Synthese

DOI 10.1007/s11229-014-0574-3

Carnapian rationality

A. W. Carus

Received: 5 May 2014 / Accepted: 22 July 2014

© Springer Science+Business Media Dordrecht 2014

Abstract It is generally thought that Carnap’s principle of tolerance cannot be inte-

grated into a coherent overall conception of rationality. The doubts come from many
sides, of which two are singled out. This paper argues (and documents) that both are
wrong, and that Carnapian rationality is a viable and perhaps quite interesting program
for further development.

Keywords Carnap’s principle of tolerance · Carnap’s inductive logic · Quine ·

Rationality · Instrumental rationality · Kantian ethics · Carnap’s meta-philosophy

Reason or the ratio of all we have already known, is not the same that it shall
be when we know more.—William Blake1

Despite the recent interest in Carnap’s principle of tolerance and its presupposi-
tions, it is still widely doubted that logical tolerance is compatible with any wider
conception of rationality. Post-Quinean empiricists, among many others, are troubled
by the frequent reference to the analytic, the a priori, and “intuition” in Carnap’s later
writings. Others worry that Carnap’s overall view allows only instrumental rational-
ity, leaving out Wertrationalität (the rationality of ultimate values, as in Max Weber’s
well-known distinction). Any conception of Carnapian rationality would therefore
seem to fall afoul of Hilary Putnam’s well-known critique of “criterial rationality”
(e.g. Putnam 1983). From both these perspectives, Carnapian rationality looks impov-
erished, incomplete, perhaps just simple-minded. This paper will argue that, while

1 Quoted by Stein (1998, p. 3). The general perspective developed in this paper owes a great deal to
Stein’s writings—not only to those cited—as well as his personal influence as a teacher.

A. W. Carus (B)
Chicago, IL, USA
e-mail: awcarus@gmail.com

123
 Synthese

Carnap did not write down much about his ultimate conception of rationality, the con-
ception he was heading toward in his later years, once misunderstandings are set aside,
is worthy of attention and perhaps further development.
Section 1 makes these two different kinds of objections to a Carnapian rational-
ity a little more specific. Section 2 addresses the first kind, showing that, correctly
understood, Carnap’s mentions of the a priori and of intuition in his later writings are
entirely consistent with the Carnap otherwise known. But it may still be thought that,
whatever his motivations, no coherent view of rationality can be developed from such
a standpoint; this worry is addressed in Sect. 3. Section 4 focusses on the second kind
of objection, and shows that Carnap took not only the norms of instrumental rational-
ity but also ultimate values to be subject to rational appraisal—which is not reducible
to a list of explicit criteria. The final section seeks to put the picture of Carnapian
rationality arising from these discussions in a broader context.

1 Two kinds of objections to a Carnapian rationality

In §41F of the Logical Foundations of Probability, Carnap addresses an issue that Feigl
had often raised to him in conversation, and that Russell had recently treated at length
in his 1948 book Human Knowledge: Its Scope and Limits—the question whether all
induction rests on the assumption of the uniformity of nature over time. Carnap agrees
that yes, it does, but if we consider it rational to take an umbrella when rain is forecast,
then it is also rational to use induction as a guide to life if the uniformity of nature is
probable, even if not certain. So far so good, but the next step raises hackles. On the
one hand, Carnap admits that the assertion of uniformity (or probable uniformity) is
synthetic, and that to ground inductive inference on it would be circular. On the other
hand, he claims his account immune to any such vicious circle, since his assertion of
the uniformity of nature is analytic.

Any statement on probability1 or estimation is, if true, analytic. This holds also
for the statements of the probability of uniformity or the estimate of uniformity. . .
Since they are not synthetic, no empirical confirmation is required. Thus the
earlier difficulty disappears. (1950a, p. 181)

Carnap anticipates the objection that the statement of uniformity has to be taken
as factual, since otherwise an agent X —whose long series of bets in accordance with
probability estimates is highly probable not to result in a loss, thereby justifying action
in accordance with probability estimates—will not be assured of success in the long
run.

Our reply is: it is not possible to give X an assurance of success even in the
long run, but only the probability of success, as in the statement that [X is
unlikely to sustain a loss], and this statement is itself analytic. But can X take
a practical decision if he has as a basis merely an analytic statement, one that
does not say anything about the world? In fact, X has as a basis for his decision
two statements: first a factual statement of his total observational evidence, and
second an analytical statement of probability1 . The latter does not add anything

123
Synthese

to the factual content of the first, but it makes explicit an inductive-logical relation
between the evidence and the hypothesis in question. (ibid.)
Carnap encountered enough skepticism about this argument that he came back to
it repeatedly,2 most explicitly in the original draft of his autobiography. “I think,” he
wrote there, “like most philosophers today in contrast to those of the last century, that
the assumption U [that the world possesses a sufficient degree of uniformity] is not
required; at most the probability of U is presupposed. However, when we proceed
further from this point of common agreement, then my path diverges from that of the
majority of contemporary empiricists. On the basis of the usually accepted frequency
conception of probability, these empiricists interpret the statement of the probability
of U as a synthetic statement.”
Whoever accepts the view, as e.g. Russell does, that this synthetic statement is
a necessary presupposition, faces the dilemma of either accepting a synthetic
presupposition that cannot be confirmed by any inductive procedure, and thus
abandoning the principle of empiricism at least in this one point, or [condemning]
the whole inductive method and thereby all of empirical science as invalid. In my
view the situation is, however, entirely different. On the basis of my conception
of inductive logic, the statement that U has a high probability on the basis of the
evidence available is interpreted as a statement of logical probability and thus as
an analytic statement. Thus the dilemma disappears. Of course it is not possible
on the basis of a purely analytic inductive logic to obtain results about the facts
of the world. But the decisive point is that in order to recognize the rationality of
an inductive method, no factual knowledge about the world is required but only
a clarification of concepts. Suppose, for example, that the evidence e says that
100 balls have been drawn from an urn and 99 of them were white and one black,
and that the hypothesis h says that the next ball will be black. Then it follows
from the meaning of the concept of logical probability that the probability of
h with respect to e is smaller than that of non-h. Suppose further that E is the
observational knowledge available to the person X at the present time and that
the probability of a certain hypothesis H with respect to E is smaller than that of
non-H. Then it follows from the meaning of the concept of rationality that it is
rational for X to decline a bet at even odds on this prediction H. It seems to me one
of the most important philosophical consequences of my concept of induction
that it makes it possible to solve the problem of the presuppositions for the
application of induction in determining a rational decision, without abandoning
the principle of empiricism. (UCLA Box 2, CM3, folder M-A5, pp. P20–P22)3
Since Quine, of course, many philosophers have taken exception to the idea that
anything is explained by classifying a sentence as “analytic.” While this passage does
not explicitly use the terminology of “a priori,” Carnap’s rhetoric seems to imply
that the mere classification of a sentence as “analytic” is sufficient to give it some

