01.

Course Intro
Two main areas of philosophy of science:
(1) Methodology of Science
(2) Philosophical Foundations of Particular Theories in Science
(1) Methodology of Science
Is there a scientific method?
What characterizes the sort of knowledge produced by such a method?
exact/hard
physics
chemistry
biology
etc.
nonexact/soft
sociology
anthropology
psychology
etc.
physics
chemistry
biology
psychology
sociology
anthropology
geology
paleontology
oceanography
meterology
archeology
etc.
sample fields in science
unified via
reduction? unified
via allegence to a
single method?
Not related at all?
Possible grouping. Other possibilities:
experimental vs. theoretical
life sciences vs. hard sciences
Sample topics
(a) Explanation
Does science aim towards providing explanations of phenomena? What is a scientific explanation, and how does
it differ, if at all, from other types of explanations?
(b) Confirmation
What is the relation between theory and evidence? What factors condition our belief in the claims made by
scientific theories?
(c) Laws of Nature
What is a law of nature?
(d) Scientific Realism
What do scientific theories tell us about the world? What would the world be like if they were true? Should we
draw a line between claims a theory makes that we could be justified in believing and claims that we cannot be
justified in believing?
(e) Science vs. Pseudoscience
The demarcation problem: Are there general characteristics we can use to distinguish science from
pseudoscience?
(f) Scientific Change
What happens when one theory in science supplants another? Can science be said to progress?
(g) Unity of Science
Is there one science or many sciences? How are different fields in science related, if at all?
1
(2) Philosophical Foundations of Particular Theories in Science
Philosophy of physics
(a) Nature of spacetime in general relativity
(b) Interpretations of quantum mechanics
(c) Determinism in classical and quantum physics
(d) Ontology of quantum field theory
Philosophy of biology
(a) Types of explanation in the theory of natural selection
(b) Demarcation: evolution vs. creationism
(c) Reductionism in biology
Sample Topics:
Philosophy of psychology
(a) Demarcation: psychoanalysis
(b) Realism vs. anti-realism: behaviorism vs. cognitive science
2
In this course, well first become aquainted with the central issues in Area (1). To do this, we will
have to go back to their original formulations, which usually means going back to the group of
philosophers in the 1920s-30s generally known as the logical positivists. All 20th century
philosophy of science is either an extension of, or a reaction (usually violent) to this group. The
first 3/4s of the class will be spent on the following topics:
I. Explanation and Laws of Nature
II. Confirmation
III. Scientific Theories, Scientific Change, and Scientific Realism
In the last quarter of the course, we'll look at some topics in Area (2).
02. Explanation. Part 1.
Topics:
I. Introduction
II. Deductive-Nomological (DN) Model
III. Laws: Preliminary Sketch
I. Introduction
First blush:
A scientific explanation is an attempt to render understandable or intelligible some particular
event, or some general fact, by appealing to other particular and/or general facts drawn from one
or more branches of empirical science.
As Salmon notes, this is pretty vague. Lets get a bit more precise.
Terminology:
1. Explanandum - fact (particular or general) to be explained.
2. Explanans - that which does the explaining
3. Explanation - (2 hair-splitting views)
(a) A linguistic object consisting of an explanandum-statement and an explanans statement.
(b) A collection of facts consisting of explanandum-facts and explanans-facts.
More Preliminaries: Arguments
Since DN views explanations as arguments, we should be clear about what arguments are.
ASIDE: Weve now introduced 2 distinct types of object:
(1) an explanation; and
(2) an argument.
These are not necessarily the same type of object! One attempt to further define what an explanation
in science amounts to is the DN account. This particular account claims an explanation is a type of
argument. We will investigate the adequacy of this account in the following lectures. But at this point,
it is very important to realize that, in general, explanations and arguments are differents sorts of things.
An argument is a collection of statements, one of which is identified as a claim
(conclusion), and the others are identified as reasons given for the claim (premises).
1
(For our purposes, nothing too important rides on this distinction. Just be aware of it.)
Two Main Types of Argument (not explanation!)
1. Ampliative: Conclusion contains information
beyond that expressed in premises.
2. Not necessarily truth preserving.
3. Not erosion-proof: Addition of premises may
strengthen/weaken argument.
1. Non-ampliative: Content of conclusion
is present in premises.
2. Truth-preserving: If premises are
true, conclusion must be true.
3. Erosion-proof: Addition of new premises
does not affect strength of argument (as long
as original premises are left alone).
4. Deductive validity is all-or-nothing.
A deductive argument is either valid or
invalid.
4. Inductive strength comes in degrees. Some
inductive arguments are stronger/weaker than
others.
This is a valid deductive argument:
1. It is non-ampliative: the conclusion is already implicit in the premises.
2. It is truth-preserving: If it is true that all animals with wings can fly, and if it is true that pigs have wings, then it
must also be true that pigs can fly.
3. It is erosion-proof: If we added the premise Wilbur has butterfly wings, the conclusion would still follow in the
required truth-preserving way.
ASIDE: Of course, pigs dont have wings, and not all animals with wings can fly (penguins, for example). Note,
however, that the truth-preserving property simply requires that it can never be the case that all the premises are true
and the conclusion false. So long as this holds, the argument is valid. This does allow any other combination of truth-
values for the premises and conclusion. For instance, a valid argument could have all false premises and a true
conclusion; or all false premises and a false conclusion; or some combination of false/true premises together with a
false/true conclusion. Again, the only combination that is prohibited by property (2) is the combination of all true
premises and a false conclusion.
ASIDE: If we explicitly added Pigs dont have wings as a third premise, then the conclusion Pigs can fly
would still be true if all the premises were true. To see this, note that this new third premise contradicts the
second premise Pigs have wings -- they cant both be true at the same time (or false at the same time). So
adding Pigs dont have wings prevents the argument from ever having all true premises and a false conclusion;
and this is just the truth-preserving property (2). Similarly, adding Pigs cant fly as a third premise would
contradict the conclusion, which is already implicit in the first two premises. So again, we could never have a
situation in which all the premises were true and the conclusion false. So again, property (2) would be upheld.
4. Finally, as weve seen, its validity is all-or-nothing. Weve established that it is valid, and shown that nothing we
can do to it (short of destroying it) obviscates this fact.
Ex1: All animals with wings can fly.
Pigs have wings.
Pigs can fly.
Deductive Inductive
2
Ex2: 95% of observed smokers developed lung cancer.
Smoking causes lung cancer.
This is an inductive argument.
1. It is ampliative: The conclusion contains information not already present in the premise.
2. It is not necessarily truth-preserving: If it is true that a certain survey found that 95% of smokers surveyed went
on to develop lung cancer, then it does not necessarily follow that smoking was to blame. There could have been other
causal factors that influenced the development of cancer in those 95%.
3. It is not erosion-proof. Suppose we added a second premise that states 100, 000 smokers were surveyed. This
would strengthen the conclusion, all things remaining equal. It would establish that the sample size of the survey was
very big. However, if we then added a fourth premise that states All smokers surveyed lived in coal mines, this
would weaken the conclusion. It would establish that the sample was pretty biased; in this case, it would lead us to
think that perhaps the large incidence of cancer was due to inhaling coal dust, as opposed to smoking.
4. Finally, (3) shows how inductive strength comes in degrees.
II. Deductive-Nomological (DN) Model of Scientific Explanation
DN explanation - an account of the explanandum that indicates how it follows deductively from a
law of nature (covering-law account).
Key characteristics are given by:
The conditions of adequacy define what a DN explanation is. In other words, an explanation is a
DN explanation if and only if it satisfies conditions 1-4.
Conditions of Adequacy
1. Must be a valid-deductive argument with premises stating the explanans and the
conclusion stating the explanandum.
2. Premises (explanans) must contain a law.
3. Explanans must have empirical content.
4. Explanans must be true.
CLAIM: Scientific explanations are DN explanations.
Hempel & Oppenheim (1948) Studies in the Logic of Explanation
3
General form of DN explanations
Ex1: Why do skaters spin faster as they bring their arms in towards their bodies?
(reducing rotational intertia).
DN explanation:
1. Angular momentum is conserved.
2. Skater doesnt interact with external objects.
3. Skater has non-zero initial angular momentum.
4. Skater brings arms in towards body
! Skater spins faster.
law
conditions
observed phenomena
Subsumption of particular fact (skater spinning faster) under a law (conservation of angular
momentum).
Ex2: Why did Jans bracelet melt when it was heated to 1063 C?
DN explanation:
1. Gold melts at 1063 C. law
2. Jans bracelet is made of gold. condition
! Jans bracelet melted at 1063 C. observation
ASIDE: Ex1 satisfies the 4 conditions of adequacy. In particular, it is a valid-deductive argument -- If the
premises are all true, then the conclusion must be true. To see this concretely, note that the argument can be
formulated mathematically in the following manner (where the angular momentum L of a spinning object is
defined as L = I!, where I is the objects moment of inertia (its rotational inertia, which is roughly a measure
of the objects tendancy to continue spinning in the absernce of external forces), and ! is its rotational velocity
(which measures how fast it is rotating)):
1. L
i
= L
f
2. L
i
= I
i
!
i
and L
f
= I
f
!
f
3. L
i
= 0
4. I
f
< I
i
! !
f
> !
i

(nothing contributes to L other than the skaters I and !)
(Intuitively, to preserve the equation I
i
!
i
= I
f
!
f
when I
f
is less
than I
i
, the quantity !
f
must be greater than !
i
to compensate)
L
1
, L
2
, ...
C
1
, C
2
, ...
O
1
, O
2
, ...
law(s)
conditions underwhich laws are applicable
observed phenomena
explanans
explanandum
4
Initial Problem for DN model: What is a law of nature?
Preliminary Sketch
Claim: Laws must
(a) Describe regularities that hold universally at all times and places.
(b) Be capable of supporting counterfactual statements.
(c) Be capable of supporting modal statements.
counterfactual statement = An if-then statement with a false if-clause.
Ex: If Abe Lincoln were alive today, then hed be clawing at the lid of his coffin.
modal statement = A statement that asserts a physical necessity or (im)possibility.
Ex: It is impossible to construct an enriched uranium sphere with mass > 100,000 kg.
Three examples of candidate laws:
(1) All the apples in my refrigerator are yellow.
(2) No gold sphere has a mass greater than 100,000 kg.
(3) No enriched uranium sphere has a mass greater than 100,000 kg.
Is (1) lawlike? (Does it satisfy (a), (b), (c)?)
It doesnt satisfy (a). It refers to a particular place (and time).
To say that a law supports a counterfactual/modal statement is to say that the law makes
the counterfactual/modal statement true.
Hempel & Oppenheim (1948) Studies in the Logic of Explanation
Is (2) lawlike?
It satisfies (a). (It's reasonable to suppose that in our universe there will never be enough gold
to assemble such a massive sphere.)
It doesnt satisfy (b). It doesnt support the following true counterfactual statement:
If two gold spheres with masses of 50,001 kg each were put together, then they would form a sphere with mass
100,001 kg.
It doesnt satisfy (c). It doesnt support the following true modal statement:
It is possible to construct a gold sphere with mass greater than 100,000 kg.
5
Accidental generalization = A true generalization that satisfies (a) but not (b) or (c).
Circularity Problem with this preliminary account of laws
This account says a law of nature is a true generalization that statisfies conditions (a), (b) and (c).
In particular, laws differ from accidental generalizations solely on the basis of the ability of laws to
support counterfactuals and modal statements.
Lawlike generalization = A true generalization that satisfies (a), (b) and (c).
BUT: Why do we think certain counterfactuals and modal statements are true in the first place?
If its because we think there are laws of nature that underlie them, then we cant use
them to define what we mean by a law, on pain of circularity.
SO: This preliminary account works only if we already have a theory of counterfactuals and
modal statements that is independent of the notion of a law and which can be used to
determine which counterfactuals/modal statements are true and which are false. Such a
theory is hard to envision. (And note that it cant simply be based on our intution; i.e.,
we can say that, intuitively, we think that the modal statement Its physically possible to
construct a 100,000 kg gold sphere is true. The question is, What underlies this
intuition?)
Is (3) lawlike?
It satisfies (a).
It satisfies (b). It supports the following true counterfactual statement:
If 100,000 kg of enriched uranium were assembled, then we would no longer have any uranium.
It satisfies (c). It supports the following true modal statement:
It is impossible to construct a sphere of enriched uranium with mass greater than 100,000 kg.
6
03. Laws of Nature: Three Accounts
Topics:
I. Regularity Account
II. Best System Account
III. Necessitarian Account
I. Regularity Account
Claim: Laws are regularites.
ASIDE: Lets make this a bit more concrete. Think of F and G as properties. In particular, let F be the
property of being in free fall near the surface of the earth, and let G be the property of experiencing an
acceleration of 9.8 m/s
2
(which is the acceleration due to gravity). Then Newtons Law of Gravity,
which, in this context, claims that all objects in free fall near the surface of the earth experience an
acceleration of 9.8m/s
2
is, on the SRT account, nothing more than the collection of all instances of objects
displaying both F and G. A law, on the SRT account, is just a report of a bunch of observations.
Problems
(1) Not all regularities are laws. In particular, accidental generalizations are not laws.
Hence being a regularity is not sufficient for being a law.
ASIDE: Bird also considers the possibility of "single-instance" laws, which
(obviously) are not regularities. His example is the Big Bang as a single
instance of the law encoded by Einstein's field equations in general
relativity (with relevant initial/boundary conditions). The idea is that the
Einstein equations are supposed to describe the (large scale) structure of
our universe, hence there can only be one instance of them; namely, that
instance that in fact does describe our universe.
Simple Regularity Theory (SRT)
It is a law that Fs are Gs if and only if all Fs are Gs.
(2) Not all laws are regularities (which means being a regularity is not necessary for being a
law). Purported examples:
1
(a) No-instance laws
(i) The ideal gas law: P = kT/V
(ii) Newtons 2nd law: F = ma
The claim here is that (i) and (ii) both are applicable only under ideal conditions, and in
nature such ideal conditions never occur. So there are no real instances of (i) or (ii).
Nevertheless, we still want to think of them as laws.
(b) Functional laws
Laws expressible as functions that can take on a continuum of values, more than the finite
number of instances that can be physically observed in nature.
P
V/T
A functional law is more than the mere sum of all
of its instances. Think of the instances as a finite
number of data points on a graph. These data
points dont pick out a unique function (curve on
the graph). There are a number of different curves
that can be made to fit to the data points.
Ex1: The ideal gas law
ASIDE: Functional laws indicate why the distinction between accidental generalizations and
lawlike generalizations based on the ability to support counterfactuals (and modal statements)
seems initially plausible. A functional law gives us information about instances that have not yet
been observed. Such a law allows us to infer what would be the case if certain conditions were met.
(c) Probabilistic laws
A law that states that Fs have a certain probability of being Gs.
Ex2. All nuclei of Type A have a half-life of 100,000 years (i.e., in 100,000 years, half of a population of Type A
nuclei will have decayed).
F = being a nucleus of type A
G = decaying after 100,000 years
Claim: A probabilistic law cannot be considered just a summary of its instances.
Why? A probabilistic law describes an average distribution of a property over a population
of individuals. So an individual might not have the property but still be governed by
the law.
P = pressure, T = temperature, V = volume, k = const.
Gives P as a function of T/V.
Allows P to take on a continuum of values; more than those values that actually occur as regularities
displayed by actual gases (under conditions approaching ideal conditions).
Hence this functional law is more than just a summary of its actually occurring instances.

