This week we consider some objections against Plantinga’s Evolutionary Argument
Against Naturalism
Planting’s Evolutionary Argument against Naturalism
‘R’ expresses the proposition that our all of our cognitive mechanisms are reliable. ‘E’
stands for evolutionary theory as it is currently understood. ‘N’ expresses the
proposition there are no supernatural entities
1)
2)
3)
4)
5)
Pr (R | E&N) is low
E&N
If person S believes 1) and 2), S has a defeater for R
If person S has a defeater for R, then S has a defeater for everything S believes
Evolutionary naturalists who accept 1) have a defeater for evolutionary
naturalism
Plantinga supports premise 3) with several arguments from analogy, the most central of
which is the XX Pill case. As Plantinga notes, an argument form analogy is necessary
here, because low probability is not sufficient for defeat: e.g., I do not gain a defeater
for believing that I live in Richmond by accepting that the probability of this is low, given
that the Earth revolves around the Sun.
Epistemic Progress
A cognitive mechanism is reliable if and only if it produces mostly true beliefs. So
defined, the reliability of a cognitive mechanism clearly changes over time, both
ontogenetically and phylogenetically.
Ontogenetic progress refers to the changes that occur over the life of an
individual; e.g., my ratio of true to false beliefs is greater now than it was when I
was a child.
Phylogenetic progress refers to population-level changes that occur over the life
of a species; e.g., modern astronomers have many more true beliefs about the
cosmos than ancient astronomers had. This is a specific instance of a more
general trend: humanity’s ratio of true to false beliefs has increased over time.
Plantinga would have accept both type of epistemic progress but would argue that the
only explanation for each one involves a God hypothesis.
Epistemic Progress
Let’s formulate Plantinga’s argument more explicitly, as follows:
1) A cognitive mechanism (CM) is unreliable if and only if it produces mostly false
beliefs.
2) The CM’s that reached fixation because of natural selection during the
Pleistocene were unreliable.
3) The CM’s found in contemporary humans are the same CM’s that existed during
the Pleistocene
4) If God does not intervene in the course of human evolution, then the reliability
of a CM is diachronically fixed (i.e., it does not change over time)
5) According to evolutionary naturalism, God does not intervene in the course of
human evolution
6) Therefore, evolutionary naturalism implies that the CM’s found in contemporary
humans are unreliable.
As premises 1) and 5) are just matters of definition, the only premises to question are
2), 3), and 4).
Evaluation of premise 2)
The CM’s that reached fixation because of natural selection during the Pleistocene were
unreliable
Plantinga cites Darwin and Patricia Churchland in support of his view that, if selection
causes cognitive mechanisms to evolve, then it’s unlikely that these mechanisms are
reliable.
The idea here is that, if a belief enhances fitness, it will evolve by natural selection,
whether or not it is true. Similarly, if a cognitive mechanism enhances fitness, it will
evolve by natural selection, whether or not it is reliable. All that matters in both cases is
fitness. And selection for fitness is distinct from selection for either truth or reliability,
since fitness enhancing behavior is just as easily caused by false belief (and unreliable
mechanisms) as it is true belief (and reliable mechanisms).
Evaluation of premise 2)
Churchland’s claim: the probability that a cognitive mechanism (CM) will evolve by
natural selection is determined by whether this CM enhances fitness, and not by
whether CM is reliable.
1) Pr ( CM evolves | CM enhances fitness and CM is reliable) =
Pr ( CM evolves | CM enhances fitness and CM is unreliable)
Notice that a CM’s effect on fitness screens off its reliability from its evolving. Plantinga
takes 1) to imply that:
2) Pr ( CM evolves | CM is reliable) = Pr ( CM evolves | CM is unreliable)
But 1) does not imply 2). As we discussed, the fact that A screens-off B from C does not
imply that B and C are probabilistically independent of each other; the barometer
example illustrates why. Plantinga’s argument assumes otherwise.
Screening-off and Probabilistic Independence
B screens-off C from A precisely when: Pr (A | B and C) = Pr (A | B and not C)
If A is independent of C, conditional of B, it does not follow that A and C are
independent
Pr (storm | drop in barometric pressure and barometer reads X) =
Pr (storm | drop in barometric pressure and barometer does not read X)
Does not imply
Pr (storm | barometer reads X) =
Pr (storm | barometer does not read X)
Selection of/Selection for
Plantinga argues that the reliability of any cognitive mechanism does not cause any
difference in fitness, which implies that reliability cannot be selected for.
•
•
•
But if there was no selection for reliability it does not follow that there was no
selection of reliability (Darwin makes a similar point against Wallace)
The reliability of a cognitive mechanism need not cause an organism to fitter in
order to be fitter.
What any trait contributes to fitness is distinct from what its fitness value is.
Evaluation of premise 3)
The CM’s found in contemporary humans are the same CM’s that existed during the
Pleistocene
Support from this premise comes from a prominent branch of evolutionary psychology
(the so-call “Santa Barbara” branch) according to which our modern skulls contain stone
age minds. This view is taken to explain why we are more afraid of snakes and spiders
than we are of guns, despite the fact that many more people die from the latter. It is
also taken to explain why we do so poorly on the Wason selection task, despite doing
quite well when problems with the same logical structure are presented through cases
that involve social cheating.
Wason Selection Task
Question: What cards must be turned over to determine if this rule is violated?
Answer: cards ‘E’ and ‘3 ‘(a vowel on one side is sufficient but not necessary for an even
number of the other, which is why card ‘2’ is irrelevant). Most people do not give the
correct answer to this question.
Non-abstract Social Cheating Version of Wason Selection Task
Rule: If a person is drinking alcohol, then this person must be 21 or older.
There are four people in a bar, you know what two are drinking, but not their age, and
the age of two, but not what they are drinking. Here’s what you know:
Beer
Water
45
15
Question: What people must be investigated to determine if this rule has been violated?
Answer: the beer drinker and the 15 year old. Most people do very well with this one,
despite the fact that it has the same logical structure as the Wason task.
Evolutionary psychologists use the data from these experiments to support their view
that our minds are massively modular, which each module dedicated to solving an
adaptive problem faced by our Pleistocene ancestors (including the problem of
identifying cheaters in systems of social cooperation).
Developmental Canalization
The view that our modern skulls contain stone age minds assumes that our cognitive
mechanisms are developmentally canalized.
But many argue that our cognitive mechanisms are not developmentally canalized. On
this view, our modern skulls do not contain stone age minds, but are instead quite
different. Empirical support from this view comes from recent work on neuroplasticity
and epigenetics.
Evaluation of premise 4)
If God does not intervene in the course of human evolution, then the reliability of a CM is
diachronically fixed (i.e., it does not change over time)
Suppose that a), we do have the same CM’s as our Pleistocene ancestors, and that b),
these CM’s were unreliable during the Pleistocene.
Does it follow from a) and b) that the CM’s of contemporary humans are unreliable? It
seems to me that it does not. That is, if a CM is unreliable at time T, it does not follow
that it is unreliable at T+n, even if we assume that this CM is developmentally canalized.
The reason for this is based on the Garbage In/Garbage out principle, according to which
external factors explain why the same mechanism can produce mostly false beliefs in
one context but in a different context produce mostly true ones. So we may have the
same CM as our Pleistocene ancestors—maybe our modern skulls do contain stone age
minds; but our CM’s, unlike theirs, may be reliable.
Evaluation of premise 4)
If God does not intervene in the course of human evolution, then the reliability of a CM is
diachronically fixed (i.e., it does not change over time)
Suppose that a), we do have the same CM’s as our Pleistocene ancestors, and
that b), these CM’s were unreliable during the Pleistocene.
External factors that improve the reliability of a cognitive mechanism include advances
in our instruments of observation (such as telescopes, microscopes, cloud chambers,
etc.), and advances in our instruments of analysis (such as calculus, geometry, statistics,
probability theory, and randomized controlled trials).
Plantinga’s argument assumes the best explanation for the unreliability of a cognitive
mechanism is an internalist one (bad wiring). But it could just as plausibly be an
externalist one (bad inputs).
Arguments from analogy
Let’s now consider Plantinga’s XX pill analogy. The strength of any argument from
analogy depends on the degree of relevant similarity between target and analogue.
Consider two cognitive mechanisms CMa and CMb
1) CMa has property P
2) CMb is similar to CMa to degree N
3) Therefore, CMb has property P
Since arguments from analogy are non-deductive, 1) and 2) do not entail 3), but rather
raise its probability. And the probability that 1) and 2) confer on 3) is equal to N. So if N
is low, then the argument is weak.
Arguments from analogy
1)
2)
3)
4)
5)
6)
7)
8)
9)
Pr (R | XX pill is consumed) is low
XX pill is consumed
If person S believes 1) and 2), S has a defeater for R
Pr (R | E&N) is low
E&N
The XX pill case and the E&N case are similar to degree N
If person S believes 4) and 5), S has a defeater for R
If S has a defeater for R, then S has a defeater for everything S believes
Evolutionary Naturalism is self-defeating
This argument succeeds only if N is high. But one might argue N is very low.
Disanalogies between XX pills and Evolution without God (EN)
The XX pill destroys with certainty cognitive reliability in 95% of cases and thus entails
skepticism.
