
implausible.
Reactions On Behalf Of The Scientific Realist
(1) Privilege - Human beings are more likely to think up hypotheses that are true than those that are false. For
otherwise, evolution would have weeded us out. False beliefs about the world make you less fit, in that you cannot
predict and control your environment, and may likely die.
Objection: The kinds of things that selected us during our evolution based on our inferences do not depend on what
we have believed to be true. They just have to not kill us, and enable us to have children. In addition, we only have to
be able to infer what is empirically adequate.
(2) Forced Choice - We cannot do without inferring what goes beyond our evidence; thus the choice between
competing hypotheses is forced. To guide our choices, we need rules of reasoning, and IBE fits the bill.
Objection - The situation may force us to choose the best we have; but it cannot force us to believe that the best we
have is true. Having to choose a research program, for example, only means that you think that it is the best one
available to you, and that you can best contribute to the advancement of science by choosing that one. It does not
thereby commit you to belief in its truth.
The problem that is evinced here, and will crop up again, is that any rule of this sort has to make assumptions about
the way the world is. The things that seem simple to us are most likely to be true; the things that seem to explain
better than any of the alternatives that we've come up with are more likely to be true that those that do not; and so
on.
Lecture 19
4/12/94
Entity Realism (Hacking & Cartwright)
Last time we looked at an argument for scientific realism based on an appeal to a putative rule of reasoning, called
Inference to the Best Explanation (IBE). We examined a case study where the scientific realist argues that
convergent but independent determinations of Avogadro's number were better explained by the truth of the molecular
hypothesis than its empirical adequacy (Salmon 1984, Scientific Explanation and the Causal Structure of the World,
pages 213-227). (Note that Salmon's treatment of these cases does not explicitly appeal to IBE, but to the Common
Cause principle.) The primary problems with that argument were (1) the realist has given no reason to think that we
as a rule infer the truth of the best explanation, rather than its empirical adequacy, and (2) it is impossible to argue for
IBE as a justified rule of inference unless one assumes that human beings are, by nature, more likely to think up true
explanations rather than ones that are merely empirically adequate. There are, of course, responses a realist can
make to these objections, and we examined two responses to problem (2), one that argues that (a) evolution
selected humans based on our ability to generate true rather than false hypotheses about the world, and the other
that (b) accepts the objection but argues that we are somehow forced to believe the best available explanation.
Neither of these responses seems very convincing (van Fraassen 1989, Laws and Symmetry, pages 142-150).
Today we will look at a moderate form of realism, which I will call "entity realism," and arguments for that view that do
not depend on IBE. Entity realists hold that what one is rationally compelled to believe the existence of some of the
unobservable entities postulated by our best scientific theories, but one is not obligated to believe that everything that
our best theories say about those entities is true. Nancy Cartwright, for example, argues that we are compelled to
believe in those entities that figure essentially in causal explanations of the observable phenomena, but not in the
theoretical explanations that accompany them. The primary reason she gives is that causal explanations, e.g., that a

change in pressure is caused by molecules impinging on the surface of a container with greater force after heat
energy introduced into the container increases the mean kinetic energy of the molecules, make no sense unless you
really think that molecules exist and behave roughly as described. Cartwright claims that you have offered no
explanation at all if you give the preceding story and then add, "For all we know molecules might not really exist, and
the world simply behaves as if they exist." Theoretical explanations, on the other hand, which merely derive the laws
governing the behavior of those entities from more fundamental laws, are not necessary to believe, since a
multiplicity of theoretical laws can account for the phenomenological laws that we derive from experiment. Cartwright
argues that scientists use often different and incompatible theoretical models based on how useful those models are
in particular experimental situations; if this is so, scientists cannot be committed to the truth of all their theoretical
models. However, scientists do not admit incompatible causal explanations of the same phenomenon; according to
Cartwright, that is because a causal explanation cannot explain at all unless the entities that play the causal roles in
the explanation exist.
Cartwright's argument depends on a certain thesis about explanation (explanations can either cite causes or can be
derivations from fundamental laws) and an associated inference rule (one cannot endorse a causal explanation of a
phenomenon without believing in the existence of the entities that, according to the explanation, play a role in
causing the phenomenon). As Cartwright sometimes puts it, she rejects the rule of Inference to the Best Explanation
but accepts a rule of Inference to the Most Probable Cause. Van Fraassen, of course, is unlikely to acquiesce in such
reasoning, since he rejects the notion that a causal explanation cannot be acceptable unless the entities it postulates
exist; on the contrary, if what is requested in the circumstances is information about causal processes according to a
particular scientific theory, it will be no less explanatory if we merely accept the theory (believe it to be empirically
adequate) rather than believe that theory. Thus, the constructive empiricist can reject Cartwright's argument since he
holds a different view of what scientific explanation consists in.
Hacking takes a different route in arguing to an entity realist position. Hacking argues that the mistake that Cartwright
and van Fraassen both make is concentrating on scientific theory rather than experimental practice. His approach
can be summed up in the slogans "Don't Just Peer, Interfere" (with regard to microscopes), and "If you can
manipulate them, they must be real" (with regard to experimental devices that use microscopic particles such as
electrons as tools). Let's look at his arguments for these two cases.
Lecture 20
4/14/94
Entity Realism And The "Non-Empirical" Virtues
Last time we discussed arguments due to Cartwright and Hacking for the entity realist position. Entity realism, as you
recall, is the view that belief in certain microphysical entities can be (and is) rationally compelling. Cartwright argues
that we are rationally required to believe in the existence of those entities that figure essentially in causal
explanations that we endorse. (Van Fraassen's response to her argument is that since the endorsement of the
explanation only amounts to accepting it--i.e., believing it to be empirically adequate--belief in the unobservable
entities postulated by the explanation is not rationally required.) By contrast, Hacking argued that we are rationally
required to believe in the existence of those entities that we can reliably and stably manipulate. He argues that once
we start using entities such as electrons then we have compelling evidence of their existence (and not merely the
empirical adequacy of the theory that postulates their existence). To make his case, he gives detailed descriptions of
how we stably and reliably interact with things shown by optical microscopes, and with electron guns in the PEGGY
series. The microscope case is especially interesting since it indicates that a person can acquire new perceptual
abilities by using new instruments and that "observability" is a flexible notion.
The question that we will examine in the first part of the lecture today is whether Hacking's arguments do not simply
beg the question against van Fraassen's constructive empiricism. Let us begin by discussing Hacking's argument
that stability of certain features of something observed using different instruments is a compelling sign of their reality.
In response, van Fraassen asks us to consider the process of developing such instruments. When we are building
various types of microscopes, we not only use theory to guide the design, but also learn to correct for various
artifacts (e.g., chromatic aberration). As van Fraassen puts it, "I discard similarities that do not persist and also build
machines to process the visual output in a way that emphasizes and brings out the noticed persistent similarities.
