Synthese (2013) 190:3209–3226

DOI 10.1007/s11229-012-0138-3

The pessimistic induction: a bad argument gone

too far

Moti Mizrahi

Received: 10 February 2012 / Accepted: 16 June 2012 / Published online: 30 June 2012
© Springer Science+Business Media B.V. 2012

Abstract In this paper, I consider the pessimistic induction construed as a deductive

argument (specifically, reductio ad absurdum) and as an inductive argument (specifi-
cally, inductive generalization). I argue that both formulations of the pessimistic induc-
tion are fallacious. I also consider another possible interpretation of the pessimistic
induction, namely, as pointing to counterexamples to the scientific realist’s thesis that
success is a reliable mark of (approximate) truth. I argue that this interpretation of the
pessimistic induction fails, too. If this is correct, then the pessimistic induction is an
utter failure that should be abandoned by scientific anti-realists.

Keywords Anti-realism · Inductive generalization · Pessimistic induction ·

Scientific realism

1 Introduction

In the context of the scientific realism/anti-realism debate, scientific realism is taken

to include one or more of the following theses (Psillos 2006, p. 135):
(MSR) The Metaphysical Thesis: The world has a definite and mind-independent
structure.
(SSR) The Semantic Thesis: Scientific theories are truth-conditioned descriptions
of their intended domain. Hence, they are capable of being true or false. The
theoretical terms featuring in theories have putative factual reference. So if
scientific theories are true, the unobservable entities they posit populate the
world.

M. Mizrahi (B)
Philosophy Department, St. John’s University, 8000 Utopia Parkway, Queens, NY 11439, USA
e-mail: [email protected]

123
3210 Synthese (2013) 190:3209–3226

(ESR) The Epistemic Thesis: Mature and predictively successful scientific theories
are well-confirmed and approximately true. So entities posited by them, or,
at any rate entities very similar to those posited, inhabit the world (Cf. Psillos
1999 and Boyd 1989).
The main argument against (ESR), known as the “pessimistic induction,” the “pes-
simistic meta-induction,” or the “disastrous historical meta-induction,” has different
versions that differ in their details (see, e.g., Poincaré 1952, p. 160; Putnam 1978,
p. 25; Laudan 1981). Generally speaking, the argument begins by recalling that many
scientific theories were once successful (in terms of their explanatory and predictive
power). However, the pessimist argues, most of these past theories are now considered
strictly false. Therefore, the pessimist concludes, current successful theories will turn
out to be false as well.
This anti-realist argument has generated an enormous literature (see, e.g., Psillos
1999, pp. 102–103) and it still plays a prominent role in the scientific realism debate
(see, e.g., the recent issue of the Journal for General Philosophy of Science, 42, 2011).
For instance, Dopplet (2011, p. 295) argues that any realist account must “overcome
the pessimistic meta-induction.” (See also Devitt 2011.) As Wray (2011) recently
pointed out, however, there seems to be little agreement about the form the pessimis-
tic induction is supposed to take. Some have construed this argument as a reductio of
the view that the apparent success of the best scientific theories of mature sciences
justifies belief in the approximate truth of these theories or in the existence of the unob-
servables they posit, i.e., as a reductio of (ESR). For example, Lewis (2001, p. 372)
reconstructs the argument as follows:
(1) Assume that the success of a theory is a reliable test for its truth.
(2) Most current scientific theories are successful.
(3) So most current scientific theories are true.
(4) Then most past scientific theories are false, since they differ from current theories
in significant ways.
(5) Many of these false past theories were successful.
(6) So the success of a theory is not a reliable test for its truth. (Cf. Psillos 1996;
Saatsi 2005.)
Similarly, Lange (2002, p. 282) argues that the pessimistic induction “takes the form
of a reductio of the view that the apparent success of some scientific theory justifies
our believing in its accuracy regarding unobservables.”
More recently, Wray (2011) has argued that to reconstruct the pessimistic induc-
tion as a reductio is a mistake. Wray (2011) says that “the argument purports to be an
inductive argument” (original emphasis; see also Held (2011)). He then elaborates as
follows (Wray 2011):
I believe that from a survey of the history of science, Laudan infers that most
past successful theories have turned out to be false. This claim then figures as a
premise in an argument whose conclusion is that our current best theories are
probably false (original emphasis).
In this paper, I examine the pessimistic induction construed as a deductive argu-
ment (specifically, reductio) and as an inductive argument (specifically, inductive

123
Synthese (2013) 190:3209–3226 3211

generalization). My overall argument has two stages. First, I will argue that the pessi-
mistic induction is a fallacious argument as follows:

(P1) Either the pessimistic induction is a deductive argument or the pessimistic induc-
tion is an inductive argument.
(P2) Construed as a deductive argument (i.e., reductio), the pessimistic induction is
an invalid argument.
(P3) Construed as an inductive argument (i.e., inductive generalization), the pessi-
mistic induction is a weak argument.
(C) Therefore, the pessimistic induction is a fallacious argument.

Then I will argue that the pessimistic induction also fails when understood as pointing
to counterexamples to the scientific realist’s thesis that success is a mark of (approxi-
mate) truth, since the alleged counterexamples miss their intended target.