P =
kT
V
Any individual Type A nucleus may not have decayed after 100,000 years (the law says that on
average, half the population will have decayed after 100,000 years, which will be true if some decay
before and some after 100,000 years). But we still want to say that all Type A nuclei are governed by
the probabilistic law.
2
To say that an individual Type A nucleus has a probability of 1/2 of decaying after 100,000 years is to say that
the nucleus has an intrinsic indeterministic property -- there is no determinant fact of the matter about the
nucleus that we could ever know that would allow us to predict if it will have decayed after 100,000. All we can
say now (or at any time) is that the nucleus has a probability of 1/2 of decaying after 100,000 years.
Claim: Under the ontic view, there can be instances of probabilistic laws, these instances being
individuals with intrinsic indeterministic properties.
But: This assumes a particular interpretation of probability.
Espistemic Interpretation of Probability = Probabilities are a measure of our ignorance.
To say that an individual Type A nucleus has a probability of 1/2 of decaying after 100,000 years is to say that,
at the present, we dont have enough information about it to predict with certainty whether it will decay after
100,000 years. But there is a determinant fact of the matter whether it will or will not decay. All we can do
with our present state of knowledge is to predict what will happen to a large ensemble of such nuclei: we can
predict that, on the average, 1/2 will have decayed after 100,000 years.
Ontic Interpretation of Probability = Probabilities refer to probabilitistic properties.
So: An STRer can respond to the probabilistic law objection by adopting an ontic view of
probabilities.
ASIDE: But there do seem to be epistemic probabilities in physics. Probabilities in Statistical Mechanics are
epistemic, for instance. The STRer will thus have to claim that there are no laws in Statistical Mechanics!
The real laws are those of Newtonian Mechanics to which Statistical Mechanics reduces. Note, also, that in
Quantum Mechanics, probabilities are not given for ensembles, but for individual states. The view of
probabilities thus is the ontic, as opposed to the epistemic, view (at least for most interpretations of QM).
3
II. Best System Account (BSA): Modified Regularity Account
BSA: A regularity is a law if and only if it appears as a theorem or axiom in that true
deductive system which achieves a best combination of simplicity and strength.
least number
of axioms
info-content
Motivations
(1) Laws systematize facts; they dont just report them.
(2) How do you systematize facts? Construct a theory in which they can be embedded. Hence,
to get at laws, write down the simpliest and strongest theory that accounts for the
phenomena. The laws will be the basic principles (theorems or axioms) of this theory.
theory
Kepler's Laws
Law #1: The orbits of the planets are in the forms of ellipses.
Law #2: The orbits of the planets sweep out equal areas in equal time intervals.
Law #3: The ratio D
3
/T
2
is constant for all planets (D = ave. distance from sun, T = period).
4
Advantages
(1) Allows a distinction between accidental generalizations and laws. Accidental
generalizations will not figure into the simpliest and strongest systematization of the facts.
(2) Allows a distinction between basic laws and derived laws.
Ex. Keplers 3 Laws of planetary motion are derived from Newtons law of gravity. The BSA will claim that
under the best system, Newtons law of gravity will appear at a lower level (maybe as one of the axioms
of the system), while Keplers laws will appear at a higher level. Note that the Simple Regularity Theory
cannot make this distinction. Under SRT a law is simply a regularity, and no provision is given for
distinguishing more fundamental regularities from less-fundamental ones.
(3) The account of laws (most) scientists take for granted. Ask a physicist what a law of nature
is and she will probably recite a law (like Newtons Law of Gravity) that appears as the
foundation of a given theory in physics (Newtons Theory of Gravity). But: What about
fields like biology or psychology?
(3) Accounts for the link between laws and counterfactuals/modal statements. What we take to
be true counterfactuals is based on what we know about the world. And what we know
about the world is given to us by our best theories. So both laws and counterfactuals have
their bases in our best theories.
5
Problems
(1) Refers to standards of simplicity and strength. Shouldnt laws be independent of such
standards? Laws are supposed to be objective features of nature.
(2) What if there is more than one best system?
One response: Insist that nature is simple and uniform and will yield to our probing in a
nice well-ordered manner (uniformity of nature assumption). This may be appealing in the
more abstract sciences (theoretical physics, for instance); but a biologist will quickly tell you
how complex and redundant some aspects of nature are (biological systems for the most part
usually make use of what they are given and usually in highly redundant, complex ways).
One response: This assumes a pre-theoretic intuition concerning what a law is. The BSA
says, Let our theories decide for us what the laws of nature are. Whatever they decide will be
a law by definition, and not an accidental generalization. (What Bird might be getting at is
the worry that we will never have available to us an ideal best system of the world, and thus
may never know what the true laws of nature really are. But this is really a different issue
than the one the BSA is intended to address. All the BSA is giving us is a definition of what
a law of nature is. Theres nothing about a law of nature that requires that, for something to
be a law, we have to have epistemic access to it.)
One response: Then there is no fact of the matter as to what the true laws of nature are.
(3) Birds objection: It may be possible for the best system to have accidental generalizations as
its axioms and theorems.
III. Necessitarian Account
It is a law that Fs are Gs if and only if Fness necessitates Gness.
universal universal relation of necessitation
between universals
Assumes the following metaphysical distinctions:
universals
particular
Fness, Gness
necessitation
The necessitarian account says: A law is a relation of necessitation between two universals.
Example:
Let Fness = being in free fall near the surface of the earth
Gness = having an acceleration of 9.8m/s
2