EN destroys with certainty cognitive reliability is 0% of cases and thus does not entail
skepticism. EN simply fails to explain reliability and thus entails only (at best)
agnosticism.
Effects of XX pill are irreversible
Effects of EN are (arguably) not
In the XX pill case, by hypothesis, there is a single cause that irreversibly destroys with
near certainty all of our cognitive mechanisms.
Our actual beliefs are caused by many mechanisms, mechanisms that differ in their
reliability. Some are more reliable than others, and many become more reliable over
time.
What the XX pill and (E & N) case have in common is a conditional probability claim:
1) Pr (R | I’ve ingested XX) is low
2) Pr (R | E & N) is low.
But there’s only one explanation for 1), whereas there are many for 2). And the more
plausible explanation for the latter would take into consideration the fact that our
beliefs are caused by many distinct mechanisms (system 1 vs. system 2), and that the
reliability of a cognitive mechanism changes over time, both phylogenetically and
ontogenetically.
Back to the main argument
1)
2)
3)
4)
5)
6)
7)
8)
9)
Pr (R | XX pill is consumed) is low
XX pill is consumed
If person S believes 1) and 2), S has a defeater for R
Pr (R | E&N) is low
E&N
The XX pill case and the E&N case are similar to degree N
If person S believes 4) and 5), S has a defeater for R
If S has a defeater for R, then S has a defeater for everything S believes
Evolutionary Naturalism is self-defeating
Evolutionary Naturalism is not self-defeating because N is low.
But perhaps accepting 4) and 5) has other epistemic implications that should trouble
the naturalist. Consider these two: a) evolutionary naturalism does not explain the
reliability of our cognitive faculties; and b) the reliability of our cognitive faculties
counts as evidence against evolutionary naturalism. Once an evolutionary naturalist
comes to accept either a) or b), it would be evidentially irrational for this individual
to continue accepting evolutionary naturalism.
Evolutionary naturalism does not explain the reliability of our cognitive faculties
Here is one argument for this view:
1) If E&N implies that R is improbable, then E&N does not explain R.
2) E&N implies that R is improbable.
3) Therefore, E&N does not explain R.
Premise 1) assumes a view of explanation (Hempl 1965) that is now widely regarded as
mistaken (Jeffrey 1969, and Salmon 1990, 1998). On Hempel’s view, hypothesis H
cannot explain observation O unless H implies that O has a probability greater than .5.
Jeffrey and Salmon argue against this view on the basis of the following type of
example: if an offspring is an AA homozygote, this would be explained if its parents are
both AB heterozygotes, even though the probability that that parental pair will produce
an AA offspring is only 0.25. The improbability that an explanans confers on its
explanandum is no reason to regard it as unexplanatory. Thus, from the mere fact that
Pr (R | E&N) is low, it does not follow that E&N fails to explain R.
The reliability of our cognitive faculties is evidence against evolutionary naturalism
Here is an argument for this view:
1) E&N implies that R is improbable
2) R
3) R is evidence against E&N
I hope that you can see that this argument is flawed because it violates the law of
likelihood.
The reliability of our cognitive faculties is evidence against evolutionary naturalism
The general formulation of the argument we’ve just considered is this: If hypothesis H
implies that evidence E is improbable, and E is observed, then E is evidence against H.
This type of argument has some intuitive plausibility, as the following example
illustrates. Lamarck’s phylogenetic theory is that each species on earth evolved
independently of other species. If this theory is true, then the probability of gill slits in
human embryos is low. But human embryos have gill slits. Thus, the fact that we
observe what Lamarck’s theory implies we shouldn’t constitutes evidence against this
theory.
Despite the intuitive plausibility of this inference, it is mistaken.
The reliability of our cognitive faculties is evidence against evolutionary naturalism
The general formulation of the argument we’ve just considered is this: If hypothesis H
implies that evidence E is improbable, and E is observed, then E is evidence against H.
The improbability of evidence E on hypothesis H is not evidence against H (Royal
1997, and Sober 2008). Sober (2008, p 52) develops the following example in
support of this position. Suppose (E) that two individuals are a heterozygous
match at each of twenty independent loci. At each locus, each individual has a
one copy of a rare allele (frequency=0.001) and one copy of the alternative
common one (frequency=0.999). According to the hypothesis (H), that these
individuals are full siblings, Pr (E | H) = [0.001)(.05)]20. H implies that E is very
improbable, but E is not evidence against H, because on the alternative
hypothesis (H*) that these individuals are unrelated (H*), Pr (E | H*) = [(0.001)
(0.001]20. Thus, from the mere fact that Pr (R | E&N) is low, it does not follow
that R is evidence against E&N.
In order to establish that R counts as evidence against E&N, or that E&N fails to
explain R, it is not enough to establish that the likelihood of E&N is low. One has
to establish instead that it is lower than some alternative, as the Law of
Likelihood states.
The Contrastive Nature of Evidence
As these considerations illustrate, questions about explanation and the epistemic
significance of evidence are contrastive. On one very common pattern of reasoning in
both science and philosophy—inference to the best explanation—we should not ask
whether hypothesis H explains observation O, we should ask instead whether H explains
O better than some specific alternative hypothesis (Lipton 2004). Similarly, whether O is
evidence for or against H depends on the alternative hypothesis H* with which H
competes (Hacking 1965). So until an alternative to E&N is considered, from Pr (R |
E&N) is low, it does follow that evolutionary naturalism fails to explain our cognitive
reliability, or that our cognitive reliability is evidence against R.
Plantinga’s Argument as a Likelihood Argument
The only alternative to evolutionary naturalism is evolutionary supernaturalism (E&S).
To establish that evolutionary naturalism fails to explain our cognitive reliability, or that
the reliability of our cognitive faculties is evidence against evolutionary naturalism, the
first step is to describe a specific version of E&S. This step is important because the
assertion that supernatural entities exist does not entail that God exists. And the
assertion that that God exists does not entail that the Christian God exists. The second
step is to establish that that likelihood of a specified version of E&S, with respect to R, is
greater than the likelihood of E&N. So an argument for the evidential irrationality of
evolutionary naturalism needs to defend the following likelihood inequality: Pr (R |
E&N) < Pr (R | E&S).
To determine whether this likelihood inequality can be defended successfully,
let’s conjoin evolution with two different supernaturalisms: Austere Theism (E&AT), and
Christianity (E&C). It seems clear that Pr (R | E&AT) is inscrutable. Pr (R | E&AT) is
inscrutable because the proposition that a God exists implies nothing about this God’s
aims (what does this God want to accomplish?) and abilities (what is this God able to
accomplish?). It also seems clear that the Pr (R | E&C) is inscrutable. If conjoining
evolutionary theory with these supernaturalist alternatives to naturalism generates
only inscrutable likelihoods, then, from Pr (R | E&N) is low, it obviously does not
follow that (R | E&N) < Pr (R | E&S).
Plantinga’s Argument as a Likelihood Argument and Sober’s Razors
Consider an evolutionary naturalist S who accepts that Pr (R | E&N) is low and R. If S
also believes that Pr (R | E&S) is inscrutable, then S has no evidential reason to reject
E&N, as Bayes’ theorem indicates.
Ratio
of posteriors
Pr (E&N | R)
Pr (E&S | R)
Likelihood Ratio
=
Pr (R | E&N)
Pr (R | E&S)
Ratio of Priors
X
Pr (E&N)
Pr (E&S)
Unless there is a warranted objective estimate for the relevant priors here, S’s evidence
has no implications for the probability of either evolutionary naturalism or evolutionary
supernaturalism.
PACIFIC
Original
PROBABILITY
Article
PHILOSOPHICAL
DEFEATERS
QUARTERLY
Blackwell
Oxford,
Pacific
PAPQ
©
0031-5621
310September
84
002003Philosophical
University
UK
Publishing
2003 AND
of Quarterly
Southern
Ltd
California and Blackwell Publishing Ltd.
PROBABILITY AND
DEFEATERS
BY
ALVIN PLANTINGA
Abstract: Branden Fitelson and Elliott Sober raise several objections to
my evolutionary argument against naturalism; I reply to four of them.
My thanks to Branden Fitelson and Elliott Sober (hereafter ‘F&S’) for
their comments on my evolutionary argument against naturalism.1 F&S
devote most of their attention to what I called “the preliminary argument” (WPF, pp. 228–29). This argument as stated in WPF contains an
error: it confuses the unconditional objective or logical probability of R
with its probability conditional on our background knowledge.2 The main
argument, happily, is unaffected, and here I’ll comment only on what
F&S have to say about the main argument. F&S start several hares, most
of which seem to me to run rather badly. I can’t chase them all, so I’ll
restrict myself to the following four.
1.
P(R/N&E) low or inscrutable?