2 E.g. in his later paper “The Aim of Inductive Logic” (Carnap 1962).
3 This passage, originally in Sect. 12 of Carnap’s (1963) autobiography, was omitted from the published
version.

123
 Synthese

explanatory force.4 Those like Putnam (e.g. 1983), on the other hand, who reject all
“criterial rationality,” cannot countenance “the meaning of the concept of rationality,”
or accept that any meaning one might agree on could by itself establish that “it is
rational for X to decline a bet at even odds on this prediction H.” Again, Carnap’s
rhetoric appears to imply something that is not fully spelled out, in this case that there
is a single unquestioned concept of rationality for which explicit criteria can be stated,
and—since such criteria must be internal (in the sense of Carnap 1950b)—that this
unambiguous meaning cannot extend beyond instrumental reason.
These two kinds of objections are by no means exhaustive, but they cover a wide
range of possible critiques, and come from opposite ends of the spectrum, in a sense;
while they do not contradict each other directly, it seems that if one of them is right,
the other must be wrong. This paper will argue they are both wrong, as they both
misinterpret Carnap’s admittedly often clumsy rhetoric, and his way of framing these
issues in his later years.

2 First complaint: a priori intuitions

One thing that has made people particularly uneasy about the later Carnap’s invocation
of the “analytic” or the “a priori” is that it often occurs in conjunction with the notion
of “intuition.” Here is an example from the Schilpp volume that has been singled out
for such criticism:

It seems to me that the reasons to be given for accepting any axiom of inductive
logic have the following characteristic features. . .:

(a) The reasons are based upon our intuitive judgements concerning inductive
validity, i.e., concerning inductive rationality of practical decisions (e.g.
about bets).
Therefore:
(b) It is impossible to give a purely deductive justification of induction.
(c) The reasons are a priori. (Carnap 1963, p. 978)

And Carnap’s last major talk, at Popper’s London conference in 1965, was about
“Inductive Intuition” (much to Popper’s dismay), which continues to arouse conster-
nation. John Earman, for instance, is willing in principle to consider Carnap’s goal of
making “the constraints of rationality extend beyond the requirements of coherence,
which entails that degrees of belief must conform to the standard axioms of prob-
ability. According to Carnap the requirements of rationality do not suffice to single
out a unique probability function, but they do significantly constrain the choice of a
probability function . . .” (Earman 1993, pp. 28–29) Carnap’s attempt fails, though, in
Earman’s view:

4 Of course one might accept Carnapian accounts of both “analytic” and “a priori” and still reject Carnap’s
argument here, e.g. because one rejects his conception of probability. This is bracketed here; Carnap’s
conception of probability and induction is used in this paper only to illuminate his larger conception of
rationality, and nothing depends on it.

123
Synthese

However, I do not think that Carnap managed to stake out a defensible position, as
can be brought out by the question of how one recognizes which probability func-
tions are “rational.” Carnap’s answer was to appeal to what he variously called
“inductive intuition” and “inductive common sense.” The trouble, of course, is
that one person’s inductive common sense is another person’s inductive non-
sense. So the appeal to intuition reveals very different opinions as to whether it
is rational to learn from experience at all and, if so, at what rate. (ibid., p. 29)
I will argue, against this view, that perfectly good sense can be made of these uses of
“a priori” and “intuition,” even to John Earman. The above passage from the Schilpp
volume is followed immediately by some further clarification:
By (c) I mean that the reasons are independent both of universal synthetic prin-
ciples about the world, e.g. the principle of the uniformity of the world, and of
specific past experiences, e.g. the success of bets which were based on the pro-
posed axioms. If e represents the total experience of X so far, then the value of
c(h, e) is certainly dependent on e, and thus upon the experiences of X . However,
the acceptability of the axioms, by which the value of c(h, e) is determined or by
which certain restrictions are placed on admissible c-functions, is independent
of these experiences. (Carnap 1963, p. 979)
This represents a change of mind from a view Carnap had held only a few years
previously, e.g. in §18 of the Continuum of Inductive Methods:
It is psychologically difficult to change a faith supported by strong emotional
factors (e.g. a religious or political creed). The adoption of an inductive method
is neither an expression of belief nor an act of faith, though either or both may
come in as motivating factors. An inductive method is rather an instrument for
the task of constructing a picture of the world on the basis of observational
data and especially of forming expectations of future events as a guidance for
practical conduct. X may change this instrument just as he changes a saw or an
automobile, and for similar reasons. If X , after using his car for some time, is
no longer satisfied with it, he will consider taking another one, provided that he
finds one that seems to him preferable. . . (Carnap 1952, p. 55)
This earlier view puts everything at the same level; the tools we use to choose
tools are chosen in the same way and with the same tools that we use for object-level
choices among goods or bets. This is essentially the attitude familiar from the Logical
Syntax, where “usefulness” or “convenience” became the only criteria for the choice
of language.
Further reflection clearly made Carnap uncomfortable with this assimilation, since
by 1957, when he wrote the above passages from the Schilpp volume, he was saying
that, while this one-level approach was not downright wrong, it could be dispensed
with. Rationality, he now began to think, should not actually need to use experience to
decide on the tools for learning from experience. Decisions about rationality should
be taken at a different level from the everyday decisions that use and presuppose
rationality. Inductive logic was no different from deductive logic in this respect; it was
a different aspect of rationality, but no less a priori. Of course, as Carnap stresses in the

