Laws of Nature

Author(s): Fred I. Dretske

Reviewed work(s):
Source: Philosophy of Science, Vol. 44, No. 2 (Jun., 1977), pp. 248-268
Published by: The University of Chicago Press on behalf of the Philosophy of Science Association
Stable URL: http://www.jstor.org/stable/187350 .
Accessed: 18/05/2012 15:27

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

The University of Chicago Press and Philosophy of Science Association are collaborating with JSTOR to
digitize, preserve and extend access to Philosophy of Science.

http://www.jstor.org
 LAWS OF NATURE*

FRED I. DRETSKEt
Universityof Wisconsin

It is a traditionalempiricistdoctrinethat naturallaws are universaltruths.

In order to overcome the obvious difficulties with this equation most
empiricistsqualify it by proposingto equate laws with universaltruthsthat
play a certain role, or have a certain function, within the larger scientific
enterprise.This view is examinedin detail and rejected; it fails to account
for a variety of featuresthat laws are acknowledgedto have. An alternative
view is advanced in which laws are expressed by singular statements of
fact describingthe relationshipbetweenuniversalpropertiesand magnitudes.

It is tempting to identify the laws of nature with a certain class

of universal truths. Very few empiricists have succeeded in resisting
this temptation. The popular way of succumbing is to equate the
fundamental laws of nature with what is asserted by those universally
true statements of nonlimited scope that embody only qualitative
predicates.' On this view of things a law-like statement is a statement
of the form "(x)(Fx D Gx)" or "(x)(Fx Gx)" where "F" and
"G" are purely qualitative (nonpositional). Those law-like statements
that are true express laws. "All robins' eggs are greenish blue,"
"All metals conduct electricity," and "At constant pressure any gas
expands with increasing temperature" (Hempel's examples) are law-
like statements. If they are true, they express laws. The more familiar
sorts of things that we are accustomed to calling laws, the formulae
and equations appearing in our physics and chemistry books, can
supposedly be understood in the same way by using functors in place
of the propositional functions "Fx" and "Gx" in the symbolic
expressions given above.
I say that it is tempting to proceed in this way since, to put it
bluntly, conceiving of a law as having a content greater than that
expressed by a statement of the form (x)(Fx D Gx) seems to put

*ReceivedJune, 1976.
t For their helpful comments my thanks to colleagues at Wisconsin and a number
of other universities where I read earlier versions of this paper. I wish, especially,
to thankZaneParks,RobertCausey, MartinPerlmutter,NormanGillespie,and Richard
Aquillafor their critical suggestions, but they should not be blamed for the way I
garbledthem.
'This is the position taken by Hempeland Oppenheim([10]).

Philosophy of Science, 44 (1977) pp. 248-268.

Copyright ? 1977 by the Philosophy of Science Association.

248
 LAWS OF NATURE 249

it beyond our epistemological grasp.2 We must work with what we

are given, and what we are given (the observational and experimental
data) are facts of the form: this F is G, that F is G, all examined
F's have been G, and so on. If, as some philosophers have argued,3
law-like statements express a kind of nomic necessity between events,
something more than that F's are, as a matter of fact, always and
everywhere, G, then it is hard to see what kind of evidence might
be brought in support of them. The whole point in acquiring instantial
evidence (evidence of the form "This Fis G") in support of a law-like
hypothesis would be lost if we supposed that what the hypothesis
was actually asserting was some kind of nomic connection, some
kind of modal relationship, between things that were F and things
that were G. We would, it seems, be in the position of someone
trying to confirm the analyticity of "All bachelors are unmarried"
by collecting evidence about the marital status of various bachelors.
This kind of evidence, though relevant to the truth of the claim that
all bachelors are unmarried, is powerless to confirm the modality
in question. Similarly, if a hypothesis, in order to quality as a law,
must express or assert some form of necessity between F's and G's,
then it becomes a mystery how we ever manage to confirm such
attributions with the sort of instantial evidence available from observa-
tion.
Despite this argument, the fact remains that laws are not simply
what universally true statements express, not even universally true
statements that embody purely qualitative predicates (and are, as a
result, unlimited in scope). This is not particularly newsworthy. It
is commonly acknowledged that law-like statements have some pecu-
liarities that prevent their straightforward assimilation to universal
truths. That the concept of a law and the concept of a universal
truth are different concepts can best be seen, I think, by the following
consideration: assume that (x)(Fx D Gx) is true and that the predicate
expressions satisfy all the restrictions that one might wish to impose
in order to convert this universal statement into a statement of law.4

2Whenthe statementis of nonlimitedscope it is alreadybeyond our epistemological

grasp in the sense that we cannot conclusivelyverify it with the (necessarily) finite
set of observationsto which traditionaltheories of confirmationrestrict themselves.
When I say (in the text) that the statement is "beyond our epistemologicalgrasp"
I have somethingmore serious in mindthan this rathertriviallimitation.
3Mostprominently,WilliamKneale in [12] and [13].
41eliminatequoteswhentheirabsencewill cause no confusion. I will also, sometimes,
speakof laws and statementsof law indifferently.I think, however, that it is a serious
mistake to conflate these two notions. Laws are what is expressed by true lawlike
statements(see [1], p. 2, for a discussion of the possible senses of "law" in this
regard).I will returnto this point later.
250 FRED I. DRETSKE

Consider a predicate expression "K" (eternally) coextensive with

"F"; i.e., (x)(Fx = Kx) for all time. We may then infer that if
(x)(Fx D Gx) is a universal truth, so is (x)(Kx D Gx). The class
of universal truths is closed under the operation of coextensive
predicate substitution. Such is not the case with laws. If it is a law
that all F's are G, and we substitute the term "K" for the term
"F" in this law, the result is not necessarily a law. If diamonds
have a refractive index of 2.419 (law) and "is a diamond" is coextensive
with "is mined in kimberlite (a dark basic rock)" we cannot infer
that it is a law that things mined in kimberlite have a refractive
index of 2.419. Whether this is a law or not depends on whether
the coextensiveness of "is a diamond" and "is mined in kimberlite"
is itself law-like. The class of laws is not closed under the same
operation as is the class of universal truths.
Using familiar terminology we may say that the predicate positions
in a statement of law are opaque while the predicate positions in
a universal truth of the form (x)(Fx D Gx) are transparent. I am
using these terms in a slightly unorthodox way. It is not that when
we have a law, "All F's are G," we can alter its truth value by
substituting a coextensive predicate for "F" or "G." For if the
statement is true, it will remain true after substitution. What happens,
rather, is that the expression's status as a law is (or may be) affected
by such an exchange. The matter can be put this way: the statement
(A) All F's are G (understood as (x)(Fx D Gx))
has "F" and "G" occurring in transparent positions. Its truth value
is unaffected by the replacement of "F" or "G" by a coextensive
predicate. The same is true of

(B) It is universally true that F's are G.