2 The pessimistic induction as a deductive argument

In this section, I will not rehearse the familiar objections against the reductio formu-
lation of the pessimistic induction, which are well known to readers of this journal
(see, e.g., Lewis 2001 and Lange 2002). Instead, I would like to examine more closely
steps (3) and (4) of Lewis’ reductio formulation of the pessimistic induction quoted
above. These steps in the reasoning can be unpacked as follows:
(3 ) Most current scientific theories are true.
(4 ) Most past scientific theories differ from most current scientific theories in sig-
nificant ways.
(4 ) Therefore, most past scientific theories are false.
Clearly, from the fact that past and current theories differ in “significant ways,” it does
not necessarily follow that past and current theories must have different truth values.
Past theories and current theories can differ in significant ways, and yet have the
same truth value. For example, consider a past scientific theory, such as Georg Ernst
Stahl’s theory of combustion, according to which a “fire principle” called phlogiston
is expelled from metals when they are heated (which accounts for the fact that some
metals gained in weight when they were burned), and a current scientific theory, such
as string theory. Surely, these two theories differ in significant ways. One was pro-
posed by one physician, whereas the other is the result of the work of many physicists
and mathematicians. One has only one formulation, whereas the other has many. One
is supposed to be a theory about combustion, whereas the other is supposed to be a
theory of everything (i.e., a theory of the fundamental forces and particles). One had
some experimental backing at the time (e.g., the ash of a metal like lead—called a
calx—weighed more than the original metal), whereas the other currently has none.
Even though these two theories differ in significant ways, string theory could still turn
out to be false, just as Stahl’s combustion theory did.
Or consider a past scientific theory, such as William Harvey’s theory of blood circu-
lation, according to which the heart pumps blood through the arteries to the veins and
back to the heart in a continuous circulation (which accounts for the fact that, when a

123
3212 Synthese (2013) 190:3209–3226

corpse is dissected, the arteries are nearly empty, whereas the veins may still contain
some blood), and a current scientific theory, such as string theory. Surely, these two
theories differ in significant ways. One was proposed by one physician, whereas the
other is the result of the work of many physicists and mathematicians. One has only
one formulation, whereas the other has many. One is supposed to be a theory about
blood circulation, whereas the other is supposed to be a theory of everything (i.e., a
theory of the fundamental forces and particles). One had some experimental backing
at the time (e.g., the relation of the heart’s contractions to the pulsing of blood into the
arteries), whereas the other currently has none. Even though these two theories differ
in significant ways, string theory could still turn out to be true, just as Harvey’s theory
of blood circulation did.
The point, then, is this: from the fact that past and current theories differ in signif-
icant ways, it does not necessarily follow that they must have different truth values,
regardless of whether past theories were true or false. So a deductive argument with
(3 ) and (4 ) as premises and (4 ) as its conclusion is an invalid argument. If this is
correct, then the reductio formulation of the pessimistic induction, which is supposed
to be a reductio, and hence a deductive argument, is an invalid argument.
Pessimists might respond to this charge by trying to clarify the ambiguous phrase
‘differ in significant ways’ in Lewis’ reductio formulation of the pessimistic induction.
They might claim that ‘differ in significant ways’ means “differ in truth value.” With
this clarification, then, the aforementioned argument can be reformulated as follows:

(3 ) Most current scientific theories are true.

(4 ) Most past scientific theories differ from most current scientific theories in their
truth value (i.e., most past scientific theories are false).
(4 ) Therefore, most past scientific theories are false.

However, this argument seems circular. It purports to establish that most past sci-
entific theories are false, but it assumes as one of its premises, i.e., (4 ), that most past
theories are false, since ‘differ in significant ways’ here just means “having a differ-
ent truth value.” Of course, the question is precisely whether most past theories are
false—a claim that scientific realists will undoubtedly deny—and so pessimists cannot
assume that claim as a premise in their argument against scientific realism without
begging the question. Pessimists must provide an independent reason for thinking that
(4 ) is true, rather than assume a paraphrase of (4 ), i.e., (4 ), as a premise in an
argument that purports to establish that (4 ) is true.
Another possible formulation of these crucial steps in Lewis’ reductio formulation
of the pessimistic induction quoted above is the following:
(3 ) Most current scientific theories are true.
(4 ) If most current theories are true, then most past scientific theories are false.
(4 ) Therefore, most past scientific theories are false.
But what reasons do we have for accepting (4 )? One might think that, for any aban-
doned theory T , assuming our current theories to be approximately true is just to
assume that T is false. In other words, in order to show that (4 ) is true, pessimists
could argue as follows:

123
Synthese (2013) 190:3209–3226 3213

(V1) Scientists abandon theory T1 in favor of T2 only if they think that T1 is false
and T2 is true.
(V2) T1 has been abandoned by scientists in favor of T2 .
(V3) Therefore, T1 is considered false and T2 is considered true by scientists.
If this argument were sound, it would show that, of two competing theories, the one that
is currently accepted is considered true by scientists, and the one that was abandoned
is considered false by scientists. In other words, this argument would lend support to
(4 ). Then, based on this argument, pessimists could argue that, since the “history
of science is a graveyard of [abandoned] theories” (Lipton 2005, p. 1265), there is a
good reason to think that most of our current theories will end up in that graveyard.
However, this argument depends on the assumption that abandonment of a theory
T by the scientific community is an indication that T is considered false. But why
should we assume that? As a general claim about scientific change, (V1) seems false,
since it is open to counterexamples. For example, Newlands’ law of octaves has been
abandoned in favor of the periodic law, not because it was strictly false, but because
it did not apply to elements of higher atomic weights. Both laws share the essentially
correct insight that the properties of the elements are a function of their atomic weight
(Caven 1921, p. 69).
To this pessimists might reply that examples such as Newlands’ law are rare, and so
(V1) is likely to be generally true. In other words, pessimists might argue as follows:
(V1*) Usually, when scientists abandon theory T1 in favor of T2 it is because they
think that T1 is false and T2 is true.
(V2*) T1 has been abandoned by scientists in favor of T2 .
(V3*) Therefore, (probably) T1 is considered false and T2 is considered true by
scientists.
I do not think that Newlands’ law is a rare example from the history of science of
a law that was abandoned, not because it was considered strictly false, but because
it was not comprehensive enough. But even if it is a rare example, it is enough to
break the relation of entailment between abandonment of a theory and strict falsity as
a reason for that abandonment. Pessimists need this relation between abandonment of
a theory and strict falsity to be a relation of entailment for the reductio formulation
of the pessimistic induction to be a valid argument. If pessimists want to abandon the
reductio formulation and construe the pessimistic induction as an inductive argument
instead, then it seems that we need a larger and more diverse sample of theories in
order to examine the proportion of current accepted theories to past competitors that
were abandoned. I will consider such a sample in Sect. 3.
Pessimists might also try to appeal to inference to the best explanation (IBE is not
a deductive form of inference; Psillos 2007, pp. 442–443) to support the claim that
most past scientific theories are false and argue as follows:
(I1) The history of science is a graveyard of abandoned theories.
(I2) The best explanation for (I1) is that abandoned theories are strictly false.
(I3) No other hypothesis can explain (I1) as well as strict falsity does.
(I4) Therefore, abandoned theories are strictly false.
Then, based on this argument, pessimists could argue that this historical trend will
continue, and so our current theories will end up in the graveyard as well. (I recognize