123
 Synthese

further elaboration of the above passage from the Schilpp volume, he does not mean
synthetic a priori, and the principle of tolerance continues to govern the choice among
“a priori” tools. But the considerations by which we shape the basic components of
our rationality should not be at the same level as the choices made by means of our
reason, once constructed. They should be more principled and structural than mere
iterated learning from empirical error; they need not be “blind,” but can be informed
by what we know.
But what does “intuition” have to do with it? That would seem—has seemed, to
Popper, Earman, and many others—to go in precisely the opposite direction, away
from anything more principled and structural than mere learning from error. But here
again, we need to look more closely at what Carnap was actually saying, specifi-
cally in response to his sustained immersion in the probability literature from the
early 1940s. As long as he had focussed largely on deductive logic, the pragmatics
of notation systems did not overly concern him. One could try out different logics
and not worry too much about the motivations underlying this or that axiom sys-
tem. But inductive logic was different; the justifications of various axioms were much
more concretely rooted in scenarios of practical rationality, human gain and loss. The
Dutch Book arguments underlying the Ramsay-De Finetti representation of coher-
ence on the basis of the axioms of probability, for instance, obviously rely on intu-
itions about the structure of certain bets and the advisability of taking them; “the
question of the validity of inductive reasoning goes back to—or, at least, can use-
fully be based on—considerations about the rationality of decisions.” (Carnap 1968,
p. 264)
But this does not make the “rationality of decisions” any less a matter of logic for
Carnap. This is easy to misunderstand. With respect to the distinction between deduc-
tive reasoning (a priori, analytic), and what is usually called inductive or experiential
reasoning (a posteriori, synthetic), we expect inductive reasoning to fall on the experi-
ential side. This is not how Carnap saw it; for him, all reasoning, all of what he thinks
of as “rationality,” falls on the a priori and analytic side, including both inductive and
deductive reasoning (e.g. Carnap 1950a, §§41–45, esp. §43B).
Decisions about the basic structure of our rationality are, of course, based on “rel-
ativized” a priori considerations, as Michael Friedman would say. In fact, they are
voluntaristic a priori; we decide on them. But how? If we are not to use experience,
if we are not to think of them as tools at the same level as all the other tools we
employ, then how are we supposed to proceed? Well, the suggestion is not that we
ignore experience, but that we use it “heuristically”;5 experience is unable (by itself)
to decide (as it decides questions with empirical content). We can, after all, foresee
possible paths of experience, which means that we can incorporate our response to
those paths into a more general method, which would thus obviously be more robust
and flexible. Carnap gives a (still not very concrete) example in an unpublished paper
of 1957:

5 As Carnap puts it in the unpublished typescript discussed further down in this paragraph (ASP/RC
082-07-01, p. 10).

123
Synthese

At the time t0 , before X had the actual sequence of experiences, say e2 , which
induced him at the time t1 to change from cA1 to cA2 , he might have considered
the possible sequences e1 , e2 , e3 , etc. for the period from t0 to t1 and reasoned as
follows: “I shall re-examine my method cA1 at time t1 in view of the experiences
made up to that time. If the experience sequence e1 occurs, I shall find that the
method cA1 had been more successful than any other of the methods cA , and
I shall therefore decide to continue using it. If, however, e2 occurs, I shall see
that cA2 would have been more successful than cA1 , and therefore I shall change
over to cA2 . If e3 occurs, I shall instead change to cA3 , etc. Since now, at t0 ,
I can foresee these hypothetical changes to be made at t1 , it seems preferable
not to begin with cA1 or any other method of the class cA , but instead to use a
more general method, say cB , which incorporates the desired influence of the
experiences up to t1 , whatever they will be, into its rules. Then, for any h, the value
of cB (h, e1 ) will be similar to that of cA1 (h, e1 ), likewise cB (h, e2 ) similar
to cA2 (h, e2 ), etc.” This example illustrates the following point: although it
is not wrong to abandon an inductive method and choose instead another one
in view of past experiences, this procedure does not seem necessary. It should
in principle be possible to construct a generalized method and thereby avoid
changing methods in view of previous experiences. And if we succeed in doing
so, such a method would be preferable. (ASP/RC 082-07-01, p. 11)

This suggests two things. First, experience is not left out; Carnap’s point is not
that only a priori considerations are relevant to the decision, it is that experience by
itself does not determine the decision. Obviously without a store of experience (and
a science based on it) there would be no way to describe the possible paths e1 , e2 ,
e3 , . . .
Second, this passage suggests that a different kind of “experience” does actually
influence our reasoning, deductive as well as inductive, which is our cumulative expe-
rience, as a community of reasoners, of working with our formal instruments—our
experience with the “a priori.” Carnap often comes back to this:

For fruitful work in a field it is not necessary to be in possession of a well-

founded epistemological theory about the sources of knowledge in the field.
Both arithmetic and geometry were successfully developed for more than two
thousand years before the first detailed epistemological theories for both were
offered by Kant; and although some important insights have since been gained,
there is even today no complete agreement. On the other hand, it is certainly
helpful, also for work on substantive problems within a new field, if we can
achieve even a little, more explicit awareness of what we are doing and how
new knowledge in the field is gained. When we face a difficult problem, this
insight may help us to see in which way and in which direction we should look
for a possible solution. Thus our working procedure, which in the beginning
is inevitably more or less instinctive, may slowly become more consciously
directed. (Carnap 1968, pp. 266–267)