Eventually the refined products of these processes are strikingly similar when initiated in similar circumstances ...

Since I have carefully selected against non-persistent similarities in what I allow to survive the visual output
processing, it is not all that surprising that I have persistent similarities to display to you" (Images of Science, page
298). In other words, van Fraassen argues that we design our observational instruments (such as microscopes) to
emphasize those features that we regard as real, and de-emphasize those we regard as artifacts. If that is so,
however, we cannot point to convergence as evidence in the reality of those features, since we have designed the
instruments (optical, ultraviolet, electron microscopes) so that they all converge on those features we have
antecedently decided are real, not artifacts.
The principle that Hacking is using in his argument from stability across different observational techniques is that if
there is a certain kind of stable input-output match in our instruments, we can be certain that the output is a reliable
indicator of what is there at the microphysical level. Van Fraassen notes that, given the constraints that we place on
design and that the input is the same (say, a certain type of prepared blood sample), it is not surprising that the
output would be the same, even if the microstructure that we "see" through the microscope has no basis in reality.
Thus, Hacking is correct in seeking a more striking example, which he attempts to provide with his Grid Argument.
(Here, as you recall, a grid is drawn, photographically reduced, and that reduction is used to manufacture a tiny metal
grid. The match between the pattern we see through the microscope at the end of the process and the pattern
according to which the grid was drawn at the beginning indicates both that the grid-manufacturing process is a
reliable one and that the microscope is a reliable instrument for viewing microphysical structure. Thus, by analogy,
Hacking argues that we should believe it what the microscope reveals about the microphysical structure of things that
we have not manufactured.) Van Fraassen objects on two levels. First, he argues that argument by analogy is not
strong enough to support scientific realism. For analogy requires an assumption that if one class of things resembles
another in one respect, they must resemble it in another. (To be concrete, if the microscopic grid and a blood cell
resemble one another in being microscopic, and we can be sure that the former is real because its image in the
microscope matches the pattern that we photographically reduced, then we can by analogy infer that what the
microscope shows us about the blood cell must be accurate as well.) To this van Fraassen replies, "Inspiration is
hard to find, and any mental device that can help us concoct more complex and sophisticated novel hypotheses is to
be welcomed. Hence, analogical thinking is welcome. But it belongs to the context of discovery, and drawing
ingenious analogies may help to find, but does not support, novel conjectures" (Images of Science, page 299).
Discuss. What could Hacking reply? Consider the following: to show that an analogical inference is unjustified, you
have to argue that there is an important disanalogy between the two that blocks the inference. Here the obvious
candidate is that the grid is a manufactured object, whereas the blood cell, if it exists, is not. Since we would not
accept any process as reliable that did not reproduce the macroscopic grid pattern we drew, it is no surprise that
what we see through the microscope matches the pattern we drew--for we developed the process so that things
would work out that way. However, we cannot say the same thing about the blood cell. Is this convincing?
Second, van Fraassen argues that to make Hacking's argument work we have to assume that we have successfully
produced a microscopic grid. How do we know that? Well, because there is a coincidence between the pattern we
observe through the microscope and the pattern we drew at the macroscopic level. Hacking's argument is that we
would have to assume some sort of cosmic conspiracy to explain the coincidence if the microscopic pattern did not
reflect the real pattern that was there, which would be unreasonable. Van Fraassen's response is that not all
observable regularities require explanation.
Discuss. Might Hacking fairly object that, while it is true that not all coincidences require explanation, it is true that
coincidences require explanation unless there is positive reason to think they cannot be explained. For this reason,
he might argue that van Fraassen's counterexamples to a generalized demand for explanation, e.g., those based on
the existence of verified coincidences predicted by quantum physics (the EPR type), are not telling, since we have
proofs (based on the physical theories themselves) that there can be no explanations of those kinds of events. (Also,
discuss van Fraassen's argument that to explain a coincidence by postulating a deeper theory will not remove all
coincidences; eventually, "why" questions terminate.)
Now, we turn to the question of whether Hacking's claim that the fact that we manipulate certain sorts of microscopic
objects speaks to their reality, not just the empirical adequacy of the theory that postulates those entities. Van
Fraassen would want to ask the following question: How do you know that your description of what you are

manipulating is a correct one? All you know is that if you build a machine of a certain sort, you get certain regular
effects at the observable level, and that your theory tells you that this is because the machine produces electrons in a
particular way to produce that effect. The constructive empiricist would accept that theory, too, and so would accept
the same description of what was going on; but for him that would just mean that the description is licensed by a
theory he believes to be empirically adequate. To infer that the observable phenomena that are described
theoretically as "manipulating electrons in such-and-such a way to produce an effect" compels belief in something
more than empirical adequacy requires assuming that the theoretical description of what he is doing is not merely
empirically adequate, but also true. However, that is what the constructivist would deny.
Non-Empirical Virtues As A Guide To Truth: A Defense?
We have already examined in some detail van Fraassen's reasons for thinking that non-empirical features of theories
that we regard as desirable (simplicity, explanation, fruitfulness) are not marks of truth, nor even of empirical
adequacy, though they are certain pragmatic reasons for preferring one theory over another (i.e., using that theory
rather than another). Let us look briefly at a defense of the view that "non-empirical virtues" of this sort are indeed
guides to truth, due to Paul Churchland.
Churchland's view is basically that van Fraassen's argument against the non-empirical virtues as guides to truth can
be used against him. Since we are unable to assess all the empirical evidence for or against a particular theory, we
have in the end to decide what to believe based on what is simplest, most coherent or explanatory. This would sound
much like the "forced choice" response that we looked at last time, except that Churchland is arguing something
subtly different. He argues that van Fraassen cannot decide which of two theories is more likely to be empirically
adequate without basing his decision on which of the two theories is simplest, most fruitful and explanatory, and so
on. That is because the arguments that apply to the theory's truth with regard to its subject matter (available evidence
in principle can never logically force the issue one way or another when it comes to unobservable structure), apply
just as well to empirical adequacy. We'll never have complete knowledge of all observable evidence either, and so
nothing compels us one way or another to accept one theory as empirically adequate rather than another (when both
agree on what we have observed so far, or ever will observe). However, we cannot completely suspend belief
altogether. Since that is the only choice that van Fraassen's reasoning gives us in the end, it ought to be rejected,
and so we can conclude that non-empirical virtues are just as much a guide to truth as are what we have observed.