individual = a piece of chalk that I just dropped

The piece of chalk (the individual) has both the properties of Fness and Gness.
6
individual
general property
property of general property
Or, according to the diagram, a law is a particular property that holds of two general properties
that hold of an individual (i.e., a law is a particular relation (a 2-place property) that holds
between two general properties of an individual). This particular relation is called necessitation.
ASIDE: Think of necessitation like youd think of the 1st-order relation taller than. Two individual
objects can stand in the relation taller than (the Empire State Building and a grass hut, for
instance). Similarly, two general properties of an object can stand in the relation of necessitation.
The necessitarian says: These properties also stand in the relation of necessitation to each
other: Fness necessitates Gness. This is a manifestation of Newtons Law of Gravity. The law is
just the relation of necessitation between being in free fall near the surface of the earth and
having an acceleration of 9.8m/s
2
.
Advantages
(1) Allows the distinction between accidental regularities and laws. All instantiations of laws are
regularities; but a regularity need not be an instantiation of a law. (Only those regularities
that are characterized by the necessitation relation are instantiations of laws.)
(3) Provides a basis for induction. Bird claims that the SRT and the BSA do not. Well come
back to the problem of induction later in the course. The idea is that the necessitation
relation provides the metaphysical connectivity that justifies causal inductive inferences. If
we dont allow for some form of connectivity, we cant justify such inferences. And the SRT
and the BSA dont give us this. Big Problem: What justifies the claim that such connectivity
exists? In other words, How can we ever be sure when the relation of necessitation is present
and when it is not? This is probably the central problem facing this accout...
(2) Provides an explanation for lawlike regularities. The intuition is that the necessitation
relation explains the presence of those regularities that are instantiations of laws: Why do
pieces of chalk in free fall near the earths surface fall with an acceleration of 9.8m/s
2
?
Because those pieces of chalk also have the 2nd-order property of necessitation. This
additional property explains this regularity.
7
ASIDE: Bird claims that the SRT and BSA cannot explain regularities in the same way. The SRT
cannot: you cant explain a regularity by refering back to the regularity. But arguably the BSA does
provide an explanation for lawlike regularities; in particular, it provides what is called a unifying
explanation (which well look at in more detail later): it explains lawlike regularities by demonstrating
how they are embedded in a unifying systematization of the facts. In any event, we cant really view
(2) as an advantage until we have a grasp of what the notion of explanation is. (Recall, we are
looking at different accounts of laws in order to get a grasp of what an explanation is.)
Problems
(1) What is the necessitation relation? Any description of it fails to distinguish it from the BSA
acocunt.
To see the problem that (2) presents, consider two worlds W
1
, W
2
that agree on all
observable regularities. The necessitarian must claim that they could still disagree on what
the laws of nature are. (Note that the BSAer must claim that W
1
and W
2
must agree on
what the laws of nature are (as long as their standards of simplicity and strength are the
same).)
(2) How is the necessitation relation known? It has to go beyond observable regularities (if we
stop at observable regularities, we have the BSA). It is an in-principle unobservable property
(unlike, say, the property being in free fall near the surface of the earth).
The question for the necessitarian is, if the laws of W
1
and W
2
are different, how could we
ever come to know this? If our only access to laws is through empirical observations of
regularities in nature, then we could never come to know this.
One response: We have a built-in intuition that lets us directly grasp what the laws
of nature are.
Reply: Dont fortune-tellers make similar claims?
8
04. Explanation. Part 2.
Topics:
I. Problems with DN
II. DN Variants
General Characteristics
(a) Argument Thesis: DN explanations are arguments.
(b) Nomic Expectability Thesis: DN explanations demonstrate how the explanandum is
nomically expected; i.e., how it follows necessarily from a law.
(c) Explanation/Prediction Symmetry Thesis: Any DN explanation of a particular fact
could have been used to predict the fact if the explanans had been available prior to
the facts occurrance. So:
(i) Every DN explanation is a potential prediction.
(ii) Every prediction is a potential DN explanation.
Note that (c) seems problematic right off
the bat: Predictions can be based on mere
correlations, whereas DN explanations
cannot: they have to based on real laws
What this entails:
(A) Anything that satisfies the Adequacy Conditions is a scientific explanation.
(Conditions 1-4 are sufficient conditions for something to count as a scientific explanation.)
(B) All scientific explanations satisfy Adequacy Conditions.
(Conditions 1-4 are necessary conditions for something to count as a scientific explanation.)
DN Adequacy Conditions
(1) Valid deductive argument.
(2) Explanans must contain law(s).
(3) Explanans must have empirical content.
(4) Explanans must be true.
The general opinion is that this Claim is false. There are purported counterexamples to both
Claim (A) and Claim (B). Lets look at some of them.
Recall:
DN explanations that satisfy these conditions have 3 general characteristics:
Claim: Scientific explanations are DN explanations.
1
Counterexamples to Claim (A)
Consider the following DN explanation of the length of the shadow of a flagpole:
(1) Flagpole and Shadow
Explains the length l of the flagpoles shadow by showing how l follows from a law and the
conditions that make the law applicable. Satisfies Adequacy Conditions 1-4, so it's DN.
Explains the height h of the flagpole by showing how h follows from a law and the conditions
that make the law applicable. Satisfies Conditions 1-4, so it's a DN explanation.
I. Problems with DN
(1) Light propagates rectilinearly.
(2) Sun is at certain elevation e.
(3) <statements describing ambient atmospheric conditions>
(4) Flagpole has height h.
! Shadow has length l.
law
conditions
observed fact
Examples that attempt to demonstrate that the Adequacy Conditions are not sufficent for something to count as a
scientific explanation. They are examples of things that satisfy the Adequacy Conditions but should not be considered
legitimate scientific explanations.
But: We can also construct the following DN explanation that also satsifies Conditions 1-4:
(1) Light propagates rectilinearly.
(2) Sun is at certain elevation e.
(3) <statements describing ambient atmospheric conditions>
(4) Shadow has length l.
! Flagpole has height h.
law
conditions
observed fact
But: Is it a legitimate explanation? It explains a cause (the height h of the flagpole) by means of
its effect (the length l of the shadow). Effects are normally explained in terms of their
causes, and not vice-versa.
ASIDE: Explaining why a flagpole has a certain height by refering to the length of its shadow is analogous to explaining
why you approached an automatic sliding door (in a grocery store, say) by refering to the door sliding open. Your approach
to the door caused it to slide open; its sliding open is the effect of your approaching it. You dont normally say Why did I
approach the door? Because it slid open. You do normally say Why did the door slide open? Because I approached it.
Moral: The DN model does not account for causal factors in explanations.
2
(2) Man & Birth-Control Pill
Consider the following DN explanation of why the man Wes failed to become pregnant during the
past year:
(1) If a human takes pill Y consistently over the course of a year, then that
human will fail to become pregnant during that year.
(2) Wes is a human and Wes took pill Y consistently over the past year.
! Wes failed to become pregnant during the past year.
Satisfies Conditions 1-4. So it's a DN explanation.
Counterexamples to Claim (B)
Ink Stain
Consider the following explanation for why there is an ink stain in the carpet in Wes office:
Because Wes knocked over a bottle of ink.
This explains the ink stain by referring to its immediate cause. Note that it is not a DN
explanation (it doesnt satisfy Adequacy Conditions 1-4).
But: Is it a legitimate scientific explanation? Taking birth-control pills is irrelevant to whether
or not a man becomes pregnant. Hence premise (2) is irrelevant to the fact (that Wes did
not become pregnant) being explained.
Moral: In scientific explanations, the explanans should be relevant to the explanandum.
ASIDE: Note that the law in (1) might be questioned. If it does not count as a law, then the
example does not count as a DN explanation. However, under the regularity account, it is a legitimate
law (it simply states a regularity). And under the BSA, it might be considered a derived law; a more
fundamental law being something like: Chemical X in pill Y inhibits the production of human egg cells.
Examples that attempt to demonstrate that The Adequacy Conditions are not necessary for something to be a
legitimate scientific explanation. They are examples of legitimate scientific explanatoins that do not satisfy the
Adequacy Conditions.
Claim: Causal explanations that do not appeal to laws are legitimate scientific explanations.
ASIDE: We could construct a detailed, complicated DN explanation of the ink stain. It might appeal to laws in
Newtonian dynamics that would describe Wes's elbow knocking over the ink bottle; and to Newtons law of gravity to
describe the bottle falling to the carpet; and to laws of diffusion in chemistry to describe how the ink spread into the
carpet; and possibly even to laws in chaos theory to describe the pattern of the ink stain. (Moreover, it would also have
to appeal to laws in psychology and/or neuro-physiology to describe the mental processes that caused Wes to move his
elbow in a certain way.) The assumption underlying the causal explanation approach is, Do scientists really always
appeal to such complex, convoluted DN explanations? Or are they willing to use much more simple causal explanations?
3
II. DN Variants
(1) DS - Deductive Statistical: Explains a statistical regularity by subsuming it under a
statistical law.
(2) IS - Inductive Statistical: Explains a particular fact by subsuming it under a statistical
law.
Important distinction:
Universal generalization: All Fs are Gs
Statistical generalilzation: X% of Fs are Gs
The laws that appear in DN explanations are in the form of universal generalizations.
The laws that appear in DS and IS explanations are in the form of statistical generalizations.
explanandum
law
universal
statistical
DN
DS
DN
IS
particular
fact
general
regularity
Problem of Irrelevant Conjunctions: A general regularity (law) can be derived from the
conjunction of that law with any other (true, empirical) law. This conjunction is itself a law and
can be used in the law premise of DN-type explanations. But this conjunction doesnt always
explain the original law/general relgularity.
Ex: Keplers Laws K can be derived from the conjunction (K & B), where B is Boyles Law (which is a true,
empirical law). But the conjunction (K & B) doesnt explain Keplers Laws. In particular, the following DN
explanation of Keplers Laws isnt a legitimate scientific explanation (note that it does satisfy all 4 DN conditions):
(1) (K & B)
! K
law
DN-type explanations of general regularities (DN and DS) face...
general regularity
ASIDE: Note that DN-type explanations of particular facts (DN and IS) dont face this
irrelevant conjunction problem. If you conjoin a particular fact to a law, the result is not a
law that could be used in the Law Premise of a DN-type explanation. (Intuitively, the
particular fact refers to a specific time and place, so the conjunction would, too; hence it would
not count as a law.) Note, also, that this Problem may only be a problem for the regularity
account of laws, underwhich it seems legitimate to consider (K & B) a law if K and B are laws
separately. Under the BSA, (K & B) will probably not be considered a legitimate law.
4
Given that DN-type explanations of general regularities face serious problems, we can still ask, do
DN-type explanations of particular facts work? Weve seen some problems with the original DN
explanation of particular facts. Lets now look at the IS model.
Inductive Statistical (IS) model of explanation
IS Adequacy Conditions
(1) Strong inductive arugment.
(2) Explanans must contain statistical law(s).
(3) Explanans must have empirical content.
(4) Explanans must be true.
(5) Explanans must satisfy the Requirement of Maximal Specificity (RMS).
ASIDE: Why is RMS needed in addition to Condition 1? Consider the following two strong inductive
arguments:
(1) Almost all cases of strep-throat clear up quickly after penicillin treatment.
(2) Jane Jones had strep-throat.
(3) Jane Jones recieved penicillin.
! Jane Jones recovered quickly.
(A)
(1) Almost all cases of penicillin-resistent strep-throat do not clear up quickly after
penicillin treatment.
(2) Jane Jones had penicillin-resistent strep-throat.
(3) Jane Jones recieved penicillin.
! Jane Jones did not recover quickly.
(B)
Both (A) and (B) are strong inductive arguments: Their premises give high probabilities to their conclusions.
But their conclusions are contradictory. If we only knew the premises of both, which conclusion are we
warranted in believing? Intuitively, we are more warranted in believing the conclusion of (B), since its premises
contain information that is more relevant to its conclusion than do the premises in (A) to its conclusion. In
other words, while (A) and (B) both satisfy (1) above, (B) also satisfies RMS, whereas (A) does not.
A strong inductive argument is one in which the premises give a high probability to the conclusion.
RMS: All relevant information must be present in the explanans that would
have an effect on the explanandum.
5
The IS conditions can be combined with the DN conditions into the following:
DN/IS Adequacy Conditions
(1) Must be either a valid-deductive argument, or a strong inductive argument.
(2) Explanans must contain a law(s) (universal or statistical).
(3) Explanans must have empirical content.
(4) Explanans must be true.
(5) Explanans must satisfy the Requirement of Maximal Specificity (RMS).
Modified Claim: The DN/IS Adequacy Conditions are necessary and sufficient conditions
for scientific explanations of particular facts.
In other words: All scientific explanations, and only scientific explanations, of particular facts
are either DN explanations or IS explanations.
Weve seen that the DN leg of this modified claim is problematic. Lets look at the IS leg. The
general opinion is that it fails, too.
Note: Both IS and DN are characterized by the Argument Thesis, the Nomic Expectibility Thesis,
and the Explanation/Prediction Symmetry Thesis. (In the IS model the argument is
inductive instead of deductive, and the law is statistical instead of universal.)
ASIDE: DN explanations automatically satisfy RMS: Characteristics
(3) and (4) of valid-deductive arguments indicate that the premises of a
valid-deductive argument are maximally relevant to the conclusion.
6
Counterexamples to IS
Examples that attempt to demonstrate that the IS conditions are neither necessary nor sufficient for scientific
explanations of particular facts.
(1) Vitamin C and the Common Cold
Consider the following IS explanation for why Wes recovered quickly from his cold.
(1) 85% of people with common colds who take massive doses of Vitamin C recover quickly.
(2) Wes is a person with a cold who took massive doses of Vitamin C.
[0.85]
! Wes recovered quickly.
Satisfies IS Conditions 1-5. So it's an IS explanation.
Moral: The argument thesis plus RMS still allows irrelevant information to be included in
the premises of an IS explanation.
(2) Syphilis and Paresis
This example is supposed to show that there are legitimate scientific explanations of particular facts that are not IS
explanations (or DN explanations).
(1) 25% of all victims of untreated latent syphilis develop paresis.
(2) The only way to get paresis is if you had untreated latent syphilis.
(3) Smith had untreated latent syphilis.
[.25]
! Smith developed paresis.
Legitimate scientific explanation of why Smith developed paresis.
But: Is it a legitimate scientific explanation of why Wes recovered quickly? Fails to take into
consideration that people tend to recover quickly from colds regardless of whether or not
they take massive doses of Vitamin C. Question left unanswered: Is Vitamin C
consumption statistically relevant to cold recovery? If it isnt, then Premise (2) is irrelevant
to the conclusion.
But It's not a strong inductive argument: The premises give a very low probability to the
conclusion.
ASIDE: This is not only a criticism of the Argument Thesis, but also of the Nomic Expectibility Thesis. The
example does not show how the explanandum is expected from a statistical law. It shows the opposite: that
the explanandum is very unlikely, given the statistical law. Nevertheless, the example is a legitimate scientific
explanation. Note that one could uphold the Nomic Expectability Thesis by claiming that the explanation is
incomplete; that the statistical law in (1) is reducible to an as yet undiscovered universal law, and this law
would show how the explanandum is expected nomically. Note that to claim this is the case for all such
counterexamples is essentially to claim that all statistical laws are ultimately reducible to universal laws (hence
all IS explanations are just incomplete DN explanations). This entails that there are no fundamentally
probabilistic facts. This is a bit hasty: this is denied under the accepted interpretation of quantum mechanics.
7
05. Explanation. Part 3: Alternatives to DN.
Topics:
I. Unification Account
II. Causal Account
III. Pragmatics Account
I. Unification Account
A scientific explanation of a fact (particular or general) is a demonstration of how the fact can be
derived from a unifying set of argument patterns.
(1) Scope: The greater the scope of T, the greater the number of conclusions that can be drawn
from T.
(2) Simplicity: The greater the simplicity of T, the smaller the number of argument patterns in T.
(3) Stringency: The greater the stringency of T, the smaller the range of applicability of T.
Set of argument patterns = basic principles (axioms, theorems, etc) that (may) underlie a
theory.
Ex. General relativity can be thought of as a unifying set of argument patterns that can be used to describe a
certain class of phenomena. Arguably, the set has great scope, great simplicity, and great stringency (it only
applies to certain phenomena; namely, phenomena that experience the gravitational force; and it prescribes the
behavior of such phenomenon in very restricted ways). Astrology, on the other hand, is not stringent: you can
apply its descriptions to almost any phenomenon you experience. (Any event you experience in the course of a
day is bound to have been predicted by your daily horoscope, given a flexible enough interpretation.)
General Idea: To scientifically explain a fact, you have to demonstrate how it can be embedded
in a unifying theory. This explains the fact by showing how it is related to other facts.
Four Characteristics:
(1) Unification explanations are derivations.
Friedman (1974) Explanation and Scientific Understanding
Kitcher (1981) Explanatory Unification
1
Unifying power: a set of argument patterns T is unifying if it scores high on the following
properties:
A derivation = A sequence of justified steps; each step being explicitly shown to follow from
the preceding ones.
Note: Contrast with DN-type explanations, which are arguments (recall, arguments are
sets of setences with one being a claim and the others reasons given for the claim).
In an argument, you dont have to explicitly show how each sentence follows directly from the last.
This allows irrelevant premises to crop up; in a derivation, there can be no step which is not
relevant to the other steps.
(2) The unification account is committed to an Expectibility Thesis: A unifying
explanation must show how the explanandum is to be expected from the explanans.
Problems with the Unification Account
(1) Problem of subjective standards: How are we to judge which explanations are more
unifying than others?
(3) Unifying explanations are not necessarily reductionistic. One might think that to
provide a unifying explanation of a fact is to show how that fact can be reduced to the
fundamental facts that underlie the ultimate grand-unifying theory of everything. In
particular, to provide a unifying explanation of a biological fact, you have to show how it can
be reduced to facts in chemistry, say, or physics.
(4) The unification account is global: A unifying explanation embeds a local fact in a
larger, global theory.
Note: This is not necessarily nomic expectibility, as with DN. In comparison to DN, one
might say that unification replaces law with unifying systematization (i.e., theory").
But note the other main difference with DN given in Characteristic 1.
But: The unification account is compatible with the possibility that biology, say, ultimately
can never be reduced to physics. If this is so, you can still construct unifying explanations of
biological facts; theyll just refer to unifying theories in biology and make no reference to
physics.
(same problem that the BSA account of laws faces)
ASIDE: Our text claims that the unification account also suffers from the Problem of
Irrelevant Conjunctions (just as it claims the general regularity DN & DS accounts do). The
claim here is that (K & B) can be thought of as a set of argument patterns (K = Keplers Laws, B
= Boyles Law); and K can be derived from (K & B). But K isnt explained by (K & B). This is
aimed at Friedmans original account of unification. The account given above is due to Kitcher
and it has a ready response to this criticism. According to it, one can say that K can also be
derived from N (Newtons theory of gravity), and N is a better explanation of K than (K & B)
because N is more unifying than (K & B). One might claim that (K & B) in fact isnt a unifying
explanation of K at all: (K & B) has poor simplicity. But this response in turn faces the problem
of subjective standards: Can we say that N is objectively more unifying than (K & B)?
2
(2) Problem of probabilistic explanations: Some legitimate explanations give a low
probability to their explananda, hence their explananda are not expected from their explanans
(recall the syphilis and paresis example). Since the unification account is commited to an
Expectability Thesis, it faces this problem.
To see how this folds out, consider the following distinction:
Two Types of Probabilistic Explanations:
(a) reducible: Given enough information, these reduce to explanations in which the
explanandum can be logically deduced from the explanans.
(b) irreducible: The explanandum cannot be logically deduced from the explanans,
regardless of how much further information is provided.
Deductive chauvinism claims: All probabilistic explanations can be reduced to
deductive explanations. There are no legitimate irreducible probabilistic explanations.
Ex1: Suppose an electron beam impinges on a potential barrier (think of a beam of electrons focused on a wall).
The Schrdinger equation in quantum mechanics gives the probability for each electron in the beam to be
reflected or to tunnel through. Suppose a given electron, e
1
, tunnels through the barrier. We can ask:
Why did e
1
tunnel through the barrier?
We cannot construct a derivation with the conclusion e
1
tunneled through the barrier. All the Schrdinger
equation gives us is the probability that e
1
will tunnel through (say its 0.80). The Schrdinger equation does not
predict with certainty whether e
1
will or will not tunnel through.
What this means: We cannot construct a unifying explanation of why e
1
tunneled through.
e
1
prob of 0.2 of
being reflected
prob of 0.8 of
tunneling through
barrier
One response: Deductive Chauvinism -- Claim that there are no legitimate explanations
of inherantly probabilistic facts.
In other words: While there may be inherantly probabilistic events, Deductive Chauvinism claims such events
cannot be explained (to the extent that inherantly probabilistic events cannot be predicted with certainty).
3
But: The unificationist who is also a Deductive Chauvinist will respond that this is fine, since there are no
legitimate explanations of inherantlyprobabilisitic events, and the actual event of e
1
tunneling through the
barrier is just such an inherantly probabilistic event.
ASIDE: Such a unificationist can explain why e
1
had an 80% chance of tunneling through the
barrier (instead of, say, a 50% chance). This is entailed by the Schrdinger equation. But
such a unificationist, again, says there is no explanation for why e
1
did in fact tunnel through.
So: A unificationist can claim that there are no explanations of inherantly probabilistic events.
A physicist might be satisfied with the claim that there is no explanation for why a particular
electron tunnelled through a barrier.
But: Does this work for explanations in the social sciences?
Ex2: Suppose an anthropologist studying the Yanomami indians of Brazil seeks an explanation
of why the Yanomami attacked village A. The anthropologist has determined the following:
(a) The Yanomami tend to attack when resources are scarce;
(b) The Yanommami tend to attack when the military advantage is theirs; and
(c) The Yanomami tend to attack when their social influence is threatened.
Note: There are no factors that determine with certainty when the Yanomami will attack.
So: The event of such an attack is an inherantly probabilistic event.
So: A unificationist who is a deductive chauvinist must claim that there is no explanation for why the Yanomami
did in fact attack village A.
But: The anthropologist certainly will not be satisfied with this and will indeed claim that some form of
explanation for the attack can be constructed.
Moral: Deductive chauvinism is a high price to pay as a response to the problem of
probabilistic explanations. But if the unificationist does not adopt it, she is faced with the
same sorts of problems that afflict the IS account.
II. Causal Account
To explain an event is to provide information about what caused it.
Two Characteristics
(1) The causal account is local.
(2) Basic causal account claim: Causal structure underlies laws and theories. This is
what gives them explanatory power. So all DN-type and unification explanations are causal
explanations, but not all causal explanations can be viewed as DN-type or unification
explanations.
Problems with the Causal Account
(1) Problem of the nature of causality: How are legitimate causal explanations
distinguished from illegitimate explanations based on mere statistical correlations?
Salmon (1984) Scientific Explanation and the Causal Structure of the World
Lewis (1986) Causal Explanation
ASIDE: The above example comes from: Steel, D. (1998) Warfare and Western Manufactures:
A Case Study of Explanation in Anthropology, Philosophy of Science 65, pp. 649-671.
4
(2) Problem of purely theoretical explanations: Some theoretical explanations do not
explicitly refer to causes.
Ex: Why cant you fit a left-handed glove on your right hand?
Theoretical explanation: Due to the topological properties of the left-handed glove and the right hand.
Causal explanation: Due to the resistance of the inner surface of the left-handed glove with your right hand.
Claim: Purely theoretical explanations count as legitimate scientific explanations. So the
causal account cannot be a complete account of scientific explanation.
(3) Problem of irreducible probabilistic explanations: What caused the Yanomami to
attack village A? What caused e
1
to tunnel? To provide causal explanations of irreducibly
probabilistic events, we need a theory of probabilistic causation (and a theory of simple
causation is hard to come by).
III. Pragmatics Account
Motivation: Explanations are context-sensitive.
van Fraassen (1980) The Scientific Image
5
Pragmatic Features of Explanations
(1) Clarification of explanation-seeking question
What is the explanandum for which an explanation is being sought?
Ex1: In the glove example, is the relevant question What prevents me right now from putting this left-handed
glove on my right-hand? or is it In general what prevents objects with opposite handedness from matching
up? The causal explanation is a more appropriate answer to the first question; the theoretical explanation is
more appropriate to the second question.
Ex2: The question Why did Adam eat the apple? will be responded to in different ways, depending on how it is
interpreted:
(a) Why did Adam eat the apple? (As opposed to a grape or an orange.)
(b) Why did Adam eat the apple? (As opposed to Eve or the snake.)
(c) Why did Adam eat the apple? (As opposed to throwing it at the snake, etc.)
In the glove example, you would offer the causal explanation to someone unfamiliar with topology, whereas you
might offer the theoretical explanation to a mathematician.
(2) Knowledge-context of the explanation-seeker
The information content of the explanans of an adequate explanation will depend on the
knowledge of the questioner.
Suppose a congressional committee is seeking an explanation for a plane crash in order to modify existing safety
regulations. It will be more interested in explanations that refer to the procedures the crew went through (or
failed to go through), as opposed to explanations that refer to principles in Newtonian dynamics.
Relevant Distinction
(a) Ideal Explanatory Text: a complete description (nomic, systematic, causal) of the
explanandum. The full, gory details. (Non-pragmatic aspect.)
Claim: Scientific explanations are almost always requests for explanation
information, and not for the ideal explanatory text.
(b) Explanation Information: Accounts of particular aspects of the ideal explanatory text.
(Pragmatic aspect.)
(3) Interests of the explanation-seeker
The content of the explanans of an adeuate explanation will depend on the purpose for which
the explanation is being sought.
6
06. Confirmation
Relation between theory and evidence
Fundamental question: How does evidence contribute to the credibility of a theory?
General Considerations (epistemological)
Notion of justification (reasons for belief)
(1) What justifies a scientific claim?
(2) Can such claims be justified? (Hume)
(3) What is the sense of justification involved?
Particular Considerations
(1) What is a hypothesis H?
(2) What is evidence E?
(3) Can a general relation between E and H be defined that describes how E supports/lends
credence to H?
E confirms H means __________.
Possible Claims
(1) E confirms H just when E is entailed by H. Symbolically: H ! E
Subject matter of confirmation theory: Analysis and assessement of such claims.
Well come back to these
considerations later
I. Introduction
Topics:
I. Introduction
II. Hypothetico-Deductive Method
III. Instance Confirmation
(2) Evidence can never confirm single, isolated Hs; evidence can only confirm theories-as-a-
whole/belief systems/conceptual schemes. (strong holism)
(3) E confirms H only relative to background knowledge K. (weak holism)
(4) Some types of evidence confirm a given H more directly than other types.
etc...
1
Useful distinctions
2 Parts to Language in which claims about objects are made
1. Observational vocabulary - terms in language that refer to objects
of types (1) and (2).
2. Theoretical vocabulary - terms in language that refer to objects of
type (3).
2 Types of Inference
I. Inferences from evidence to observational claims.
II. Inferences from evidence to theoretical claims.
2 Types of Claims about Objects
(i) Observational Claims
All swans are white.
Electron A has momentum p.
(ii) Theoretical Claims
Quarks come in 3 flavors.
3 Types of Objects
(1) Directly observable objects
swan
(2) Indirectly observable objects
dinosaur
electron
(3) In-principle unobservable objects
quark
Aside: An anti-realist answers yes to (A) and no to (B).
A realist answers yes to both (A) and (B).
Questions for Confirmation Theory
(A) Are inferences of Type I justified? If so, how?
(B) Are inferences of Type II justified? If so, how?
2
Step 2: Conduct an experiment to see if our prediction O holds true. Take a sample of gas at the
stated initial volume and pressure (in a compression chamber, say), and then vary the
pressure to its final value (by either compressing the gas or allowing it to expand), while
keeping its temperature constant. Now measure its final volume and see if it is identical
(within the appropriate error analysis) of the predicted value O.
II. Hypothetico-Deductive (HD) Method of Confirmation
Step 3: Suppose O is true. This constitutes a positive outcome, and HD requires us to say O
confirms H.
3 Steps to HD Method:
1. Given hypothesis H, deduce an observational consequence O (i.e., an observational claim).
2. Test O (i.e., check to see if it is true or false).
3. If O is true, then H is confirmed. If O is false, then H is disconfirmed.
This is required to be a valid-deductive argument; i.e., if the premises are true (if H and I and A are
all true), then the conclusion must be true (the prediction must follow). The D in HD refers to
this deductive inference described in Step 1. (The H in HD refers to the H being tested.)
Form of deduction in Step 1:
H hypothesis
I initial conditions
A auxiliary hypotheses
! O observational prediction
hypothesis to be tested
required to deduce O from H
H : P
i
V
i
= P
f
V
f
(Boyles Law: PV = constant, at constant temperature)
I : values for V
i
= initial volume, P
i
= initial pressure, P
f
= final pressure
constant value for temperature
A : principles underlying instruments used to measure gas
principles underlying composition of gas chamber
etc...
! O : predicted value for final volume V
f
= P
i
V
i
/P
f
Example 1: HD Test of Boyles Law
Step 1:
3
General argument pattern of HD method (all 3 steps) for positive outcomes:
Not H is true. Note that even saying H is confirmed is
misleading. The truth of O lends support not just to H, but
to the conjunction (H & I & A). So the postive outcome
case of HD testing is ambiguous -- it doesnt really tell us
how the evidence O lends credibility to H by itself.
An inductive argument!
General argument pattern of HD method (all 3 steps) for negative outcomes:
4
Example 2: HD Test of Corpuscular Theory of Light
Step 3: O is false! A small bright spot will appear in the center of the disk ("Poissons Bright
Spot"). A negative outcome, so HD requires us to say O disconfirms H.
Step 1: H : Corpuscular Theory of Light (claim: light consists of corpuscles traveling in straight lines)
I : Cicular disk is lit from behind
A : Assumptions about composition of disk, medium between light source and disk, etc.
! O : Cicular disk lit from behind will cast uniform shadow
Step 2: Perform experiment!
But, again: What is disconfirmed? Not just H, but (H & I & A)!
It has the valid form
of modus tollens:
(p " q)
#q
! #p
Valid-Deductive
argument!
If (H & I & A) are all true, then O is true.
O is true.
H is confirmed.
If (H & I & A) are true, then O is true.
O is not true.
! (H & I & A) are not true.
O
I & A
H
Question: To what extent does a positive outcome O confirm H? Absolutely?
No!: We cant infer that H is true solely on the basis of a single correct prediction. But the
single correct prediction does lend some support to H.
Thus: While the inference in step 1 from H to O is deductive, the complete HD inference from O
back to H is inductive:
Claim (Falsificationism):
Science progresses deductively by falsification. We are not warranted in believing hypotheses that
have been HD-confirmed, because they are based on inductive inference. Rather, we are only
warranted in believing hypotheses that have withstood attempts to HD-disconfirm them. Such
reasoning is based on valid-deductive inferences.
Duhem-Quine Problem:
In the light of a disconfirming prediction O, which of (H & I & A) is to blame? Which gets
falsified by disconfirming evidence?
5
Popper (1959) The Logic of Scientific Discovery
Duhem-Quine Thesis
Hypotheses are never tested in isolation. They are always tested in the context of background
assumptions.
Popper: Believe only in Hs that have withstood severe tests.
But: What if the reason for a passing grade rests not in the H, but in the I and/or the A?
Duhem (1906) The Aim and Structure of Scientific Theories
Quine (1953) "Two Dogmas of Empiricism"
Note: This is a major problem for the HD method, which assumes that confirmation can be localized;
i.e., that hypotheses can, at least to some extent, be tested in isolation from other hypotheses.
ASIDE: Popper has been extremely influential, and typically if a scientist knows anything about the
philosophy of science, she will have heard of Popper and Falsificationism. Unfortunantly, many scientists
only know about Popper and falsificationism. But confirmation theory did not stop in the 1950's...
The HD notion of confirmation essentially is the following (the first possible claim from page 1 of
these lecture notes):
HD: E confirms H just when E is entailed by H (symbolically: H ! E).
The 3 Steps of the HD method just indicate how to apply this definition of confirmation in actual
situations.
Apart from the Duhem-Quine Problem, the HD method faces two other problems:
Two Further Problems with HD Method
(1) Problem of Alternative Hypotheses (Underdetermination of Theory by Evidence)
For any prediction O, there are any number of incompatible Hs that can be constructed that
all entail O. So if O is true, which H does it confirm?
Ex. H = Boyles Law: PV = ! or V(P) = !/P, for constant !.
H : V(P) = !/P
H' : V(P) = AP
2
+ BP + C
O : P
1
V
1
= !
P
1
P
2
V
2
P
V
1
V
Moral: The confirmation relation cant be simple entailment (H ! E).
(2) Problem of Statistical Hypotheses
If H is a statistical hypothesis, then no O can be deduced from it to test it. At most, we have:
If H is true, then O is highly-probable.
O is true.
Moral: The confirmation relation needs to account for probabilistic inductions.
O is entailed by both H and H' (and any other curve that intersects the point (P
1
, V
1
)).
So: If O is true, which H should we believe?
6
In fact, in the graph above, H and H' also both entail the prediction P
2
V
2
= !. Its not hard to
imagine a curve H'' that differs from H but agrees with H on any indefinite number of
predictions.
H is _______?
III. Instance Confirmation.
Recall: The task is to define a relation between E and H. In other words, to fill in the following
blank: E confirms H means ________.
Instance Confirmation fills it in thusly: E confirms H means E is an instance of H.
Ex: H = All ravens are black
An instance of H, according to Nicods Criterion, is a black raven.
So E confirms H just when E = a black raven.
What does E is an instance of H mean? One definition is known as Nicods Criterion:
Nicods Criterion: For a hypothesis H of the form All S are P, an
instance of H is an individual that is both an S and a P.
One initial problem with Nicods version of Instance Confirmation:
Its applicable only to hypotheses of the general form All S are P (i.e., universal hypotheses).
7
ASIDE: Technically, a hypothesis of the form All R are B is a universal affirmative sentence, which
can be written in 1st order predicate logic as (x)(Rx " Bx) (For all individuals x, if x is R then x is B).
A Nicod-instance of H then is an individual a that satisfies both the antecedent and the consequent of the
conditional being quantified over. Such an instance would be written as the singular sentence Ra & Ba.
ASIDE: Note that Instance Confirmation is different from HD Confirmation. HD bases confirmation on the
entailment relation (it says E confirms H just when H entails E). Instance Confirmation bases confirmation on
the notion of an instance of a hypothesis. Instances need not be entailed by their hypotheses. For example,
the Nicod-instance above of a black raven is technically not entailed by the statement All ravens are black.
For those familiar with predicate logic, the universal sentence (x)(Rx " Bx) (which says All ravens are black)
does not entail the singular sentence Ra & Ba (which says Individual a is a raven, and individual a is black).
Some hypotheses in science are existential (i.e., Quarks exist); or a combination of universal and
existential (i.e., Every human dies a finite number of years after his/her birth). Hempel
addressed this problem with a more complex definition of instance. In general, however, there
are three problems that all versions of instance confirmation face (Hempels as well as Nicods):
8
3 Problems with Instance Confirmation
(1) Ravens Paradox
Claim: Instance confirmation allows evidence that is seemingly irrelevant to a given H to
confirm H.
H
1
= All ravens are black
H
2
= All non-black things are non-ravens
E
1
= a black raven
E
2
= a piece of white chalk
Note: H
1
and H
2
are logically equivalent -- whenever H
1
is true, so is H
2
and vice-versa.
Hempel (1945) Studies in the Logic of Confirmation
Equivalence condition:
If E confirms H, and H is logically equivalent to H', then E confirms H'.
Then: E
2
(a piece of white chalk) confirms H
1
(all ravens are black)! "Armchair ornithology"!
Moral: Instance-confirmation does not take relevancy into consideration.
(E
1
seems more relevant to H
1
than does E
2
.)
(2) Goodmans Paradox
Claim: A piece of evidence E can be a positive instance of two incompatible hypotheses
(depending on how E is interpreted).
Goodman (1955) Fact, Fiction and Forecast
Let:
Now: Suppose we adopt the
Def. 1: Individual a is blite just when a is black if observed before 12/31/2006
and white if observed thereafter.
H
1
= All ravens are black
H
2
= All ravens are blite
Let:
Now: Suppose we observe a black raven on 11/31/2006. Does it instance-confirm H
1
or H
2
?
Response #1: Wait until 12/31/2006 to determine what the evidence is whether it instance-
confirms H
1
or H
2
.
Reply: Time-dependence of a property is a relative characterisitic. Blite can be viewed as time-
independent by defining black in the following explicitly time-dependent manner:
9
Reply: Consider H
3
= All ravens are blite
3000
. After 12/31/2006, the evidence fails to distinguish
H
1
from H
3
.
Def. 2: Individual a is blite
3000
just when a is black if observed before 12/31/3000
and white if observed thereafter.
In general: For any time t, we can define a property blite
t
that is identical to black up till t and
white thereafter.
So: There are an infinite number of hypotheses that are instance-confirmed by what appears now,
or at any time in the past or future, as a black raven. Given the purely formal notion of
evidence under Instance Confirmation, there is never a point at which we know for certain
what any given piece of evidence is and what hypothesis it instance-confirms.
Response #2: The property blite (and its variants) is illegitimate because it is explicitly time-
dependent.
Def. 3: Individual a is black just when a is blite if observed before 12/31/2006
and whack if observed thereafter.
Def. 4: Individual a is whack just when a is white if observed before 12/31/2006
and black if observed thereafter.
Note: The issue is what appears formally in the definition of a property. The claim is simply that whether or
not time-dependence appears in the definition is a purely formal matter. Of course what we really are concerned
with is whether or not the property we call black is time-independent: is it really "projectible" into the future?
ASIDE: Hume's Problem of Induction (as we shall see) claims: There is no justification for projecting past
uniformities into the future (how do we know that all ravens will continue to be black in the future?).
Goodman's New Problem of Induction claims: Even if we allow that there are projectable uniformities (ravens
do have some projectable property), there is no way to establish what they are (but is it black or blite?).
Moral: A purely syntactic (i.e., formal) description of E confirms H is
problematic. We need to take semantic (i.e., interpretational)
considerations into account in defining the confirmation relation.
(3) Theoretical Hypotheses
Claim: Instance confirmation cannot by itself explain how purely theoretical hypotheses can be
confirmed.
A purely theoretical hypothesis is a theoretical claim; i.e., a claim about an in-principle
unobservable object. An instance of such a claim will apparently be an in-principle unobservable
object. Hence we can never have evidence E for a purely theoretical hypothesis H if our E is
restricted to instances of H.
10
07. Probability and Confirmation
Motivation: Confirmation is an epistemic notion; cant be described in purely syntactic (formal
logical) terms. Fundamental Question: How does evidence contribute to the credibility of a
theory?
HD describes the confirmation relation as deductive entailment (syntactic relation).
Instance Confirmation describes the confirmation relation in terms of the formal notion of
an instance of a universal sentence (syntactic notion).
The Bayesian Account describes the confirmation relation in terms of the notions of
probability and degrees of belief.
Probability (Pr) function: A function that assigns a number between 0 and 1 to events.
Pr(B/A) means the probability of event B given event A
Topics:
I. Probability Axioms
II. Interpretations of Probability
III. Bayesian Confirmation Theory
A1: 0 ! Pr(B/A) ! 1. Every probability is a unique real number between 0 and 1.
A2: If A ! B ("A entails B"), then Pr(B/A) = 1.
A3: Special Addition Rule. If B and C are mutually exclusive (cant both occur
simultaneously), then Pr(B or C/A) = Pr(B/A) + Pr(C/A).
A4: General Multiplication Rule. Pr(B & C/A) = Pr(B/A)Pr(C/A & B)
I. Probability Axioms
1
(1) Negation Rule. Pr("B/A) = 1 # Pr(B/A)
Proof: Pr(B or "B/A) = 1 = Pr(B/A) + Pr("B/A)
(2) Rule of Total Probability. Pr(C/A) = Pr(B/A)Pr(C/A & B) + Pr("B/A)Pr(C/A & "B)
Proof: Pr(C/A) = Pr((B&C) or ("B&C)/A)
= Pr(B&C/A) + Pr("B&C/A) (A3)
= Pr(B/A)Pr(C/A&B) + Pr("B/A)Pr(C/A&"B) (A4)
C is logically equivalent
to (B & C) or (!B & C)
3 Theorems (Consequences of Axioms):
(3) Bayes' Theorem
Proof: Pr(B&C/A) = Pr(B/A)Pr(C/A&B) (A4)
Pr(C&B/A) = Pr(C/A)Pr(B/A&C) (A4)
2