123
3214 Synthese (2013) 190:3209–3226

that some constructive empiricists might not accept this argument, since it is an infer-
ence to the best explanation, even if they endorse the pessimistic induction. See, e.g.,
van Fraassen 1980, p. 143; van Fraassen 1989, pp. 131–150; Muller 2008, p. 143).
It is worth noting here a rather peculiar feature of the scientific realism debate. Pes-
simists look at the historical record of science and see failure, whereas realists look at
the same historical record and see success (particularly, predictive success). Putting
aside the question of whether the track record of science is really as bad as pessimists
claim, or as good as realists claim, the important question for present purposes is
whether (I3) is true. Again, it seems that strict falsity is not always the best explana-
tion for why a theory has been abandoned. For example, one might reasonably think
that Wegener’s theory of continental drift was replaced by the theory of plate tectonics,
not because the former was strictly false, but rather because the latter provides a more
comprehensive account of certain geophysical phenomena. In other words, there are
many reasons why theories are abandoned or replaced by other theories, and strict
falsity is not always the best explanation for such scientific change.
If this is correct, then, in order to show that (3 ) and (4 ) entail (4 ), which must
be the case if Lewis’ reductio formulation of the pessimistic induction is to be a
valid argument, pessimists actually have to show that strict falsity is the only expla-
nation (Bird 2007b) for the abandonment of a theory. To show that strict falsity is
the only explanation for the abandonment of a theory, pessimists have to eliminate
all the alternative explanations for the abandonment of a theory, such as that the
theory is not comprehensive enough, that the theory is not well-confirmed, that the
theory is incomplete, that the theory has no forceful and influential proponents, and
so on. These alternative explanations haven’t been eliminated. In fact, in the case of
Wegener’s theory and Newlands’ law, the comprehensiveness explanation seems more
plausible than the strict falsity explanation.
Finally, pessimists might try to clarify the ambiguous phrase ‘differ in significant
ways’ by saying that ‘significant ways’ means that current theory T1 is incompatible
with past theory T2 , where T1 and T2 are competing theories about the same domain in
nature. For example, Stahl’s theory of combustion is incompatible with Lavoisier’s the-
ory of combustion, since they postulate the existence of different theoretical entities,
phlogiston and oxygen, respectively. So, if current theories are true, their competing
past theories must be false.
There are several problems with this pessimistic move. First, it is not obvious
that competing theories that postulate different theoretical entities must be (logically)
incompatible. For example, suppose that Stahl’s theory of combustion is true, and so
there is phlogiston, which is the substance that is responsible for combustion. But it
could still be the case that there is oxygen as well. In fact, it might actually be a useful
addition to Stahl’s theory of combustion, since one of its problems was to explain
why, when some metals were calcined, the resulting calx was heavier than the original
metal. Some tried to explain this by saying that, in some metals, phlogiston has nega-
tive weight. Instead, they could have said that phlogiston is lost but oxygen is gained,
which would explain the heavier weight. Second, even if it is granted to pessimists
that current theories are incompatible with their competing past theories, it does not
follow that most past theories are false. For that conclusion to follow, pessimists have
to assume that current theories are true. But this assumption would be inconsistent