Let P be the proposition my cognitive faculties are reliable, N be philosophical naturalism (i.e., the proposition that there is no such person as
God or anything like God) and E the proposition that we and our cognitive faculties have come to be by way of the processes favored in contemporary evolutionary theory. In WPF I argue that P(R/N&E) is low or
inscrutable, and do so by considering several mutually exclusive and
jointly exhaustive subcases Pi. These 4 subcases are ways in which belief
can be related to behavior: (1) epiphenomenalism (belief isn’t causally
connected with behavior at all), (2) semantic epiphenomenalism (belief is
causally connected with behavior, but just by virtue of its neurophysiological properties and not by virtue of its content), (3) belief is causally
Pacific Philosophical Quarterly 84 (2003) 291– 298
0279–0750/00/0100–0000
©2003 University of Southern California and Blackwell Publishing Ltd. Published by
Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and
350 Main Street, Malden, MA 02148, USA.
291
292
PACIFIC PHILOSOPHICAL QUARTERLY
related to behavior by virtue of content as well as neurophysiological
properties, but is maladaptive, and (4) the common sense possibility (call
it ‘C’) according to which belief is both adaptive and also causally efficacious by way of content as well as neurophysiological properties. I argued
that P(R/N&E&Pi) is low or inscrutable on all but C; I also said that P(R/
N&E&C) is not very high. The next step of the argument was to try to
arrive at a value for P(R/N&E), given the values for P(R/N&E&Pi) for
each of the Pi. This proceeds in terms of a weighted average principle:
P(R/N&E) is equal to the weighted average of P(R/N&E&Pi), weighted in
each case by P(Pi/N&E). The probability calculus gives us a formula here:
P(R/N&E) = (P(R/N&E&P1) × P(P1/N&E)) + (P(R/N&E&P2)
× P(P2/N&E)) + (P(R/N&E&P3) × P(P3/N&E))
+ (P(R/N&E&P4) × P(P4/N&E)).
I argued that P(R/N&E&C) is not very high (or inscrutable) and that
P(R/N&E&Pi) for each of the other Pi is low (or inscrutable); the result is
that P(R/N&E) is itself low or inscrutable.
F&S don’t quarrel with my suggestions here, except in the case of P(R/
N&E&C). There they mistakenly represent me (p. 121) as arguing that
that probability is low; what I actually said, however, is that “Here one
does not quite know what to say about the probability that their cognitive
systems would produce mainly true beliefs, but perhaps it would be reasonable to estimate it as somewhat greater than 1/2” (WPF p. 227). Then
later I amend this by adding that suggestion that it is also reasonable
to take this probability to be inscrutable. More important, F&S consider
only P(R/N&E&C). But given the structure of the argument, it might be
that P(R/N&E) is low or inscrutable even if P(R/N&E&C) is very high –
indeed, as high as you please. So even if their animadversions on my argument for at best a moderately high value for P(R/N&E&C) were on target, the overall argument wouldn’t suffer.
But they aren’t on target. My argument is inconclusive, say F&S,
because, so they say, I neglect to consider the fact that whether a trait is
likely to evolve depends upon the availability of the trait:
Plantinga’s mistake here is that he ignores the fact that the probability of a trait’s evolving
depends not just on its fitness, but on its availability. The reason zebras don’t have machine
guns with which to repel lion attacks is not that firing machine guns would have been less
adaptive than simply running away; the trait didn’t evolve because it was not available as a
variation on which selection could act ancestrally. . . . By ignoring the question of avail ability, Plantinga, in effect, assumes that natural selection acts on the set of conceivable variants. (pp. 120 – 21).
Well, it’s nice to know that machine guns weren’t originally available to
zebras, although even if they were it wouldn’t have helped much – their
© 2003 University of Southern California and Blackwell Publishing Ltd.
PROBABILITY AND DEFEATERS
293
hooves are much too big and clumsy to operate those little triggers. But I
entirely fail to see how my argument in effect assumes that natural selection acts on the set of conceivable variants – e.g., the conceivable variant
in which creatures from an advanced civilization elsewhere in the galaxy
take pity on our Australopithicene ancestors struggling in epistemic darkness and by a bit of genetic reengineering confer reliable cognitive faculties on them. What EAAN assumes is only that the epistemic probability
of false beliefs (or protobeliefs, or whatever would be the appropriate precursors of belief ) being available to natural selection is not much lower
than that of true beliefs being available. And that seems clearly plausible:
the natural thing to think is that any time at which beliefs are available is
a time at which both true and false beliefs are available. On the other
hand, one who disputes this assumption will presumably think P(R/
N&E&C) is inscrutable, which is also in accord with EAAN.
2.
Is (N&E) & P(R/N&E) is low or inscrutable
a defeater for R?
F&S say no: “After all, people who believe E&N might have other reasons
for believing R. For example, they might argue that R is a basic proposition that does not need theoretical support, or that R derives its epistemic
credentials from something other than the thesis of evolutionary naturalism” (p. 124). They might indeed argue for both those theses, and if they
did, I think they’d be right. But that fact does not insulate R from defeat
for the naturalist. My belief that I see a sheep is a basic proposition that
does not require theoretical support; nevertheless, it is possible to get a
defeater for it (as when you, the local authority on sheep, point out that
what I see is only a sheep dog that looks like a sheep from this distance).
So it is entirely possible to acquire a defeater for a belief you hold in the
basic way. In the same way, R (we may suppose) is a basic proposition not
needing theoretical support: nevertheless the naturalist’s belief (N&E) &
P(R/N&E) is low or inscrutable can perfectly well provide him with a
defeater for it. Similarly, the fact that R does not derive its epistemic credentials from N&E does not deliver it from defeat. For example, I believe
I’ve taken XX, a substance I believe sometimes induces massive unreliability; I also believe that P(R/I’ve taken XX) is low or inscrutable. I then
have a defeater for R with respect to myself; but of course R does not
derive its epistemic credentials from my belief that I’ve taken XX. So it is
entirely possible to acquire a defeater D for a belief, even if the belief does
not derive its epistemic credentials from D. And just this is the case, if
EAAN is right, with respect to R. R has its warrant in the basic way, and
does not derive it from N&E; nevertheless (N&E) & P(R/N&E) is low or
inscrutable provides the naturalist with a Humean defeater3 for R.
© 2003 University of Southern California and Blackwell Publishing Ltd.
294
PACIFIC PHILOSOPHICAL QUARTERLY
F&S also suggest that:
People who believe E&N should not regard the fact that this probability
is “inscrutable” to them as a reason to reject R. We suspect that many
people who are well acquainted with the theory of special relativity
and who think that birds fly still don’t know what value to assign to
Pr(Special relativity/birds fly).
Here the point seems to be that:
(1) For any propositions A and B, if P(B/A) is inscrutable, then A is a
defeater for B
is false. Indeed it is; but of course my argument doesn’t for a moment presuppose that it is true.4 There are cases like this where A is a defeater for
B, and cases where it is not; this case, so I say, is one of the former.
F&S seem also to think that ignorance of probabilities is never a guide
to belief: “The Principle of Indifference is flawed because it claims to
obtain probabilities from ignorance; the start of Plantinga’s main argument makes the complementary mistake of holding that ignorance of
probabilities is a guide to belief ” (p. 124). But surely ignorance sometimes
is a guide to belief. In the trivial case, your ignorance of P(A/B) is a good
reason for not believing, e.g., that P(A/B) is .23. But there are many less
trivial cases. I assume, as usual, that the thermometer T I’ve just bought
is reliable. You then tell me that this thermometer was made in a factory
F whose Luddite owner aims to do his best to frustrate modern industrial
society by fabricating instruments many of which are unreliable, but you
don’t know the ratio of reliable to unreliable instruments produced by the
factory. P(T is reliable/T was fabricated in F) is then inscrutable for you,
as it will be for me if I believe you. This constitutes a defeater, for me, of
my initial assumption that T is reliable, and it is a defeater, in part,
because of ignorance of a probability.
3.
Conditionalization
“Notice that Plantinga assumes that evolutionary naturalists have no
basis for deciding what to think about R, other than the proposition E&N
itself. This crucial assumption is never defended in either Warrant and
Proper Function or ‘Naturalism Defeated’” (p. 125). Now the fact is I
think R has intrinsic warrant, warrant in the basic way. That is because I
believe human beings have been made in God’s image, part of which
involves our being able to form true beliefs and acquire knowledge; and
the module of the design plan governing the production of the belief that
© 2003 University of Southern California and Blackwell Publishing Ltd.
PROBABILITY AND DEFEATERS
295
R is aimed at the production of true belief. So of course I don’t think that
the naturalist’s only basis for what to think about R is N&E. (But I do
think that neither the naturalist nor anyone else can give a cogent argument for R.) So how to understand F&S here? Perhaps the best way to
understand this complaint is as an attempt to raise the conditionalization
problem: what, in this context, can the naturalist properly conditionalize
on? Just N&E itself ? Presumably not: but then what further? This is
indeed a tough problem, but not, so far as I can see, in such a way as to
give the naturalist an escape route.5
4.
A defeater for all of my beliefs?
F&S also suggest that even if I have a defeater for R, I don’t necessarily
have a defeater for the bulk of my beliefs: “Even if E&N defeats the claim
that ‘at least 90% of our beliefs are true’, it does not follow that E&N also
defeats the more modest claim that ‘at least 50% of our beliefs are true’.