If, however, we look at

(C) It is a law that F's are G.

we find that "F" and "G" occur in opaque positions. If we think
of the two prefixes in (B) and (C), "it is universally true that . . ."
and "it is a law that. . .," as operators, we can say that the operator
in (B) does not, while the operator in (C) does, confer opacity on
the embedded predicate positions. To refer to something as a statement
of law is to refer to it as an expression in which the descriptive
terms occupy opaque positions. To refer to something as a universal
truth is to refer to it as an expression in which the descriptive terms
occupy transparent positions. Hence, our concept of a law differs
 LAWS OF NATURE 251

from our concept of a universaltruth.5

Confronted by a difference of this sort, many philosophers have
argued that the distinction between a natural law and a universal
truth was not, fundamentally, an intrinsic difference. Rather, the
difference was a difference in the role some universal statements
played within the larger theoretical enterprise. Some universal state-
ments are more highly integratedinto the constellation of accepted
scientificprinciples,they play a more significantrole in the explanation
and prediction of experimental results, they are better confirmed,
have survived more tests, and make a more substantialcontribution
to the regulation of experimental inquiry. But, divorced from this
context, stripped of these extrinsic features, a law is nothing but
a universal truth. It has the same empirical content. Laws are to
universal truths what shims are to slivers of wood and metal; the
latter become the former by being used in a certain way. There is
a functional difference, nothingelse.6
Accordingto this reductionisticview, the peculiaropacity (described
above) associated with laws is not a manifestation of some intrinsic
differencebetween a law and a universaltruth. It is merelya symptom
of the special status or function that some universalstatements have.
The basic formulais: law = universaltruth + X. The "X" is intended
to indicate the special function, status or role that a universal truth
must have to qualify as a law. Some popular candidates for this
auxiliaryidea, X, are:
(1) High degree of confirmation,
(2) Wide acceptance (well established in the relevant community),

5Popper([17]) vaguely perceives, but fails to appreciatethe significance of, the

same (or a similar)point. He distinguishesbetween the structureof terms in laws
and universalgeneralizations,referringto their occurrence in laws as "intensional"
and their occurrence in universal generalizationsas "extensional." Popper fails to
develop this insight, however, and continues to equate laws with a certain class of
universaltruths.
6Nelson Goodmangives a succinct statement of the functionalistposition: "As a
first approximationthen, we might say that a law is a true sentence used for making
predictions.IThatlaws are used predictivelyis of course a simple truism, and I am
not proposingit as a novelty. I want only to emphasizethe Humeanidea that rather
thana sentence being used for predictionbecause it is a law, it is called a law because
it is used for prediction;and that ratherthan the law being used for predictionbecause
it describesa causalconnection,the meaningof the causalconnectionis to be interpreted
in terms of predictiveityused laws" ([7], p. 26). Among functionalistsof this sort
I would include Ayer ([2]), Nagel ([16]), Popper([17]), Mackie ([14]), Bromberger
([6]), Braithwaite([3]), Hempel ([10], [11]) and many others. Achinstein is harder
to classify. He says thatlaws express regularitiesthat can be cited in providinganalyses
andexplanations([1], p. 9), but he has a ratherbroadidea of regularities:"regularities
mightalso be attributedto properties"([1], pages 19, 22).
252 FRED I. DRETSKE

(3) Explanatorypotential (can be used to explain its instances),

(4) Deductive integration(within a largersystem of statements),
(5) Predictiveuse.
To illustratethe way these values of Xare used to buttressthe equation
of laws with universal truths, it should be noted that each of the
concepts appearingon this list generates an opacity similar to that
witnessed in the case of genuine laws. For example, to say that it
is a law that all F's are G may possibly be no more than to say
that it is well established that (x)(Fx D Gx). The peculiar opacity
of laws is then explained by pointing out that the class of expres-
sions that are well established (or highly confirmed) is not closed
under substitution of coextensive predicates: one cannot infer that
(x)(Kx D Gx) is well establishedjust because "Fx" and "Kx" arecoex-
tensive and (x)(Fx D Gz) is well established (for no one may know
that "Fx" and "Kx" arecoextensive). It may be supposed, therefore,
that the opacity of laws is merely a manifestationof the underlying
fact that a universal statement, to qualify as a law, must be well
established, and the opacity is a result of this epistemic condition.
Or, if this will not do, we can suppose that one of the other notions
mentionedabove, or a combinationof them, is the source of a law's
opacity.
This response to the alleged uniqueness of naturallaws is more
or less standard fare among empiricists in the Humean tradition.
Longstanding(= venerable) epistemologicaland ontological commit-
ments motivate the equation: law = universal truth + X. There is
disagreement among authors about the differentia X, but there is
near unanimity about the fact that laws are a species of universal
truth.
If we set aside our scruples for the moment, however, there is
a plausible explanationfor the opacity of laws that has not yet been
mentioned. Taking our cue from Frege, it may be argued that since
the operator "it is a law that .. ." converts the otherwise transparent
positions of "All F's are G" into opaque positions, we may conclude
that this occurs because within the context of this operator (either
explicitly present or implicitly understood)the terms "F" and "G"
do not have their usual referents. There is a shift in what we are
talking about. To say that it is a law that F's are G is to say that
"All F's are G" is to be understood (in so far as it expresses a
law), not as a statementabout the extensions of the predicates "F"
and "G," but as a singularstatementdescribinga relationshipbetween
the universal properties F-ness and G-ness. In other words, (C) is
to be understoodas having the form:
 LAWS OF NATURE 253