123
Synthese (2013) 190:3209–3226 3215

with the conclusion of the pessimistic induction, which purports to show that current
theories are false (Cf. Devitt 2011, p. 288). Third, for this pessimistic move to work,
pessimists have to assume that every current theory has an actual corresponding past
theory that is in direct competition with it. I do not think that this assumption is true.
Consider, for example, Ehrlich’s side-chain theory. What is the past theory that is
incompatible with Ehrlich’s side-chain theory? Or consider Boyle’s gas law. What is
the past theory that is incompatible with Boyle’s law? Or consider Guth’s inflation
theory. What is the past theory that is incompatible with it? The point, then, is that not
every theory has a past competitor that is incompatible with it.
Pessimists might reply by saying that most current theories have past competitors,
which is compatible with the claim that some current theories do not have past com-
petitors. The question, then, becomes what is the proportion of current theories that are
accepted theories to past competitors that were abandoned. To answer this question, it
seems that we need a larger and more diverse sample of theories. I will examine such
a sample in Sect. 3.
If the aforementioned considerations are correct, then it seems that the reductio for-
mulation of the pessimistic induction is an invalid argument. Perhaps roughly similar
considerations have led Saatsi (2005, p. 1092) to argue that the following statistical
argument should be added as a supplement to Lewis’ reductio formulation of the
pessimistic induction:
(1*) Of all the successful theories, current and past, most are taken to be false by the
current lights.1
(2*) The current theories are essentially no different from the past successful the-
ories with respect to their “observable” properties. (Viz. properties potentially
figuring in the realist’s explanatory argument.)
(3*) Success of a current theory is not a reliable indicator of its truth (by the reduction
argument above), and there is no other reliable indicator of truth for the current
theories.
(4*) Therefore any current successful theory is probably false by statistical reasoning.
For present purposes, I would like to make two points about Saatsi’s statistical argu-
ment. First, notice the tension between premise (4) in Lewis’ reductio formulation of
the pessimistic induction and premise (2*) of Saatsi’s statistical argument. The former
says that past theories differ from current theories in significant ways, whereas the
latter says that current theories are essentially no different from past theories. This
seems to be a problem if, as Saatsi (2005, p. 1092) argues, his statistical argument
is supposed to supplement Lewis’ reductio formulation of the pessimistic induction.
Second, why think that the statistical premise (1*) is true? Presumably, it is based on a
sample of theories. But what is that sample? Is it a representative sample of successful
theories?
In the next section, I show why the pessimistic induction is a weak inductive argu-
ment. To do so, I discuss Godfrey-Smith’s account of random sampling as a form

1 It seems that (1*) should read as follows: “Of all the past successful theories, most are taken to be false
by the current lights.” After all, it seems to be inconsistent to say that current successful theories are taken to
be false by the current lights, for the current light are supposed to be current theories. Thanks to a perceptive
reviewer for noticing this inconsistency.

123
3216 Synthese (2013) 190:3209–3226

of inductive inference that is in-principle justifiable. I then argue that the pessimistic
induction fails to measure up to the standards of this kind of inductive inference, since
it does not provide what Godfrey-Smith calls “a bridge from observed to unobserved.”
If this is correct, then the pessimistic induction is a weak inductive argument.

3 The pessimistic induction as an inductive argument

To see why the pessimistic induction is a weak inductive argument, we need to under-
stand the sort of inductive argument that the pessimistic induction is supposed to be.
Godfrey-Smith (2011, p. 34) argues that “the familiar philosophical concept of ‘induc-
tion’ has conflated two kinds of inference, each of which is successfully exploited by
science” (Cf. Okasha 2001). As he (2011, p. 34) writes:
For each of the two inference patterns, an account can be given of its in-princi-
ple reliability. That account is a kind of philosophical justification. The package
usually known as “induction” does not have that kind of justification, however.
It combines elements from each method without combining parts that give rise
to an in-principle reliable combination in its own right.
Godfrey-Smith is concerned with the pattern of argument used to answer questions
of the form ‘How Many Fs are G?’ This is a question about proportion or frequency,
which could be expressed by asking ‘What is the rate of G in the Fs?’ For example
(2011, p. 33):
1. How many teenagers smoke?
2. How many ravens are black?
3. How many emeralds are green?
4. How many people in this room are third sons?
5. How many organisms have the amount of bases cytosine and guanine equal, and
the amount of adenine and thymine equal, in their DNA?
6. How many electrons have charge of approximately—1.6 × 10−19 coulombs?
In induction, these questions are answered by making a generalization based on the
relation between F and G in the observed cases. So the number of observed Fs is
supposed to have epistemic significance. The classic case of inductive inference is thus
the one where all observed Fs are G, which is then used to infer that all unobserved
Fs are also G.
One form of inductive inference that is justifiable in principle, according to God-
frey-Smith, is inference based on random sampling. Random sampling means that
“every member of the population you are drawing conclusions about has the same
chance of making its way into the sample.” He (2011, p. 40) argues that, in these
sampling cases, “the power of randomness is what gives us a ‘bridge’ from observed
to unobserved.” For example, if we want to know how many teenagers smoke, then we
collect a random sample of teenagers, measure the rate of smoking in the sample, and
then extrapolate to the teenager population. Why should we follow this method rather
than any other? Because we have a statistical model that tells us why this method is
reliable in principle. According to Godfrey-Smith (2011, p. 39):

123
Synthese (2013) 190:3209–3226 3217

The model tells us how samples of different sizes will be distributed, in rela-
tion to the actual properties of the population being sampled. It tells us when,
and the extent to which, the properties of a sample are reliable indicators of the
properties of the underlying population.
In collections that can be randomly sampled, then, the “bridge” from the observed
to the unobserved is random sampling (i.e., every individual in the collection has the
same chance of ending up in the sample).
But what do we do when the collections we want to draw conclusions about cannot
be randomly sampled? In that case, there is another “bridge” from the observed to the
unobserved. As Godfrey-Smith (2011, p. 39) writes:
If we want to make inferences about a population that cannot be sampled, we
must ask: what kind of collection is this? Are these objects the products of a
common origin? Do they have a common internal structure? What sort of causal
relationship is there likely to be between properties we are projecting from and
properties we are projecting to? There need not be “laws of nature” overtly on
the scene here, but we are basing the inference on some kind of natural connec-
tion—some combination of laws, mechanisms, and etiologies.
Unlike in inferences from random samples, numbers are not epistemically signifi-
cant in inferences based on causal structures and kinds. According to Godfrey-Smith
(2011, p. 41):
In the purest examples of this sort of investigation, one instance of an F would
be enough, in principle, if you picked the right case and analyzed it well. Ronald
Reagan is supposed to have said “once you’ve seen one redwood, you’ve seen
them all.” When something like this is true, it is a powerful basis for inference.
In practice, we may wish to find more instances in order to find out how F and G
are related. In principle, however, assuming we have picked the right case, we do not
need more cases to find out that there is an F − G association.
So, Godfrey-Smith proposes two distinct “bridges” from the observed to the unob-
served. The first is generalization from random samples. According to Godfrey-Smith
(2011, p. 42), “This form of inference has the following features: sample size mat-
ters, randomness matters, and ‘law-likeness’ or ‘naturalness’ does not matter.” The
second is the “seen one, seen them all” kind of inference or generalization based
on causal structures and kinds. According to Godfrey-Smith (2011, p. 42), “In these
cases sample size per se does not matter, randomness does not matter, but the status
of the kinds matters enormously.” He (2011, p. 42) argues that “These two strategies
of inference involve distinct ‘bridges’ between observed and unobserved cases: one
goes via the power of random sampling, the other via reliable operation of causes and
mechanisms.”
Now, it seems clear that the pessimistic induction is not a “seen one, seen them all”
kind of inference or generalization based on causal structures and kinds, for even pes-
simists concede that past scientific theories “differ from current theories in significant
ways” [See premise (4) in Lewis’ reductio formulation of the pessimistic induction
quoted above; Lewis 2001, p. 373]. As (Bird, 2007a, p. 80) puts it:

123
3218 Synthese (2013) 190:3209–3226

The falsity of earlier theories is the very reason for developing the new ones—
with a view to avoiding that falsity. It would be folly to argue that because no
man has run 100 m in under 9.5 seconds no man ever will. On the contrary,
improvements in times spur on other competitors, encourage improvements in
training techniques and so forth, that make a sub 9.5 second 100 m quite a high
probability in the near future. The analogy is imperfect, but sufficiently close
to cast doubt on Laudan’s pessimistic inference. Later scientific theories are not
invented independently of the successes and failures of their predecessors. New
theories avoid the pitfalls of their falsified predecessors and seek to incorporate
their successes. Even if the successor theory is false also, we cannot apply a
simple enumerative induction. […] we cannot make any good inference from
the premise ‘the succession of theories T1 , …Tn are all false’ to the conclusion
‘later theories in this sequence will also be false’ without additional information.
Likewise, Lipton (2000, p. 197) argues that we cannot infer ‘future theories are
likely to be false’ from ‘past theories turned out to be false’ by enumerative induction
because of the “Darwinian” evolution of theories. Whether or not these realist replies
to the pessimistic induction succeed is not important for present purposes. For present
purposes, the important point is that the sample of theories is not uniform in the way
that a generalization based on causal structures and kinds requires.
In a way that is somewhat similar to the attempts to disambiguate ‘significant ways’
discussed in Sect. 2 above, pessimists might respond that the pessimistic induction
can still be construed as a “seen one, see them all” inductive inference as follows:

(i) For most past successful theories, if theory T1 postulates entity E 1 , then there
is a competing theory T2 that postulates entity E 2 .
(ii) Therefore, for most successful theories, if T1 postulates E 1 , then there is a
competing theory T2 that postulates E 2 .
(iii) For most past successful theories, if T1 is now considered false because it is
believed that E 1 does not exist, then T2 is now considered true because it is
believed that E 2 does exist.
(iv) Therefore, for most successful theories, if T1 is now considered false because
it is believed that E 1 does not exist, then T2 is now considered true because it
is believed that E 2 does exist.

As I have argued in Sect. 2 above, it is not at all obvious that most current theories
have actual past competing theories. But even if most current theories have actual past
competitors, theories that postulate different entities do not have to be incompatible,
such that if one is true, the other has to be false. That is why a sample of theories is
not uniform enough to serve as a basis for an inductive generalization based on causal
structures and kinds. A uniform—as opposed to diverse—sample might be a sample
of, say, copper rods. From a sample of just a few copper rods that are tested for elec-
trical conductivity, it is reasonable to conclude that all copper rods conduct electricity
because, if you have seen one or two copper rods, you have seen them all (given their
uniform atomic structure). Scientific theories, however, are not as uniform as copper
rods. The point, then, is that any sample of theories is not going to be uniform in a
way that is required for a “seen one, seen them all” inductive generalization.

123
Synthese (2013) 190:3209–3226 3219

Magnus and Callender (2004, p. 324) also reconstruct the pessimistic induction
as an argument based on a sample, and add that “the sample includes past scientific
inferences, whether inductive or otherwise” (footnote 4). If the pessimistic induction
is not a generalization based on causal structures and kinds (“seen one, seen them all”),
then perhaps it is an inductive generalization from random samples. The pessimistic
induction can be construed as an inductive generalization as follows:
(PIG1) Most past successful theories are false.
(PIG2) Therefore, most successful theories are false.
As Godfrey-Smith points out, it is random sampling that allows us to project a
property from the observed sample onto the general population. However, in the case
of this pessimistic inductive generalization, the problem is that the sample of success-
ful theories from the history of science is not random. To see why this is a problem,
consider the form of this kind of inductive generalization:
X % of observed Fs are G.
Therefore, X % of all Fs are G.
In the pessimistic inductive generalization, the reference class (F) is successful theo-
ries. The attribute class (G) is being false, i.e., the property of being false is attributed
to most past successful theories, and, by induction, to most successful theories.
Now, the pessimistic inductive generalization is a weak inductive argument because,
as an inductive generalization, it fails to provide grounds for projecting the property of
the observed members of the reference class to unobserved members of the reference
class. And this is so because the pessimistic inductive generalization does not provide
what Godfrey-Smith calls a “bridge” between observed and unobserved.
To put it another way, pessimists argue that there is “a good basis for inductively
inferring that any presently accepted scientific theory is actually false” (Held 2011).
This good basis is supposed to be Laudan’s list of successful but false theories (1981,
p. 33):
• the crystalline spheres of ancient and medieval astronomy;
• the humoral theory of medicine;
• the effluvial theory of static electricity;
• ‘catastrophic’ geology, with its commitment to a universal (Noachian) deluge;
• the phlogiston theory of chemistry’
• the caloric theory of heat;
• the vibratory theory of heat;
• the vital force theories of physiology;
• the electromagnetic aether;
• the optical aether;
• the theory of circular inertia;
• theories of spontaneous generation.
If the aforementioned considerations are correct, however, then this list does not pro-
vide a good basis for a pessimistic inductive generalization. Laudan’s list does not
provide a good basis for a pessimistic inductive generalization because Laudan’s list
is not a good sample. It is not a good sample because it is not a random sample,