Churchland is a philosopher who works in cognitive psychology (with an emphasis on neurophysiology). He argues
that we know that "values such as ontological simplicity, coherence, and explanatory power are some of the brain's
most basic criteria for recognizing information, for distinguishing information from noise ... Indeed, they even dictate
how such a framework is constructed by the questing infant in the first place" ("The Ontological Status of
Observables," Images of Science, page 42). Thus, he concludes that since even our beliefs about what is observable
are grounded in what van Fraassen calls the merely "pragmatic" virtues (so-called because they give us reason to
use a theory without giving us reason to believe that it is true), then it cannot be unreasonable to use criteria such as
simplicity, explanatory power, and coherence to form beliefs about what is unobservable to us. Van Fraassen's
distinction between 'empirical, and therefore truth-relevant' and 'pragmatic, and therefore not truth-relevant' is
therefore not a tenable one.
Churchland concludes with a consideration that I will leave you as food for thought. Suppose that a humanoid race of
beings were born with electron microscopes for eyes. Van Fraassen would then argue that since for them the
microstructure of the world would be seen directly, they would unlike us have different bounds on what they regard as
"observable." Churchland regards this distinction as wholly unmotivated. He points out that van Fraassen's view
leads to the absurd conclusion that they can believe in what their eyes tell them but we cannot, even though if we put
our eyes up to an electron microscope we will have the same visual experiences as the humanoids. There is no
differences between the causal chain leading from the objects that are perceived and the experience of perception in
both cases, but van Fraassen's view leads to the conclusion that nevertheless we and the humanoids must embrace
radically different attitudes towards the same scientific theories. This he regards as implausible.

Lecture 21
4/19/94
Laudan on Convergent Realism
Last time, we discussed van Fraassen's criticism of Hacking's arguments for entity realism based on the Principle of
Consistent Observation (as in the microscope case, most notably the grid argument). We also talked about
Churchland's criticisms of van Fraassen's distinguishing empirical adequacy from the "pragmatic" or non-empirical
virtues. Churchland argued that, based on what we know about perceptual learning, there's no principled reason to
think that empirical adequacy is relevant to truth but simplicity is not.
Today we're going to look at an argument for scientific anti-realism that proceeds not by arguing a priori that
observability has a special epistemological status, but by arguing that, based on what we know about the history of
science, scientific realism is indefensible. Laudan discusses a form of realism he calls "convergent" realism, and
seeks to refute it. Central to this position is the view that the increasing success of science makes it reasonable to
believe (a) that theoretical terms that remain across changes in theories refer to real entities and (b) that scientific
theories are increasing approximations to the truth. The convergent realist's argument is a dynamic one: it argues
from the increasing success of science to the thesis that scientific theories are converging on the truth about the
basic structure of the world. The position thus involves the concepts of reference, increasing approximations to the
truth, and success, which we have to explicate first before discussing arguments for the position.
Sense vs. Reference. Philosophers generally distinguish the sense of a designating term from its reference. For
example, "the current President of the U.S." and "the Governor of Arkansas in 1990" are different descriptive phrases
with distinct senses (meanings), but they refer to the same object, namely Bill Clinton. Some descriptive phrases are
meaningful, but have no referent, e.g., "the current King of France." We can refer to something by a description if the
description uniquely designates some object by virtue of a certain class of descriptive qualities. For example,
supposing that Moses was a real rather than a mythical figure, descriptions such as "the author of Genesis," "the
leader of the Israelites out of Egypt" and "the author of Exodus" might serve to designate Moses (as would the name
"Moses" itself). Suppose now that we accept these descriptions as designations of Moses but subsequently discover
that Moses really didn't write the books of Genesis and Exodus. (Some biblical scholars believe the stories in these
books were transmitted orally from many sources and weren't written down until long after Moses had died.) This
would amount to discovering that descriptions such as "the author of Genesis" do not refer to Moses, but this
wouldn't mean that Moses didn't exist, simply that we had a false belief about him. While we may pick out Moses
using various descriptions, it doesn't mean that all the descriptions we associate with Moses have to be correct for us
to do so.
Realists hold that unobservable entities are like that between changes in scientific theories. Though different theories
of the electron have come and gone, all these theories refer to the same class of objects, i.e., electrons. Realists
argue that terms in our best scientific theories (such as "electron") typically refer to the same classes of unobservable
entities across scientific change, even though the set of descriptions (properties) that we associate with those
classes of objects change as the theories develop.
Approximate Truth. The notion of approximate truth has never been clearly and generally defined, but the intuitive
idea is clear enough. There are many qualities we attribute to a class of objects, and the set of qualities that we truly
attribute to those objects increases over time. If that is so, then we say that we are moving "closer to the truth" about
those objects. (Perhaps in mathematical cases we can make clearer sense of the notion of increasing closeness to
the truth; i.e., if we say a physical process evolves according to a certain mathematical equation or "law," we can
sometimes think of improvements of that law converging to the "true" law in a well-defined mathematical sense.)
The "Success" of Science. This notion means different things to different people, but is generally taken to refer to
the increasing ability science gives us to manipulate the world, predict natural phenomena, and build more
sophisticated technology.
Convergent realists often argue for their position by pointing to the increasing success of science. This requires that
there be a reasonable inference from a scientific theory's "success" to its approximate truth (or to the thesis that its
terms refer to actual entities). However, can we make such an inference? As Laudan presents the convergent

realist's argument in "A Confutation of Convergent Realism," the realist is arguing that the best explanation of a
scientific theory's success is that it is true (and its terms refer). Thus, the convergent realist uses "abductive"
arguments of the following form.
If a scientific theory is approximately true, it will (normally) be successful.
[If a scientific theory is not approximately true, it will (normally) not be successful.]
Scientific theories are empirically successful.
Scientific theories are approximately true.
If the terms in a scientific theory refer to real objects, the theory will (normally) be successful.
[If the terms in a scientific theory do not refer to real objects, the theory will (normally) not be successful.]
Scientific theories are empirically successful.
The terms in scientific theories refer to real objects.
Laudan does not present the argument quite this way; in particular, he does not include the premises in brackets.
These are needed to make the argument even prima facie plausible. (Note, however, that the premises in brackets
are doing almost all the work. That's OK since, as Laudan points out, the convergent realist often defends the first
premise in each argument by appealing to the truth of the second premise in that argument.)
Question: Could the convergent realist use weaker premises, e.g., that a scientific theory is more likely to be
successful if it is true than if it is false (with a similar premise serving the role in the case of reference). Unfortunately,
this would not give the convergent realist what he wants since he could only infer from this that success makes it
more likely that the theory is true--which is not to say that success makes it likely that the theory is true. (This can be
seen by examining the probabilistic argument below, where pr(T is successful) _ 1 and pr(T is true) _ 1, and PR(-) =
pr(-|T is successful).)