(6) F-ness -> G-ness.7

To conceive of (A) as a universal truth is to conceive of it as expressing
a relationship between the extensions of its terms; to conceive of
it as a law is to conceive of it as expressing a relationship between
the properties (magnitudes, quantities, features) which these predicates
express (and to which we may refer with the corresponding abstract
singular term). The opacity of laws is merely a manifestation of this
change in reference. If "F" and "K" are coextensive, we cannot
substitute the one for the other in the law "All F's are G" and
expect to preserve truth; for the law asserts a connection between
F-ness and G-ness and there is no guarantee that a similar connection
exists between the properties K-ness and G-ness just because all
F's are K and vice versa.8
It is this view that I mean to defend in the remainder of this essay.
Law-like statements are singular statements of fact describing a
relationship between properties or magnitudes. Laws are the relation-
ships that are asserted to exist by true law-like statements. According
to this view, then, there is an intrinsic difference between laws and
universal truths. Laws imply universal truths, but universal truths
do not imply laws. Laws are (expressed by) singular statements
describing the relationships that exist between universal qualities and

7I attach no special significance to the connective "--." I use it here merely as

a dummy connective or relation. The kind of connection asserted to exist between
the universalsin question will depend on the particularlaw in question, and it will
vary dependingon whetherthe law involves quantitativeor merely qualitativeexpres-
sions. For example, Ohm's Law asserts for a certain class of situations a constant
ratio (R) between the magnitudesE (potential difference) and I (currentintensity),
a fact that we use the "=" sign to represent: E/I = R. In the case of simple qualitative
laws (thoughI doubt whetherthere are many genuinelaws of this sort) the connective
"--" merely expresses a link or connection between the respectivequalitiesand may
be read as "yields." If it is a law that all men are mortal, then humanity yields
mortality(humanity-> mortality).Incidentally,I am not denying that we can, and
do, express laws as simply "All F's are G" (sometimes this is the only convenient
way to express them). All I am suggestingis that when lawlikestatementsare presented
in this formit maynot be clearwhatis beingasserted:a law or a universalgeneralization.
When the context makes it clear that a relation of law is being described, we can
(withoutambiguity)express it as "All F's are G" for it is then understoodin the
mannerof (6).
8 On the basis of an
argumentconcernedwith the restrictionson predicateexpressions
that may appear in laws, Hempel reaches a similar conclusion but he interprets it
differently. "Epitomizingthese observations we might say that a lawlike sentence
of universal nonprobabilisticcharacteris not about classes or about the extensions
of the predicateexpressions it contains, but about these classes or extensions under
certaindescriptions"([11], p. 128).I guess I do not know what being about something
undera descriptionmeans unless it amountsto being about the propertyor feature
expressedby that description.I returnto this point later.
254 FRED I. DRETSKE

quantities; they are not universal statements about the particular objects
and situations that exemplify these qualities and quantities. Universal
truths are not transformed into laws by acquiring some of the extrinsic
properties of laws, by being used in explanation or prediction, by
being made to support counterfactuals, or by becoming well estab-
lished. For, as we shall see, universal truths cannot function in these
ways. They cannot be made to perform a service they are wholly
unequipped to provide.
In order to develop this thesis it will be necessary to overcome
some metaphysical prejudices, and to overcome these prejudices it
will prove useful to review the major deficiencies of the proposed
alternative. The attractiveness of the formula: law = universal truth
+ X, lies, partly at least, in its ontological austerity, in its tidy portrayal
of what there is, or what there must be, in order for there to be
laws of nature. The antidote to this seductive doctrine is a clear
realization of how utterly hopeless, epistemologically and functionally
hopeless, this equation is.
If the auxiliary ideas mentioned above (explanation, prediction,
confirmation, etc.) are deployed as values of X in the reductionistic
equation of laws with universal truths, one can, as we have already
seen, render a satisfactory account of the opacity of laws. In this
particular respect the attempted equation proves adequate. In what
way, then, does it fail?
(1) and (2) are what I will call "epistemic" notions; they assign
to a statement a certain epistemological status or cognitive value.
They are, for this reason alone, useless in understanding the nature
of a law.9 Laws do not begin to be laws only when we first become
aware of them, when the relevant hypotheses become well established,
when there is public endorsement by the relevant scientific community.
The laws of nature are the same today as they were one thousand
years ago (or so we believe); yet, some hypotheses are highly confirmed
today that were not highly confirmed one thousand years ago. It
is certainly true that we only begin to call something a law when
it becomes well established, that we only recognize something as
a statement of law when it is confirmed to a certain degree, but
that something is a law, that some statement does in fact express
a law, does not similarly await our appreciation of this fact. We
discover laws, we do not invent them-although, of course, some
invention may be involved in our manner of expressing or codifying
these laws. Hence, the status of something as a statement of law

9Molnar([15]) has an excellent brief critiqueof attemptsto analyze a law by using