This suggests that what Carnap meant by the “inductive intuition” (and “deductive
intuition”) that has so confused people is the instinct referred to here, developed over

123
 Synthese

generations of instrument-users gaining experience with it. The “intuition” here is

just the embedded and encoded first-hand experience of the formal-instrument-user
community.6 Those who have not been socialized into the use of these instruments will
not share these intuitions; nothing would have surprised Carnap less than the Tversky
and Kahnemann research about preference reversals and other “mistakes” untrained
humans are prone to when faced with questions of probability.7
“Intuition,” then, is just a source (perhaps the main source) of explicanda. Spelling
out intuitions of the formal-language user community, making them explicit, arriving
at provisional agreement on them, and finding room for them in existing formal sys-
tems, are essential parts of the explication process; they constitute the crucial step of
“clarification of the explicandum” (Carnap 1950a, §2).

3 First complaint reasserted: a priori presuppositions

The first kind of critic will now perhaps admit that Carnap’s heart was in the right
place. But can his strategy actually be made to work? Or is the aspiring Carnapian
forced, despite these good intentions, to rely on unacknowledged “a priori” presuppo-
sitions in a more substantive sense like Kant’s? This was just what a number of 1950s
commentators on Carnap’s inductive logic argued, and it was evidently in response
to these that Carnap’s view had changed as described in Sect. 2. When the Princeton
philosopher John Lenz published a critique to this effect in Philosophy of Science in
1956, Feigl (who had evidently been urging the point to Carnap in conversation since
1942)8 wrote to Carnap that Lenz

6 Where the emphasis is on “first-hand”—the relevant intuitions are those of active users, not passive
commentators or consumers. Carnap suggests this in his reply to Ernest Nagel when he says that Nagel’s
intuitions would change if he actually tried to construct a logic of confirmation himself: “. . . since the time
I began to construct a system of inductive logic, I myself had sometimes to change intuitive judgments
on certain properties of logical probability. And this occurs still today when I begin to think about the
extension of inductive logic to a new, more complex language form. If Nagel were to try to construct a
system of [degree of confirmation] on the basis of his present intuitive judgments, I am convinced that he
would change quite a number of these judgments because he would find that the system would yield many
results which would appear to him either as undesirable or even as entirely unacceptable. From experiences
with my own intuitive judgments and those of friends whom I often asked for reactions to particular results,
I have learned that isolated intuitive judgments are often very unreliable. Of course, the development of
inductive logic must be guided by intuitive inductive judgments. But these judgements are more useful if
they are made, not on isolated points, but in the context of the tentative construction of a system.” (Carnap
1963, p. 994)
7 Any more than he was the least fazed by Arne Naess’s research showing that half the population did not
use the truth predicate in accordance with Tarski.
8 Though Carnap may well have written the section on uniformity of nature (§41F, quoted above, Sect. 1)
in 1949, as he says in the original draft of his autobiography, the discussions with Feigl go back to at least
April 1942, i.e. before he wrote the rest of the book; a diary entry says “Mit Feigl ausführlich über dc
[degree of confirmation]. (Er: darin steckt immer faktische Annahme über Gleichförmigkeit der Natur.)”
(ASP/RC 025-82-08, 26 Apr 1942) Carnap also made several notes on uniformity of nature at around this
time, mentioning Feigl’s question about what factual assumptions underlie Carnap’s proposed degree of
confirmation (ASP/RC, folder 078-25). So the fact that Carnap wrote section 41F later likely meant that he
was not yet satisfied he had worked out this problem to his own satisfaction by 1944–1946.

123
Synthese

. . . raises (without reference to Burks or to my “Scientific Method without Meta-

physical Presuppositions”)9 exactly (and more explicitly) the questions I’ve tried
to put to you for many years. Maybe he and I are stupid or confused—but I think
the questions are serious and important. Of course, he doesn’t answer them—and
my answer may not be good enough. But do let us have your answer! (ASP/RC
089-57-02, Letter of Feigl to Carnap of 3 August 1956)

Carnap felt obliged to comply, and wrote several drafts of a response to Lenz over the
next year or two, which remained unpublished. The first draft was a restatement of the
previous position laid out in the above quotation from Continuum of Inductive Methods.
Only when Hempel pointed out that Carnap appeared not really to be addressing Lenz’s
critique head-on10 did Carnap start over, and begin to revise his view as he revised
the paper. The result was a new draft “How Can Induction be Justified?” in which
he distances himself explicitly from the position, quoted above, that he had taken in
Continuum of Inductive Methods. Also, he now sharply distinguishes between two
steps—the choice of axioms for inductive logic and the choice of a particular method
within the λ-class defined by the chosen axioms. Different standards apply to these
two steps. The choice of axioms, he argues, can (or should) make use only of deductive
and inductive reasoning, not of experience and not (obviously) of a synthetic a priori
principle. The second step (the choice of method) can also make use of experience
(in addition to deductive and inductive reasoning), but need not; this is where Carnap
distances himself from §18 of Continuum of Inductive Methods) and argues (Sect. 2
above) that various possible paths of experience can be anticipated without actual trial
and error.
Lenz claimed that Carnap’s use of the uniformity of nature was circular in just
the way that Carnap had denied in the quotation we began with (from Carnap 1950a,
§41F). For present purposes, we are concerned not with the uniformity of nature or
with inductive logic, but only with Carnap’s view of rationality as revealed by his
argument. So with that focus in mind, just two points in Carnap’s response to Lenz
will be singled out.
First, Carnap denies the circularity on the grounds of his two-step procedure in
justifying an inductive method. If we have first established a λ-class by choosing
axioms, then for purposes of choosing a method within that class, those axioms can
be regarded as provisionally fixed. Assuming that the axioms apply equally to all the
methods under review, there is nothing circular about using the axioms in justifying
the choice of method c over method c . Among the justifiable axioms, Carnap says,
is the axiom of instantial relevance, or as he puts it here, “the axiom of learning from
experience,” which he paraphrases as “If one of two properties (of the same logical
form) has occurred more frequently than the other among the observed cases, then,