(1) pr(T is successful | T is true) > pr(T is successful | T is false)
(2) pr(T is true | T is successful) > pr(T is true)
(3) PR(T is successful) = 1
PR(T is true) > pr(T is true)
That said, let's go with the stronger argument that uses the conditionals stated above. Without the premises in
brackets, these arguments are based on inference to the best explanation of the success of science. As such, it
would be suspect, for the reasons given before. Leaving that issue aside, Laudan argues that any premises needed
for the argument to go through are false. Let us consider the case of reference first. Laudan argues that there is no
logical connection between having terms that refer and success. Referring theories often are not successful for long
periods of time (e.g., atomic theory, Proutian hydrogen theory of matter, and the Wegnerian theory of plates), and
successful theories can be non-referring (e.g., phlogiston theory, caloric theory, and nineteenth century ether
theories).
Next, we consider the premise that if a scientific theory is approximately true, it is (normally) successful. However, we
wouldn't expect theories that are approximately true overall to result in more true than false consequences in the
realm of what we have happened to observe. (That's because a false but approximately true scientific theory is likely
to make many false predictions, which could for all we know "cluster" in the phenomena we've happened to observe.)
In addition, since theories that don't refer can't be "approximately true," the examples given above to show that there
are successful but non-referring theories also show that there can be successful theories that aren't approximately
true (e.g., phlogiston theory, caloric theory, and nineteenth century ether theories).
Retention Arguments. Because the simple arguments given above are not plausible, convergent realists often try to
specify more precisely which terms in successful scientific theories we can reasonably infer refer to real things in the
world. The basic idea is that we can infer that those features of a theory that remain stable as the theory develops
over time are the ones that "latch onto" some aspect of reality. In particular, we have the following two theses.
Thesis 1 (Approximate Truth). If a certain claim appears an initial member of a succession of increasingly
successful scientific theories, and either that claim or a claim of which the original claim is a special case appears in

subsequent members of that succession, it is reasonable to infer that the original claim was approximately true, and
that the claims that replace it as the succession progresses are increasing approximations to the truth.
Thesis 2 (Reference). If a certain term putatively referring to certain type of entity occurs in a succession of
increasingly successful scientific theories, and there is an increasingly large number of properties that are stably
attributed to that type of entity as the succession progresses, then it is reasonable to infer that the term refers to
something real that possesses those properties.
There are many ways a theory can retain claims (or terms) as it develops and becomes more successful. (We will
allow those claims and terms that remain stable across radical change in theories, such as "paradigm shifts" of the
sort described by Kuhn.) For example, the old theory could be a "limiting case" of the new, in the formal sense of
being derivable from it (perhaps only with auxiliary empirical assumptions that according to the new theory are false);
or, the new theory may reproduce those empirical consequences of the old theory that are known to be true (and
maybe also explain why things would behave just as if the old there were true in the domain that was known at the
time); finally, the new theory may preserve some explanatory features of the old theory. Convergent realists argue
from retention of some structure across theoretical change that leads to greater success that whatever is retained
must be either approximately true (in the case of theoretical claims such as Newton's Laws of Motion, which are
"limiting cases" of laws that appear in the new, more successful theory) or must refer to something real (in the case of
theoretical terms such as "electron," which have occurred in an increasingly successful succession of theories as
described by Thesis 2).
Next time, we will examine an objection to retention arguments, as well as several replies that could be made on
behalf of the convergent realist.
Lecture 22
4/21/94
Convergent Realism and the History of Science
Last time, we ended by discussing a certain type of argument for convergent realism known as retention arguments.
These arguments rely on the following two theses.
Thesis 1 (Approximate Truth). If a certain claim appears an initial member of a succession of increasingly
successful scientific theories, and either that claim or a claim of which the original claim is a special case appears in
subsequent members of that succession, it is reasonable to infer that the original claim was approximately true, and
that the claims that replace it as the succession progresses are increasing approximations to the truth.
Thesis 2 (Reference). If a certain term putatively referring to certain type of entity occurs in a succession of
increasingly successful scientific theories, and there is an increasingly large number of properties that are stably
attributed to that type of entity as the succession progresses, then it is reasonable to infer that the term refers to
something real that possesses those properties.
As I noted last time, there are many ways a theory can retain claims (or terms) as it develops and becomes more
successful. (We will allow those claims and terms that remain stable across radical change in theories, such as
"paradigm shifts" of the sort described by Kuhn.) For example, the old theory could be a "limiting case" of the new, in
the formal sense of being derivable from it (perhaps only with auxiliary empirical assumptions that according to the
new theory are false); or, the new theory may reproduce those empirical consequences of the old theory that are
known to be true (and maybe also explain why things would behave just as if the old there were true in the domain
that was known at the time); finally, the new theory may preserve some explanatory features of the old theory.
Convergent realists argue from retention of some structure across theoretical change that leads to greater success
that whatever is retained must be either approximately true (in the case of theoretical claims such as Newton's Laws
of Motion, which are "limiting cases" of laws that appear in the new, more successful theory) or must refer to
something real (in the case of theoretical terms such as "electron," which have occurred in an increasingly successful
succession of theories as described by Thesis 2).
Warning: Now I should note that I am presenting the convergent realist's position, and retention arguments in
general, somewhat differently than Laudan does in "A confutation of convergent realism." In that article, Laudan
examines the "retentionist" thesis that new theories should retain the central explanatory apparatus of their
predecessors, or that the central laws of the old theory should provably be special cases of the central laws of the
new theory. This is a prescriptive account of how science should proceed. According to Laudan, convergent realists
also hold that scientists follow this strategy, and that the fact that scientists are able to do so proves that the

successive theories as a whole are increasing approximations to the truth. He objects, rightly, that while there are
certain cases where retention like this occurs (e.g., Newton-Einstein), there are many cases in which retention of this
sort does not occur (Lamarck-Darwin; catastrophist-uniformitarian geology; corpuscular-wave theory of light; also,
any of the examples that involve an ontological "loss" of the sort emphasized by Kuhn, e.g., phlogiston, ether,
caloric). Moreover, Laudan points out that when retention occurs (as in the transition from Newtonian to relativistic
physics), it only occurs with regard to a few select elements of the older theory. The lesson he draws from this is that
the "retentionist" strategy is generally not followed by scientists, so the premise in the retentionist argument that says
that scientists successfully follow this strategy is simply false. I take it that Laudan is correct in his argument, and
refer you to his article for his refutation of that type of "global" retentionist strategy. What I'm doing here is slightly
different, and more akin to the retentionist argument given by Ernan McMullin in his article "A Case for Scientific
Realism." A retentionist argument that uses Theses 1 and 2 above and the fact that scientific theories are
increasingly successful to argue for a realist position is not committed to the claim that everything in the old theory
must be preserved in the new (perhaps only as a special case); it is enough that some things are preserved.