Plantinga must show that E&N not only defeats R, but also defeats the
claim that ‘at least a non-negligible minority of our beliefs are true’
(p. 125). But why must I show that? I agree that (E&N) & P(R/N&E) is low
or inscrutable does not offer a direct defeater for the belief that at least
50% of our beliefs are true. But how does that help the naturalist? F&S
seem to make the mistake, here, of thinking that if you don’t have a
defeater for the proposition that 50% of your beliefs are true, then 50% of
your beliefs are such that you don’t have a defeater for them. But that
doesn’t follow at all. I still have a defeater for each of my beliefs, even if I
also believe that 50% of them are true, and even if 50% of them are true.
I’ve been reading an authoritative book on evolutionary biology, naturally enough believing what I read. The author then tells me that up to
50% of the propositions written on p. 45 of the book are false, but does
not tell me which. If I have no further relevant information I will not
believe any of these propositions, even though I believe that at least 50%
of them are true. Another example: I’ve been keeping records, relying on
that thermometer T of a few paragraphs ago. You then tell me that the
factory owner designed T in such a way that 50% of its readings are true
and 50% false. If I believe you and have no other source of relevant information, I have a defeater for each of the beliefs I formed on the basis of the
readings of the thermometer, even though I believe half of them are true.
Finally, I’d like to comment on F&S’s paragraph 2.3, p. 125. This paragraph illustrates what seems to me a couple of unhappy features of their
essay: (1) consistent misunderstanding, and (2) attempts to show that my
arguments are uncogent by finding invalid argument forms of which they
are instances, or false propositions that entail their premises (or both).
But every argument is an instance of an invalid argument form (e.g.,
© 2003 University of Southern California and Blackwell Publishing Ltd.
296
PACIFIC PHILOSOPHICAL QUARTERLY
P1, . . . , Pn; therefore Q) and every proposition is a consequence of some
false proposition.6 In para 2.3 they begin by saying that they have “interpreted the main argument . . . as asserting that a low value for Pr(R/E&N)
suffices for E&N to defeat R”, which is near enough true. But then they
say that in “Naturalism Defeated” “Plantinga denies that this is what he
meant and tries to develop an account of defeat that clarifies how the
argument is supposed to go.” I don’t know what they mean here. I do
think (near enough) that a low value for Pr(R/E&N) suffices for E&N to
defeat R; but I certainly did not (in “Naturalism Defeated” or anywhere
else) deny that this is what I meant. So what do they have in mind? I think
perhaps they are referring to my rejecting (in “Naturalism Defeated”) the
idea that
(2) For any A and B, if P(R/A&B) is low or inscrutable, then B is a
defeater for A.
But I never so much as dreamt that that proposition is true, and certainly
did not propose or presuppose it in my original statement of EAAN.
Next, F&S note that I develop three principles of defeat in “Naturalism
Defeated” and complain that “he never explains how they are relevant to
establishing that E&N defeats R. In fact their logical form renders them
incapable of closing the gap between premises 1 and 2 [their statement of
my argument on p. 123].” Now to close the gap between those premises, I
take it, would be to show that if,
(3) P(R/N&E) is low,
then,
(4) E&N is a defeater for R.
But of course those three principles were not designed to show that (3)
implies (4). Thus consider the First Principle of Defeat:
(FPD) If S rationally believes that the warrant a belief B has for him is derivative from the
warrant a belief A has for him, then B is not a defeater, for him, of A (“Naturalism
Defeated”, p. 48).
(FPD) was designed to summarize a number of cases in which (2), above,
is false: those cases where the warrant of B is derivative, in the appropriate way, from the warrant of A. Similarly, the Second Principle of Defeat
is designed to show some other circumstances under which a belief B does
not constitute a defeater for a belief A; and the third principle is designed
to exhibit a set of circumstances under which a proposition B does not
© 2003 University of Southern California and Blackwell Publishing Ltd.
PROBABILITY AND DEFEATERS
297
constitute a defeater-defeater for a proposition A. Hence the complaint
that they fail to fill the gap between premises (1) and (2) is based just on
misunderstanding.
Next, F&S propose that (FPD), far from filling the gap between (3) and
(4) “apparently helps establish that E&N is not a defeater of R”; that is
because they “suspect that many evolutionary naturalists rationally
believe that their warrant for believing E&N depends on their being warranted in believing that their cognitive faculties are highly reliable”. But it
is one thing for the warrant B has for me to depend on A’s having warrant
for me, and quite another for the warrant B has for me to be derivative
from the warrant A has for me. As I said in “Naturalism Defeated”, the
paradigm case of the latter is where I infer B from A. But the naturalist
does not, one hopes, infer N&E from R. I don’t know what, if anything,
he does infer N from, but presumably he infers E (perhaps by way of an
“argument to best explanation”) from premises about the fossil record,
the distribution of species across the world, certain results from molecular
biology, and the like. This argument certainly does not require the additional premise that his cognitive faculties are reliable, and will not be the
least improved by its addition.
They next correctly point out that X’s being a defeater of Y does not
depend upon P(Y/X) being low, at any rate if self-defeat is possible; but
again, I never for a moment thought otherwise. They go on to claim that
it is difficult to connect low or inscrutable probability with defeat. As we
have seen, however, for each there are plenty of unproblematic cases
where it is indeed connected with defeat. Finally, they suggest (p. 126)
that I waffle on the proposition high probability is necessary for rational
belief and suggest that “What Plantinga is coming up against here is a
close relative of the phenomenon that Kyburg’s lottery paradox made
vivid”. They then add that “This connection with the lottery paradox suggests that the task of repairing the main argument is formidable”. But I
can’t see that F&S have done anything to show that the main argument
needs repair. Further, I fail to see any connection between the lottery paradox and my argument. The lottery paradox arises when one asserts quite
generally that rational belief that A depends upon A’s being probable with
respect to some body of propositions – what you know, perhaps, or what
you know immediately, or what is certain for you, or something of the
sort. I certainly don’t believe and as far as I know have never suggested
that high probability (with respect to what?) is in general necessary for
rational belief. Thus I believe I’ve just thrown three heads in a row with
this fair coin; even if that belief is improbable with respect to the rest of
what I believe it is nevertheless perfectly rational. Still, further, EAAN, as
far as I can see, doesn’t even depend on the proposition that naturalists
are rational in believing R only if P(R/N&E) is high. A naturalist might
rationally believe R in the basic way, for example, and never so much as
© 2003 University of Southern California and Blackwell Publishing Ltd.
298
PACIFIC PHILOSOPHICAL QUARTERLY
consider P(R/N&E); then she will not (or at any rate need not) have a
defeater for R. It is only the reflective naturalist, one who sees that this
probability is low or inscrutable, and thinks about the bearing of this on
R, who gets a defeater here. This is another case where F&S find some
false general proposition from which what I say perhaps follows, and then
complain that what I say is therefore false.7
Department of Philosophy
University of Notre Dame
NOTES
1
Pacific Philosophical Quarterly 79 (1998). My argument can be found in Logos 12, 1991,
pp. 27 – 50, and in chapter 12 of Warrant and Proper Function (New York: Oxford University Press, 1993) (hereafter ‘WPF’).
2
I thank Fitelson and Sober for helping me see this. For a correction and repair of the
preliminary argument, see my Warranted Christian Belief (New York: Oxford University
Press, 2000), pp. 229–31.
3
See my Warranted Christian Belief (New York: Oxford University Press, 2000), p. 225.
4
As I point out in “Naturalism Defeated” (unpublished but available on the web at philofreligion.homestead.com/Plantingapage.html), pp. 46 ff.
5
See Richard Otte’s “Conditional Probabilities in Plantinga’s Argument Against Naturalism” in Naturalism Defeated? Essays on Plantinga’s Evolutionary Argument Against Naturalism, ed. James Beilby (Ithaca: Cornell University Press, 2001), which is largely devoted
to the conditionalization problem. I’ve said what I can about this problem in Section IC of
“Reply to Beilby’s Minions” in the Beilby volume.
6
And not just some necessarily false proposition; every possibly true proposition is a
consequence of a false contingent proposition.
7
One more thing: F&S argue that I am mistaken in holding that “if naturalism is true,
then, surely, so is evolution” and in holding that the probability of evolution on naturalism
is high: “Neither of these claims is right. Recall that proposition E adverts to the mechanisms described in contemporary evolutionary theory. If that theory were found wanting, it
would not entail the falsehood of naturalism . . .” Quite right: naturalism does not entail
evolution. What I meant is that the epistemic probability of evolution is high, given naturalism together with our current evidence. That is of course quite compatible with the
thought that if new evidence showed up, naturalists could sensibly move to some other
theory of evolution, or even give it up altogether.
© 2003 University of Southern California and Blackwell Publishing Ltd.
PLANTINGA’S
PROBABILITY
ARGUMENTS AGAINST
EVOLUTIONARY
NATURALISM
BY
BRANDEN FITELSON AND ELLIOTT SOBER*
Abstract: In Chapter 12 of Warrant and Proper Function, Alvin Plantinga
constructs two arguments against evolutionary naturalism, which he construes
as a conjunction E&N. The hypothesis E says that “human cognitive faculties
arose by way of the mechanisms to which contemporary evolutionary thought
directs our attention” (p. 220).1 With respect to proposition N, Plantinga
(p. 270) says “it isn’t easy to say precisely what naturalism is,” but then adds
that “crucial to metaphysical naturalism, of course, is the view that there is
no such person as the God of traditional theism.” Plantinga tries to cast doubt
on the conjunction E&N in two ways. His “preliminary argument” aims to
show that the conjunction is probably false, given the fact (R) that our psychological mechanisms for forming beliefs about the world are generally reliable.