Pr(B/A&C) =
Pr(B/A)Pr(C/A&B)
Pr(C/A)
, Pr(C/A) ! 0
=
Pr(B/A)Pr(C/A&B)
Pr(B/A)Pr(C/A&B) +Pr(" B/A)Pr(C/A& " B)
Example: Frisbee Factory. Part 1.
Machine #1 (M1): produces 800 frisbees/day, 1% of which are defective.
Machine #2 (M2): produces 200 frisbees/day, 2% of which are defective.
A = getting a frisbee produced on May Day "B = getting a frisbees produced by M2
B = getting a frisbee produced by M1 C = getting a defective frisbee
Note: Pr(C/A) = Pr(B/A) $ Pr(C/A & B) + Pr("B/A) $ Pr(C/A & "B)
$ $ + =
Prob of getting a
defective frisbee,
given it's a May
Day frisbee
Prob of getting
an M1 frisbee,
given it's a May
Day frisbee
Prob of getting a
defective frisbee,
given it's an M1
May Day frisbee
Prob of getting
an M2 frisbee,
given it's a May
Day frisbee
Prob of getting a
defective frisbee,
given it's an M2
May Day frisbee
= 0.8 $ 0.01 + 0.2 $ 0.02 = 0.012
Note: Bayes' Theorem lets us calculate "inverse" probabilities; i.e., probabilities of past events,
based on present events.
Question: What is the probability of getting a defective May Day frisbee? What is Pr(C/A)?
Example: Frisbee Factory. Part 2.
Question: Given a defective frisbee, what is the probability that it's an M1 frisbee? What is Pr(B/A & C)?

Pr(B/A &C) =
Pr(B/A)Pr(C/A & B)
Pr(B/A)Pr(C/A & B) + Pr(! B/A)Pr(C/A& ! B)
=
0.8 "0.01
0.8 "0.001 + 0.2"0.02
= 2/3
Prediction: Given a cause, what is the effect?
"Retrodiction": Given an effect, what is the cause?
3
II. Interpretations of Probability
1. Classical Interpretation Laplace 1814 A Philosophical Essay on Probabilities
Ex1. A = getting an even number on roll of standard die
favorable cases of A = {2, 4, 6}
equally possible cases of A's type = {1, 2, 3, 4, 5, 6}

Pr(A) =
3
6
Ex2. A = getting two heads on the flip of two non-biased coins
favorable cases of A = {HH}
possible cases of A's type = {HH, HT, TT}
equally possible cases of A's type = {HH, HT, TH, TT}
Why "equally" possible?

Pr(A) =
1
4
Pr(A) =
# of favorable cases of A
# of equally possible cases of A's type
Principle of Indifference (PI):
Two outcomes are equally possible if we have no reason to prefer one to the other.
Problem: PI may lead to violations of the Probability Axioms!
Ex3. Joe the Sloppy Bartender
Joe's sloppy mix for a 3:1 martini: Anywhere from 2:1 to 4:1
Two properties associated with Joe's sloppy martini:
(1) Ratio of gin to vermouth: From 2:1 to 4:1
(2) Proportion of vermouth: From 1/3 = 20/60 to 1/5 = 12/60
3:1 martini = 3 parts gin, 1 part vermouth
Now: Consider the two following outcomes for Property (1):
Next martini will have a ratio of gin to vermouth of 2:1 to 3:1.
Next martini will have a ratio of gin to vermouth of 3:1 to 4:1.
PI: Both outcomes are equally possible!
Now: Consider two outcomes for Property (2):
Next martini will have a proportion of vermouth of 20/60 to 16/60.
Next martini will have a proportion of vermouth of 16/60 to 12/60.
PI: Both outcomes are equally possible!
But! The PI has now given us contradictory predictions!
(a) According to the PI applied to Property (1), there's a 50% chance that Joe's next
martini will have a proportion of vermouth between 20/60 and 15/60 (=1/4).
(b) According to the PI applied to Property (2), there's a 50% chance that Joe's next
martini will have a proportion of vermouth between 20/60 and 16/60.
Two different
probabilties for the
same outcome:
Violation of
Axiom 1
2 non-linearly
related properties!
4
2. Frequency Interpretation
Pr(A) = the limit of the sequence of relative frequencies of A.
Ex1. A = getting a head on flip of non-biased coin
Actual flips: H T H T T T H H T T H T T T H ....
Relative frequencies of A: 1/1, 1/2, 2/3, 2/4, 2/5, 2/6, 3/7, ...
Pr(A) = limit of the sequence {1/1, 1/2, 2/3, 2/4, 2/5, 2/6, 3/7, ...} = 1/2.
Relative frequency of A =
# of actual occurances of A
# of actual occurances of events of A's type
What this means: As the members of the sequence go to infinity, if there is a member afterwhich all other
members are very close to some number L, then Pr(A) = L. If not, then Pr(A) is not defined. (And we
believe, in this case, that the limit exists and is 1/2.)
So: A sequence of relative frequencies = sequence of real numbers!
Def. A sequence of real numbers has a limit L just when there is a member of the sequence
afterwhich all other members are very close to L. If there is no such member, then the
sequence has no limit.
Two Problems
finite data
(behavior of finite
initial sequence)
limit
(behavior of
infinite sequence)
extrapolate
How? And what justifies this process?
ASIDE: Mathematically, the
problem is to determine a unique
formula that describes a given
sequence of relative frequencies.
Once we have such a formula,
determining if it has a limit, and
what it is, is not difficult. But
an infinite number of formulas
are compatible with any finite
initial sequence of real numbers!
Claim: Any limit L is compatible with any finite initial data sequence
(i) How are limiting frequencies determined?
5
3. Propensity Interpretation
Ex. Coin flips
Let

m
n
!
"
#
#
$
#
#
%
&
#
#
'
#
#
=
#heads
#total flips
!
"
#
#
$
#
#
%
&
#
#
'
#
#
be a finite sequence of relative frequencies.
Suppose limiting frequency = 1/2.
Then the sequence

m +a
n +b
!
"
#
#
$
#
#
%
&
#
#
'
#
#
also has limiting frequence = 1/2, where a ! b.
What this means:
(a) Add any sequence of b flips, a of which are heads, to {m/n} and limiting frequency remains unchanged.
(b) Chop off any sequence of b flips, a of which are heads, from {m+a/n +b} and limiting frequency remains
unchanged.
So:
Strong Claim: The observed relative frequencies in any finite sample are irrelevant to
whether a limiting frequency exists and what it is.
(ii) How are Single Case Probabilities Explained?
How does the Frequentist explain the probability of a single occurance?
Pr(A) = a measure of the causal tendancy (propensity) to produce A.
Ex. A = getting a 6 on roll of a standard die
Pr(A) = 1/2 = measure of the propensity in the die to produce 6 upon one roll
Ex. Outcome of 2008 election. What does it mean to say "The probability of a democrat winning is 85%"?
explains single case probability
Problem: How are "Inverse" Probabilities Explained?
Note: The Propensity Interpretation explains the probability of an outcome in terms of its cause.
But: Bayes' Theorem lets us calculate the probability of a cause in terms of its outcome/effect!
So: The Propensity Interpretation cannot explain all types of probabilities.
6
4. Personalist Interpretation
Pr(A) = the degree of belief in A of an ideal rational person whose set of beliefs
conform to the Probability Axioms.
Def 1. An incoherent set of beliefs is a set that does not conform to the Probability
Axioms.
Claim: A person is subject to a Dutch Book if and only if that person holds an incoherent
set of beliefs.
Def 2. A Dutch Book is a set of bets such that the subject loses no matter what the
outcome of the event wagered on.
Example
Let A = getting heads on next flip of coin.
"A = getting tails on next flip of coin.
Suppose: Joe's degree of belief in A = 2/3. % 2:1 odds for heads
Joe's degree of belief in "A = 2/3. % 2:1 odds for heads
"2:1 odds" means
"Joe is willing to
risk $2 to gain $1"
(A) Heads
Heads bet pays off: +$1
Tails bet fails: #$2
#$1 net
(B) Tails
Heads bet fails: #$2
Tails bet pays off: +$1
#$1 net
Only 2 possible outcomes:
Joe loses no matter what! Joe has succumbed to a Dutch Book!
Why Did Joe Succumb To a Dutch Book?
Because his set of beliefs did not conform to the Probability Axioms. In particular, he failed to abide by
the Negation Rule: Pr(A/K) + Pr("A/K) = 1, where K = "background knowledge".
Important Distinction
Propensity Interp: Probabilities measure the tendancy of a mechanism to produce a
single outcome.
Frequency Interp: Probabilities are properties of sequences of outcomes.
Both agree that probabilities are objective features of physical systems.
7
5. Logical Interpretation
Problem: Too "subjective"?
Any set of beliefs will be judged rational so long as it conforms to the Probability Axioms.
(Intuition: You can subscribe to really strange, WEIRD beliefs, so long as your belief system is
consistent!)
This means that the Personalist Interpretation does not explain or prescribe how the values of
probabilities are to be set/determined (a problem similar to one faced by the Frequency Interpretation).
Pr(A) = a weighted sum of all state descriptions of the universe in which A is true.
state description = a description of a possible way the universe could be ordered.
Example: Simple universe with 3 individuals and one property
Individuals: a, b, c
Property: F
How to assign weights? Carnap's Method:
(a) Group state descriptions according to similar structure.
(b) Assign equal weights to structure descriptions.
(c) Assign equal weights to all state descriptions with same structure.
State Description Weight Structural Description Weight
1. Fa & Fb & Fc 1/4 All F 1/4
2. Fa & Fb & "Fc 1/12
3. Fa & "Fb & Fc 1/12 2 F, 1 "F 1/4
4. "Fa & Fb & Fc 1/12
5. Fa & "Fb & "Fc 1/12
6. "Fa & Fb & "Fc 1/12 1 F, 2 "F 1/4
7. "Fa & "Fb & Fc 1/12
8. "Fa & "Fb & "Fc 1/4 No F 1/4
State Descriptions:
1. Fa & Fb & Fc 5. Fa & "Fb & "Fc
2. Fa & Fb & "Fc 6. "Fa & Fb & "Fc
3. Fa & "Fb & Fc 7. "Fa & "Fb & Fc
4. "Fa & Fb & Fc 8. "Fa & "Fb & "Fc
8
Now: Suppose A is Fa (i.e., "Individual a has property F").
Then: A is true in state descriptions 1, 2, 3, 5, which have weights 1/4, 1/12, 1/12, 1/12.
So: Pr(A) = 1/4 + 1/12 + 1/12 + 1/12 = 1/2
Note: Logical probabilities are not unique! They depend on how weights are assigned to state
descriptions. Call Carnap's method of assignment m*. (So m*(A) = 1/2.)
Def. c*(H/E) = m*(H & E)/m*(E)
where c*(H/E) means "The degree of confirmation of H given E"
Example:
Let: H = Fc, E = Fa
Then: m*(E) = 1/2
m*(H & E) = 1/4 + 1/12 = 1/3 (Fa & Fc are both true in state descriptions 1 and 3)
So: c*(H/E) = 2/3
Now: Suppose evidence was E' = Fa & Fb
Then: m*(E') = 1/4 + 1/12 = 1/3 (Fa & Fb are both true in state descriptions 1 and 2)
m*(H & E') = 1/4
So: c*(H/E') = 3/4
Thus: c*(H/E') > c*(H/E)!
What this means:
H is confirmed by E' to a greater extent than it is confirmed by E. This makes sense: The claim
that c has property F gains more support from the knowledge that a and b have property F than
it does from the knowledge that just a has property F.
But: Why pick Carnap's method m*?
Claim: It makes possible a confirmation function c* that "learns from experience".
Now: Suppose we had chosen a method of assigning weights that spread them equally over
state descriptions, instead of structure descriptions (Wittgenstein's proposal).
Claim: This particular method, call it m