(McMullin, in particular, uses a variation on Thesis 2, where the properties in question are structural properties, to
argue for scientific realism. See the section of his article entitled "The Convergences of Structural Explanation.") That
is because the convergent realist whom I am considering claims only that it is reasonable to infer the reality (or
approximate truth) of those things that are retained (in one of the senses described above) across theoretical
change. Thus, it is not an objection to the more reasonable convergent realist position that I'm examining here (of
which McMullin's convergent realism is an example) to claim that not everything is retained when the new theory
replaces the old, and that losses in overall ontology occur with theoretical change along with gains.
That said, there are still grounds for challenging the more sensible, selective retentionist arguments that are based
on Theses 1 and 2. I will concentrate on the issue of successful reference (Thesis 2); similar arguments can be given
for the case of approximate truth (Thesis 1). (Follow each objection and reply with discussion.)
Objection: The fact that a theoretical term has occurred in a succession of increasingly successful scientific theories
as described by Thesis 2 does not guarantee that it will continue to appear in all future theories. After all, there have
been many theoretical terms that appeared in increasingly successful research programs (phlogiston, caloric, ether)
in the way described by Thesis 2, but those research programs subsequently degenerated and went the way of the
dinosaurs. What the convergent realist needs to show to fully defend his view is that there is reason to think that
those terms and concepts that have been retained in recent theories (electron, quark, DNA, genes, fitness) will
continue to be retained in all future scientific theories. However, there is no reason to think that: indeed, if we
examine the history of science we should infer that any term or concept that appears in our theories today is likely to
be replaced at some time in the future.
Reply 1: The fact that many terms have been stably retained as described by Thesis 2 in recent theories is the best
reason one could have to believe that they will continue to be retained in all future theories. Of course, there are no
guarantees that this will be the case: quarks may eventually go the way of phlogiston and caloric. Nevertheless, it is
reasonable to believe that those terms will be retained, and, what is more important, it becomes more reasonable to
believe this the longer such terms are retained and the more there is a steady accumulation of properties that are
stably attributed to the type of entities those terms putatively designate. Anti-realists such as Laudan are right when
they point out that one cannot infer that the unobservable entities postulated by a scientific theory are real based
solely on the empirical success of that theory, but that is not what scientists do. The inference to the reality of the
entities is also a function of the degree of stability in the properties attributed to those entities, the steadiness of
growth in that class of properties over time, and how fruitful the postulation of entities of that sort has been in
generating accumulation of this sort. (See McMullin, "The Convergences of Structural Explanation" and "Fertility and
Metaphor," for a similar point.)
Reply 2: It's not very convincing to point to past cases in the history of science where theoretical entities, such as
caloric and phlogiston, were postulated by scientists but later rejected as unreal. Science is done much more
rigorously, rationally, and successfully nowadays than it was in the past, so we can have more confidence that the
terms that occur stably in recent theories (e.g., electron, molecule, gene, and DNA) refer to something real, and that
they will continue to play a role in future scientific theories. Moreover, though our conception of these entities has
changed over time, there is a steady (and often rapid) accumulation of properties attributed to entities postulated by
modern scientific theories, much more so than in the past. This shows that the terms that persist across theoretical
change in modern science should carry more weight than those terms that persisted across theoretical change in
earlier periods.
Reply 3: The objection overemphasizes the discontinuity and losses that occur in the history of science. Laudan, like
Kuhn, wants to argue there are losses as well as gains in the ontology and explanatory apparatus of science.

However, in doing so he overemphasizes the importance of those losses for how we should view the progression of
science. The fact that scientific progress is not strictly cumulative does not mean that there is not a steady
accumulation of entities in scientific ontology and knowledge about the underlying structural properties that those
entities possess. Indeed, one can counter Laudan's lists of entities that have been dropped from the ontology of
science with equally long lists of entities that were introduced and were retained across theoretical change. Our
conception of these entities has changed, but rather than focusing on the conceptual losses, one should focus on the
remarkable fact that there is a steadily growing class of structural properties that we attribute to the entities that have
been retained, even across radical theoretical change. (See McMullin, "Sources of Antirealism: History of Science,"
for a similar point.)
Lecture 23
4/26/94
The Measurement Problem, Part I
For the next two days we will be examining the philosophical problems that arise in quantum physics, specifically
those connected with the interpretation of measurement. Today's lecture will provide the necessary background and
describe how measurement is understood according to the "orthodox" interpretation of quantum mechanics, and
several puzzling features of that view. The second lecture will describe various "non-orthodox" interpretations of
quantum mechanics that attempt to overcome these puzzles by interpreting measurement in a quite different way.
A Little Historical Background
As many of you know, the wave theory of light won out over the corpuscular theory by the beginning of the nineteenth
century. By the end of that century, however, problems had emerged with the standard wave theory. The first problem
came from attempts to understand the interactions of matter and light. A body at a stable temperature will radiate and
absorb light. The body will be at equilibrium with the light, which has its energy distributed among various
frequencies. The frequencies of light emitted by the body varies with its temperature, as is familiar from experience.
However, by the end of the nineteenth century physicists found that they could not account theoretically for that
frequency distribution. Two attempts to do so within the framework of Maxwell's theory of electromagnetic radiation
and statistical mechanics led to two laws--Wien's law and the Rayleigh-Jeans law--that failed at lower and higher
frequencies, respectively.
In 1900, Max Planck postulated a compromise law that accounted for the observed frequency distribution, but it had
the odd feature of postulating that light transmitted energy in discrete packets. The packets at each frequency had
energy E = hn, where h is Planck's constant and n is the frequency of the light. Planck regarded his law as merely a
phenomenological law, and thought of the underlying distribution of energy as continuous. In 1905, Einstein proposed
that all electromagnetic radiation consists of discrete particle-like packets of energy, each with an energy hn. He
referred to these packets, which we now call photons, as quanta of light. Einstein used this proposal to explain the
photoelectric effect (where light shining on an electrode causes the emission of electrons).
After it became apparent that light, a wave phenomenon, had particle-like aspects, physicists began to wonder
whether particles such as electrons might also have wave-like aspects. L. de Broglie proposed that they did and
predicted that under certain conditions electrons would exhibit wave-like behavior such as interference and
diffraction. His prediction turned out to be correct: a diffraction pattern can be observed when electrons are reflected
off a suitable crystal lattice (see Figure 1, below).
Figure 1
Electron Diffraction By A Crystal Lattice
Schrdinger soon developed a formula for describing the wave associated with electrons, both in conditions where
they are free and when they are bound by forces. Famously, his formula was able to account for the various possible
energy levels that had already been postulated for electrons orbiting a nucleus in an atom. At the same time,
Heisenberg was working on a formalism to account for the various patterns of emission and absorption of light by
atoms. It soon became apparent that the discrete energy levels in Heisenberg's abstract mathematical formalism
corresponded to the energy levels of the various standing waves that were possible in an atom according to
Schrdinger's equation.