9 Feigl presumably has in mind Burks (1953) and Feigl (1954).

10 Hempel makes specific reference to the passage from §18 of Continuum of Inductive Methods quoted
above, and asks whether “while the choice of a c-function, to be sure, need not rest on any empirical
assumption, the problem remains whether the justification of such a choice as adequate would require
reference to empirical fact. . . I wonder whether you could not add a word about that question, because
otherwise the reader might feel that your reply sidesteps a crucial issue at stake.” (ASP/RC 089-57-03,
p. 2)

123
 Synthese

other things being equal, its occurrence in the next case is more probable than that of
the other property.”

Lenz is, of course, right that it would be illegitimate to try to justify the preference
of c to c by inductive reasoning based specifically on c. But the use of the axioms
is legitimate because they are satisfied by both c and c . Now suppose that the
evidence available to X shows that with respect to the observed cases c was more
successful than c . Then it follows from the axiom [of instantial relevance] that
it is more probable than not, other things being equal, that in the next case, or in
any class of future cases, c will be more successful than c . (ASP/RC 082-07-01,
p. 18)

Carnap concludes that “it is possible to make an inductive inference, i.e. a compar-
ative probability determination, for the future successfulness of two methods c and c ;
the result may then be a factor in the justification of the preference of c over c and
thus of the choice of c.” He further concludes “that this inductive inference does not
make use of c and thus does not involve a vicious circle; and further it does not use
any synthetic assumption concerning the future.” (ibid., p. 19)
The new preference, discussed in the previous section, for a more “principled” way
of arriving at axioms of inductive logic than blind trial and error, amounts then, to a
hierarchical form of reasoning, corresponding to the “levels” at which different choices
are made. Since the choice of axioms for inductive and deductive logic is fundamental
to rationality, they were necessarily prerequisite to the choice of an inductive method
within the bounds permitted by that choice of axioms. So for the purposes of the latter
choice, the axioms may be held constant.
This illustrates vividly the second point in the above argument from 1957 to be
highlighted here (as exemplifying Carnap’s view of rationality): in terms of his (1950b)
distinction between “internal” and “external” questions, this argument reveals that
the justification of constitutive axioms—of inductive no less than deductive logic—
is an external matter. Burks (1963) pointed this out long ago, and Carnap not only
confirmed that Burks was right, but also reaffirmed that these external questions were
to be understood as practical ones. Moreover, he adds,

. . . first, that it is an important task of the philosophical clarification of induc-

tion to specify the factors which are relevant for a rational decision of a prac-
tical external question; and, second, that important theoretical questions are
involved in the relevant deliberations leading to a practical decision of this kind.
(Carnap 1963, p. 982)

So while it may well appear, if we look only at Carnap’s work on inductive logic, that
his concern is exclusively with a rather narrow conception of instrumental rationality,
we now see that Carnap regarded the norms of instrumental rationality to be subject to
rational appraisal—in which he specifically included theoretical considerations. This
discussion of Carnap’s reliance on the a priori leads naturally, then, into a discus-
sion of the other, opposite, objection of the two considered in Sect. 1—the objection
that only Zweckrationalität could fall within Carnap’s purview, to the exclusion of
Wertrationalität.

123
Synthese

4 Second complaint: instrumental rationality

The suspicion that Carnap could entertain only instrumental rationality is wide-
spread.11 This suspicion usually takes one of two forms. In the context of the Carnap-
Quine debate it takes the form of a kind of category-mistake diagnosis. Carnap’s
conception of reason, it is held, requires any assertion to be relative to a linguistic
framework. Therefore there can be no rational assertion, no “judgement,” outside any
linguistic framework,12 e.g. bearing on the choice among frameworks. The choice
among frameworks (and thus also—more locally—the choice among explications for
a particular explicandum) must therefore take place outside any rational framework.13
So for Carnap, rationality can only be instrumental, the story goes, since any form of
meta-rationality, outside a specified framework, is by definition not part of rationality.
Another form of this same suspicion is to discern an infinite regress in Carnap’s view
that the choice among linguistic frameworks, or explications, cannot be made within
one of the frameworks to be chosen among. It must be, according to Carnap, external
to any of those frameworks. But this, it seems (Steinberger 2014), leaves Carnap with
two choices: either the choice among frameworks is arbitrary and irrational or there
must be a rational “selection framework” in the metalanguage in which the choice
among object-level frameworks is made. But which framework? How can we make a
choice among meta-frameworks without invoking a selection framework at the meta-
meta-level, and so on? Where does this regress stop? And if we cut it off at some level,
we again leave Carnap with only framework-internal, i.e. instrumental, rationality.
A different perspective on this problem was introduced by Stein (1992), who saw
that in the later Carnap, the relation between the theoretical and practical was implicitly
dialectical. There is an interplay or mutual feedback relation, in this view, between
formal media of reasoning, on the one hand, and the realm of practical decisions—the
values by which we choose or decide among candidate explications or languages—on
the other. The knowledge we articulate in our formal languages constrains the decision