, produces a confirmation function that does not

learn from experience!
9
Now:
If: H = Fc, E = Fa
Then: m

(E) = 4 $ 1/8 = 1/2

m

(H & E) = 2 $ 1/8 = 1/4 (Fa & Fc are both true in state descriptions 1 and 3)
So: c

(H/E) = 1/2
Now: Suppose evidence was E' = Fa & Fb
Then: m

(E') = 2 $ 1/8 = 1/4 (Fa & Fb are both true in state descriptions 1 and 2)
m

(H & E') = 1/8

So: c

(H/E') = 1/2
Thus: c

(H/E') = c

(H/E)!
What this means:
H is confirmed by E' to the same extent as it is confirmed by E. This does not makes sense: The
claim that c has property F should gain more support from the knowledge that a and b have
property F than it should from the knowledge that just a has property F.
General Problem for Logical Interpretation:
Vaules for logical probabilities are not uniquely fixed!
Constraints on weight assignments:
(a) Sum of weights for all state descriptions must equal 1.
(b) Weight for any state description must be greater than 0.
So: Besides Carnap's method m*, there are in principle an infinite number of ways to assign
weights. (And really why should "learning from experience" be a criterion when it comes to
defining the very meaning of probability?)
State Description Weight
1. Fa & Fb & Fc 1/8
2. Fa & Fb & "Fc 1/8
3. Fa & "Fb & Fc 1/8
4. "Fa & Fb & Fc 1/8
5. Fa & "Fb & "Fc 1/8
6. "Fa & Fb & "Fc 1/8
7. "Fa & "Fb & Fc 1/8
8. "Fa & "Fb & "Fc 1/8
Wittgenstein's Method m

Let c

(H/E) = m

(H & E)/m

(E)
But! If probabilities really are measures of a rational agent's degree of belief, then
"learning from experience" may be a desireable characteristic...
III. Bayesian Confirmation Theory
Let H = hypothesis, E = evidence, K = background knowledge
Bayes' Theorem becomes:
Exactly what we are looking for in the context of confirmation theory! Provides a quantitative
measure of how belief in a theory is affected by evidence and background knowledge.
Claim: Under a Personalist Interpretation, Bayes' Theorem gives us exactly such an account!

Pr(H/K &E) =
Pr(H/K)Pr(E/K &H)
Pr(E/K)
, Pr(E/K) ! 0
=
Pr(H/K)Pr(E/K &H)
Pr(H/K)Pr(E/K &H) +Pr(" H/K)Pr(E/K& " H)
Pr(E/K&H) = likelihood of E given K and H = Degree of belief in E, given H is true.
Pr(E/K) = expectedness of E given K = Degree of belief in E regardless of whether H is true
or false.
Pr(H/K&E) = posterior probability of H given K and E = Degree of belief in H given
background knowledge and evidence.
Personalist Interpretation of Bayes' Theorem:
Pr(H/K) = prior probability of H given K = Degree of belief in H before E is obtained.
10
Recall: We want a quantitative account of how evidence affects our belief in a theory.
Bayesian Confirmation Theory Claims:
1. Belief in a theory is rational if it conforms to the Probability Axioms (Personalism)
and, in the light of evidence, it is "up-dated" by Bayes' Theorem (Bayesianism).
2. "E confirms H relative to K" means "Pr(H/K & E) > Pr(H/K)"
(a) Let E
1
= 1 head observed after 1 flip. Then
(i) likelihood Pr(E
1
/K & H) = 1 A2 and (K & H) ! E
1
(ii) Pr("H/K) = 1 # 0.01 = 0.99 Negation rule
(iii) Pr(E
1
/K & "H) = 0.5 Pr(head on 1 flip/fair coin)
So for John, Pr(H/K & E
1
) = = 0.02
(0.01)(1)
(0.01)(1) + (0.99)(0.5)
(b) Let E
2
= 2 heads observed after 2 flips. Then
(i) Pr(E
2
/K & H) = 1
(ii) Pr(E
2
/K & "H) = Pr(heads on 2 flips/fair coin)
= Pr(heads on flip2 and heads on flip1/fair coin)
= Pr(heads on 1 flip/fair coin)Pr(heads on 1 flip/fair coin)
= (0.5)(0.5) = 0.25
So for John, Pr(H/K & E
2
) = = 0.04
(0.01)(1)
(0.01)(1) + (0.99)(0.25)
(c) Similarly, Pr(H/K & E
10
) = 0.91
since Pr(E
10
/K&"H) = (0.5)
10
= 1/1024)
SO: As more heads are reported, Johns belief in H converges to 1 (certainty).
Now: Suppose Wes' prior probability in H given K is Pr(H/K) = 0.5. Then for Wes,
Pr(H/K & E
1
) = 0.67
Pr(H/K & E
2
) = 0.80
Pr(H/K & E
10
) = 0.99
Wes' belief in H is strengthened
as more evidence accumulates.
11
Important Fact: John and Wes start out with different priors,
but eventually their beliefs about H converge.
"Washing
out" of priors
Ex: Coin flips
H = Penny p is 2-headed
K = Penny p is either 2-headed or fair
E = <report of series of coin flips>
Suppose prior Pr(H/K) = 0.001 for John
Johns initial degree of belief that p is 2-headed.
Advantages of Bayesian Confirmation
(1) Models convergence of beliefs to certainty on positive evidence
(2) Wash-out of priors: the convergence in (1) occurs regardless of initial prior probabilities.
(3) Provides a quantitative description of confirmation. (It gives us a numerical value for the
degree of belief in a hypothesis given evidence.)
3 Problems
(1) Problem of Subjectivity
12
So: Belief in intelligent design, for instance, will be just as rational as belief in evolution, given
that the creationist makes appropriate assignments to his priors and likelihoods.
But: This consistency constraint does not prescribe what the values to the priors, likelihoods
and expectedness must be.
Note: A belief in a theory will be deemed rational just when it obeys the Probability Axioms
(Personalism) and is up-dated in the light of evidence and background knowledge via
Bayes Theorem (Bayesianism).
(2) Problem of Old Evidence
Suppose: Pr(E/K) = 1 E is known prior to H.
And: Pr(E/K & H) = 1 H entails E.
Then: Pr(H/K & E) = Pr(H/K) (Bayes' Theorem)
But: This is problematic since there are cases in science in which old evidence did affect belief in
a theory. The perihelion advance in Mercurys orbit was known before Einstein constructed
his general theory of relativity (GR); it constituted old evidence with respect to GR.
However, the perihelion advance was taken as confirming evidence for GR; it provided
additional proof that most physicists at the time used to condition their beliefs in the
theory. The upshot is that the Bayesian account of confirmation cannot describe how such
old evidence conditions belief.
So: E plays no role in how beliefs get assigned to H.
So: Cant say that E confirms H.
E is "old evidence".
(3) Problem of Non-Zero Priors
Poppers Claim: Pr(H/K) = 0 for all non-tautological universal H's (H's that make
universal claims and that are not trivially true).
Proof: Let H = (i)Pa
i
(A generic universal H: "Every individual a
i
has property P", where i runs from 1
to some total number)
13
Note: For Bayes' Theorem to work, cant assign Pr(H/K) = 0. There must be some initial degree
of belief in H to begin with.
Now assume:
(I) Pr(Pa
1
& Pa
2
& ... & Pa
n
/K) = Pr(Pa
1
/K) Pr(Pa
2
/K) ... Pr(Pa
n
/K), for all n
(E) Pr(Pa
m
/K) = Pr(Pa
n
/K), for any m, n
Note: H entails (Pa
1
& Pa
2
& ... & Pa
n
), for any n.
Then: From (I),
Thus: By (A1), Pr(H/K) = 0.

Pr(H/K) ! lim
n"#
Pr(Pa
1
&Pa
2
& ... &Pa
n
/K)
Now: From (A1), any particular factor Pr(Pa
i
/K) in the limit on the RHS of (&&) satisfies
0 ! Pr(Pa
i
/K) ! 1.
So:
So: Either Pr(a
i
/K) = 0, or Pr(a
i
/K) = 1/r for some integer r ' 1. (The case r = 1
represents a trivially true (i.e., tautologous) universal claim.)

lim
n!"
Pr(Pa
1
& ... &Pa
n
/K) = lim
n!"
Pr(Pa
1
/K)...Pr(Pa
n
/K) (&&)
Now: From (E), all factors in the limit on the right-hand-side of (&&) are equal.
(&)

Pr(H/K) ! lim
n"#
Pr(Pa
1
&Pa
2
& ... &Pa
n
/K) = 0
So: Either they are all 0, hence the limit is zero, or they are all equal to 1/r for some integer
r ' 1. In the latter case, the right-hand-side becomes

lim
n!"
(1/r)
n
= 0.
So: So either way, the right-hand-side of (&&) vanishes and we have,
Response:
To block the conclusion, we can challenge either of assumptions (I) or (E).
(1) (E) claims that instances of a property P are exchangeable. This assumption guarantees
"Weak Hume Projectability", which might be appealing to a Bayesian.
14
(2) (I) claims that all instances of a property P are probabilistically independent of each other.
Popper says: If (I) doesn't hold, this implies some form of causal glue between events.
A Bayesian may respond: The "glue" need not be causal, but merely probabilistic!
Def. A property P is Weakly Hume Projectible relative to background knowledge K just
when, for the sequence ... a
#2
, a
#1
, a
0
, a
1
, a
2
, ... and any n,