His “main argument” aims to show that the conjunction E&N is self-defeating
– if you believe E&N, then you should stop believing that conjunction.
Plantinga further develops the main argument in his unpublished paper
“Naturalism Defeated” (Plantinga 1994). We will try to show that both arguments contain serious errors.
1. The Preliminary Argument
Plantinga constructs his preliminary argument within a Bayesian framework. His goal is to establish that Pr(E&N|R) – the probability of E and
N, given R – is low. To do this, Plantinga uses Bayes’ Theorem, which
says that this conditional probability is a function of three other quantities:
Pacific Philosophical Quarterly 79 (1998) 115–129
0031–5621/98/0200–0000
© 1998 University of Southern California and Blackwell Publishers Ltd. Published by
Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and
350 Main Street, Malden, MA 02148, USA.
115
116
PACIFIC PHILOSOPHICAL QUARTERLY
Pr(E & N | R ) =
Pr(R | E & N ) ⋅ Pr(E & N )
⋅
Pr(R )
Plantinga says you should assign to Pr(R) a value very close to 1 on the
grounds that you believe R (p. 228). He argues that Pr(R|E&N) is low.
Although Plantinga doesn’t provide an estimate of the prior probability
Pr(E&N), he says that it is “comparable” to the prior probability of
traditional theism (TT) (p. 229), meaning, we take it, that their values
aren’t far apart.
This last claim should raise eyebrows, not just among evolutionary
naturalists who reject the idea that their theory and traditional theism are
on an equal footing before proposition R is taken into account, but also
among critics of Bayesianism, who doubt that there is an objective basis
for such probability assignments. Plantinga says (p. 220, n. 7) that his
probabilities can be interpreted either “epistemically” or “objectively,”
but that he prefers the objective interpretation. However, Bayesians have
never been able to make sense of the idea that prior probabilities have
an objective basis. The siren song of the Principle of Indifference has
tempted many to think that hypotheses can be assigned probabilities
without the need of empirical evidence, but no consistent version of this
principle has ever been articulated. The alternative to which Bayesians
typically retreat is to construe probabilities as indicating an agent’s
subjective degree of belief. The problem with this approach is that it
deprives prior probabilities (and the posterior probabilities that depend
on them) of probative force. If one agent assigns similar prior probabilities
to evolutionary naturalism and to traditional theism, this is entirely
consistent with another agent’s assigning very unequal probabilities to
them, if probabilities merely reflect intensities of belief.
Although Plantinga’s Bayesian framework commits him to making
sense of the idea that the conjunction E&N has a prior probability, his
argument does not depend on assigning any particular value to this quantity. As Plantinga notes (p. 228), if Pr(R) ⊕ 1 and Pr(R|E&N) is low, then
Pr(E&N|R) also is low, no matter what value Pr(E&N) happens to have.
1.1 PROPOSITION R
For the sake of clarity, it is worth spelling out proposition R more
precisely. What does it mean for our psychological mechanisms for
forming beliefs to be “generally” reliable? In his unpublished manuscript,
Plantinga says that R means that “the great bulk” of our beliefs are true
(Plantinga 1994, p. 2). Aside from questions about how beliefs are to be
counted, we don’t want to challenge the truth of this summary statement.
However, it drastically underspecifies the data that need to be explained.
For the fact of the matter is that our cognitive mechanisms are reliable
© 1998 University of Southern California and Blackwell Publishers Ltd.
PLANTINGA’S PROBABILITY ARGUMENTS
117
on some subjects, unreliable on others, and of unknown reliability on still
others. We should divide our beliefs into categories and associate a characteristic degree of reliability with each of them.2 Perhaps certain simple
perceptual beliefs are very reliable, while beliefs about other subjects are
less so. Rather than trying to obtain a summary statement about all these
mechanisms and the beliefs they generate, it would be better to consider
a conjunction R1&R2& . . . &Rn, which specifies the degree of reliability
that human belief formation devices have with respect to different subjectmatters, or in different problem situations. Plantinga (pp. 216–17, 227,
231–32, and in a personal communication) does not object to this partitioning and uses it himself to discuss the probability that E&N confers
on R.
If R is true, why should one bother to spell it out in more detail? This
wouldn’t matter if Plantinga’s argument were deductive. A sound argument stays sound when the premisses are supplemented with more (true)
details. However, probability arguments don’t have this property. Even
if Pr(R|E&N) is less than Pr(R|TT), it remains to be seen whether Pr(R1&
. . . &Rn|E&N) is less than Pr(R1& . . . &Rn|TT).
Before we get to that comparative question, let’s consider whether the
conditional probability Pr(R1&R2& . . . Rn|E&N) is high or low. Suppose
that evolutionary naturalism does a good job of predicting each of the
conjunct Ri’s, conferring on each a probability, say, of 0.99. It still could
turn out that E&N confers a low probability on the conjunction. If
Pr( R1 & R2 & . . . & R n | E & N ) = Pr( R1 | E & N ) ⋅ Pr( R2 | E & N )
. . . Pr( Rn | E & N )
(i.e., if the Ri’s are probabilistically independent of each other, conditional
on E&N), then the left-hand term may have a low value, even though each
product term on the right has a high value. Multiply 0.99 times itself sufficiently often and you get a number close to zero. This can happen to any
good theory; it may confer a low probability on a massive conjunction of
observations even though it confers a very high probability on each conjunct.
Once we decompose proposition R into a conjunction of claims, it is
far from obvious that evolutionary theory does a worse job of predicting
this conjunction than traditional theism does. Plantinga says the traditional theist “believes that God is the premier knower and has created
us human beings in his image, an important part of which involves his
endowing them with a reflection of his powers as a knower (p. 237).”
However, an influential point of view in cognitive science asserts that
human reasoning is subject to a variety of biases. It isn’t just that people
occasionally make mistakes, but that the human reasoning faculty seems
to follow heuristics that lead to systematic error (Kahnemann, Tversky
© 1998 University of Southern California and Blackwell Publishers Ltd.
118
PACIFIC PHILOSOPHICAL QUARTERLY
and Solvic 1982). It would be no surprise, from an evolutionary point of
view, if human beings had highly reliable devices for forming beliefs about
practical issues that affect survival and reproduction, but are rather less
gifted when it comes to matters of philosophy, theology, and theoretical
science. Does traditional theology also predict this result? No doubt, a
theology can be specified that makes any prediction one wants. However,
it is not at all clear that Plantinga’s traditional theology does a good job
predicting the varying levels of reliability that the human mind exhibits.
Plantinga must address the same problem that Paley’s design argument
faces: Why would an omniscient, omnipotent, and benevolent deity produce organisms who seem to be so manifestly imperfect in the adaptations
they exhibit (Sober 1993)?
1.2 SETTING PR(R) ⊕ 1
We mentioned earlier that Plantinga sets Pr(R) ⊕ 1 because he believes
proposition R. Within the context of Bayesian confirmation theory,
assigning the evidence a probability close to unity has a peculiar consequence, as we now will show.
Bayesians define confirmation in terms of probability raising; an observation O confirms a hypothesis H if and only if the posterior probability
Pr(H|O) is greater than the prior probability Pr(H). If we rewrite Bayes’
theorem as follows
Pr(H | O ) Pr(O | H )
=
,
Pr(H )
Pr(O )
it is clear that O cannot confirm H, if Pr(O) = 1. With this assignment, the
right-hand ratio can’t be greater than unity, so the left-hand ratio can’t
either. Plantinga’s stipulation that Pr(R) is close to unity doesn’t quite insure
that R can’t confirm a hypothesis H. After all, it is possible that Pr(R|H)
should be even closer to unity than Pr(R) is. Let us say that a hypothesis
H is quasi-deterministic with respect to R if Pr(R|H) > Pr(R) ⊕ 1. If evolutionary naturalism isn’t quasi-deterministic in this sense, then R can’t
confirm it, given Plantinga’s assignment. Proposition R may leave the probability of E&N unchanged, or it can lower its probability; there is nowhere
to go but down. Unless traditional theism is quasi-deterministic with respect
to R, it too cannot be confirmed by proposition R, if Pr(R) ⊕ 1.
Bayesians like to point out that it is a consequence of Bayes’s theorem
that an observation is incapable of confirming a hypothesis when the
observation is completely unsurprising. However, for most predictions of
any interest, a Bayesian agent isn’t certain in advance that they will come
true; when surprising predictions do come true, they provide confirmation.
The wet sidewalk (W) confirms the hypothesis that it has been raining.
© 1998 University of Southern California and Blackwell Publishers Ltd.
PLANTINGA’S PROBABILITY ARGUMENTS
119
The fact that you believe that the sidewalk is wet shouldn’t lead you to
assign this evidence a probability of unity. A reasonable assignment of
value to Pr(W) is given by the fact that the sidewalk is rarely wet. The
observation is therefore somewhat surprising, and so is capable of
confirming the hypothesis. Thus, Plantinga’s claim that Pr(R) is close to
unity is very odd; it is crucial to his argument that Pr(E&N|R) is low.