11 Since Carnap is an emotivist, according to one such critic, “value statements lack cognitive content;
they do not have any truth-value. But then the justification of science on the basis of ethics is not really a
justification at all since the claim about the good life merely expresses Carnap’s emotive state or disposition
toward life.” So “emotivism in ethics severed the crucial link between science and the good life as he
envisioned it. Perhaps this is the ultimate aporia of logical empiricism.” (Irzik 2003, pp. 343–344)
12 This widespread view projects a Wittgensteinian absolutism regarding the “boundaries” of language
onto Carnap, who explicitly rejected it and endorsed a gradualist conception, e.g. in Carnap (1963, p. 934).
13 One of the clearest statements of the tradeoffs between such a position, attributed to Carnap, and
a Quinean position where we are simply inside whatever framework we are in, in mediis rebus (Dreben
1994), without access to its basic framework principles, let alone the ability to choose them, is that of George
(2012), concluding: “But for Carnap, as we have seen, such considerations can at best exert a non-rational
influence on an agent who is choosing which framework to adopt. Talk of rational constraint only has its
place within a framework, once a language and rules of reasoning and inquiry have been settled upon. For
Carnap, this observation is critical in understanding why traditional philosophical disputes have proven to be
so frustratingly irresolvable and so different from scientific disagreements: philosophers, unlike scientists,
typically dispute about which framework to adopt, which language to speak, and no facts about the world
can rationally bear on such disagreements. The empirical facts only come into focus, and talk of rational
relevance only gets a grip, once a particular linguistic framework has been adopted.” (George 2012, p. 10)

123
 Synthese

space of possible actions (including verbal actions such as explication), it informs us of

the consequences of various choices within that space, and shapes our values in many
other respects. But on the other hand, our values (which, though shaped, informed, and
constrained in many ways by this knowledge, are not determined by it) in turn guide
our choices among explications and formal languages, and thereby also shape our
knowledge by determining what we mean by “knowledge.” So there is a continuous
mutual feedback: values ultimately determine what we define as reason, but the reason
thus provisionally defined informs and enlightens our values; reason is not the slave
of the passions, as in Hume, but nor can reason determine our values, as in Kant.14
This is not an empirical claim about how the practical and theoretical realms ordi-
narily interact, but a regulative or ideal portrayal of how they could interact (and have
sometimes interacted) to optimize the beneficial consequences of human knowledge
to the species. In this regulative ideal, our ultimate values are not dogmatically fixed,
but open to correction by what we as a species have progressively come to learn about
the world we inhabit (including ourselves). This ideal portrayal is essentially that of
the Enlightenment, in which knowledge self-consciously arrived at by empirical crit-
icism of hypotheses programatically replaces passive acceptance of traditional and
folk knowledge as the basis for individual and collective action (Carus 2007, Chap. 1
and 11).15
But in the traditional Enlightenment view of the eighteenth and nineteenth centuries,
even of the Vienna Circle in its heyday, the new knowledge replacing traditional and
folk knowledge was authoritative; it was the final word, even if in some accounts it
was portrayed as permanently evolving, never entirely finished—scientific knowledge
was still the ultimate tribunal. This cognitive authoritarianism made the Enlightenment
vulnerable to criticism from the start.16 Carnap, however, suggested a way of over-
coming it. In the Carnap-Stein view described above, scientific knowledge becomes
framework-relative (and thus, in a certain dimension, value-relative). Moreover, where
room is left for such a virtuous dialectic, there will be a continuous mutual adjustment
between provisionally-chosen principles of rationality and the knowledge we arrive at
within the framework defined by those principles. And if we regard these principles
of rationality as a hierarchy of the kind we observed in Sect. 3 within the principles of
inductive logic, then the highest level of this hierarchy includes principles of practical
reason articulating ultimate values. There is no infinite regress of selection frame-
works since these ultimate values, just as in a Kantian context, determine choices of
frameworks and explications, even as they continuously adjust to the better knowledge
articulated in those frameworks.

14 Cf. the discussion of Stein’s conception of pragmatics more generally in Carus (2010).
15 Which is again not to claim that the non-traditional, non-conformist origins of knowledge or values make
them inherently “better,” certainly not by any absolute standard, though presumably there is a better chance
for knowledge generated within the constraints of a given value system to be “better” by the standards of
that value system.
16 Among the better-known milestones in this tradition of Enlightenment critique are the Romantic
(and Goethean) rebellion against the portrayal of nature in terms of “Zahlen und Figuren,” Adorno’s and
Horkheimer’s “dialectic of enlightenment,” and Richard Rorty’s rejection of science as a cognitive exemplar
(Carus 2007, Ch. 1).

123
Synthese

This view may perhaps be found attractive, and perhaps even consistent with Car-
nap’s late perspective, but did the actual Carnap hold anything like it? Some of his
1930s rhetoric about value statements “having no meaning,” and his caginess about
the nature of the choice among frameworks or explications (George 2012, pp. 5–6),
suggest a negative answer. But there is also considerable evidence that Carnap did in
fact maintain a dialectical view of the kind suggested by Stein regarding the interrela-
tion of knowledge and values (Carus 2007). And surprisingly, it turns out there is even
evidence that Carnap held ultimate values to be subject to rational considerations. This
is made explicit in a shorthand fragment in Carnap’s Nachlass on “value statements”
that Carnap had originally sketched as part of the final section of his replies in the
Schilpp volume (ASP/RC 089-14-01). He evidently decided that this reply, however,
already by far the longest in a thousand-page book, was getting out of hand, and saved
up this part for a separate publication (which he never got around to).
In this fragment, Carnap first discusses rationality relatively to a value function,
which is a function of an entire world history (state description with arbitrarily long
time dimension). “Good,” “better,” and other value words are defined relatively to
such value functions. A person has many different value functions of different scope,
ranging from those focussed on a particular value aspect, such as health or good food, to
the most comprehensive value function in which all aspects are weighted according to
their relative priority within an overall system of values. These comprehensive value
functions, Carnap says, might appropriately be called the moral ones. The narrow-
scope value functions are essentially expected utility functions restricted to a particular
decision problem as addressed in decision theory; the “possible worlds” or “states of
nature” that frame the decision problem are restricted to a limited number of parameters
(e.g. whether it is raining or not, what to eat), and the expected utility function ranges
over a limited number of arguments (staying dry, good health, pleasure in eating).
Confirmational success is one example of a narrow-scope value; the expected utility
function in an inductive decision problem ranges over arguments such as fidelity
to the facts or empirical robustness. Carnap notes that in the case of induction (as
correspondingly in other narrow-scope decision problems) it is customary to suppose
that certain inductive methods or axioms could be criticized as irrational, by a priori
standards of inductive rationality (“a priori” understood as in Sect. 2 above). He then
asks whether it makes sense to suppose that there could analogously be standards of
rationality for (comprehensive) value functions themselves. And he argues that yes,
in fact, we can impose standards of rationality on moral value functions. He makes
some rather throwaway suggestions for the kind of thing he has in mind, involving
formal constraints on the shape of the value function (e.g. that it not fluctuate too
drastically over time),17 but backs off immediately and says he has nothing invested
in these examples, and is adducing them just as illustrations.