epistemicconditionsof the kind being discussed.
 LAWS OF NATURE 255

does not depend on its epistemological status. What does depend

on such epistemologicalfactors is our ability to identify an otherwise
qualified statement as true and, therefore, as a statement of law.
It is for this reason that one cannot appealto the epistemic operators
to clarify the nature of laws; they merely confuse an epistemological
with an ontologicalissue.
What sometimes helps to obscure this point is the tendency to
conflate laws with the verbal or symbolic expression of these laws
(what I have been calling "statements of law"). Clearly, though,
these are different things and should not be confused. There are
doubtless laws that have not yet (or will never) receive symbolic
expression, and the same law may be given different verbal codifica-
tions (thinkof the variety of ways of expressingthe laws of thermody-
namics). To use the language of "propositions" for a moment, a
law is the proposition expressed, not the vehicle we use to express
it. The use of a sentence as an expression of law depends on
epistemologicalconsiderations,but the law itself does not.
There is, furthermore, the fact that whatever auxiliary idea we
select for understandinglaws (as candidates for X in the equation:
law = universaltruth + X), if it is going to achieve what we expect
of it, should help to account for the variety of other features that
laws are acknowledged to have. For example, it is said that laws
"support" counterfactuals of a certain sort. If laws are universal
truths, this fact is a complete mystery, a mystery that is usually
suppressed by using the word "support." For, of course, universal
statements do not imply counterfactuals in any sense of the word
"imply" with which I am familiar. To be told that all F's are G
is not to be told anything that implies that if this x were an F, it
would be G. To be told that all dogs born at sea have been and
will be cocker spaniels is not to be told that we would get cocker
spaniel pups (or no pups at all) if we arrangedto breed dachshunds
at sea. The only reason we might think we were being told this is
because we do not expect anyone to assert that all dogs born at
sea will be cocker spaniels unless they know (or have good reasons
for believing)that this is true; and we do not understandhow anyone
could know that this is true without being privy to informationthat
insures this result-without, that is, knowing of some bizzare law
or circumstancethat preventsanythingbut cocker spanielsfrom being
born at sea. Hence, if we accept the claim at all, we do so with
a certain presumptionabout what our informantmust know in order
to be a serious claimant. We assume that our informant knows of
certain laws or conditions that insure the continuance of a past
regularity, and it is this presumed knowledge that we exploit in
256 FRED I. DRETSKE

endorsingor acceptingthe counterfactual.But the simple fact remains

that the statement "All dogs born at sea have been and will be cocker
spaniels" does not itself supportor imply this counterfactual;at best,
we support the counterfactual(if we support it at all) on the basis
of what the claimant is supposed to know in order to advance such
a universalprojection.
Given this incapacity on the part of universal truths to support
counterfactuals,one would expect some assistance from the epistemic
condition if laws are to be analyzed as well established universal
truths.But the expectationis disappointed;we are left with a complete
mystery. For if a statement of the form "All F's are G" does not
support the counterfactual, "If this (non-G) were an F, it would
be G," it is clear that it will not support it just because it is well
established or highly confirmed. The fact that all the marbles in the
bag are red does not supportthe contention that if this (blue) marble
were in the bag, it would be red; but neither does the fact that we
know (or it is highly confirmed) that all the marbles in the bag are
red support the claim that if this marble were in the bag it would
be red. And making the universal truth more universalis not going
to repairthe difficulty. The fact that all the marbles in the universe
are(havebeen andwillbe) reddoes not implythatI cannotmanufacture
a blue marble; it implies that I will not, not that I cannot or that
if I were to try, I would fail. To represent laws on the model of
one of our epistemic operators, therefore, leaves wholly unexplained
one of the most important features of laws that we are trying to
understand.They are, in this respect, unsatisfactory candidates for
the job.
Though laws are not merely well established general truths, there
is a related point that deserves mention: laws are the sort of thing
that can become well establishedprior to an exhaustive enumeration
of the instances to which they apply. This, of course, is what gives
laws their predictive utility. Our confidence in them increases at a
much more rapidrate than does the ratio of favorableexaminedcases
to total numberof cases. Hence, we reach the point of confidently
using them to project the outcome of unexamined situations while
there is still a substantialnumberof unexaminedsituationsto project.
This feature of laws raises new problems for the reductionistic
equation. For, contrary to the argumentin the second paragraphof
this essay, it is hard to see how confirmationis possible for universal
truths. To illustrate this difficulty, consider the (presumablyeasier)
case of a general truth of finite scope. I have a coin that you have
(by examinationandtest) convinced yourself is quite normal.I propose
to flip it ten times. I conjecture (for whatever reason) that it will
 LAWS OF NATURE 257

land heads all ten times. You express doubts. I proceed to "confirm"
my hypothesis. I flip the coin once. It lands heads. Is this evidence
that my hypothesis is correct? I continue flipping the coin and it
turns up with nine straightheads. Given the opening assumptionthat
we are dealing with a fair coin, the probability of getting all ten
heads (the probability that my hypothesis is true) is now, after
examinationof 90% of the total population to which the hypothesis
applies, exactly .5. If we are guided by probability considerations
alone, the likelihood of all ten tosses being heads is now, after nine
favorable trials, a toss-up. After nine favorable trials it is no more
reasonableto believe the hypothesis than its denial. In what sense,
then, can we be said to have been accumulatingevidence (during
the first nine trials) that all would be heads? In what sense have
we been confirmingthe hypothesis?It would appearthat the probability
of my conjecture's being true never exceeds .5 until we have exhaus-
tively examined the entire populationof coin tosses and found them
all favorable. The probabilityof my conjecture's being true is either:
(i) too low (_ .5) to invest any confidence in the hypothesis, or
(ii) so high (= 1) that the hypothesis is useless for prediction. There
does not seem to be any middle ground.
Our attempts to confirm universal generalizations of nonlimited
scope is, I submit, in exactly the same impossible situation. It is
true, of course, that after nine successful trials the probabilitythat
all ten tosses will be headsis greatlyincreasedover the initialprobability
that all would be heads. The initial probability(assuminga fair coin)
that all ten tosses would be heads was on the order of .002. After
ninefavorabletrialsit is .5. In this sense I have increasedthe probability
that my hypothesis is true; I have raised its probabilityfrom .002
to .5. The importantpoint to notice, however, is that this sequence
of trials did not alter the probability that the tenth trial would be
heads. The probabilitythatthe unexaminedinstancewouldbe favorable
remainsexactly what it was before I began flipping the coin. It was
originally .5 and it is now, after nine favorable trials, still .5. I am
in no better position now, after extensive sampling, to predict the
outcome of the tenth toss than I was before I started. To suppose
otherwise is to commit the converse of the Gambler'sFallacy.
Notice, we could take the first nine trials as evidence that the
tenth trial would be heads if we took the results of the first nine
tosses as evidence that the coin was biased in some way. Then,
on this hypothesis, the probabilityof getting heads on the last trial
(and, hence, on all ten trials) would be greater than .5 (how much
greater would depend on the conjectured degree of bias and this,
in turn, would presumablydepend on the extent of sampling). This
258 FRED I. DRETSKE