Thus, Schrdinger proposed that electrons could be understood as waves in physical space, governed by his
equation. Unfortunately, this turned out to be a plausible interpretation only in the case of a single particle. With
multiple-particle systems, the wave associated with Schrdinger's equation was represented in a multi-dimensional
space (3n dimensions, where n is the number of particles in the system). In addition, the amplitudes of the waves
were often complex (as opposed to real) numbers. Thus, the space in which Schrdinger's waves exist is an abstract
mathematical space, not a physical space. Thus, the waves associated with electrons in Schrdinger's wave

mechanics could not be interpreted generally as "electron waves" in physical space. In any case, the literal
interpretation of the wave function did not cohere with the observed particle-like aspects of the electron, e.g., that
electrons are always detected at particular points, never as "spread-out" waves.
The Probability Interpretation Of Schrdinger's Waves; The Projection Postulate
M. Born contributed to a solution to the problem by suggested that the square of the amplitude of Schrdinger's wave
at a certain point in the abstract mathematical space of the wave represented the probability of finding the particular
value (of position or momentum) associated with that point upon measurement. This only partially solves the
problem, however, since the interference effects are physically real in cases such as electron diffraction and the
famous two-slit experiment (see Figure 2).
Figure 2
The Two-Slit Experiment
If both slits are open, the probability of an electron impinging on a certain spot on the photographic plate is not the
sum of the probabilities that it would impinge on that spot if it passes through slit 1 and that it would impinge on that
spot if it passes through slit 2, as is illustrated by pattern (a). Instead, if both slits are opinion you get an interference
pattern of the sort illustrated by pattern (b). This is the case even if the light source is so weak that you are in effect
sending only one photon at a time through the slits. Eventually, an interference pattern emerges, one point at a time.
Interestingly, the interference pattern disappears if you place a detector just behind one of the slits to determine
whether the photon passed through that slit or not, and you get a simple sum of the waves as illustrated by pattern
(a). Thus, you can't explain the interference pattern as resulting from interaction among the many photons in the light
source.
A similar sort of pattern is exhibited by electrons when you measure their spin (a two-valued quantity) in a SternGerlach device. (This device creates a magnetic field that is homogeneous in all but one direction, e.g., the vertical
direction; in that case, electrons are deflected either up or down when they pass through the field.) You get
interference effects here just as in the two-slit experiment described above. (See Figure 3)
Figure 3
Interference Effects Using Stern-Gerlach Devices
When a detector is placed after the second, left-right oriented Stern-Gerlach device as in (d), the final, up-down
oriented Stern-Gerlach device shows that half of the electrons are spin "up" and half of the electrons are spin "down."
In this case, the electron beam that impinges on the final device is a definite "mixture" of spin "left" and spin "right"
electrons. On the other hand, when no detector is placed after the second, left-right oriented Stern-Gerlach device as
in (e), all the electrons are measured "up" by the final Stern-Gerlach device. In that case, we say that the beam of
electrons impinging on the final device is a "superposition" of spin "left" and spin "right" electrons, which happens to
be equivalent to a beam consisting of all spin "up" electrons. Thus, placing a device before the final detector destroys
some information present in the interference of the electron wave, just as placing a detector after a slit in the two-slit
experiment destroyed the interference pattern there.
J. von Neumann generalized the formalisms of Schrdinger and Heisenberg. In von Neumann's formalism, the state
of a system is represented by a vector in a complex, multi-dimensional vector space. Observable features of the
world are represented by operators on this vector space, which encode the various possible values that that
observable quantity can have upon measurement. Von Neumann postulated that the "state vector" evolves
deterministically in a manner consistent with Schrdinger's equation, until there is a measurement, in which case
there is a "collapse," which indeterministically alters the physical state of the system. This is von Neumann's famous
"Projection Postulate."
Thus, von Neumann postulated that there were two kinds of change that could occur in a state of a physical system,
one deterministic (Schrdinger evolution), which occurs when the system is not being measured, and one
indeterministic (projection or collapse), which occurs as a result of measuring the system. The main argument that
von Neumann gave for the Projection Postulate is that whatever value of a system is observed upon measurement
will be found with certainty upon subsequent measurement (so long as measurement does not destroy the system, of
course). Thus, von Neumann argued that the fact that the value has a stable result upon repeated measurement
indicates that the system really has that value after measurement.
The Orthodox Copenhagen Interpretation (Bohr)
What about the state of the system in between measurements? Do observable quantities really have particular
values that measurement reveals? Or are there real values that exist before measurement that the measurement
process itself alters indeterministically? In either case, the system as described by von Neumann's state vectors (and
Schrdinger's wave equation) would have to be incomplete, since the state vector is not always an "eigenvector," i.e.,
it does not always lie along an axis that represents a particular one of the possible values that a particular observable

quantity (such as spin "up"). Heisenberg at first took the view that measurement indeterministically alters the state of
the system that existed before measurement. This implies that the system was in a definite state before
measurement, and that the quantum mechanical formalism gives an incomplete description of physical systems.
Famously, N. Bohr proposed an interpretation of the quantum mechanical formalism that denies that the description
given by the quantum mechanical formalism is incomplete. According to Bohr, it only makes sense to attribute values
to observable quantities of a physical system when system is being measured in a particular way. Descriptions of
physical systems therefore only make sense relative to particular contexts of measurement. (This is Bohr's solution to
the puzzling wave-particle duality exhibited by entities such as photons and electrons: the "wave" and "particle"
aspects of these entities are "complementary," in the sense that it is physically impossible to construct a measuring
device that will measure both aspects simultaneously. Bohr concluded that from a physical standpoint it only makes
sense to speak about the "wave" or "particle" aspects of quantum entities as existing relative to particular
measurement procedures.) One consequence of Bohr's view is that one cannot even ask what a physical system is
like between measurements, since any answer to this question would necessarily have to describe what the physical
system is like, independent of any particular context of measurement.
It is important to distinguish the two views just described, and to distinguish Bohr's view from a different view that is
sometimes attributed to him:
A physical system's observable properties always have definite values between measurement, but we can never
know what those values are since the values can only be determined by measurement, which indeterministically
disturbs the system. (Heisenberg)
It does not make sense to attribute definite values to a physical system's observable properties except relative to a
particular kind of measurement procedure, and then it only makes sense when that measurement is actually being
performed. (Bohr)
A physical system's observable properties have definite values between measurement, but these values are not
precise, as is the case when the system's observable properties are being measured; rather, the values of the
system's observable quantities before measurement are "smeared out" between the particular values that the
observable quantity could have upon measurement. (Pseudo-Bohr)
Each of these views interprets a superposition differently, as a representation of our ignorance about the true state of
the system (Heisenberg), as a representation of that values that the various observable quantities of the system
could have upon measurement (Bohr), or as a representation of the indefinite, imprecise values that the observable
quantities have between measurements (Pseudo-Bohr). Accordingly, each of these views interprets projection or
collapse differently, as a reduction in our state of ignorance (Heisenberg), as the determination of a definite result
obtained when a particular measurement procedure is performed (Bohr), or as an instantaneous localization of the
"smeared out," imprecise values of a particular observable quantity (Pseudo-Bohr).