123
3220 Synthese (2013) 190:3209–3226

i.e., it is not a sample where every individual in the population has the same chance of
ending up in the sample. The theories in this list were not randomly selected. Rather,
they were selected precisely because they are considered to be successful but strictly
false. In other words, Laudan’s list is a biased sample, which makes the pessimistic
inductive generalization a weak inductive argument. As Park (2011, p. 83) puts it:
the pessimistic induction is a fallacy of biased statistics. The pessimistic inducer
took samples only from science before the twentieth century. […] The mini-
mum requirement for fair samples is that they be randomly selected from the
sciences of both before and after the year 1900. Laudan’s samples do not meet
this requirement.
I have argued that Laudan’s list of successful but false scientific theories is a biased
sample because not all theories had an equal chance of making it into the sample. In
fact, there are only distant-past successful theories in Laudan’s list, but no recent-past
successful theories, and thus it cannot serve as a sample in an inductive generalization
about most successful theories. Doing so would be analogous to making the following
inductive generalization:
(A1) Most American politicians are rich.
(A2) Therefore, most politicians are rich.
(A2) cannot be inferred from a sample that contains only American politicians, for
it is not representative of the general population of politicians, which includes Euro-
pean, African, and other politicians. Similarly, (PIG2) cannot be inferred from a sam-
ple that contains only distant-past theories, for it is not representative of the general
population of successful theories, which includes recent-past successful theories as
well.
If this is correct, then Laudan’s list is a biased sample because it is unrepresenta-
tive of the general population of scientific theories. Laudan’s list is unrepresentative
because, as a sample, it is not diverse enough to support the conclusion that most suc-
cessful theories are false. The theories in Laudan’s list were not randomly selected,
but rather were cherry-picked in order to argue against a thesis of scientific realism.
If this is correct, then the pessimistic inductive generalization is a weak inductive
argument.
To this pessimists might object that, if we simply do the required random sampling,
then the pessimistic inductive generalization would be vindicated and shown to be a
strong inductive generalization. So, to get a random of sample of scientific theories
(i.e., a sample where theories have an equal chance of being selected for the sample),
I used the following methodology:
• Using Oxford Reference Online, I searched for instances of the word ‘theory’ in the
following titles: A Dictionary of Biology, A Dictionary of Chemistry, A Dictionary
of Physics, and The Oxford Companion to the History of Modern Science.
◦ I limited myself to these reference sources to make the task more manageable.
◦ Since it is not clear how to individuate theories (e.g., is the Modern Evolution-
ary Synthesis a theory or is each of its theoretical claims, such as the claims
about natural selection and genetic drift, a theory in its own right?), I limited
myself to instances of the word ‘theory’.

123
Synthese (2013) 190:3209–3226 3221

• After collecting 124 instances of ‘theory’ and assigning a number to each instance,
I used a random number generator to select 40 instances out of the 124.
• I divided the sample of 40 theories into three categories: accepted theories (i.e.,
theories that are accepted by the scientific community), abandoned theories (i.e.,
theories that were abandoned by the scientific community), and debated theories
(i.e., theories whose status as accepted or rejected is in question) (See Table 1).
Based on this sample, pessimists could construct the following inductive generaliza-
tion:

Table 1 A random sample of 40 scientific theories divided into three categories

Accepted theories Abandoned theories Debated theories

Conformal field theory Einstein theory of specific heat Lattice gauge theory
Density functional theory Bohr theory Supermembrane theory
Lowry-Bronsted theory Phlogiston theory Unified-field theory
Quantum theory of radiation Projective relativity theory Sliding filament theory
BCS theory Higgs field Kaluza-Klein theory
Atomic theory Brans-Dicke theory
Perturbation theory
Free-electron theory
Molecular-orbital theory
Quantum electrodynamics
Clonal selection theory
RRKM theory
Endosymbiont theory
Trichromatic theory
Chemiosmotic theory
Frontier-orbital theory
Resonating valence bond theory
Ligand-field theory
Crystal-field theory
Acid growth theory
Electroweak theory / GWS model
Kinetic theory
Transport theory
Debye-Huckel-Onsager theory
Big-bang theory
VSEPR theory
Quantum chromodynamics
Band theory
McMillan-Mayer theory
Total 29 6 5

123
3222 Synthese (2013) 190:3209–3226

15 % of sampled theories are abandoned theories (i.e., considered false).