new hypothesis, however, is something quite different than the original

one. The original hypothesis was of the form: (x)(Fx D Gx), all
ten tosses will be heads. Our new conjuecture is that there is a physical
asymmetry in the coin, an asymmetry that tends to yield more heads
than tails. We have succeeded in confirming the general hypothesis
(all ten tosses will be heads), but we have done so via an intermediate
hypothesis involving genuine laws relating the physical make-up of
the coin to the frequency of heads in a population of tosses.
It is by such devices as this that we create for ourselves, or some
philosophers create for themselves, the illusion that (apart from
supplementary law-like assumptions) general truths can be confirmed
by their instances and therefore qualify, in this respect, as laws of
nature. The illusion is fostered in the following way. It is assumed
that confirmation is a matter of raising the probability of a hypothesis. 10
On this assumption any general statement of finite scope can be
confirmed by examining its instances and finding them favorable.
The hypothesis about the results of flipping a coin ten times can
be confirmed by tossing nine straight heads, and this confirmation
takes place without any assumptions about the coin's bias. Similarly,
I confirm (to some degree) the hypothesis that all the people in the
hotel ballroom are over thirty years old when I enter the ballroom
with my wife and realize that we are both over thirty. In both cases
I raise the probability that the hypothesis is true over what it was
originally (before flipping the coin and before entering the ballroom).
But this, of course, isn't confirmation. Confirmation is not simply
raising the probability that a hypothesis is true, it is raising the
probability that the unexamined cases resemble (in the relevant respect)
the examined cases. It is this probability that must be raised if genuine
confirmation is to occur (and if a confirmed hypothesis to be useful
in prediction), and it is precisely this probability that is left unaffected
by the instantial "evidence" in the above examples.
In order to meet this difficulty, and to cope with hypotheses that
are not of limited scope, 11the reductionist usually smuggles into his

'?Brody argues that a qualitativeconfirmationfunction need not require that any

E that raises the degree of confirmationof H thereby (qualitatively)confirms H.
We need only require (perhapsthis is also too much) that if E does qualitatively
confirm H, then E raises the degree of confirmationof H. His argumentstake their
pointof departurefromCarnap'sexamplesagainstthe specialconsequenceandconverse
consequence condition ([4], pages 414-418). However this may be, I think it fair
to say that most writerson confirmationtheory take a confirmatorypiece of evidence
to be a piece of evidence that raises the probabilityof the hypothesis for which
it is confirmatory.How well it must be confirmedto be acceptable is anothermatter
of course.
" If the hypothesis is of nonlimitedscope, then its scope is not known to be finite.
Hence, we cannot know whether we are getting a numericalincrease in the ratio:
 LAWS OF NATURE 259

confirmatoryproceedings the very idea he professes to do without:

viz., a type of law that is not merely a universal truth. The general
truth then gets confirmed but only through the mediation of these
supplementarylaws. These supplementaryassumptions are usually
introduced to explain the regularities manifested in the examined
instances so as to provide a basis for projecting these regularities
to the unexamined cases. The only way we can get a purchase on
the unexamined cases is to introduce a hypothesis which, while
explainingthe data we already have, implies something about the
datawe do not have. To suppose that our coin is biased (first example)
is to suppose something that contributes to the explanation of our
extraordinaryrun of heads (nine straight)and simultaneouslyimplies
somethingabout the (probable)outcome of the tenth toss. Similarly
(second example) my wife and I may be attendinga reunion of some
kind, and I may suppose that the other people in the ballroom are
old classmates. This hypothesis not only explains our presence, it
implies that most, if not all, of the remainingpeople in the room
are of comparable age (well over thirty). In both these cases the
generalizationcan be confirmed, but only via the introduction of
a law or circumstance (combined with a law or laws) that helps to
explain the data already available.
One additionalexample should help to clarify these last remarks.
In sampling from an urn with a population of colored marbles, I
can confirm the hypothesis that all the marbles in the urn are red
by extractingat random several dozen red marbles (and no marbles
of any other color). This is a genuine example of confirmation,not
because I have raised the probabilityof the hypothesis that all are
red by reducingthe numberof ways it can be false (the same reduction
would be achieved if you showed me 24 marbles from the urn, all
of which were red), but because the hypothesis that all the marbles
in the urn are red, together with the fact (law) that you cannot draw
nonred marbles from an urn containing only red marbles, explains
the result of my random sampling. Or, if this is too strong, the law
that assures me that random sampling from an urn containing a
substantialnumberof nonred marbleswould reveal (in all likelihood)
at least one nonred marble lends its supportto my confirmationthat
the urn contains only (or mostly) red marbles.Withoutthe assistance
of such auxiliary laws a sample of 24 red marbles is powerless to

examined favorable cases/total number of cases. If an increase in the probability

of a hypothesis is equated with a (known) increase in this ratio, then we cannot
raise the probabilityof a hypothesis of nonlimitedscope in the simple-mindedway
describedfor hypothesesof (known)finite scope.
260 FRED I. DRETSKE

confirm a hypothesis about the total population of marbles in the

urn. To suppose otherwise is to suppose that the same degree of
confirmation would be afforded the hypothesis if you, whatever your
deceitful intentions, showed me a carefully selected set of 24 red
marbles from the urn. This also raises the probability that they are
all red, but the trouble is that it does not (due to your unknown
motives and intentions) raise the probability that the unexamined
marbles resemble the examined ones. And it does not raise this
probability because we no longer have, as the best available explanation
of the examined cases (all red), a hypothesis that implies that the
remaining (or most of the remaining) marbles are also red. Your careful
selection of 24 red marbles from an urn containing many different
colored marbles is an equally good explanation of the data and it
does not imply that the remainder are red. Hence, it is not just the
fact that we have 24 red marbles in our sample class (24 positive
instances and no negative instances) that confirms the general
hypothesis that all the marbles in the urn are red. It is this data
together with a law that confirms it, a law that (together with the
hypothesis) explains the data in a way that the general hypothesis
alone cannot do.
We have now reached a critical stage in our examination of the
view that a properly qualified set of universal generalizations can
serve as the fundamental laws of nature. For we have, in the past
few paragraphs, introduced the notion of explanation, and it is this
notion, perhaps more than any other, that has received the greatest
attention from philosophers in their quest for the appropriate X in
the formula: law = universal truth + X. R. B. Braithwaite's treatment
([3]) is typical. He begins by suggesting that it is merely deductive
integration that transforms a universal truth into a law of nature.
Laws are simply universally true statements of the form (x)(Fx D
Gx) that are derivable from certain higher level hypotheses. To say
that (x)(Fx D Gx) is a statement of law is to say, not only that
it is true, but that it is deducible from a higher level hypothesis,
H, in a well established scientific system. The fact that it must be
deducible from some higher level hypothesis, H, confers on the
statement the opacity we are seeking to understand. For we may
have a hypothesis from which we can derive (x)(Fx D Gx) but from
which we cannot derive (x)(Kx D Gx) despite the coextensionality
of "F" and "K." Braithwaite also argues that such a view gives
a satisfactory account of the counterfactual force of laws.
The difficulty with this approach (a difficulty that Braithwaite
recognizes) is that it only postpones the problem. Something is not
a statement of law simple because it is true and deducible from some
 LAWS OF NATURE 261