Next time we will discuss several problematic aspects of the Copenhagen understanding of measurement, along with
several alternative views.
Lecture 24
4/28/94
The Measurement Problem, Part II
Last time, we saw that Bohr proposed an interpretation of the quantum mechanical formalism, the Copenhagen
interpretation, that has become the received view among physicists. Today we will look more closely at the picture
that view gives us of physical reality and the measurement process in particular. Then we will examine several
alternative interpretations of measurement, proceeding from the least plausible to the most plausible alternatives.
The Copenhagen Understanding Of Measurement
According to Bohr, it only makes sense to attribute values to observable quantities of a physical system when system
is being measured in a particular way. Descriptions of physical systems therefore only make sense relative to
particular contexts of measurement. (This is Bohr's solution to the puzzling wave-particle duality exhibited by entities
such as photons and electrons: the "wave" and "particle" aspects of these entities are "complementary," in the sense
that it is physically impossible to construct a measuring device that will measure both aspects simultaneously. Bohr
concluded that from a physical standpoint it only makes sense to speak about the "wave" or "particle" aspects of
quantum entities as existing relative to particular measurement procedures.) One consequence of Bohr's view is that
one cannot even ask what a physical system is like between measurements, since any answer to this question would
necessarily have to describe what the physical system is like, independent of any particular context of measurement.
On Bohr's view (which is generally accepted as the "orthodox" interpretation of quantum mechanics even though it is
often misunderstood), the world is divided into two realms of existence, that of quantum systems, which behave

according to the formalism of quantum mechanics and do not have definite observable values outside the context of
measurement, and of "classical" measuring devices, which always have definite values but are not described within
quantum mechanics itself. The line between the two realms is arbitrary: at any given time, one can consider a part of
the world that serves as a "measuring instrument" (such as the Stern-Gerlach device) either as a quantum system
that interacts with other quantum systems according to the deterministic laws governing the state vector
(Schrdinger's equation) or as a measuring device that behaves "classically" (i.e., always has definite observable
properties) though indeterministically.
There are several difficulties with this view, which together constitute the "measurement problem." To begin with, the
orthodox interpretation gives no principled reason why physics should not be able to give a complete description of
the measurement process. After all, a measuring device (such as a Stern-Gerlach magnet) is a physical system, and
in performing a measurement it simply interacts with another physical system such as a photon or an electron.
However, (a) there seems to be no principled reason why one particular kind of physical interaction is either
indescribable within quantum physics itself (as Bohr suggests), or is not subject to the same laws (e.g., Schrdinger's
equation) that governs all other physical interactions, and (b) the orthodox view offers no precise characterization that
would mark off those physical interactions that are measurements from those physical interactions that are not.
Indeed, the orthodox interpretation claims that whether a certain physical interaction is a "measurement" is arbitrary,
i.e., a matter of choice on the part of the theorist modeling the interaction. However, this hardly seems satisfactory
from a physical standpoint. It does not seem to be a matter of mere convenience where we are to draw the line
between the "classical" realm of measuring devices and the realm of those physical systems that obey the
deterministic laws of that formalism (in particular, Schrdinger's equation). For example, Schrdinger pointed out that
the orthodox interpretation allows for inconsistent descriptions of the state of macroscopic systems, depending on
whether we consider them measuring devices. For example, suppose that you placed a cat in an enclosed box along
with a device that will release poisonous gas if (and only if) a Geiger counter measures that a certain radium atom
has decayed. According to the quantum mechanical formalism, the radium atom is in a superposition of decaying and
not decaying, and since there is a correlation between the state of the radium atom and the Geiger counter, and
between the state of the Geigen counter and the state of the cat, the cat should also be in a superposition,
specifically, a superposition of being alive and dead, if we do not consider the cat to be a measuring device. If the
orthodox interpretation is correct, however, this would mean that there is no fact of the matter about whether the cat
is alive or dead, if we consider the cat not to be a measuring device. On the other hand, if we consider the cat to be a
measuring device, then according to the orthodox interpretation, the cat will either be definitely alive or definitely
dead. However, it certainly does not seem to be a matter of "arbitrary" choice whether (a) there is no fact of the
matter about whether the cat is alive or dead before we look into the box, or (b) the cat is definitely alive or dead
before we look into the box.
Wigner's Idealism: Consciousness As The Cause Of Collapse
Physicist E. Wigner argued against Bohr's view that the distinction between measurement and "mere" physical
interaction could be made arbitrarily by trying to emphasize its counterintuitive consequences. Suppose that you put
one of Wigner's friends in the box with the cat. The "measurement" you make at a given time is to ask Wigner's friend
if the cat is dead or alive. If we consider your friend as part of the experimental setup, quantum mechanics predicts
that before you ask Wigner's friend whether the cat is dead or alive, he is in a superposition of definitely believing the
cat is dead and definitely believing that the cat is alive. Wigner argued that this was an absurd consequence of
Bohr's view. People simply do not exist in superposed belief-states. Wigner's solution was that, contrary to what Bohr
claimed, there is a natural division between what constitutes a measurement and what does not--the presence of a
conscious observer. Wigner's friend is conscious; thus, he can by an act of observation cause a collapse of the wave
function. (Alternately, if we consider the cat to be a conscious being, the cat would be definitely alive or definitely
dead even before Wigner's friend looked to see how the cat was doing.)
Few physicists have accepted Wigner's explanation of what constitutes a measurement, though his view has "trickled
down" to some of the (mainly poor quality) popular literature on quantum mechanics (especially the type of literature
that sees a direct connection between quantum mechanics and Eastern religious mysticism). The basic source of
resistance to Wigner's idealism is that it requires that physicists solve many philosophical problems they'd like to
avoid, such as whether cats are conscious beings. More seriously, Wigner's view requires a division of the world into
two realms, one occupied by conscious beings who are not subject to the laws of physics but who can somehow
miraculously disrupt the ordinary deterministic evolution of the physical systems, and the other by the physical
systems themselves, which evolve deterministically until a conscious being takes a look at what's going on. This is
hardly the type of conceptual foundation needed for a rigorous discipline such as physics.