lim
k!"
Pr Pa
n
/Pa
n#1
&Pa
n#2
&...Pa
n#k
&K
( )
=1
"If all a's have been P's for as far back as we care to consider, then the next a will be a P
with certainty."
Claim. Exchangeability entails Weak Hume Projectability.
08. Humes Problem of Induction
Two types of objects of knowledge, according to Hume:
(I) Relations of ideas = Products of deductive (truth-preserving)
inferences; negation entails a contradiction.
Recall: Subject of confirmation = How scientific claims are justified.
This assumes that they are capable of justification in the first place.
Hume asks: Is there a rational basis for inductive inferences?
Hume responses: No!.
All observed ravens are black.
All ravens are black.
Example:
Hume asks, Can we ever be justified
in believing the conclusion?
1740 - Treatise of Human Nature
1748 - Inquiry Concerning Human Understanding
Outline of Humes Argument:
(1) Matters of fact can only be known through experience (a posteriori).
(2) Therefore matters of fact can only be justified by recourse to experience.
(3) But any attempt to do so is circular.
! There is no justification for inductive inferences.
Consequence: To the extent that scientific claims are based on inductive inferences, they
cannot be justified.
Ex: 2 + 2 = 4
Ex: All ravens are black.
(II) Matters of fact = Products of inductive inferences; negation
does not entail a contradiction.
ASIDE:
1. Hume is not just saying that we can never be certain about inductive inferences (i.e., we can never be 100%
certain that all ravens are black). This would be uncontentious: Most people would agree that theres
always room for error in making an inductive inference. However, most people would at the same time claim
that we are justified in making (some) inductive inferences, even though they arent 100% guaranteed to
work (i.e., we think there are standards by which we can judge good inductive inferences from bad ones).
Hume is saying that this is wrong: we are not justified in believing any type of inductive inference.
2. Premise (1) is the fundamental claim of Empiricism. Rationalism, on the other hand, claims that some
matters of fact can be known a priori (without recourse to experience). So a Rationalist can block Humes
argument by rejecting the first premise.
3. Hume demonstrates Premise (3) for a particular notion of justification. Briefly, for Hume, to justify a
method of inference requires knowing with certainty that it works. There are weaker notions of justification.
1
Examples of Circular Reasoning
Main question: What justifies inductive inferences?
(1) Causal relationships: We are justified in inferring general claims from finite amounts of
evidence just when theres a causal relationship involved.
Problem: Causal relationships are matters of fact, known only through experience; i.e.,
they are established by means of induction (we never directly observe causal
connections - we inductively infer their existence based on our observations of
correlations). So to say that causal relations justify induction is to say that
induction justifies induction.
Humes analysis of cause/effect relationships:
Three parts - cause, effect, causal connection.
What we observe:
(1) Temporal priority: cause comes before effect.
(2) Spatiotemporal proximity: cause and effect are close to each other.
(3) Constant conjunction: same cause-effect sequence on numerous observations.
What we dont observe:
Causal connection. This is only inductively inferred from (1) - (3).
(2) Uniformity of Nature: We think a causal relation is present given past observations and
the assumption that the future will be like the past (uniformity).
Problem: Why should we believe that nature is uniform? If its because this has been our
past experience (i.e., nature has appeared to be uniform in the past), then we are
using circular reasoning: the inference from past to future is an inductive
inference. So to to say that the uniformity of nature justifies induction is to say
that induction justifies induction.
(3) Appeal to Track Record: Induction has worked in the past, so we are justified in
believing it will work in the future.
Problem: Same as for (2) above.
2
Some Responses to Hume
(1) Success of Science: Scientific forms of inductive inference are justified by their success.
(Scientific forms of inductive inference are based on highly controlled experiments. Since these
experiments are very stringent, their success in identifying correlations justifies us in believing
that these correlations are evidence for underlying causal connections.)
So: In the context of induction, Hume is just asking: Is it reasonable to be reasonable?
(i.e., Is it reasonable to base our beliefs on inductive evidence?) And this by definition
is trivially true.
So: Response (2) says it makes no sense to ask if induction can be validated -- by
definition, induction forms one of the basic principles by which we justify our beliefs.
But: Humes question is, Can induction be vindicated? And the answer to this is, No:
We cant demonstrate how it achieves its goals without circularity.
Goal of induction - to successfully project past regularities into the future.
(2) Ordinary Language Dissolution: Hume doesnt understand what it means to be "rational".
So: The real question Hume is asking is: Is it reasonable (i.e., does it serve our goal of
making successful predictions) to be reasonable (i.e., to use inductive inferences)?
And this is not trivially true.
Problem: This is an appeal to track record (successful in past, therefore successful in future),
and hence is circular.
Claim: To be reasonable (rational) is by definition to base ones beliefs on evidence (indutive
or deductive).
Problem: There are two ways to justify a method:
Validation: Appeal to more basic principles.
Vindication: Indicate how method achieves its goals.
3
(3) Deductivism: Give up inductive inferences.
Claim (Popper): Scientific inferences are deductive; so science does not face a problem of
induction.
Recall the simple negative-outcome form of HD reasoning:
If H is true, then O is true.
O is not true.
! H is not true.
If H does not yield false Os (i.e., if it passes severe tests), then Popper says that H is
corroborated (as opposed to confirmed).
Problems:
(a) Recall that this simplistic account faces the Duhem-Quine Problem: In realistic
situations, H cannot be tested in isolation; we need additional assumptions (aux.
hypotheses; initial conditions) in order to derive O from H. Hence if O is not true, we
can always retain H and blame these additional assumptions.
Moral: The falsification method on which the Deductivism response to Hume is
based, is too simplistic.
(b) Furthermore, the falsification method only describes the past performance of H. It says
nothing about the future success of H in making predictions. So Popper must claim
that science is concerned entirely with explanation, not prediction. Otherwise, he still
faces Humes Problem: What informs us about which H to use if we want to make a
prediction? Successful past performance does not guarantee successful future
performance.
4
(4) Pragmatic Vindication:
What vindicates induction? Ans: Given the alternatives, its the best option.
Reichenbachs decision matrix:
Let our goal be to predict the future with reliability. Reichenbach lists our options in
achieving this goal in the following table:
Why failure here?
Claim: Any successful method can provide the basis for the inductive method. In other
words, if any method works, then so does induction. Or: If induction doesnt work,
then no method works.
possible
states
options
use induction
(sci. methd)
dont use
induction
(crystal ball)
success
success or
failure
failure
failure
Nature is
uniform
Nature is
not uniform
Ex: Crystal ball forecasting - If it works, it establishes a uniformity. And we can use this uniformity as
the basis for induction. So if crystal ball forecasting can produce consistent successful predictions,
then so can induction.
Conclusion: We have nothing to lose and everything to gain in using the
inductive method.
Problems:
(a) Notion of uniformity - How much is needed to allow induction to work?
(b) What type of induction is being vindicated?
induction = projection of
past regularities into future
09. Kant and Logical Positivism
Topics:
I. Two Types of Skepticism
II. Kants Proposal
III. Logical Positivism
I. Two Types of Skepticism
Inductive skepticism = Skepticism about claims about the world that go beyond the
currently available evidence.
Arises whenever there are two or more possible ways of describing the world and no
evidence currently available is sufficient to determine which is true.
Example: We are brains in vats.
Example: All ravens are black.
Note: In the terms of a previous lecture, metaphysical skepticism is skepticism about theoretical
claims (in-principle unobservable claims); inductive skepticism is skepticism about in-
principle observational claims.
II. Kants Proposal
(1) Response to metaphysical skepticism: Argue that we should not be concerned with
claims that go beyond all possible experience. Make the following distinction:
World of Experience
(Phenomenal World)
object of knowledge
Things-in-Themselves
(Noumenal World)
no knowledge possible
Noumenal world:
raw data
cannot be directly
experienced.
Claim: Metaphysical skepticism is generated by claims about the noumenal world (claims
that go beyond all possible experience). Thus it is untroublesome -- such claims are
not the proper objects for knowledge.
(1787) Critique of Pure Reason
Metaphysical skepticism = Skepticism about claims that go beyond all possible evidence.
Arises whenever there are two or more possible ways of describing the world and no possible evidence
is sufficient to determine which is true.
Phenomenal world:
product of experience
conditioned by the way we
process/filter the raw data
Experience:
act of filtering/processing raw data
dependent on "built-in" intrinsic
"filters"
1
(2) Response to inductive skepticism: Recall Hume's two assumptions:
(a) (Empiricist Claim) Matters of fact are known only through experience.
(b) To know means knowing with certainty that you know.
Kant rejects (a). He does this by distinguishing between four types of claims:
Four Types of Statements (according to Kant):
3. a priori: - truth/falsity can be established without recourse to experience
4. a posteriori: - truth/falsity can only be established by recourse to experience
In the sense of ruling out certain possibilities;
i.e., not true in all possible worlds
Recall: Hume only distinguishes two types - Relations of Ideas vs. Matters of Fact.
Heres how they relate to Kants distinctions:
This cell is empty. All analytic claims are a
priori - their truth is established by logical
considerations alone
Matters of Fact are synthetic claims that
are a posteriori - All ravens are black
cant be known through reason alone.
analytic
synthetic
a priori a posteriori
Relations of Ideas
Matters of Fact
?
Are there synthetic a priori claims?
Kant's Claim: Yes.
(1) and (2) are mutually exclusive (a claim cant be both analytic and synthetic at the same
time). Likewise, (3) and (4) are mutually exclusive.
Ex: 7 + 5 = 12. Kant argues that this statement is certain, but is not based on experience (so it
is a priori). But it is not analytic: The concept of 12 is not contained in the concepts of
7 and 5. In this sense, the negation of 7 + 5 = 12 does not entail a contradiction.
Hence, in addition to being a priori, it is also synthetic; i.e., it is a synthetic a priori truth.
2
1. analytic: - logical truth or definition
- negation entails contradiction
- devoid of factual content
2. synthetic: - contingent (i.e., could be either true or false)
- negation does not entail contradiction
- contains factual content
Ex: All ravens are either black or not black.
Ex: All ravens are black.
True by definition; true by virtue of the
meaning of the terms that appear in it.
Three of Kants examples:
(1) Every event has a cause.
(2) Statements in Euclidean geometry.
(3) Statements in arithmetic.
Synthetic a priori truths: True claims about the world knowable through reason alone.
The concept of causality is a pure form of understanding
that preconditions how we comprehend natural phenomena.
Euclidean geometry provides a pure form of intuition
that preconditions our sensation of space.
Claim: Forms of "intuition" and "understanding" are necessary in order for knowledge and
experience to be possible. Such forms are preconditions for the possibility of knowledge
and experience.
So: Kant rejects Humes assumption (a).
But: Kant still retains assumption (b).
Concepts in arithmetic (continuity and infinity) provide the pure
form of intuition that preconditions our senstation of time.
Summary:
(1) Knowledge has both a form, which is conceptual and mind-dependent, and a content, which is
contributed by the world.
(2) The formal aspect consists of unrevisable synthetic a priori truths.
(3) These truths include, among other things, Euclidean geometry, arithmetic, and the claim that
all events have a cause.
3
ASIDE: In other words, Kant claims that Euclidean geometry and arithmetic must be true of the world, because
they preconfigure the way we experience the world. One way to think about this is the following: According to
Kant, the phenomenal world is literally constructed by us. It is the result of prossessing the raw data of experience
(associated with the unknowable noumenal world) into a general Form that we can comprehend. This Form depends
exactly on the way the data gets prossessed by us; it is prefigured by how we prossess data. And there are a number
of ways we prossess data. Kant calls these ways pure forms of intuition (which prefigure our senses; he identifies 2:
Euclidean geometry and arithmetic) and pure forms of understanding (which prefigure how we analyze the data from
our senses; he calls these categories and identifies 12 including the concept of causality listed above).
So: The general Form our experience takes (the phenomenal world) is the result of a number of sub-forms working in
conjunction. Again, these prefiguring forms include the notion of causality, and Euclidean geometry and arithmetic.
ASIDE: One can disagree with Kant on any or all of these points. Kants argument
for Claim (2) is known as the transcendental deduction. He thought Claim (2) was an
a priori truth. Kants argument for Claim (3) is known as the metaphysical deduction.
Ex: Kant thought the hypothesis H = All objects are infinitely divisible is unknowable because it cannot be
verified with certainty (which is the criterion Hume adopts in assumption (b)). The Demon and the Banana
example indicates why H cannot be known with certainty: No matter how many cuts you make, you will never
know with certainty that the next cut of the banana will succeed (the demon may decide then and there to
thwart your attempt).
ASIDE: As indicated earlier, there are weaker notions of justification than verification with certainty. For example,
you could maintain that to be justified in believing a claim, it must have been produced by a method that is reliable in
the limit (at some point after umpteen many pieces of evidence, it has to converge to the truth). One can show that
Kants H is refutable in the limit; hence, on this notion of justification, we can have knowledge that it is false.)
The Demise of Kants Proposal
Kants claim that there are synthetic a priori truths has been questioned. Some take the following
two historical developments of the late 19th/early 20th century to show that his examples from
Euclidean geometry and arithmetic fail.
This distinction arises because we now have a choice (Euclidean vs. non-Euclidean) of
descriptions, and which description really holds is an empirical matter; it cannot be decided based
on pure reason alone.
Note: The claim that nevertheless human perception is based on Euclidean geometry does not
defuse this objection to Kant (someone might observe that Euclidean geometry is no
longer a necessary precondition for describing the world, but it still is a precondition for
experiencing the world). The mere logical possibility of alternatives to Euclidean
geometry is enough to question his claim that the latter contains synthetic a priori truths.
Euclidean
Spherical Hyperbolic
Sum of angles = 180 Sum of angles > 180 Sum of angles < 180
2 lines are
parallel if they
never cross
A
B
A
B
Aside: Euclidean geometry can be axiomatized. The standard scheme includes the postulate:
(1) Euclidean 5th postulate: Through any given point only one line can be drawn parallel to a given line.*
You get a non-Euclidean geometry by tweaking this 5th postulate and leaving all the other axioms alone.
There are two basic types of non-Euclidean geometry, depending on the tweaking:
(2) Spherical geometry 5th postulate: Through any given point no lines can be drawn parallel to a given
line.
(3) Hyperbolic geometry 5th postulate: Through any given point infinitely many lines can be drawn that
are parallel to a given line.
(1) The development of non-Euclidean geometry.
Non-Euclidean geometries indicate that Euclidean geometry is not a necessary precondition for
describing the physical world. This introduces a distinction between:
(a) pure geometry: consists of analytic a priori statements.
(b) applied geometry: consists of synthetic a posteriori statement.
*This is actually a consequence of Euclids 5th postulate.

4
(2) The development of modern axiomatic numerical analysis and modern logic.
Indicates that statements in arithmetic can be considered analytic a priori (as opposed to
synthetic a priori). In general, allows a distinction between applied arithmetic vs pure
arithmetic similar to the one in geometry.
III. Logical Positivism
School of thought in 1920s-30s associated (primarily) with philosophers in Vienna and Berlin
Characterisitcs:
(a) Logical analysis of science.
(b) Empirical (positivist) basis for scientific knowledge.
positivism from August Comte
(19th cent.) - Claim: sense
experience is the basis for knowledge
4 Logical Positivist Attempts to Address Metphysical Skepticism:
(1) Logical Constructivism
Reduce experience (sense-data) to logical constructions (atomic propositions)
Replace Kant's Phenomenal World with a logical construction.
Result: A logical basis for all meaningful claims about the world.
Russell (1918) Logical Atomism, Carnap (1928) The Logical Structure of theWorld
sense datum = atomic proposition
unit of experience linguistic unit
Russells Dictum: Whenever possible, logical constructions are to be substituted for inferred
entities.
Translation of world of
experience into a logical
system (a language)
Consequence: Resolution of metaphysical skepticism! Claims that go beyond all possible
evidence are claims about inferred entities. The goal of logical constructivism is to reduce
such metaphysical claims to logical relations between claims that can be adjudicated by the
available evidence (i.e., observational claims).
Aside: The philosopher Michael Friedman views the development of modern logic and analysis as a foil
to Kant in the following way: Kant only had Aristotelian logic available to him, which does not provide
the concepts of infinity and continuity that underlie numerical analysis. Thus, such concepts must be
intutive for Kant. The development of modern logic and analysis in the late 19th and early 20th centuries
demonstrated that such concepts need not be left to the intuition, but can be given rigorous formulations.
5
ASIDE: Russells version adopts phenomenalism -- the emphasis is on immediate sense-data.
Problem: How can an objective world of experience be founded on immediate personal subjective
experience? Because of this problem, Carnaps version drops phenomenalism for physicalism --
the basic terms in the language refer to middle-sized objects (not immediate sense-data).
Problem with logical constructivism: Not all meaningful concepts can be reduced to logical
constructions.
Typical example: The logical reconstruction of numbers using set theory faced Russells
Paradox.
How to define numbers using sets:
Let 0 be the empty set (i.e., the set with no members): 0 = { } or !.
Let 1 be the set with exactly one member; namely the empty set: 1 = {!} = {0}
Let 2 be the set with exactly two members: 2 = {!, {!}} = {0, 1}
3 = {!, {!}, {!, {!}}} = {0, 1, 2}
4 = {!, {!}, {!, {!}}, {!, {!}, {!, {!}}}} = {0, 1, 2, 3}
But: Without any constraints on what counts as a set, there should be sets that do not belong to
themselves:
- The set of all cats
- The set {0, 1, 2, 3}
- The set {0, 1, 2, 3, {0, 1}}
- The set x = {a, b, c}
And there should be sets that do belong to themselves:
- The set of sets
- The set of all non-cats
- The set y = {a, b, c, y}
6
Russells Paradox
Let R be the set of all sets that do not belong to themselves.
Claim: R belongs to itself if and only if R does not belong to itself.
Proof: (i) Suppose R belongs to R.
Then R is a set that does not belong to itself.
So R does not belong to R.
(ii) Suppose R does not belong to R.
Then R is a set that belongs to itself.
So R does belong to R.
ASIDE: Russells solution was to claim that sets of sets are different types of objects than sets.
(2) Verifiability Principle of Meaning (VPM)
A claim is meaningful if and only if it can be verified (i.e., if and only if a set of observations
exists that would establish the truth of the claim).
Consequence: Metaphysical claims about in-principle unobservable objects are meaningless.
This dissolves the problem of metaphysical skepticism given we adopt:
Schlick ~1920s
Main Problem with VPM: Its too strict. No scientific claim is verifiable in the sense that its
truth can be conclusively established on the basis of observation alone. Thus a strict application
of VPM would result in all of science being deemed meaningless.
(3) Meaning Postulates
Attempt to demonstrate how terms in a theory aquire their meanings (i.e., by means of
"meaning postulates"), and hence how claims incorporating such terms aquire credibility.
Carnap (1936) Testability & Meaning
Important Point: Metaphysical skepticism is a concern with criteria for belief (it is an
epistemological concern). It asks: Do we have any reasons for believing all the claims of a
description of the world that goes beyond all possible evidence? The Verifiability Principle of
Meaning is a linguistic principle. The above assumption is the link between the epistemological
concern and the linguistic principle.
Motivation: We can reject the VPM as too strict, but still retain the assumption that
meaningless claims are not objects for belief. Then, in order to demonstrate what sorts of claims
are believable and what arent, we need to be able to determine what sorts of claims are
meaningful and what arent. To do this, we need to know how the sentences and terms in a
theory are given their meanings.
ASIDE: Its important to distinguish between two types of question:
(A) Are the claims a theory makes believable? (i.e., Are they well-confirmed? Should we believe them?)
(B) Are the claims a theory makes meaningful? (i.e., Are the actual sentences the claims of the theory is
expressed in, and the terms that appear in these sentences, meaningful?)
Question (A) is epistemological. Question (B) is linguistic. One legacy of logical positivism is this link
between linguistic concerns and epistemological concerns.
7
Assumption: Meaningless claims are not objects of belief
How terms in a theory aquire meaning (Carnap's method):
(a) First distinguish between:
(i) Observational terms = terms that refer to in-principle observable objects.
(ii) Theoretical terms = terms that refer to in-principle unobservable objects.
(b) Now claim:
(i) Observational terms aquire their meaning directly through experience.
(ii) Theoretical terms aquire their meaning by translating them into observational terms.
Meaning postulate = A sentence that contains both observational terms and
theoretical terms and acts as a translation rule that establishes the meaning of
a theoretical term by linking it directly with one or more observational terms.
Two characteristics of Meaning Postulates:
(1) A meaning postulate is an analytic sentence: It is true by definition.
(2) Meaning postulates play two roles:
(i) They establish the meaning of theoretical terms.
(ii) They allow Instance Confirmation to be extended from purely observational hypotheses to
purely theoretical hypotheses.
ASIDE: Recall that under Instance Confirmation, E confirms H just when E is an instance of, or entails
an instance of, H. E must be an observation sentence (only containing observational terms). If H is an
observation sentence, then it can share some of its terms with E, hence we can see how E might entail an
instance of it. However, if H is a purely theoretical sentence, it can never share any terms in common
with any E; so the instance confirmation account fails for such H's. Meaning postulates fill this gap --
they allow purely theoretical sentences H to be linked with purely observational sentences; hence they
allow the possibility of instances of a purely theoretical H to be entailed by a purely observational E.
8
Meaning postulates as operational definitions Bridgeman (1937) The Logic of Modern Physics
An operational definition defines a term by linking it to a series of operations and observable
outcomes that must occur in order for the term to be applicable.
Ex1: Operational definition of soluable in water:
x is soluable in water means If x is placed in water, then x dissolves
Ex2: Operational definition of simultaneity:
Events A and B are simultaneous means If light signals are emitted from A to B and B
to A, then they will be received at an event half-way between A and B at the same time
Problems with operational definitions:
(a) One definition per measuring device/operation.
(b) Some terms cannot be equated with observable outcomes of operations (Isnt there more to
pain than the sum of its effects?) The Anesthetized Baby Parable.
(c) Anything that does not satisfy the "if" clause of the definition satisfies the definition! (In Ex1,
the conditional is true of any x that is never placed in water; so according to the definition,
any x that is never placed in water is soluable in water!)
Ex3: Operational definition of a mental state (pain):
x is in pain means If x is brought into contact with certain stimuli, then certain
movements of xs bodily extremities are observed, accompanied by high-frequency vocal
emissions
9
(4) Conventionalism Reichenbach (1938) Experience and Prediction
Given two hypotheses that agree on all possible evidence but disagree on their claims, the
choice between them is made by convention. There is no fact of the matter which is true.
Motivation: Refocus concern on believability (confirmation) as opposed to meaning.
ASIDE: One way to think of conventionalism is by comparing it with Kant. Recall that Kant
held a form/content distinction in describing knowledge. For Kant, the formal element was
unrevisable and inherent in all rational beings. The conventionalist can be seen as observing that
the formal element, if it does exist, is not unrevisable -- there are many ways we can put structure
and form on the content of our experience; any one particular way is adopted by convention.
ASIDE: Operational definitions are a particular type of explicit definition in which a theoretical term is
explicitly defined by reference to observational terms. Explicit definitions take the following general form:
Cx ! (Sx " Rx)
"x is an instance of C if and only if, if x satisfies the test condition S, then x is an instance of R", where C is the
theoretical term to be defined, S is an observational term describing a test condition, and R is an observational
term describing an outcome of the test. Carnap originally suggested using what he called "bilateral reduction
sentences" as meaning postulates. They take the following general form:
Sx " (Cx ! Rx)
"If a case of x satisfies the test condition S, then x is an instance of C if and only if x is an instance of R".
Bilateral reduction sentences don't face problem (c) above; but problems (a) and (b) are still present. In more
general accounts, a meaning postulate can be any sort of generalization. The idea is to treat some subclass of
generalizations in a theory as meaning postulates: relations between observational and theoretical terms that
are stipulated as definitions and not subject to confirmation/disconfirmation (i.e., they are treated as analytic).
10
Example: The Conventionality of Simultaneity
Claim: The choice of a simultaneity relation to describe the simultaneity of distant events at rest
with respect to each other is conventional. Given an event A, there is no objective fact of
the matter as to what distant events at rest with respect to A are simultaneous with A.
Question: How can the simultaneity of distant events be established?
Einstein (1905): By setting up synchronized clocks at these events.
Question: How can distant clocks be synchronized?
Einstein (1905): Use light signals.