Plantinga needs a better reason for this assignment than the fact that he
believes R.
Plantinga’s preliminary argument might be replaced by a different argument, one that seeks to establish that Pr(TT |R) > Pr(E&N|R). The goal
now is to compare two posterior probabilities, not to estimate their absolute values. This inequality is true precisely when
Pr( R | TT ) ⋅ Pr(TT ) > Pr( R | E & N ) ⋅ Pr( E & N ).
Notice that the value of Pr(R) is now irrelevant. The argument might
begin with the assertion that it is more probable that our psychological
mechanisms for forming beliefs are reliable if TT is true than would be
the case if E&N were true. If God exists and intervenes in natural processes
to guarantee that human beings end up with reliable cognitive faculties,
this makes R more of a sure thing than would be the case if chancy natural
processes are the only causes of the mental equipment we possess.3 (Here
we ignore the problems adumbrated in section 1.1 concerning how R
should be spelled out.) If TT and E&N are assigned the same prior probabilities, it then follows that TT has the higher posterior probability.
The present argument provides a recipe for replacing any nondeterministic theory in the natural sciences. If quantum mechanics predicts
that a certain experimental outcome was merely very probable, why not
accept instead the theistic hypothesis that this outcome was the inevitable
outcome of God’s will? Theism can be formulated in such a way that it
renders what we observe as probable as you please. Those who feel the
need to appeal to God’s intervention in the case of human mentality
should explain why they do not do so across the board.
1.3 IS PR(R|E&N) LOW? RETHINKING “DARWIN’S DOUBT”
Plantinga (pp. 223–228) argues that Pr(R|E&N) is low by enumerating
several logically conceivable scenarios that describe how beliefs and
actions might be related. For each of them, he contends that it is very
unlikely that the cognitive mechanisms that evolve should be highly reliable.4 Here are the possibilities that Plantinga considers: (i) beliefs are not
causally connected with behavior; (ii) beliefs don’t cause behavior, but
are effects of behavior or are effects of events that also cause behavior;
(iii) beliefs cause behavior, but do so in virtue of their syntax, not by
© 1998 University of Southern California and Blackwell Publishers Ltd.
120
PACIFIC PHILOSOPHICAL QUARTERLY
virtue of their semantics; (iv) the semantic properties of belief cause behavior, but the behaviors are maladaptive; (v) beliefs cause adaptive behavior.
In this last category, the adaptive behaviors may be caused by true beliefs,
but they also can be caused by false ones. To illustrate this point,
Plantinga describes a prehistorical hominid named Paul who manages to
avoid being eaten by tigers even though he desires that they should
consume him. Paul gets what’s good for him by desiring what is bad; he
stays out of trouble because his beliefs are false in just the right way. In
each of these scenarios, Plantinga says that it is improbable that our cognitive faculties should have evolved to be highly reliable. So Pr(R|E&N)
is low.
In the body of his more recent unpublished manuscript, Plantinga also
says that Pr(R|E&N) is low under scenario (ii) (epiphenomenalism);
however, he draws a different conclusion in note 15 (p. 8). If beliefs and
actions have neural events as common causes, then Plantinga concludes
that the probability is “inscrutable,” meaning that he can’t figure out what
value it should be assigned. We take this to be Plantinga’s present considered view on the matter. Although Plantinga is developing the main
argument against evolutionary naturalism in this manuscript and is not
talking about the preliminary argument, it is worth noting that this conclusion undercuts the preliminary argument, which depends on assigning
a low value to Pr(R|E&N).
Whether or not Plantinga’s considered view now is that Pr(R|E&N) is
inscrutable under scenario (ii), this is what his view should be, given the
information he considers. Assuming that beliefs don’t cause actions is not
the same as assuming that they are wholly unrelated. Resistance to malaria
doesn’t cause anemia, nor does anemia cause malaria resistance; yet, the
traits are correlated in a number of human populations because they are
phenotypes caused by the same gene. There is no way to tell a priori how
probable R is under scenario (ii).
When Plantinga turns his attention to category (v) – the case in which
beliefs cause adaptive action – he argues that false adaptive beliefs are
just as likely to evolve as true adaptive beliefs. The reason is that the
behaviors produced by a set of true beliefs also could be produced by a
set of false ones. The example of Paul shows one way this could be true.
Plantinga describes another in the unpublished manuscript. If “that is
a tree” is a true belief, then “that is a witch-tree” is a false belief that
would lead to the same behavioral consequences, and so be equally fit.
Plantinga’s mistake here is that he ignores the fact that the probability of
a trait’s evolving depends not just on its fitness, but on its availability. The
reason zebras don’t have machine guns with which to repel lion attacks
is not that firing machine guns would have been less adaptive than simply
running away; the trait didn’t evolve because it was not available as a
variation on which selection could act ancestrally (see also, Fodor 1997).
© 1998 University of Southern California and Blackwell Publishers Ltd.
PLANTINGA’S PROBABILITY ARGUMENTS
121
This means that Plantinga’s argument that Pr(R|E&N) will be low in
category (v) situations is inadequate. Plantinga might reply that witchbeliefs and other systems of adaptive false beliefs were available ancestrally. However, we don’t see any reason to think that this substantive
claim about the past can be successfully defended. By ignoring the question of availability, Plantinga, in effect, assumes that natural selection
acts on the set of conceivable variants. This it does not do; it acts on the
set of actual variants.
In general, the way to have two (logically independent) properties be
well-correlated is to have one cause the other, or to have each trace back
to a common cause. If belief and action failed to be causally connected
in either of these two ways, then it would be surprising for selection on
action to lead cognitive mechanisms to evolve that are highly reliable.
However, if belief and action are causally connected, then it takes a more
detailed argument than Plantinga provides for concluding that reliable
belief formation devices are unlikely to evolve via selection on actions.
Proposition R is improbable under scenario (i), but that’s about all one
can say.
1.4 THE PRINCIPLE OF TOTAL EVIDENCE
Suppose Plantinga is right in saying that Pr(E&N|R) is low. It does
not follow that E&N is improbable relative to all relevant evidence.
Evolutionary naturalists can happily accept the idea that the conditional
probability just mentioned is low; it does not follow that they should have
less confidence than they presently do in the truth of evolutionary theory
and naturalism.
If you draw a card at random from a standard deck of cards, the probability is only 1 in 52 that you will draw the seven of diamonds. If you
do draw this card, that doesn’t mean that you should conclude that the
deck isn’t standard or that the card wasn’t drawn at random. If you have
independent evidence that the deck is standard and that the draw was
random, you simply accept the fact that some of the things that happen
don’t have high probabilities. Even if it turns out that there are features
of human cognitive makeup that are improbable on the hypothesis that
human beings evolved, there is lots of evidence that the human mind is
a product of evolution. In this light, the sensible thing to do is to accept
evolutionary theory and come to terms with the fact that evolutionary
processes sometimes have improbable outcomes.5
As for the separate hypothesis of naturalism, by which Plantinga means
atheism together with some other claims that he does not spell out, these
too must be evaluated in the light of all the evidence, not just with respect
to proposition R.
© 1998 University of Southern California and Blackwell Publishers Ltd.
122
PACIFIC PHILOSOPHICAL QUARTERLY
1.5 A CONTRADICTION AND TWO WAYS OUT
We have mentioned that Plantinga thinks Pr(R) is close to 1, Pr(R|E&N)
is low, Pr(R|TT ) is high, and that Pr(E&N) and Pr(TT ) are “comparable.”
Plantinga’s preliminary argument also includes the assumption that N
and TT are the only two “significant alternatives” (p. 228).
If the claim that Pr(E&N) and Pr(TT ) are “comparable” means that
their values are not far apart, and if the claim that E&N and TT are the
only two “significant alternatives” means that they are the only possibilities that have non-negligible prior probabilities, then this set of probability claims is contradictory. To see why, let’s expand Pr(R):
Pr( R ) = Pr( R | E & N ) ⋅ Pr( E & N ) + Pr( R | TT ) ⋅ Pr(TT ).
If E&N and TT are exhaustive, then Plantinga’s claim that they have
“comparable” priors means that they are each close to 0.5. Substituting
this and the other values that Plantinga assigns to the component expressions, we obtain
1 ≈ ( low ) ⋅ ( 0.5) + ( high ) ⋅ ( 0.5).
This is impossible – a contradiction of the axioms of probability.
There are a couple of ways out of this difficulty. One is to retain all
the probability assignments, but deny that E&N and TT exhaust the significant alternatives. If a third possibility, theory X, is countenanced, then
Pr(R) expands to
Pr( R ) = Pr( R | E & N ) ⋅ Pr( E & N ) + Pr( R | TT ) ⋅ Pr(TT )
+ Pr( R | X ) ⋅ Pr( X ).
If Pr(R) ⊕ 1, Pr(E&N) and Pr(TT ) are about the same, and Pr(R|E&N)
is low, then Plantinga must assign Pr(E&N) and Pr(TT ) negligible probabilities, so that Pr(X ) is close to 1. He also must assign Pr(R|X ) a value
close to unity. This revision in Plantinga’s argument thus requires the
existence of an alternative to traditional theism that is vastly more probable a priori, and which entails that proposition R is very probable indeed.