The Many-Worlds Interpretation

Another view, first developed by H. Everett (a graduate student at Princeton) in his Ph.D. thesis, is called the manyworlds interpretation. It was later developed further by another physicist, B. de Witt. This view states that there is no
collapse when a measurement occurs; instead, at each such point where a measurement occurs the universe
"branches" into separate, complete worlds, a separate world for every possible outcome that measurement could
have. In each branch, it looks like there the measuring devices indeterministically take on a definite value, and the
empirical frequencies that occur upon repeated measurement in almost every branch converge on the probabilities
predicted by the Projection Postulate. The deterministically evolving wave function correctly describes the quantum
mechanical state of the universe as a whole, including all of its branches.
Everett's view has several advantages, despite its conceptual peculiarity. First, it can be developed with
mathematical precision. Second, it postulates that the universe as a whole evolves deterministically according to
Schrdinger's equation, so that you only need only type of evolution, not Schrdinger's equation plus collapse during
a measurement. Third, the notion of a measurement (which is needed to specify where the branching occurs) can be
spelled out precisely and non-arbitrarily as a certain type of physical interaction. Fourth, it makes sense to speak of
the quantum state of the universe as a whole on Everett's view, which is essential if you want to use quantum
mechanical laws to develop cosmological theories. By contrast, on the orthodox, Copenhagen view, quantum
mechanics can never describe the universe as a whole since quantum mechanical description is always relativized to
an arbitrarily specified measuring device, which is not itself given a quantum mechanical description.
On the other hand, the many worlds interpretation has several serious shortcomings. First, it's simply weird
conceptually. Second, it does not account for the fact that we never experience anything like a branching of the
world. How is it that our experience is unified (i.e., how is it that my experience follows one particular path in the
branching universe and not others)? Third, as noted above, the many worlds interpretation predicts that there will be
worlds where the observed empirical frequencies of repeated measurements will not fit the predictions of the
quantum mechanical formalism. In other words, if the theory is true, there will be worlds at which it looks as if the
theory is false! Finally, the theory does not seem to give a clear sense to locutions such as "the probability of this
electron going up after passing through this Stern-Gerlach device is 1/2." What exactly does that number 1/2 mean, if
the universe simply branches into two worlds, in one of which the electron goes up and in the other of which it goes
down? The same branching would occur, after all, if the electron were in a state where the probability of its going up
were 3/4 and the probability of its going down were 1/4.
Because of problems like this, the many worlds interpretation has gained much notoriety among physicists, but
almost no one accepts it as a plausible interpretation.
The Ghirardi-Rimini-Weber (GRW) Interpretation
In 1986, three physicists (G.C. Ghirardi, A. Rimini, and T. Weber) proposed a way of accounting for the fact that
macroscopic objects (such as Stern-Gerlach devices, cats, and human beings) are never observed in superpositions,
whereas microscopic systems (such as photons and electrons) are. (Their views were developed further in 1987 by
physicist John Bell.) According to the Ghirardi-Rimini-Weber (GRW) interpretation, there is a very small probability
(one in a trillion) that the wave functions for the positions of isolated, individual particles will collapse spontaneously
at any given moment. When particles couple together to form an object, the tiny probabilities of spontaneous collapse
quickly add up for the system as a whole, since when one particle collapses so does every particle to which that
particle is coupled. Moreover, since even a microscopic bacterium is composed of trillions and trillions of subatomic
particles, the probability that a macroscopic object will have a definite spatial configuration at any given moment is
vanishingly close to 100 percent. Thus, GRW can explain in a mathematically precise way why we never observe
macroscopic objects in superpositions, but often observe interference effects due to superposition when we're
looking at isolated subatomic particles. Moreover, they too can give a precise and non-arbitrary characterization of
the measurement process as a type of physical interaction.
There are two problems with the GRW interpretation. First, though subatomic particles are localized to some degree
(in that the collapse turns the wave function representing the position of those particles into a narrowly focused bell
curve), they never do have precise positions. The tails of the bell curve never vanish, extending to infinity. (Draw
diagram on board.) Thus, GRW never give us particles with definite positions or even with a small but spatially
extended positions: instead, what we get are particles having "mostly" small but spatially extended positions.
Importantly, this is also true of macroscopic objects (such as Stern-Gerlach devices, cats, and human beings) that
are composed of subatomic particles. In other words, according to GRW you are "mostly" in this room, but there's a
vanishingly small part of you at every other point in the universe, no matter how distant! Second, GRW predicts that
energy is not conserved when spontaneous collapses occur. While the total violation of conservation of energy
predicted by GRW is too small to be observed, even over the lifetime of the universe, it does discard a feature of
physical theory that many physicists consider to be conceptually essential.

Bohm's Interpretation
In 1952, physicist David Bohm formulated a complete alternative to standard (non-relativistic) quantum mechanics.
Bohm's theory was put into a relatively simple and elegant mathematical form by John Bell in 1982. Bohm's theory
makes the same predictions as does standard (non-relativistic) quantum mechanics but describes a classical,
deterministic world that consists of particles with definite positions. The way that Bohm does this is by postulating that
besides particles, there is a quantum force that moves the particles around. The physically real quantum force, which
is represented mathematically by Schrdinger's wave equation, pushes the particles around so that they behave
exactly as standard quantum mechanics predicts. Bohm's theory is deterministic: if you knew the initial configuration
of every particle in the universe, applying Bohm's theory would allow you to predict with certainty every subsequent
position of every particle in the universe. However, there's a catch: the universe is set up so that it is a physical
(rather than a merely practical) impossibility for us to know the configuration of particles in the universe. Thus, from
our point of view the world behaves just as if it's indeterministic (though the apparent indeterminism is simply a
matter of our ignorance); also, though the wave function governing the particle's motion never collapses, the particle
moves around so that it looks as if measurement causes it to collapse.
Despite its elegance and conceptual clarity, Bohm's theory has not been generally accepted by physicists, for two
reasons. (1) Physicists can't use Bohm's theory, since it's impossible to know the configuration of all particles in the
universe at any given time. Because of our ignorance of this configuration, the physical theory we have to use to
generate predictions is standard quantum mechanics. Why then bother with Bohm's theory at all? (2) What is more
important, in Bohm's theory all the particles in the universe are intimately connected so that every particle
instantaneously affects the quantum force governing the motions of the other particles. Many physicists (Einstein
included) saw this feature of Bohm's theory as a throwback to the type of "action at a distance" expunged by relativity
theory. Moreover, Bohm's theory assumes that simultaneity is absolute and so is inconsistent with relativity theory in
its present form. (There is no Bohmian counterpart to relativistic quantum mechanics, though people are currently
working on producing one.)
Conclusion
Each of the various interpretations of quantum mechanics we have examined, from the orthodox Copenhagen
interpretation to the four non-orthodox interpretations (idealistic, many-worlds, GRW, and Bohm) has some sort of
conceptual shortcoming. This is what makes the philosophy of quantum mechanics so interesting: it shows us that
the fact that experiments agree with the predictions of quantum mechanics, under any of these interpretations,
indicates one thing with certainty--the world we live in is a very weird place!