T
A-emit
T
B-reflect
T
!
T
!
T
A-return
x
t
A B
Claim: The standard definition is conventional. We could also stipulate that Clocks A and B
are in synchrony just when
T
B-reflect
= T
!
# T
A-emit
+ !(T
A-return
$ T
A-emit
) , 0 < ! < 1
Standard Definition of Simultaneity
The event at T
B-reflect
is simultaneous with the
event at T
!
= T
A-emit
+ !(T
A-return
$ T
A-emit
).
Non-Standard Definition of Simultaneity
The event at T
B-reflect
is simultaneous with the event at T
!
.
ASIDE: According to special relativity, the simultaneity of events in constant motion with respect to each other
(i.e., in different inertial frames) is relative, and thus non-absolute (this is the relativity of simultaneity). The
conventionality of simultaneity claims that simultaneity is non-absolute for events in the same inertial frame, too.
Operational Procedure for Clock Synchronization
(1) Emit a light signal from Clock A to Clock B and record the time T
A-emit
on Clock A.
(2) Have Clock B reflect the signal back to Clock A. Record the Clock B time T
B-reflect
.
(3) Record the Clock A time T
A-return
when the light signal returns.
Clocks A and B are stipulated to be in synchrony just when
T
B-reflect
= T
!
# T
A-emit
+ !(T
A-return
$ T
A-emit
)
11
Note: There are an infinite number of non-standard definitions! One for every value of ! between
0 and 1. The standard definition is obtained by setting ! = 1/2. The conventionalist
claims that we choose the value of ! by convention.
Why? Why isn't there a fact of the matter what the value of ! should be? Key assumption of
Eintein's ! = 1/2 "choice": The light signal travels at the same speed from Clock A to
Clock B as it does from Clock B back to Clock A.
Notice: If we choose the T
!
event in the diagram as simultaneous with the T
B-reflect
event,
this entails that, according to Clock A, the light signal takes more time to travel from A to
B than it does to travel from B back to A. But the distance traveled is the same in both
directions. So the light signal's speed must be slower outward-bound than inward-bound.
And now the catch: In order to measure the speed of light from point A to point B, say (the
"one-way" speed of light), we first need to have synchronized clocks set up at A and B.
Problem of Circularity:
(a) To measure the one-way speed of light, we need synchronized clocks.
(b) But we can only synchronize our clocks if we have prior knowledge of distant simultaneity,
which requires prior knowledge of the one-way speed of light.
Note: The round-trip speed of light is unproblematic. You just need one clock to measure it.
Upshot: An Infinte Number of Competing Hypotheses that Agree on All Possible
Evidence:
Every value of ! between 0 and 1 represents a particular claim about the nature of simultaneity.
All values agree on all possible evidence. The conventionalist says we pick one such value as a
matter of convention.
Scientific Change
Topics:
I. Recieved View
II. Historicist View
III. Social Constructivism
IV. Practical Science View
First: Somewhat important distinction (a la Salmon)
justification:
knowledge:
method of analysis:
units of analysis:
verifiability
certainty/incorrigibility
reduction to logical
constructs
sense data
(phenomenalism)
Russell
Vienna Circle: Carnap,
Schlick, early Wittgenstein,
Neurath
confirmability
probability
logical analysis of concepts (explanation,
confirmation, theory, etc.)
middle-sized objects
(physicalism/realism)
Reichenbach, Hempel, later Carnap
Logical Positivism Logical Empiricism
I. Recieved View - Characteristics
(Hacking Scientific Revolutions)
(1) Realism: Science yields objective knowledge of the world
(2) Demarcation: Clear distinction between science and non-science.
Ex: Poppers criterion - falsifiability
psychoanalysis,
Marxism
vs. general relativity
(4) Observation/Theory Distinction: 3 levels:
a. directly observable objects
b. indirectly observable objects
c. in-principle unobservable objects
theoretical objects
observable objects
(3) Cumulative Nature of Progress: Science progresses by building on previous developments.
(5) Foundationalism: Hierarchical view of scientific knowledge with observation and experiment
at base.
Weak version: The world exists independently of the scientist, whos role is to discover its properties.
(Compare with Kant)
Strong version (anti-nominalism):
The world has a given structure that science aims to uncover. Science cuts Nature at its joints.
1
ASIDE: Nominalism is the claim that there are no essenses; no natural kinds. There are only individuals. A
natural kind is like a species, but more permanent. A realist (strong version) may claim that electrons, for instance,
form a natural kind. There are electrons as individuals, and there is the natural kind to which all electrons belong.
A nominalist thinks there are no such permanent groupings in Nature; there are only individual objects.)
(6) Deductive nature of theories:
theories are representable in a formal language
observational predictions can be deduced from hypotheses
strong claim: theories in science can be axiomatized
(7) Meanings of scientific terms are precise, fixed, objective.
(8) Distinction between Discovery and Justification
For any hypothesis H, we can ask:
(a) How was H discovered?
(b) How is H justified?
Claim: Factors influencing discovery may be psychological, social, economic, political, etc.
Claim: Justification, unlike discovery, is subject to a purely logical analysis. Philosophy of science
should restrict its attention to the context of justification only.
Examples:
(a) Kekules discovery of the structure of benzene (firey snakes).
(b) Ramanujan and the smallest integer expressible as a sum of two cubes (goddess in dream).
(c) Newton and the apple.
(9) Unity of Science Thesis: 2 views
(a) Hierarchy of fields plus reductionism
(b) Unity of method
Motivation from physics:
Quantum
Gravity (?)
General
Relativity
Special
Relativity
Newtonian
Dynamics
Newtonian
Gravity
Non-relativistic
Quantum Mechanics
Quantum
Field Theory
Quantized
Newtonian
Gravity
G
h
1/c
3 fundamental constants of nature:
G = Newtonian gravitational constant
h = Plancks constant
c = speed of light
Major fields in physics are
related to each other via 3
constants of nature G, h, c.
Question 1: Is physics representative of all fields in science?
Question 2: Is View (a) compatible with strong realism?
Is View (b) compatible with strong realism?
Question 3: Is View (a) compatible with nominalism?
Is View (b) compatible with nominalism?
social sciences
psychology
biology
chemistry
physics
!
!
!
!

2
II. Historicist (Globalist) View
(Kuhn, Lakatos, Laudan, Feyerabend)
Kuhn (1962) The Structure of Scientific Revolutions
1. Change comes in stages:
normal science crisis revolution new normal science
classical physics quantum physics problems
Ex:
2. Underlying aim of science: Problem-solving
General Characteristics
(A) Normal Science vs. Revolutionary Science
Brief Overview:
(B) Paradigm: 2 senses
(a) World-view associated with a particular episode of normal science.
(b) Collection of problem-solving techniques that characterize a particular episode of
normal science.
(C) Crisis: A buildup of anomalies (problems that the current paradigm cannot solve) that leads
to a revolution.
(D) Incomensurability Thesis: New and old paradigms are incomensurate; they are
completely different. (One sense of completely different: non-translatable.)
Example: Classical physics at the turn of the century could not solve two small problems: (1) consistent
description of black body radiation (leads to quantum theory); (2) consistent description of motion
with respect to the aether (leads to relativity theory).
Example: The term mass in Newtonian physics is completely different from
(cannot be translated into) the term mass in Relativistic physics.
Note: The incomensurability thesis presupposes meaning holism: the claim is that there is no neutral
observation language by means of which we can compare paradigms.
(E) Noncumulative nature of change: Paradigms determine what the appropriate questions
are that drive research. When you change paradigms, you change what the appropriate
questions are; hence you change fundamentally the focus of research.
(F) Gestalt-Switch: New paradigms completely replace old ones. Paradigm change involves a
complete switch in ways of thinking about and viewing the world.
3
" " "
" "
(G) Social Character of Science: Scientific change is modeled after social change (scientific
revolutions are literally revolutions (except (usually) without the guns)). The emphasis is on
social, institutional, political, etc., factors.
Main Problem: How to reconcile meaning holism and incommensurability with change?
We need some common frame of reference for comparing old and new paradigms; i.e., we need to
be able to coherantly explain how a new paradigm solves the anomalies of an old one.
But: If there is a common frame of reference, then change cannot be completely noncumulative,
and paradigms cannot be completely incommensurate.
McGuire: Kuhn is careless in slipping from a linguistic claim (meaning holism) to an ontological claim (scientists live
in different worlds).
P. Feyerabend (1978) Against Method
(Feyerabend = Popper + meaning holism)
Claims:
(1) Scientific practice is not rational (no underlying method).
(2) Thus progress cannot be characterized by reference to an underlying method.
More precisely, Feyerabend argues that:
(a) there is no meaning invariance over theory change;
(b) there is no consistency over theory change;
But: This is good!
Claim: (a) & (b) lead to theoretical monism: the dominance of a single theory or method to
the exclusion of all others. Theoretical monism leads to dogmatism and stagnation.
i.e., there is incomensurability over theory
change. Evidence: Historical examples
(Newtonian mass vs. Relativistic mass, etc)
Ex: accupuncture, creationism, astrology, etc.
Feyerabends motto: Anything goes.
Problem: Theoretical pluralism is itself a method; so it should go, too.
ASIDE: This is a criticism of Kuhn. On Kuhns view, science is rational in the sense of possessing a method; namely, the
method associated with a given paradigm. For Kuhn, progress involves the replacement of one paradigm with another.
One Option: View Kuhn as subscribing to a Kantian metaphysics. Kant claims the way we construct the
phenomenal world is fixed. Kuhn claims the way we construct the phenomenal world is not fixed,
but changes as paradigms change.
ASIDE: Feyerabend is motivated by Popper:
If you allow a free arena of competition, the
strongest theories/methods will survive; i.e.,
those that have passed severe tests.
Progress (critical comparison) occurs via theoretical pluralism: allow as many competing
conflicting theories/methods as possible.
4
III. Social Constructivism
(Strong Programme - Eddingburgh)
D. Bloor (1976) Knowledge and Social Imagery
B. Latour and S. Woolgar (1986) Laboratory Life
S. Shapin and S. Schaffer (1985) Leviathan and the Air Pump
A. Sokal and Social Text (1996)
Claim: Science is a social phenomena in 2 respects:
(1) The manner in which it produces results.
(2) The results themselves: scientific facts are socially constructed.
Big Unresolved Problem: What does socially constructed mean?
One option: Scientific facts are socially constructed in so far as they can be explained solely in
terms of social parameters.
(a) Methodological claim -- Methods used by science are socially influenced (peer review, grant
application process, institutional politics, etc.)
(b) Ontological claim -- Products of science (i.e., facts) are social constructs.
Claim (a) is not controversial. But evidence for Claim (a) is not evidence for Claim (b). To
substantiate Claim (b) requires arguments based on ontological premises.
One option: Must show that scientific facts supervene on social facts - for every scientific fact,
there are social facts that are necessary and sufficient conditions for it.
Another option: Note that the distinction between (1) and (2) in terms of methodology and
ontology is blurred if we subscribe to a Kantian notion of fact. Recall that Kant maintains that
the world of experience (the phenomenal world) is essentially, literally constructed by us. But
Kant still retains the noumenal world as the unknowable objective substrate of experience. The
social constructivists reject the noumenal world, but still retain the Kantian subjective description
of the phenomenal world. So socially constructed means constructed in the Kantian sense.
5
Problem: Requires a theory of social causation; a causal account of how knowledge arises in
which only social parameters occur. In particular, such an account must be able to
distinguish social causes from mere correlations -- highly problematic.
General Problem: Risk is great of running the above 2 claims together. This conflates
methodological claims with ontological claims:
Claim: Experimental procedures are independent of theoretical practices.
Strong Claim: Scientific change is driven by experimental practices.
Weak Claim: Scientific change must be understood both in terms of theory succession and in
terms of how experimental practices change.
IV. Practical Science View
P. Galison (1987) How Experiments End
D. Hull (1987) Science as a Process
S. Pickering (1984) Constructing Quarks
Emphasis on experimental science and technology as opposed to pure (theoretical) science.
Exs: Particle beam accelerators
PCR (polymerase chain reaction) technology
DNA chip technology
Pickering (1984, Constructing Quarks): Notions of resistence and accomodation. Nature
resists experimental probing, and experimentalists accomodate. (Cooking the data in chem lab.)
The Practical Science View sometimes overlaps the Social Constructivist view:
Practical Science Claim: Accomodation first occurs at the experiemental/technological level
and then percolates up to the theoretical level.
Further Claim: There are many ways to accomodate resistence, and nothing inherant in the
phenomena that selects one way over others.
Implicit Nominalist Assumption: There is no fundamental kind-structure to the world; we
can describe it in many ways (we can cut Nature along any lines we see fit). Each way is
underdetermined by our practices (i.e., read evidence). Hence physicists can be said to
construct quarks in the sense of choosing one mode of description over others (note: such modes
really constitute the phenomenal world for a Kantian).
(Note further: This underdetermination thesis is not exactly identical to the Duhem-Quine thesis.
The DQ thesis is compatible with realism.)
6
11. Conceptual Foundations of Quantum Information Theory
Consider two physical properties of electrons:
Color -- black, white
Hardness -- hard, soft
only 2 possible values
How to measure Color and Hardness:
Color
electrons
enter
black
electrons
white
electrons
electrons
enter
Hardness
hard
electrons
soft
electrons
Experimental Result #1: No correlation between Color and Hardness properties
ASIDE: Color and Hardness are used just to
make the discussion more concrete. The actual
properties being referred to are the spin of an
electron along two different directions (axes). For
electrons, spin along a given direction can have
only one of two values (up or down). All the
following results for Color and Hardness have
been experimentally confirmed for spin.
Color
all hard
(or all soft)
half
black
half
white
Hardness
all white
(or all black)
half
hard
half
soft
Experimental Result #2:
Consider 3 box set-up:
Color
a bunch of
electrons
black
electrons
white
electrons
Hardness
half
hard
half
soft
Color
half
black
half
white
This is strange! If only white electrons enter the middle Hardness box, wed expect only white
electrons to exit the last Color box. But they dont!
1
spin-up along
direction n
spin-down along
direction n
spin-up along
direction m
spin-down along
direction m
I. Classical vs. Quantum Systems
2
Questions:
(1) Can we build a Hardness box that doesnt disrupt color?
(2) Which electrons get their color disrupted by the Hardness box?
Experimental
evidence says: NO!
Experimental evidence
says: No way to tell!
Upshot: Cant simultaneously measure both Hardness and Color!
Suppose we feed white electrons into the device.
What should we expect to emerge at h and s?
(i) 50% will be hard and thus will follow the h path. At h and s, they will emerge as hard electrons, so upon a
Color measurement, 50% will be white and 50% will be black.
(ii) 50% will be soft and thus will follow the s path. At h and s, they will emerge as soft electrons, so upon a Color
measurement, 50% will be white and 50% will be black.
Hardness
soft
electrons
hard
electrons
h
mirror
mirror
h and s
black
box
electrons
in
h
s
s
Seems consistent with earlier
experiments: The above
device is essentially a
Hardness box with some
mirrors that simply divert the
direction of the electrons in it.
BUT: Experiments demonstrate that we get 100% white at the end!
The mirrors dont
affect hardness:
they just divert the
path of an electron.
The black box doesnt
affect hardness: it
just redirects paths h
and s to coincide.
SO: We would expect that, of the initial white electrons fed
in, 50% will emerge white and 50% will emerge black.
Experimental Result #3: The 2-Path Device (essentially a Hardness box)
3
Now insert a barrier in the s path:
Hardness
soft
electrons
hard
electrons
h
mirror
mirror
h and s
black
box
white
electrons
h
s
s
barrie
r
What should we expect to emerge at h and s?
(1) 50% less electrons at h and s.
(2) What Color?
Experiments demonstrate that 50% are white and 50% are black!
Recap:
Without barrier: 100% of electrons at h and s are white.
With barrier: 50% are white and 50% are black.
Big Question: What path does an individual white electron take without the barrier present?
(C) Both h and s? No! Experimentally, feed in an electron and check which path it takes.
50% will be on path h, and 50% will be on path s.
(D) Neither h nor s? No! Block both paths and nothing is registered at h and s.
(A) Path h? No! Electrons that take path h are hard, so their Color statistics are 50/50.
But without the barrier, the Color statistics of electrons are 100% white.
(B) Path s? No! (Same reason.) Electrons that take path s are soft, so their color statistics
are 50/50. But without the barrier, the Color statistics are 100% white.
4
SO: What can we say about a white electron in our device?
One option: We can say Hardness is an indeterminate property of white electrons. A white
electron is in a superposition of being hard and soft.
A white electron just is an electron in a superposition of hard and soft.
or
Moral: Experiments suggest that quantum properties are very different from
classical properties. So the mathematical description of quantum
properties will have to be different from that of classical properties.
How to Describe Physical Phenomena: 5 Essential Notions
Lets get just a bit more general and then a bit more specific:
(a) Physical system
Classical example: a baseball Quantum example: an electron
(b) State of a physical system: A description of the system at an instant in time in terms of its
properties.
Classical examples:
(i) A baseball moving at 95 mph, 5 feet from batter.
(ii) A baseball moving at 50 mph, dropping over left-field wall.
(d) State space of a physical system: The collection of all possible states of a physical system.
(c) Properties of a physical system
Classical examples:
momentum
position
energy
Quantum examples:
(i) An electron entering a Hardness box.
(ii) An electron exiting a Hardness box.
Quantum examples:
hardness (spin along certain direction)
color (spin along another direction)
momentum
position
energy
(e) Dynamics of a physical system: Description of how the states of the system change with time.
ex: Newtonian dynamics (Newtons equations of motion)
- describe how the states of a classical system change with time
Given an initial state (baseball leaving the pitchers hand with given momentum
and position), what is the state of the baseball at a later time (as it crosses
homeplate; as a given force is imparted on it; as it soars over the left-field wall)?
5
Mathematical Description of Classical Physical System
Baseball example:
(i) A state of the baseball is specified by giving the baseballs momentum (its mass times its
velocity) and its position. (Newtons equations of motion uniquely determine the way such
states change in time.) To specify the baseballs momentum at any given time requires 3
numbers p
1
, p
2
, p
3
(its velocity has 3 components); to specify its position also requires 3
numbers q
1
, q
2
, q
3
. (So the baseball has 6 degrees of freedom.)
(ii) The state space of the baseball can thus be represented by a 6-dimensional set of points
(called the phase space of the system):