The effect of these assignments is to make Pr(E&N|R) and Pr(TT |R) low
and Pr(X |R) high. If a low value for Pr(E&N|R) suffices to reject evolutionary naturalism, then one should reject traditional theism as well. This
revision of his argument indicates that the most acceptable alternative is
theory X.
Another way to avoid contradiction is to reinterpret what it means for
Pr(E&N) and Pr(TT ) to be “comparable.” Plantinga has suggested to us
(personal communication) that this should be taken to mean that the agent
doesn’t believe that the two theories have very different probabilities. For
© 1998 University of Southern California and Blackwell Publishers Ltd.
PLANTINGA’S PROBABILITY ARGUMENTS
123
example, suppose that the agent places E&N and TT in the same wide
interval of probabilities (say, between 0.05 and 0.95) and isn’t able to be
more specific than this.
Plantinga’s various claims can be rendered consistent by this revision.
To see why, let’s return to the expansion of Pr(R) and see what happens
when we let Pr(E&N ) and Pr(TT ) be unknown, save for the fact that
each must fall between 0.05 and 0.95:
Pr( R ) = Pr( R | E & N ) ⋅ Pr( E & N ) + Pr( R | TT ) ⋅ Pr(TT )
1 ≈ ( low ) ⋅ (?) + ( high ) ⋅ (?)
The insertion of question marks insures that no contradiction arises.
However, values for the question mark quantities are not left open;
Pr(E&N) must be very close to zero and Pr(TT ) must be very close to
unity. The argument is now consistent, but is entirely deprived of its probative force. Consistency (and the revised interpretation of what “comparable” means) require the assumption that traditional theism is virtually
certain a priori, and that evolutionary naturalism is almost certainly false,
again a priori. Those not already convinced before proposition R is considered that traditional theism is vastly more probable than evolutionary
naturalism will reject the argument at the outset. In addition, this revision
of Plantinga’s preliminary argument undermines its original motivation.
Plantinga’s thought was to develop what he calls “Darwin’s doubt” – that
Pr(R|E&N) is low. However, once Pr(R) is set close to 1, and Pr(E&N) is
assumed to be small, it automatically follows that Pr(E&N|R) is still
smaller, no matter what value Pr(R|E&N) happens to have.
2. Plantinga’s Main Argument Against E&N
The argument just described is preliminary to the main event, in which
Plantinga (pp. 234–35) argues that E&N is self-defeating. The main argument doesn’t aim to show that the conjunction E&N is probably false (or
that E&N is less probable than TT ), but that people shouldn’t believe E&N:
1.
2.
Pr(R|E&N) is low or its value is inscrutable.
Therefore, E&N is a defeater of R – if you believe E&N, then you
should withhold assent from R.
3. If you should withhold assent from R, then you should withhold
assent from anything else you believe.
4. If you believe E&N, then you should withhold assent from E&N
(E&N is self-defeating).
∴ You should not believe E&N.
© 1998 University of Southern California and Blackwell Publishers Ltd.
124
PACIFIC PHILOSOPHICAL QUARTERLY
Plantinga goes further, in both the book and the manuscript. He argues
that naturalism by itself is self-defeating. In the book, he says that “if
naturalism is true, then, surely so is evolution” (p. 236). In the manuscript,
he says that Pr(E|N) is high (p. 11). Neither of these claims is right. Recall
that proposition E adverts to the mechanisms described in contemporary
evolutionary theory. If that theory were found wanting, it would not entail
the falsehood of naturalism; naturalists could quite consistently cast about
for a better scientific theory. In the manuscript Plantinga makes the point
that E is “the only game in town” for a naturalist; this may or may not
be true (now), but that hardly shows that naturalism on its own makes
1990’s evolutionary theory probable.
2.1 PROBLEMS THAT THE MAIN ARGUMENT INHERITS FROM THE
PRELIMINARY ARGUMENT
We have already discussed why we are unconvinced by Plantinga’s argument that Pr(R|E&N) is low. We also argued that even if Pr(E&N|R) were
low, that would not entail that R suffices to reject E&N. The symmetrical
point that pertains to the main argument is that even if Pr(R|E&N) were
low, that would not oblige people who believe E&N to withhold belief
from R. After all, people who believe E&N might have other reasons for
believing R. For example, they might argue that R is a basic proposition
that does not need theoretical support, or that R derives its epistemic credentials from something other than the thesis of evolutionary naturalism.
The same point holds if you don’t know what value to assign to
Pr(R|E&N). People who believe E&N should not regard the fact that this
probability is “inscrutable” to them as a reason to reject R. We suspect
that many people who are well acquainted with the theory of special relativity and who think that birds fly still don’t know what value to assign
to Pr(Special relativity|birds fly), especially if probability has to be an
objective quantity; however, that doesn’t show that they should withhold
belief in special relativity. The Principle of Indifference is flawed because
it claims to obtain probabilities from ignorance; the start of Plantinga’s
main argument makes the complementary mistake of holding that ignorance of probabilities is a guide to belief.
In the light of these points, consider the following passage from Warrant
and Proper Function (p. 229, emphasis added) in which Plantinga justifies
the first step of the main argument:
Someone who accepts E&N and also believes that the proper attitude toward Pr(R|E&N)
is one of agnosticism [or, one of low degree of belief] clearly enough has good reason for
being agnostic about [or, having a low degree of belief with respect to] R as well. She has
no other information about R . . . but the source of information she does have gives her no
reason to believe R and no reason to disbelieve it.
© 1998 University of Southern California and Blackwell Publishers Ltd.
PLANTINGA’S PROBABILITY ARGUMENTS
125
Notice that Plantinga assumes that evolutionary naturalists have no basis
for deciding what to think about R, other than the proposition E&N itself.
This crucial assumption is never defended in either Warrant and Proper
Function or “Naturalism Defeated.”
2.2 WHAT DEFEATING R MEANS
In the second step of the main argument, Plantinga says that E&N’s defeat
of R means that evolutionary naturalists should withhold assent from
anything else they believe – for example, from E&N itself. This goes
beyond what the defeat of proposition R really entails. Proposition R says
that “the great bulk” of the beliefs we have are true (Plantinga 1994, p.
2). If evolutionary naturalists should withhold assent from R, this does
not mean that they should withhold assent from most of what they believe,
much less from everything they believe. Even if E&N defeats the claim
that “at least 90% of our beliefs are true,” it does not follow that E&N
also defeats the more modest claim that “at least 50% of our beliefs are
true.” Plantinga must show that E&N not only defeats R, but also defeats
the claim that “at least a non-negligible minority of our beliefs are true.”
2.3 CONDITIONAL PROBABILITY AND DEFEAT
Although, like a number of other commentators, we have interpreted the
main argument of Warrant and Proper Function as asserting that a low
value for Pr(R|E&N) suffices for E&N to defeat R, Plantinga (1994) denies
that this is what he meant and tries to develop an account of defeat that
clarifies how the argument is supposed to go. However, Plantinga still
spends time arguing in this later manuscript that Pr(R|E&N) is low or
inscrutable, so he presumably still holds that the value of this probability
is relevant to establishing that E&N defeats R.
Plantinga develops three principles that he thinks govern the defeat
relation. Although he explains why he thinks these principles are correct,
he never explains how they are relevant to establishing that E&N defeats
R. In fact, their logical form renders them incapable of closing the gap
between premisses 1 and 2. The first and third principles assert sufficient
conditions for X ’s not defeating Y. The second states a necessary
condition for X ’s defeating Y.
In addition, Plantinga’s “First Principle of Defeat” apparently helps
establish that E&N is not a defeater of R. Substituting R and E&N for
A and B in this principle yields
If S rationally believes that the warrant that E&N has for him is derivative from the warrant that R has for him, then E&N is not a defeater,
for him, of R.
© 1998 University of Southern California and Blackwell Publishers Ltd.
126
PACIFIC PHILOSOPHICAL QUARTERLY
We suspect that many evolutionary naturalists rationally believe that their
warrant for believing E&N depends on their being warranted in believing
that their cognitive faculties are highly reliable.6
Not only is a low value for Pr(X |Y) not sufficient for Y ’s defeating X;
it also is not necessary, if defeaterhood is to ground the idea of self-defeat.
The reason is that Pr(Y |Y) = 1, for all Y. And as difficult as it is to connect low probability to defeaterhood, it seems even harder to see why the
inscrutability of Pr(X |Y) should help establish that Y defeats X.
In the preliminary argument, Plantinga assigns to Pr(R) a value close
to unity because he believes R to be true. In the main argument as formulated in Warrant and Proper Function, he gives the impression that he
thinks evolutionary naturalists should withhold belief in R because E&N
fails to confer a sufficiently high probability on R. These are two ways
of expressing the same sentiment: high probability is necessary for rational
belief. However, the more recent manuscript “Naturalism Defeated” repudiates the idea that there is any such simple relation between probability
and acceptance.
What Plantinga is coming up against here is a close relative of the
phenomenon that Kyburg’s (1961) lottery paradox made vivid. Suppose
there are 10,000 tickets in a fair lottery; one ticket will win and each has
the same chance of winning. Suppose you adopt the following criterion
for belief – you accept a proposition if you think it has a high probability.