well established higher level hypothesis. For every generalization

implies another of smaller scope (e.g. (x)(Fx D Gx) implies (x)(Fx
Hx D Gx)), but this fact has not the slightest tendency to transform
the latter generalization into a law. What is required is that the higher
level hypothesis itself be law-like. You cannot give to others what
you do not have yoursalf. But now, it seems, we are back where
we started from. It is at this point that Braithwaite begins talking
about the higher level hypotheses having explanatory force with respect
to the hypotheses subsumed under them. He is forced into this
maneuver to account for the fact that these higher level hypotheses-
not themselves law-like on his characterization (since not themselves
derivable from still higher level hypotheses)-are capable of conferring
lawlikeness on their consequences. The higher level hypotheses are
laws because they explain; the lower level hypotheses are laws because
they are deducible from laws. This fancy twist smacks of circularity.
Nevertheless, it represents a conversion to explanation (instead of
deducibility) as the fundamental feature of laws, and Braithwaite
concedes this: "A hypothesis to be regarded as a natural law must
be a general proposition which can be thought to explain its instances"
([3], p. 303) and, a few lines later, "Generally speaking, however,
a true scientific hypothesis will be regarded as a law of nature if
it has an explanatory function with regard to lower-level hypotheses
or its instances." Deducibility is set aside as an incidental (but, on
a Hempelian model of explanation, an important) facet of the more
ultimate idea of explanation.
There is an added attraction to this suggestion. As argued above,
it is difficult to see how instantial evidence can serve to confirm
a universal generalization of the form: (x)(Fx D Gx). If the generali-
zation has an infinite scope, the ratio "examined favorable cases/total
number of cases" never increases. If the generalization has a finite
scope, or we treat its probability as something other than the above
ratio, we may succeed in raising its probability by finite samples,
but it is never clear how we succeed in raising the probability that
the unexamined cases resemble the examined cases without invoking
laws as auxiliary assumptions. And this is the very notion we are
trying to analyze. To this problem the notion of explanation seems
to provide an elegant rescue. If laws are those universal generalizations
that explain their instances, then following the lead of a number of
current authors (notably Harman ([8], [9]); also see Brody ([4]))
we may suppose that universal generalizations can be confirmed
because confirmation is (roughly) the converse of explanation; E
confirms H if H explains E. Some universal generalizations can be
confirmed; they are those that explain their instances. Equating laws
262 FRED I. DRETSKE

with universal generalizations having explanatory power therefore

achieves a neat economy: we account for the confirmabilityof laws
in terms of the explanatorypower of those generalizationsto which
laws are reduced.
To say that a law is a universal truth having explanatory power
is like saying that a chair is a breath of air used to seat people.
You cannot make a silk purse out of a sow's ear, not even a very
good sow's ear; and you cannot make a generalization, not even
a purely universal generalization,explain its instances. The fact that
every F is G fails to explain why any F is G, and it fails to explain
it, not because its explanatoryefforts are too feeble to have attracted
ourattention,but because the explanatoryattemptis never even made.
The fact that all men are mortal does not explain why you and I
are mortal; it says (in the sense of implies) that we are mortal, but
it does not even suggest why this might be so. The fact that all
ten tosses will turn up heads is a fact that logically guarantees a
head on the tenth toss, but it is not a fact that explains the outcome
of this final toss. On one view of explanation, nothing explains it.
Subsumingan instance under a universal generalizationhas exactly
as much explanatorypower as deriving Q from P Q. None.
If universal truths of the form (x)(Fx D Gx) could be made to
explaintheir instances, we might succeed in makingthem into natural
laws. But, as far as I can tell, no one has yet revealed the secret
for endowing them with this remarkablepower.
This has been a hasty and, in some respects, superficial review
of the doctrine that laws are universal truths. Despite its brevity,
I think we have touched upon the major difficulties with sustaining
the equation: law = universal truth + X (for a variety of different
values of "X"). The problems center on the following features of
laws:
(a) A statement of law has its descriptive terms occurringin opaque
positions.
(b) The existence of laws does not await our identification of them
as laws. In this sense they are objective and independent of
epistemic considerations.
(c) Laws can be confirmed by their instances and the confirmation
of a law raises the probabilitythat the unexaminedinstances will
resemble (in the respect described by the law) the examined
instances. In this respect they are useful tools for prediction.
(d) Laws are not merely summariesof their instances; typically, they
figure in the explanation of the phenomena falling within their
scope.
 LAWS OF NATURE 263