17 These examples suggest that he would have liked the general approach, at least, of Nozick’s list of 23—
progressively stronger—conditions for constraining “rational” utility functions in the section “Rational
Preferences” of a chapter entitled “Instrumental Rationality and its Limits” (Nozick 1993, pp. 139–151).
Nozick does not argue explicitly that there is no genuine distinction between instrumental and substantive
rationality, but by the end of this list it is hard to deny that the boundary is much less clear-cut than the
rhetoric surrounding this debate usually suggests.

123
 Synthese

In any case, wherever he might ultimately have taken this, it is worth emphasiz-
ing that he was not only willing to entertain a broader instrumental rationality, as
in the replies to Lenz and Burks cited in Sect. 3, but far beyond that, argues here
for the rationality of ends. An echo of Kant—especially the hierarchy of Vernunft
and Verstand—is unmistakable. For Carnap says, strikingly, that “all logic, including
inductive logic, and factual knowledge are irrelevant” to the question of the rationality
of (comprehensive) value functions; we must rather draw on “other, purely valuational
criteria by which to judge a value function as more or less rational than another.” (ibid.,
p. 4) So comprehensive value functions are not only a component of rationality; they
govern the choice of instrumental values (narrow-scope or partial value functions) that
constitute the reason by which we judge axioms of deductive and inductive logic—
they govern the application of the understanding to its objects. The understanding,
asymmetrically, has no bearing on the principles of reason.
It is clear that this sweeping exclusion cannot be taken quite literally; in his exam-
ples of possible “purely valuational” constraints on comprehensive value functions,
Carnap uses both empirical concepts and concepts of analysis such as continuity and
differentiability. Indeed, it is hard to imagine how “all logic and factual knowledge”
could be avoided in the specification of such constraints. As in the case of the choice
of inductive axioms and inductive methods, Carnap obviously does not mean that
experience has no relevance whatever to the choice of values. He evidently means,
as in the inductive case, that it cannot determine the choice of values. Experience,
i.e. empirical knowledge (called “descriptive pragmatics” in this context), forms the
essential background to all deliberation about values (Carnap 1963, p. 83), just as it
does in deliberation about inductive axioms and explication more generally. Its role in
the pragmatic sphere is not determinative, though, it is (as he says in the unpublished
1957 paper quoted from in Sects. 2 and 3) “heuristic” (footnote 5 above).
And as in our discussion in Sect. 3 of Carnap’s hierarchical justification of the
axioms of inductive logic—holding them fixed for the purposes of justifying the choice
of a particular inductive method within a λ-class—it seems clear that the principles of a
given comprehensive value function, once decided on, are also to be held provisionally
fixed as a basis for choices of axioms for inductive logic, deductive logic, and other
components of our “rationality.” In other words, the hierarchy of rationality has at
least three levels, in the case of inductive logic, and could in other cases have more.
While such a hierarchy of reason and understanding, and the primacy of practical
reason, are certainly reminiscent of Kant, the resulting perspective can hardly be called
Kantian in any substantial sense. It lacks the keystone of Kant’s normative structure—
a “supreme principle of morality.” Practical reason is allowed for in Carnap’s view,
but imposes insufficient constraints to generate any such positive principle.18 And
the Carnap-Stein regulative ideal described above, where reason and understanding
mutually inform each other in a virtuous dialectic, is more that of the western Enlight-

18 Wood (2006) portrays Kant as rejecting the idea that there could be fundamentally different “supreme
principles of morality,” and as arguing that practical reason shows all such supreme principles to be equiva-
lent. This would make Carnap’s loose and pluralistic practical reason even more un-Kantian than suggested
above.

123
Synthese

enment, of Diderot, Hume, or Condorcet, than of Kant, who stressed the complete
independence of practical reason from the understanding.
But if Carnapian practical reason can generate no supreme principle of morality, if
it can only constrain, not determine, the choice among frameworks, how can it decide
on the cognitive framework to be employed for science? How can it stop the infinite
regress of framework choices? Well, how is the regress stopped in other practical
decisions? For Carnap, this question has two components, since action-guiding nor-
mative sentences, in his view, have not only a normative component (the province of
pragmatics), but also a cognitive (or descriptive) one (belonging to semantics). The
cognitive framework for decision-making does generate a potentially infinite hierarchy
of meta-languages, but in practice our ultimate meta-language is a form of the “mother
tongue,” as Quine called it.19 There is a fundamental difference, though, in the status
of this retreat to ordinary language in Carnap and Quine, reflecting their differing
appraisals of the availability of an external perspective.20 For Quine the recourse to
the mother tongue is a kind of acquiescence in a default standpoint that we take “at
face value,”21 while for Carnap the ordinary language in which we make practical
decisions, including the decisions about which language to use for what, is precisely
not to be taken “at face value,” but rather as a rough and ready Neurathian “universal
vernacular”22 in which all the concepts that have scientific explications are to be taken
in their explicated senses. In the Syntax, it is true, Carnap had aspired, as he himself
testifies (Carnap 1963, p. 55) to make the philosophical metalanguage completely
precise. But he later gave up this ideal; indeed he admitted that “even when I proposed
them” none of his explications of the term “philosophy”—and thus of the pragmatic
meta-language—“seemed fully satisfactory to me . . . and I did not like the explica-