If so, you will accept each proposition of the form “ticket i won’t win.”
However, the conjunction of these contradicts the starting assumption
that the lottery is fair. Therefore, high probability is not sufficient for
rational belief. A similar counterexample can be constructed to show that
high probability is also not necessary for rational belief. Consider any n
propositions P1, . . . ,Pn such that (i) you accept each of the Pi, and (ii)
each of the Pi is very highly probable. The conjunction P1& . . . &Pn may
turn-out to be quite improbable (see section 1.1 for a salient example of
this probabilistic phenomenon). Nonetheless, it apparently would be
rational for you to accept the conjunction P1& . . . &Pn. Hence, high
probability is not necessary for rational belief (see, also, Maher 1993,
section 6.2.4). Philosophers of probability have extracted from these
paradoxes one of two lessons – either the concepts of acceptance and
rejection are suspect, or they are more subtly related to the concept of
probability than the threshold criterion just described.
This connection with the lottery paradox suggests that the task of
repairing the main argument is formidable. That argument begins with
claims about probability, moves to claims about defeat, and then concludes with a claim about self-defeat. Each step along the way requires
principles quite different from the ones that Plantinga has so far described.
Whether plausible principles exist that forge the requisite connections we
leave to the reader to conjecture.
© 1998 University of Southern California and Blackwell Publishers Ltd.
PLANTINGA’S PROBABILITY ARGUMENTS
127
3. A Question for Evolutionism
Although Plantinga’s arguments don’t work, he has raised a question that
needs to be answered by people who believe evolutionary theory and who
also believe that this theory says that our cognitive abilities are in various
ways imperfect. Evolutionary theory does say that a device that is reliable
in the environment in which it evolved may be highly unreliable when
used in a novel environment. It is perfectly possible that our mental
machinery should work well on simple perceptual tasks, but be much less
reliable when applied to theoretical matters. We hasten to add that this
is possible, not inevitable. It may be that the cognitive procedures that
work well in one domain also work well in another; modus ponens may
be useful for avoiding tigers and for doing quantum physics.
Anyhow, if evolutionary theory does say that our ability to theorize about
the world is apt to be rather unreliable, how are evolutionists to apply this
point to their own theoretical beliefs, including their belief in evolution?
One lesson that should be extracted is a certain humility – an admission of
fallibility. This will not be news to evolutionists who have absorbed the fact
that science in general is a fallible enterprise. Evolutionary theory just
provides an important part of the explanation of why our reasoning about
theoretical matters is fallible.
Far from showing that evolutionary theory is self-defeating, this consideration should lead those who believe the theory to admit that the best
they can do in theorizing is to do the best they can. We are stuck with
the cognitive equipment that we have. We should try to be as scrupulous
and circumspect about how we use this equipment as we can. When we
claim that evolutionary theory is a very well confirmed theory, we are
judging this theory by using the fallible cognitive resources we have at
our disposal. We can do no other.
Plantinga suggests that evolutionary naturalism is self-defeating, but
that traditional theism is not. However, what is true is that neither position
has an answer to hyperbolic doubt. Evolutionists have no way to justify
the theory they believe other than by critically assessing the evidence that
has been amassed and employing rules of inference that seem on reflection
to be sound. If someone challenges all the observations and rules of inference that are used in science and in everyday life, demanding that they
be justified from the ground up, the challenge cannot be met. A similar
problem arises for theists who think that their confidence in the reliability
of their own reasoning powers is shored up by the fact that the human
mind was designed by a God who is no deceiver. The theist, like the evolutionary naturalist, is unable to construct a non-question-begging argument that refutes global skepticism.
© 1998 University of Southern California and Blackwell Publishers Ltd.
128
PACIFIC PHILOSOPHICAL QUARTERLY
Philosophy Department, University of Wisconsin–Madison
NOTES
*Gordon Barnes, Matt Davidson, Ellery Eells, Malcolm Forster, Patrick Maher, Ernan
McMullin, Alvin Plantinga, and Dennis Stampe provided valuable criticisms and suggestions. We are grateful to them for their help.
1
All page references are to Warrant and Proper Function (viz., Plantinga, 1993), unless
otherwise noted.
2
An even better strategy would be to associate a characteristic degree of sensitivity with
a mental faculty in a given environmental setting. Roughly speaking, sensitivity is a worldto-head relation, measured by the probability that the agent will believe p, conditional on
p’s being true. In contrast, reliability is a head-to-world relation, measured by the probability
that p will be true, conditional on the agent’s believing p. Sensitivity tends to be a more
stable property of measurement devices than reliability. See Sober (1994, essays 3 and 12)
for discussion.
3
Here we go along with Plantinga’s usage of the term “traditional theism,” according to
which this doctrine makes different observational predictions than evolutionary naturalism.
However, there is room to argue whether this way of seeing matters is the only one that is
available from various religious traditions (McMullin 1993). Plantinga understands traditional theism to be the idea, not just that God sets evolutionary processes in motion (where
these are understood in terms of the best theories that science now provides), but that he
occasionally intervenes in them to insure certain outcomes. The idea that God does the
former, but not the latter, confers on proposition R precisely the same probability that
evolutionary theory by itself confers on R. This can be seen more clearly by considering the
accompanying Figure.
God (G)
Evolutionary Processes (E)
Observations (O)
Figure 1: The possible relationships between G, E, and O.
If evolutionary processes (E ) “screen off” God’s activity (G) from what we can observe
(O), then Pr(O|E&G) = Pr(O|E¬-G). Plantinga thinks this equality is false; he holds that
atheistic evolutionism confers on the observations (specifically, on proposition R) a probability different from the one provided by theistic evolutionism. This means that Plantinga
is thinking of God as not simply acting through natural evolutionary processes, but as
affecting the world by a separate, “miraculous” pathway.
4
In discussing “Darwin’s doubt,” Plantinga (1994, p. 4) quotes with approval a point
made by Churchland (1987, p. 548) to the effect that natural selection “cares” only about
how adaptive the behaviors are that a set of beliefs causes; it does not care, in addition,
whether those beliefs are true. Plantinga interprets this to mean that true beliefs are no more
likely to evolve than false ones, but a probabilistic representation of Churchland’s point
(which is about conditional independence) shows that this does not follow. Churchland’s
point is that
Pr(Belief set B evolves | B produces adaptive behaviors & B is true)
= Pr (Belief set B evolves | B produces adaptive behaviours & B is false).
© 1998 University of Southern California and Blackwell Publishers Ltd.
PLANTINGA’S PROBABILITY ARGUMENTS
129
However, from this it does not follow that
Pr(Belief set B evolves | B is true) = Pr(Belief set B evolves | B is false).
In just the same way, although it is true that
Pr(it will rain tomorrow | a storm is approaching & the barometer reading is low)
= Pr(it will rain tomorrow | a storm is approaching & the barometer reading is high),
it does not follow that
Pr(it will rain tomorrow | the barometer reading is low)
= Pr(it will rain tomorrow | the barometer reading is high).
5
The point we are making here accords with what Plantinga (1994) calls “the perspiration
objection,” which he attributes to “Wykstra, DePaul, and others.” Although Plantinga
discusses how the objection should be formulated, he does not, as far as we can see, provide
an answer to it.
6
Of course, one can’t deduce E&N from R alone. But, evolutionary naturalists might
reasonably maintain that R is one of several premises which underwrite their non-deductive
inferences concerning the plausibility of E&N. It is worth pointing out that Plantinga himself
makes use of this kind of non-deductive “warrant derivation.” On page 39 of “Naturalism
Defeated,” he says, in the context of discussing an objection to his argument, that the
warrant that P has for you (where P is such that Pr[R|N&E&P] is high, but P is logically
independent of R) is “ . . . derivative from the warrant R has for you . . . it is hard to
see what other source [of warrant] there could be [for P].” We see no reason why it would
be irrational for evolutionary naturalists to say the same thing about the warrant that E&N
has for them.
REFERENCES
Churchland, P. “Epistemology in the Age of Neuroscience,” Journal of Philosophy LXXXIV,
1987, pp. 544–53.
Fodor, J. “Is Science Biologically Possible?,” unpublished (1997) manuscript of the 1998
Benjamin Lecture at the University of Missouri.
Kahnemann, D., P. Slovic and A. Tversky (eds) Judgment Under Uncertainty: Heuristics
and Biases (Cambridge: Cambridge University Press, 1982).
Kyburg, H. Probability and the Logic of Rational Belief (Wesleyan University Press, 1961).
Maher, P. Betting on Theories (Cambridge: Cambridge University Press, 1993).
McMullin, E. “Evolution and Special Creation,” Zygon 28, 1993, pp. 299–335.
Plantinga, A. Warrant and Proper Function (Oxford, Oxford University Press, 1993).
Plantinga, A. “Naturalism Defeated,” unpublished manuscript (1994).
Sober, E. Philosophy of Biology (Westview Press, 1993).
Sober, E. From a Biological Point of View (Cambridge: Cambridge University Press, 1994).
© 1998 University of Southern California and Blackwell Publishers Ltd.
Copyright of Pacific Philosophical Quarterly is the property of Wiley-Blackwell and its
content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder's express written permission. However, users may print, download, or email
articles for individual use.
Purchase answer to see full
attachment