Therefore, 15 % of all theories are abandoned theories (i.e., considered false).
Clearly, this inductive generalization hardly justifies the pessimistic claim that most
successful theories are false. Even if we consider the debated theories as false, the per-
centages do not improve much in favor of pessimists:
27 % of sampled theories are abandoned theories (i.e., considered false).
Therefore, 27 % of all theories are abandoned theories (i.e., considered false).
So, doing the required random sampling for a strong inductive generalization, it seems,
does not vindicate the pessimist’s claim that most successful theories (both distant-past
and recent-past theories) are false.
Indeed, one might think that this sample actually justifies an optimistic induction
(Cf. Nola 2008). For, based on this random sample of theories, as opposed to Laudan’s
biased sample, one could construct the following inductive generalization:
72 % of sampled theories are accepted theories (i.e., considered true).
Therefore, 72 % of all theories are accepted theories (i.e., considered true).
Scientific realists might be quite satisfied with a ± 70 % success rate for scientific
theories.
For good measure, I ran another trial with instances of the word ‘law’ instead of
‘theory’. Here are the results, which are consistent with the first trial (Table 2).
Based on this sample, pessimists could construct the following inductive gener-
alization:
12.5 % of sampled laws are abandoned laws (i.e., considered false).
Therefore, 12.5 % of all laws are abandoned laws (i.e., considered false).
Clearly, this inductive generalization hardly justifies the pessimistic claim that most
successful laws are false. Even if we consider the debated laws as false, the percentages
do not improve much in favor of pessimists:
30 % of sampled laws are abandoned laws (i.e., considered false).
Therefore, 30 % of all laws are abandoned laws (i.e., considered false).
Again, doing the required random sampling for a strong inductive generalization, it
seems, does not vindicate the pessimist’s claim that most successful laws (both dis-
tant-past and recent-past laws) are false.
Indeed, one might think that this sample actually justifies an optimistic induction
(Cf. Nola 2008). For, based on this random sample of laws, one could construct the
following inductive generalization:
70 % of sampled laws are accepted laws (i.e., considered true).
Therefore, 70 % of all laws are accepted laws (i.e., considered true).
Again, scientific realists might be quite satisfied with a ± 70 % success rate for sci-
entific laws. To be clear, I do not endorse any optimistic inductive generalization. My
present aim is simply to show that the pessimistic induction is a bad argument. But I

123
Synthese (2013) 190:3209–3226 3223

Table 2 A random sample of 40 scientific laws divided into three categories

Accepted laws Abandoned laws Debated laws

Conservation of mass Maxwell-Boltzmann law Virial equation*

Carnot principle Newton’s law of gravitation Charles’ law*
Periodic law Newton’s laws of motion Raoult’s law*
Partition law Law of octaves Joule’s law*
Hubble’s law Titius-Bode law Avogadro’s law*
Hess law Boyle’s law*
Law of conservation of energy Pressure law*
Mass-energy equation
Stefan-Boltzmann law
Planck’s radiation law
Wien’s displacement law
Law of chemical equilibrium
Law of constant proportions
Law of multiple proportions
Law of reciprocal proportions
Coulomb’s law
Law of periodicity
Fermat’s principle
2nd law of thermodynamics
Zeroth law of thermodynamics
Kepler’s law
Graham’s law
Hooke’s law
Grottius-Draper law
Bragg’s law
Balmer’s law (Balmer series)
Moseley’s law
Law of isomorphism
Total 28 5 7
Laws marked with * on the “Debated Laws” column are laws pertaining to idealized entities, such as ideal
gases and ideal solutions. Although some of these laws are accepted and used by scientists, anti-realists
and pessimists would probably object to their inclusion in the “Accepted Laws” column, which is why I
put them in the “Debated Laws” column

do think that scientific realists would be better off relying on an optimistic inductive
generalization, rather than the No-Miracles argument. For, as I have argued elsewhere,
the No-Miracles argument fails to establish what it purports to establish (See Mizrahi
2012). Of course, realists would still need an additional argument that shows that 70 %
of all theories are approximately true (or that we are justified in believing that 70 %
of all theories are approximately true).

123
3224 Synthese (2013) 190:3209–3226

4 The pessimistic induction as pointing to counterexamples

If the aforementioned considerations are correct, then the pessimistic induction turns
out to be an invalid deductive argument when construed as a reductio and a weak
inductive argument when construed as an inductive generalization. At this point, pes-
simists might insist that the pessimistic induction is not meant to be an argument at
all. Perhaps it is meant to point out counterexamples to the realist’s claim that success
is a mark of (approximate) truth. As Laudan (1981, pp. 47–48) writes: the realist’s
“epistemology is confronted by anomalies which seem beyond its resources to grapple
with.”
In order for counterexamples to succeed, however, their target claims must be
understood as universal statements. For example, finding one bird that cannot fly
would refute the claim that birds can fly only if this claim is understood as a universal
generalization, i.e., “All birds can fly.” However, finding one bird that cannot fly would
not refute the claim that a keel is a reliable indicator of flying ability in birds. There are
flightless birds, such as ostriches and kiwis, with a keel on their breastbone (though
greatly reduced). But this is consistent with the claim that birds with a keel on their
breastbone are more likely to be flying birds than flightless birds.
Similarly, to use Kitcher (1993, p. 118) example, to say that single-stranded DNA is
a counterexample to the double-helical model of DNA is to misunderstand the intended
domain of the model. The model is not meant to be a universal statement about DNA.
Now, it is not clear that the realist’s claim that success is a mark of (approximate) truth
is meant to be a universal statement. Success may be a reliable indicator of (approxi-
mate) truth, but this is compatible with some instances of successful theories that turn
out not to be approximately true. In other words, that a theory is successful is a reason
to believe that it is approximately true, but it is not a conclusive proof that the theory
is approximately true. To illustrate:
(S1) If T is successful, then it is likely that T is approximately true (since success
is a reliable indicator of approximate truth).
(S2) The big-bang theory is successful.
(S3) Therefore, it is likely that the big-bang theory is approximately true. Conversely:

(S4) If T is unsuccessful, then it is unlikely that T is approximately true (since

success is a reliable indicator of approximate truth).
(S5) The steady-state theory is unsuccessful.
(S6) Therefore, it is unlikely that the steady-state theory is approximately true.
For these reasons, to give a counterexample to (S1) is to misunderstand the realist’s
thesis. It would be as absurd as citing a few examples of ‘A’ students who have been
absent more than fifty percent of the semester as a counterexample to the thesis that
being absent on a regular basis is a reliable predictor of earning a failing grade. Sim-
ilarly, it would be absurd to cite a few examples of successful but false theories as
counterexamples to the thesis that success is a reliable predictor of (approximate)
truth. As Putnam (1975, p. 73) writes: “That terms in mature scientific theory typi-
cally refer (this formulation is due to Richard Boyd), that the theories accepted in a
mature science are typically approximately true […]—these statements are viewed by

123
Synthese (2013) 190:3209–3226 3225

the scientific realist not as necessary truth but as part of the only scientific explanation
of the success of science” (emphasis added).

5 Conclusion

In this paper, I have considered three interpretations of the pessimistic induction: as

a reductio, as an inductive generalization, and as pointing to counterexamples against
the scientific realist’s thesis that success is a reliable indicator of (approximate) truth.
If my argument is sound, then, contrary to what Saatsi (2005, p. 1098) claims, the
pessimistic induction is not “a powerful force to be reckoned with.” This is so because
the pessimistic induction is a fallacious argument and the counterexamples to which
it points miss their intended target.

Acknowledgments I am grateful to two anonymous reviewers of Synthese for helpful comments on an

earlier draft.

References

Bird, A. (2007a). What is scientific progress?. Noûs, 41, 64–89.

Bird, A. (2007b). Inference to the only explanation. Philosophy and Phenomenological Research, 74,
424–432.
Boyd, R. (1989). What realism implies and what it does not. Dialectica, 43, 5–29.
Caven, R. M. (1921). The foundations of chemical theory: The elements of physical and general
chemistry. New York: D. van Nostrand Co.
Devitt, M. (2011). Are unconceived alternatives a problem for scientific realism. Journal for General
Philosophy of Science, 42, 285–293.
Dopplet, G. D. (2011). From standard scientific realism and structural realism to best current theory
realism. Journal for General Philosophy of Science, 42, 295–316.
Godfrey-Smith, P. (2011). Induction, samples, and kinds. In J. Campbell, M. O’Rourke, & M. Slater (Eds.),
Carving nature at its joins: Topics in contemporary philosophy (pp. 33–52). Cambridge, MA: MIT
Press.
Held, C. (2011). Truth does not explain predictive success. Analysis. doi:10.1093/analysis/anr010.
Kitcher, P. (1993). The advancement of science: Science without legend, objectivity without illusions. New
York: Oxford University Press.
Lange, M. (2002). Baseball, pessimistic inductions and the turnover fallacy. Analysis, 62, 281–285.
Laudan, L. (1981). A confutation of convergent realism. Philosophy of Science, 48, 19–49.
Lewis, P. J. (2001). Why the pessimistic induction is a fallacy. Synthese, 129, 371–380.
Lipton, P. (2000). Tracking track records. Proceedings of the Aristotelian Society. Supplementary
Volumes, 74, 179–205.
Lipton, P. (2005). The truth about science. Philosophical Transactions of the Royal Society B, 360, 1259–
1269.
Magnus, P. D., & Callender, C. (2004). Realist ennui and the base rate fallacy. Philosophy of Science,
71, 320–338.
Mizrahi, M. (2012). Why the ultimate argument for scientific realism ultimately fails. Studies in History
and Philosophy of Science Part A, 43, 132–138.
Muller, F. A. (2008). In defence of constructive empiricism: Maxwell’s master argument and aberrant
theories. Journal for General Philosophy of Science, 39, 131–156.
Nola, R. (2008). The optimistic meta-induction and ontological continuity: The case of the elec-
tron. In L. Soler, H. Sankey, & P. Hoyningen-Huene (Eds.), Rethinking scientific change and
theory comparison: Stabilities, ruptures, incommensurabilities?(pp. 159–202). Dordrecht: Springer.
Okasha, S. (2001). What did hume really show about induction?. The Philosophical Quarterly, 51, 307–327.

123
3226 Synthese (2013) 190:3209–3226

Park, S. (2011). A confutation of the pessimistic induction. Journal for General Philosophy of Science,
42, 75–84.
Poincaré, H. (1952). Science and hypothesis. New York: Dover.
Psillos, S. (1996). Scientific realism and the pessimistic induction. Philosophy of Science, 63, S306–S314.
Psillos, S. (1999). Scientific realism: How science tracks truth. London: Routledge.
Psillos, S. (2006). Thinking about the ultimate argument for realism. In C. Cheyne & J. Worrall
(Eds.), Rationality & reality: Essays in honour of Alan Musgrave (pp. 133–156). Dordrecht:
Springer.
Psillos, S. (2007). The fine structure of inference to the best explanation. Philosophy and Phenomenological
Research, LXXIV(2), 441–448.
Putnam, H. (1975). Mathematics, matter, and method (Vol. 1). Cambridge: Cambridge University Press.
Putnam, H. (1978). Meaning and the moral sciences. London: Routledge.
Saatsi, J. T. (2005). On the pessimistic induction and two fallacies. Philosophy of Science, 72, 1088–1098.
van Fraassen, B. C. (1980). The scientific image. New York: Oxford University Press.
van Fraassen, B.C. (1989). Laws and symmetry. Oxford: Clarendon Press.
Wray, K. B. (2011). Success and truth in the realism/anti-realism debate. Synthese. doi:10.1007/
s11229-011-9931-7.

123