(e) Laws (in some sense) "support" counterfactuals;to know a law

is to know what would happenif certain conditionswere realized.
(f) Laws tell us what (in some sense) must happen, not merely what
has and will happen (given certain initial conditions).
The conception of laws suggested earlier in this essay, the view
that laws are expressed by singularstatements of fact describingthe
relationshipsbetween propertiesand magnitudes,proposes to account
for these features of laws in a single, unified, way: (a)-(f) are all
manifestationsof what might be called "ontologicalascent," the shift
from talking about individual objects and events, or collections of
them, to the quantities and qualities that these objects exemplify.
Insteadof talkingabout green and red things, we talk about the colors
green and blue. Instead of talking about gases that have a volume,
we talk about the volume (temperature,pressure, entropy)that gases
have. Laws eschew reference to the things that have length, charge,
capacity, internalenergy, momentum, spin, and velocity in order to
talkaboutthese quantitiesthemselves andto describe theirrelationship
to each other.
We have already seen how this conception of laws explains the
peculiar opacity of law-like statements. Once we understandthat a
law-like statement is not a statement about the extensions of its
constituent terms, but about the intensions (= the quantities and
qualities to which we may refer with the abstract singular form of
these terms), then the opacity of laws to extensional substitution is
naturaland expected. Once a law is understoodto have the form:
(6) F-ness -> G-ness
the relation in question (the relation expressed by "->") is seen to
be an extensionalrelationbetween propertieswith the terms "F-ness"
and "G-ness" occupying transparentpositions in (6). Any term
referringto the same qualityor quantityas "F-ness" can be substituted
for "F-ness" in (6) without affecting its truth or its law-likeness.
Coextensiveterms (termsreferringto the same quantitiesand qualities)
can be freely exchanged for "F-ness" and "G-ness" in (6) without
jeopardizingits truth value. The tendency to treat laws as some kind
of intensionalrelationbetween extensions, as somethingof the form
(x)(Fx FN-> Gx) (wherethe connective is some kindof modalconnec-
tive), is simplya mistakenrenditionof the fact that laws are extensional
relationsbetween intensions.
Once we make the ontological ascent we can also understandthe
modal characterof laws, the feature described in (e) and (f) above.
Althoughtrue statements having the form of (6) are not themselves
264 FRED I. DRETSKE

necessary truths, nor do they describe a modal relationshipbetween

the respective qualities,the contingentrelationshipbetween properties
that is described imposes a modal quality on the particularevents
falling within its scope. This F must be G. Why? Because F-ness
is linked to G-ness; the one property yields or generates the other
in much the way a change in the thermal conductivity of a metal
yields a change in its electricalconductivity. The patternof inference
is:
(I) F-ness -> G-ness
This is F
This must be G.
This, I suggest, is a valid pattern of inference. It is quite unlike
the fallacy committedin (II):
(II) (x)(Fx D Gx)
This is F
This must be G.
The fallacy here consists in the absorption into the conclusion of
a modality (entailment)that belongs to the relationship betweenthe
premises and the conclusion. There is no fallacy in (I), and this,
I submit, is the source of the "physical" or "nomic" necessity
generated by laws. It is this which explains the power of laws to
tell us what would happen if we did such-and-suchand what could
not happen whatever we did.
I have no proof for the validity of (I). The best I can do is an
analogy. Consider the complex set of legal relationshipsdefining the
authority, responsibilities, and powers of the three branches of
governmentin the United States. The executive, the legislative, and
the judicial branches of government have, according to these laws,
different functions and powers. There is nothing necessaryabout the
laws themselves; they could be changed.There is no law thatprohibits
scrappingall the present laws (includingthe constitution)and starting
over again. Yet, given these laws, it follows that the President must
consult Congress on certain matters, membersof the SupremeCourt
cannot enact laws nor declare war, and members of Congress must
periodically stand for election. The legal code lays down a set of
relationshipsbetween the various offices of government,and this set
of relationships(between the abstractoffices) impose legal constraints
on the individuals who occupy these offices-constraints that we
express with such modal terms as "cannot" and "must." There are
certain things the individuals (and collections of individuals-e.g.,
 LAWS OF NATURE 265

the Senate) can and cannot do. Their activities are subjected to this
modal qualification whereas the framework of laws from which this
modality arises is itself modality-free. The President (e.g., Ford) must
consult the Senate on matter M, but the relationship between the
office of the President and that legislative body we call the Senate
that makes Gerald Ford's action obligatory is not itself obligatory.
There is no law that says that this relationship between the office
of President and the upper house of Congress must (legally) endure
forever and remain indisoluble.
In matters pertaining to the offices, branches and agencies of
government the "can" and "cannot" generated by laws are, of course,
legal in character. Nevertheless, I think the analogy revealing. Natural
laws may be thought of as a set of relationships that exist between
the various "offices" that objects sometimes occupy. Once an object
occupies such an office, its activities are constrained by the set of
relations connecting that office to other offices and agencies; it must
do some things, and it cannot do other things. In both the legal
and the natural context the modality at level n is generated by the
set of relationships existing between the entities at level n + 1. Without
this web of higher order relationships there is nothing to support
the attribution of constraints to the entities at a lower level.
To think of statements of law as expressing relationships (such
as class inclusion) between the extensions of their terms is like thinking
of the legal code as a set of universal imperatives directed to a set
of particular individuals. A law that tells us that the United States
President must consult Congress on matters pertaining to M is not
an imperative issued to Gerald Ford, Richard Nixon, Lyndon Johnson,
et al. The law tells us something about the duties and obligations
attending the Presidency; only indirectly does it tell us about the
obligations of the Presidents (Gerald Ford, Richard Nixon, et al.).
It tells us about their obligations in so far as they are occupants
of this office. If a law was to be interpreted as of the form: "For
all x, if x is (was or will be) President of the United States, then
x must (legally) consult Congress on matter M," it would be incompre-
hensible why Sally Bickle, were she to be president, would have
to consult Congress on matter M. For since Sally Bickle never was,
and never will be, President, the law, understood as an imperative
applying to actual Presidents (past, present and future) does not apply
to her. Even if there is a possible world in which she becomes President,
this does not make her a member of that class of people to which
the law applies; for the law, under this interpretation, is directed
to that class of people who become President in this world, and
Sally is not a member of this class. But we all know, of course,
266 FRED I. DRETSKE

