Professional Documents
Culture Documents
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
(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') = 1/2
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):