Each point has 6 numbers

associated with it and represents
a state of the baseball.
(iii) All dynamical properties of the baseball can be represented by functions of its momentum p
i
and position q
i
(i = 1, 2, 3), and hence by functions on the phase space. These functions are
in principle always well-defined* for any point in phase space.
Ex: baseballs energy = E(p
i
, q
i
) = (p
1
2
+ p
2
2
+ p
3
2
)/2m
* Provided the phase
space is well-behaved
topologically, which is
normally the case.
Will such a mathematical description work for quantum mechanics?
NO! Experiments suggest that quantum properties are not always well-defined. So we cant
represent them mathematically as functions on a set of points.
(iv) Newtons equations of motion, which provide the dynamics for classical mechanics, can be
reformulated as equations that act on states in phase space. The result is the Hamiltonian
formulation of classical mechanics.
baseballs
mass
What mathematical representation should we use?
Classical mechanics Quantum mechanics
state space set of points (phase space) vector space
states points vectors
properties functions of points operators on vectors
6
II. 2-State Quantum Systems
Restrict attention to quantum properties with only two values (like Hardness and Color).
I.e., quantum systems with 2 degrees of freedom.
Associated state vectors are 2-dimensional:
|Q!
|0!
|1!
component of |Q! along |0!
component of
|Q! along |1!
|Q! = a|0! + b|1!
Require state vector |Q! to
have unit length:
|a|
2
+ |b|
2
= 1
Collection of all vectors decomposible in basis {|0!, |1!} forms a vector space H.
Why this is supposed to help:
Recall: Hardness seems to be an indeterminate property of white electrons.
|white!
|black!
|hard!
|soft!
Suppose Hardness basis {|hard!, |soft!} is rotated by 45 w.r.t. Color basis {|white!, |black!}:
An electron in a
white state...
... is in a superposition
of hard and soft states.
|white! = "!|hard! + "!|soft!
Then:
1. States as vectors
Standard Interpretation: A quantum system possesses a property if and only if
it is in a state associated with that property (an eigenstate).
Moreover: The probability that the white electron will be hard/soft when its Hardness is
measured is given by the square of the appropriate expansion coefficient: ("!)
2
= !. (Born Rule.)
=
O
11
a + O
12
b
O
21
a + O
22
b
O|Q! =
O
11
O
12
O
21
O
22
a
b
|Q! =
a
b
2-dim vector as 2 ! 1 matrix
O =
O
11
O
12
O
21
O
22
Operator on 2-dim vectors as 2 ! 2 matrix
2. Properties as operators
A linear operator on a vector space = An object that eats a vector and spits out another
vector.
Matrix representation:
7
3. Entangled States
Consider two quantum 2-state systems H
1
, H
2
. Combined system is represented by H
1
# H
2
.
Suppose:
{|0!
1
, |1!
1
} is a basis for H
1
and {|0!
2
, |1!
2
} is a basis for H
2
Then:
{|0!
1
|0!
2
, |0!
1
|1!
2
, |1!
1
|0!
2
, |1!
1
|1!
2
} is a basis for H
1
# H
2
An entangled state is a state in H
1
# H
2
whose expansion does not contain factors of
the component basis vectors.
Examples:
Entangled state: |"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
)
Nonentangled state: |Q! = "!(|0!
1
|0!
2
+ |1!
1
|0!
2
) = "!(|0!
1
+ |1!
1
)|0!
2
)
Standard Interpretation:
In state |"
+
!, System #1 has no determinate property, System #2 has no determinate
property, but the combined system as a whole does have determinate properties!
In state |Q!, System #1 has no determinate property, but System #2 does have a determinate
property (as does the combined system as a whole).
8
III. Quantum Information Theory
Classical Information Theory
C-bit = Either of the two states, 0 or 1, of a classical 2-state system.
Quantum Information Theory
Qubit = Either of the two states, |0! or |1!, of a quantum 2-state system.
Most general form of a qubit:
|Q! = a|0! + b|1! where |a|
2
+ |b|
2
= 1
Important Characteristic of Qubits:
A qubit |Q! = a|0! + b|1! encodes an arbitrarily large amount of information (contained in its
expansion coefficients a and b). But at most only one classical bits worth of information in a
qubit is accessible.
1. C-bits vs. Qubits
2. Transformations on Single Qubits
|0! =
1
0
|1! =
0
1
I =
1 0
0 1
Identity Negation Phase-change Negation/Phase-change
I|0! = |0!
I|1! = |1!
X|0! = |1!
X|1! = |0!
Y|0! = #|1!
Y|1! = |0!
Z|0! = |0!
Z|1! = #|1!
X =
0 1
1 0
Y =
0 1
#1 0
Z =
1 0
0 #1
H =
"! "!
"! #"!
Hadamard operator
H|0! = "!|0! + "!|1!
H|1! = "!|0! # "!|1!
Takes a basis qubit and outputs a superposition
9
3. Transformations on Two Qubits
C
NOT
|0!
1
|0!
2
= |0!
1
|0!
2
C
NOT
|1!
1
|0!
2
= |1!
1
|1!
2
C
NOT
|0!
1
|1!
2
= |0!
1
|1!
2
C
NOT
|1!
1
|1!
2
= |1!
1
|0!
2
Acts on two qubits:
changes the second if the first is |1!
leaves the second unchanged otherwise
1
0
0
0
|0!
1
|0!
2
= |0!
1
|1!
2
= |1!
1
|0!
2
= |1!
1
|1!
2
=
1 0 0 0
0 1 0 0
0 0 0 1
0 0 1 0
C
NOT
=
Controlled-NOT operator
0
1
0
0
0
0
1
0
0
0
0
1
Claim: Unknown or non-orthogonal qubits cannot be cloned.
More precisely: There is no (unitary) operator U that acts on a 2-qubit state in the
following way:
U(|v!
1
|0!
2
) = |v!
1
|v!
2
where |v!
1
is any arbitrary qubit.
What this means:
If you have a known state, say |a!
1
, you can always find a (unitary) operator U that will clone it
in the sense of creating a duplicate while leaving the original unchanged.
Ex. C
NOT
|1!
1
|0!
2
= |1!
1
|1!
2
And U will be able to clone any state that is orthogonal to |a!
1
.
BUT: Your U will not be able to clone any state that is non-orthogonal to |a!
1
.
4. No-Cloning Theorem
Proof: Suppose there is such a U. Then for any qubits |a!
1
and |b!
1
:
U(|a!
1
|0!
2
) = |a!
1
|a!
2
U(|b!
1
|0!
2
) = |b!
1
|b!
2
Now consider the non-orthogonal qubit |c!
1
= "!{|a!
1
+ |b!
1
}. Linearity entails:
U(|c!
1
|0!
2
) = "!U{|a!
1
|0!
2
+ |b!
1
|0!
2
} = "!{|a!
1
|a!
2
+ |b!
1
|b!
2
}
But, by definition, U acts on |c!
1
according to:
U(|c!
1
|0!
2
) = |c!
1
|c!
2
= !{|a!
1
|a!
2
+ |a!
1
|b!
2
+ |b!
1
|a!
2
+ |b!
1
|b!
2
}
So there is no such U.
6. Quantum Dense Coding
10
Goal: To use one qubit to transmit two classical bits.
Recall: One qubit only contains one classical bits worth of information. So how can we send 2
classical bits using just one qubit? Use entangled states!
Set-Up:
Two qubits are prepared in an entangled state |"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
).
Alice gets qubit #1, Bob gets qubit #2.
Alice manipulates her qubit #1 so that it steers Bobs qubit #2 into a state from which he can
read off the 2 classical bits Alice desires to send.
Encoder Decoder
EPR
source
Alice Bob
2 classical
bits decoded
2 classical
bits encoded
1 qubit sent
qubit #1
to Alice
qubit #2
to Bob
I, X, Y, Z BSM
|"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
Protocol
1. Alice has a number -- either 0, 1, 2, 3 -- that is encoded in 2 classical bits -- either 00, 10, 01,
11. To send it to Bob, she first encodes it in her qubit #1 by performing one of the
transformations {I, X, Y, Z} on her qubit #1 according to the following table:
Note: These are called
Bell state vectors. They
form a basis for the 2-
qubit vector space.
if number is: perform transform: to get new state:
0 (I
1
# I
2
)|"
+
! "!(|0!
1
|0!
2
+ |1!
1
|1!
2
) = |"
+
!
1 (X
1
# I
2
)|"
+
! "!(|1!
1
|0!
2
+ |0!
1
|1!
2
) = |$
+
!
2 (Y
1
# I
2
)|"
+
! "!(#|1!
1
|0!
2
+ |0!
1
|1!
2
) = |$
#
!
3 (Z
1
# I
2
)|"
+
! "!(|0!
1
|0!
2
# |1!
1
|1!
2
) = |"
#
!
2. Alice now sends her qubit #1 to Bob.
5. To distinguish between the choices in (4i) and (4ii), Bob now applies a Hadamard
transformation to qubit #1:
6. Qubit #1 now has a definite value in each of these states. Bob can now measure qubit #1 to
know exactly what number Alice sent him:
(i) In case (4i), if the outcome is |0!
1
, then Alice sent him 0. If the outcome is |1!
1
, then Alice sent him 3.
(ii) In case (4ii), if the outcome is |0!
1
, then Alice sent him 1. If the outcome is |1!
1
, then Alice sent him 2.
11
4. In each of these entangled states, qubit #1 has no definite value, but qubit #2 does. Bob can
hence measure qubit #2 with two possible results:
Were talking abstract qubit talk
here. More concretely, think of qubit
#2 as either a hard or soft electron.
Bob measures it for its Hardness.
Question: How are the 2 classical bits transferred from Alice to Bob?
Answer: They are not transferred via the single qubit (since it can only encode a bits worth of accessible info)!
Rather, they are transferred by the correlations present in the 2-qubit entangled state. This may seem strange if you
suppose that, in order to convey information between Alice and Bob, it must be physically transported from Alice to
Bob across the intervening spatial distance. But, from an information-theoretic point of view, the only thing required
to convey information is to set up a correlation between the senders data and the receivers data. It doesnt really
matter how this correlation is established.
Note: The procedure that Bob goes through to decode the entangled state is sometimes called a Bell State
Measurement (or BSM -- hence the diagram). In effect, Bob has a Bell state (one of four possible) and he performs
transformations (namely, C
NOT
and H) and measurements on it in order to extract Alices encoded information from it.
3. After reception of qubit #1, Bob now must decode it to extract Alices number (her 2
classical bits). To do this, Bob first applies a C
NOT
transformation to both qubits #1 and #2:
if 2-qubit state is: C
NOT
transform yields:
|"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
) "!(|0!
1
+ |1!
1
)|0!
2
|$
+
! = "!(|1!
1
|0!
2
+ |0!
1
|1!
2
) "!(|1!
1
+ |0!
1
)|1!
2
|$
#
! = "!(#|1!
1
|0!
2
+ |0!
1
|1!
2
) "!(#|1!
1
+ |0!
1
)|1!
2
|"
#
! = "!(|0!
1
|0!
2
# |1!
1
|1!
2
) "!(|0!
1
# |1!
1
)|0!
2
example:
C
NOT
|"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
)
= "!(|0!
1
|0!
2
+ |1!
1
|0!
2
)
= "!(|0!
1
+ |1!
1
)|0!
2
Recall: C
NOT
acts on two qubits and
changes the second just when the first is |1!.
2-qubit state: C
NOT
on 2-qubit state: H
1
on C
NOT
state:
|"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
) "!(|0!
1
+ |1!
1
)|0!
2
|0!
1
|0!
2
|$
+
! = "!(|1!
1
|0!
2
+ |0!
1
|1!
2
) "!(|1!
1
+ |0!
1
)|1!
2
|0!
1
|1!
2
|$
#
! = "!(#|1!
1
|0!
2
+ |0!
1
|1!
2
) "!(#|1!
1
+ |0!
1
)|1!
2
|1!
1
|1!
2
|"
#
! = "!(|0!
1
|0!
2
# |1!
1
|1!
2
) "!(|0!
1
# |1!
1
)|0!
2
|1!
1
|0!
2
example:
(H
1
# I
2
)("!(|0!
1
+ |1!
1
)|0!
2
)
= "!{"!(|0!
1
+ |1!
1
) + "!(|0!
1
# |1!
1
)}|0!
2
= (!|0!
1
+ !|0!
1
+ !|1!
1
# !|1!
1
)|0!
2
= |0!
1
|0!
2
(i) The outcome is |0!
2
. This means Alice send either 0 or 3.
(ii) The outcome is |1!
2
. This means Alice sent either 1 or 2.
7. Quantum Teleportation
Goal: To transmit an unknown quantum state of a system using classical bits and to reconstruct
the exact quantum state at the receiver.
12
Set-Up: Alice has an unknown qubit (qubit #0) |Q!
0
= a|0!
0
+ b|1!
0
and wants to send it to Bob.
Two qubits (#1 and #2) are prepared in an entangled state |"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
). Alice
gets qubit #1, Bob gets qubit #2. Alice manipulates her qubits #0 and #1 so that they steer
Bobs qubit #2 into the unknown state of #0.
Protocol
1. Alice starts with a 3-qubit system (qubits #0, #1, #2) in the state:
Alice Bob
unknown qubit #0
reconstructed
unknown
qubit #0
2 classical bits sent
qubit #1
to Alice
qubit #2
to Bob
EPR
source
Decoder
I, X, Y, Z
Encoder
BSM
|"
+
! = "!(|0!
1
|0!
2
+ |1!
1
|1!
2
|Q!
0
|"
+
! = "!{a|0!
0
|0!
1
|0!
2
+ a|0!
0
|1!
1
|1!
2
+ b|1!
0
|0!
1
|0!
2
+ b|1!
0
|1!
1
|1!
2
}
Alice now applies a C
NOT
transformation on qubits #0 & #1, and then a
Hadamard transformation on qubit #0:
First C
NOT
on #0 & #1:
(C
NOT
01
# I
2
)|Q!
0
|"
+
! = "!{a|0!
0
|0!
1
|0!
2
+ a|0!
0
|1!
1
|1!
2
+ b|1!
0
|1!
1
|0!
2
+ b|1!
0
|0!
1
|1!
2
}
Now H on #0:
(H
0
# I
2
# I
3
){ } = "!{a{"!(|0!
0
+ |1!
0
)}|0!
1
|0!
2
+ a{"!(|0!
0
+ |1!
0
)}|1!
1
|1!
2
+ b{"!(|0!
0
! |1!
0
)}|1!
1
|0!
2

+ b{"!(|0!
0
! |1!
0
)}|0!
1
|1!
2
}
= !|0!
0
|0!
1
(a|0!
2
+ b|1!
2
) + !|0!
0
|1!
1
(a|1!
2
+ b|0!
2
) + !|1!
0
|0!
1
(a|0!
2
# b|1!
2
) + !|1!
0
|1!
1
(a|1!
2
# b|0!
2
)
This is just the
BSM procedure
in Dense Coding!
Note: In the last line, each of the terms represents a state
in which qubits #0 and #1 have definite values, but qubit
#2 does not (according to the standard interpretation).
2. Alice now measures the 2-qubit sub-system of qubits #0 and #1 with one of 4 outcomes:
if measurement outcome is: qubit #2 is now in state:
|0!
0
|0!
1
a|0!
2
+ b|1!
2
|0!
0
|1!
1
a|1!
2
+ b|0!
2
|1!
0
|0!
1
a|0!
2
# b|1!
2
|1!
0
|1!
1
a|1!
2
# b|0!
2
3. Alice sends result of her measurement to Bob in the form of 2 classical bits: 00, 01, 10, or 11.
4. Depending on what classical bits he receives, Bob performs one of the transformations {I, X, Y,
Z} on his qubit #2. This allows him to turn it into (construct) the unknown qubit #0.
00 a|0!
2
+ b|1!
2
I
3
01 a|1!
2
+ b|0!
2
X
3
10 a|0!
2
# b|1!
2
Z
3
11 a|1!
2
# b|0!
2
Y
3
if bits
received are:
then Bobs qubit
#2 is now in state:
so to reconstruct
qubit #0, use:
13
Questions:
1. Doesnt Bob violate the No-Cloning Theorem? Doesnt he construct a clone of the
unknown qubit #0?
Answer: No violation occurs. Bob does construct a copy, in so far as his qubit #2 is now a duplicate (in all
respects) of Alices qubit #0. But Alices qubit #0 is no longer its original self anymore. After Alice is
through transforming it (and qubit #1) and measuring the system at Step 2, the original qubit #0 has now
collapsed to either |0!
0
or |1!
0
! Alice destroys qubit #0 in the process of conveying the information
contained in it to Bob! So the No-Cloning Theorem is not violated.
2. How does Bob reconstruct the unknown qubit #0 (that encodes an arbitrarily large amount of
information) from just 2 classical bits?
Answer: By using the 2 classical bits to determine the appropriate transformation on qubit #2! Thats all
there is to it. One might be puzzled if one thought (erroneously) that, in order to convey information from
Alice to Bob, it must be physically transported across the intervening spatial distance, and the only thing
being conveyed are the 2 classical bits. But, again, this is the wrong way to think of information.