that the law does not apply to individuals, or sets of individuals,

in this way; it concerns itself, in part, with the offices that people
occupy and only indirectly with individuals in so far as they occupy
these offices. And this is why, if Sally Bickle were to become President,
if she occupied this office, she would have to consult Congress on
matters pertaining to M. 12
The last point is meant to illustrate the respect and manner in
which natural laws "support" counterfactuals. Laws, being relation-
ships between properties and magnitudes, go beyond the sets of things
in this world that exemplify these properties and have these magnitudes.
Laws tell us that quality F is linked to quality G in a certain way;
hence, if object 0 (which has neither property) were to acquire property
F, it would also acquire G in virtue of this connection between F-ness
and G-ness. A statement of law asserts something that allows us
to entertain the prospect of alterations in the extension of the predicate
expressions contained in the statement. Since they make no reference
to the extensions of their constituent terms (where the extensions
are understood to be the things that are F and G in this world),
we can hypothetically alter these extensions in the antecedent of
our counterfactual ("if this were an F. ..") and use the connection
asserted in the law to reach the consequent (". . it would be G").
Statements of law, by talking about the relevant properties rather
than the sets of things that have these properties, have a far wider
scope than any true generalization about the actual world. Their scope
extends to those possible worlds in which the extensions of our terms
differ but the connections between properties remains invariant. This
is a power that no universal generalization of the form (x)(Fx D
Gx) has; this statement says something about the actual F's and
G's in this world. It says absolutely nothing about those possible
worlds in which there are additional F's or different F's. For this
reason it cannot imply a counterfactual. To do this we must ascend
to a level of discourse in which what we talk about, and what we
say about what we talk about, remains the same through alternations
in extension. This can only be achieved through an ontological ascent
of the type reflected in (6).
We come, finally, to the notion of explanation and confirmation.
I shall have relatively little to say about these ideas, not because
I think that the present conception of laws is particularly weak in

12Ifthe law was interpretedas a universal imperativeof the form described, the
most that it would permit us to infer about Sally would be a counteridentical:If
Sally were one of the Presidents (i.e. identical with either Ford, Nixon, Johnson,
. .), then she would (at the appropriate time) have to consult Congress on matters
pertainingto M.
 LAWS OF NATURE 267

this regard, but because its very real strengths have already been
made evident. Laws figure in the explanation of their instances because
they are not merely summaries of these instances. I can explain why
this F is G by describing the relationship that exists between the
properties in question. I can explain why the current increased upon
an increase in the voltage by appealing to the relationship that exists
between the flow of charge (current intensity) and the voltage (notice
the definite articles). The period of a pendulum decreases when you
shorten the length of the bob, not because all pendulums do that,
but because the period and the length are related in the fashion
T = 27r /L/g . The principles of thermodynamics tell us about the
relationships that exist between such quantities as energy, entropy,
temperature and pressure, and it is for this reason that we can use
these principles to explain the increase in temperature of a rapidly
compressed gas, explain why perpetual motion machines cannot be
built, and why balloons do not spontaneously collapse without a
puncture.
Furthermore, if we take seriously the connection between explana-
tion and confirmation, take seriously the idea that to confirm a
hypothesis is to bring forward data for which the hypothesis is the
best (or one of the better) competing explanations, then we arrive
at the mildly paradoxical result that laws can be confirmed because
they are more than generalizations of that data. Recall, we began
this essay by saying that if a statement of law asserted anything
more than is asserted by a universally true statement of the form
(x)(Fx D Gx), then it asserted something that was beyond our
epistemological grasp. The conclusion we have reached is that unless
a statement of law goes beyond what is asserted by such universal
truths, unless it asserts something that cannot be completely verified
(even with a complete enumeration of its instances), it cannot be
confirmed and used for predictive purposes. It cannot be confirmed
because it cannot explain; and its inability to explain is a symptom
of the fact that there is not enough "distance" between it and the
facts it is called upon to explain. To get this distance we require
an ontological ascent.
I expect to hear charges of Platonism. They would be premature.
I have not argued that there are universal properties. I have been
concerned to establish something weaker, something conditional in
nature: viz., universal properties exist, and there exists a definite
relationship between these universal properties, if there are any laws
of nature. If one prefers desert landscapes, prefers to keep one's
ontology respectably nominalistic, I can and do sympathize. I would
merely point out that in such barren terrain there are no laws, nor
268 FRED I. DRETSKE

is there anything that can be dressed up to look like a law. These

are inflationary times, and the cost of nominalism has just gone up.

REFERENCES
[1] Achinstein,P. Law and Explanation.Oxford:ClarendonPress, 1971.
[2] Ayer, A. J. "Whatis a Law of Nature." In [5], pages 39-54.
[3] Braithwaite,R. B. ScientificExplanation.Cambridge,England:CambridgeUni-
versity Press, 1957.
[4] Brody, B. A. "Confirmationand Explanation."Journalof Philosophy65 (1968):
282-299, Reprintedin [5], pages 410-426.
[5] Brody, B. A. Readings in the Philosophy of Science. Englewood Cliffs, N.J.:
PrenticeHall, 1970.
[6] Bromberger,S. "Why-Questions."In [5], pages 66-87.
[7] Goodman,N. Fact, Fiction and Forecast. London:The Athlone Press, 1954.
[8] Harman, G. "The Inference to the Best Explanation." Philosophical Review
74 (1965):88-95.
[9] Harman,G. "Knowledge, Inferenceand Explanation."PhilosophicalQuarterly
18 (1968): 164-173.
[10] Hempel, C. G., and Oppenheim, P. "Studies in the Logic of Explanation."
In [5], pages 8-27.
[11] Hempel, C. G. "MaximalSpecificity and Lawlikenessin ProbabilisticExplana-
tions." Philosophyof Science35 (1968):116-133.
[12] Kneale,W. "NaturalLawsandContrary-to-Fact Conditionals."Analysis10(1950):
121-125.
[13] Kneale, W. Probabilityand Induction.Oxford:OxfordUniversityPress, 1949.
[14] Mackie, J. L. "Counterfactualsand Causal Laws." In Analytical Philosophy.
(First Series). Editedby R. J. Butler. Oxford:Basil Blackwell, 1966.
[15] Molnar,G. "Kneale'sArgumentRevisited."PhilosophicalReview78 (1969):79-89.
[16] Nagel, E. The Structureof Science. New York: HarcourtBrace, 1961.
[17] Popper,K. "A Note on NaturalLaws and So-Called'Contrary-to-Fact Condition-
als." Mind58 (1949):62-66.