19 In precisely this context of a threatening regress of languages in which to specify quantifiers of object
languages lower in the hierarchy: “. . . in practice, we end the regress of background languages, in discussions
of reference, by acquiescing in our mother tongue and taking its words at face value.” (Quine 1969, p. 9)
20 Ricketts (2004) disagrees, citing Carnap’s reply to Evert Beth: “Since the metalanguage ML serves as
a means of communication between author and reader or among the participants in a discussion, I always
presupposed, both in syntax and in semantics, that a fixed interpretation of ML, which is shared by all
participants, is given.” (Carnap 1963, p. 929) But Carnap himself, in the same text, makes explicit that he
does not intend this shared default language to be taken at face value. “It seems to be obvious that , if two
men wish to find out whether or not their views on certain objects agree, they must first of all use a common
language to make sure that they are talking about the same objects. . . It is of course not quite possible to
use the ordinary language with a perfectly fixed interpretation, because of the inevitable vagueness and
ambiguity of ordinary words. Nevertheless it is possible at least to approximate a fixed interpretation to a
certain extent, e.g. by a suitable choice of less vague words and suitable paraphrases.” (ibid., pp. 929–930)
21 Though a very different one from that held out by the ordinary language philosophers, who consid-
ered ordinary language the philosopher’s “sole and essential point of contact with reality” (Strawson 1963,
p. 518). For Quine, the status of ordinary language is much more problematic, since it has no indepen-
dent authority (the parts of it found defective, within the conceptual scheme as a whole—in the light of
regimentation—are to be eliminated), but is nonetheless a kind of default standpoint.
22 Here again, as in the case of Stein’s extrapolation from Carnap above, some interpretive leeway is
required that some readers may find too wide, and reject as incompatible with their conception of Carnap’s
overall perspective. In this case, the leeway in question is that suggested by Thomas Uebel (2007, p. xvi)
regarding what he sees as the “emerging consensus on metaphilosophical matters on the so-called left wing
of the Vienna Circle,” which includes not only Carnap and Neurath but also Philipp Frank and Hans Hahn,
among others; see also Uebel (2004), and further discussion on the “scientific vernacular” in Carus (2007,
Ch. 11).

123
 Synthese

tions proposed by others any better. Finally, I gave up the search.” (Carnap 1963,
p. 862) So while there is indeed a sense in which both Carnap and Quine accepted a
version of the “mother tongue” as the practically necessary ultimate metalanguage,
there remains the fundamental difference that Quine took his version of the “mother
tongue” at face value and insisted that truth is truth and reality is reality (Dreben 1990,
p. 85), while Carnap saw no reason for accepting such traditional terms at face value.
They occur in the scientific vernacular we have to use in ordinary communication,
but are to be understood only in their explicated senses—which can dispense with
“ontological commitment” along with many other folk notions embedded in ordinary
language.
What about the normative component of our decision-making framework? Does
an infinite regress threaten here? In addition to first-order preferences, people have
preferences about those preferences, and so on.23 But this hierarchy, like the semantic
one, is limited by human processing capacity, and in effect people mostly have stable
long-term preferences that function as comprehensive value functions of the kind
Carnap sketches, corresponding to the longer-term preferences he takes for granted
in his writings on inductive logic.24 It is in the light of this comprehensive function,
trading off short-term needs and desires against longer-term values, that people make
the actual choices constituting their “revealed preferences.”
These still do not determine the choice of framework, since a framework, or even a
particular explication, is not chosen by an isolated individual, but by a social process
of some kind. Individual choices translate into social choices via economic, social, and
political interactions. The pluralism of this Carnapian, open-ended form of practical
reason is fundamental, then, reflecting that the choice of frameworks for knowledge
cannot be decided by impersonal reason alone but is—as in John Stuart Mill—a matter
for (reason-guided) controversial, political interaction.
In fact, the Carnap-Stein regulative ideal described above can also be seen as
nudging individual value-functions toward a degree of convergence, at least within
the subset of values that concern the framework for social and political interaction
(“framework values”).25 For insofar as individual value functions incorporate this
regulative ideal, they will be framed within the same “scientific vernacular,” and thus
be influenced by roughly the same empirical facts and theories.26

23 Frankfurt (1971) introduced this idea into the moral philosophy literature, where it has been current ever
since, but never high profile, any more than in economics, despite the occasional efforts of Amartya Sen.
24 For instance, in Carnap (1962, p. 312): “In the sphere of human action we have first concepts describing
overt behavior, say of a boy who is offered the choice of an apple or an ice cream cone and takes the
latter; then we introduce the concept of an underlying momentary inclination, in this case the momentary
preference of ice cream over apple; and finally we form the abstract concept of an underlying permanent
disposition, in our example the general utility function of the boy. . . When we wish to judge the morality of
a person, we do not simply look at some of his acts, we study rather his character, the system of his moral
values, which is part of his utility function. Single acts without knowledge of motives give little basis for a
judgment. . .”
25 On the distinction between “framework” and “content” values as characteristic of the liberal, Anglophone
wing of the Enlightenment tradition (in contrast to the continental, engineering-oriented wing), see Carus
(2007, pp. 296–7ff.).
26 “Roughly” since not everyone accepts every current theory (or alleged fact) in every scientific and
historical discipline. But there will be an “overlapping consensus” among all those who use the scientific

123
Synthese

5 Carnapian rationality

Critiques of Carnapian rationality come in many forms, so it is hard to address them

comprehensively. But one widely-discussed form of the first criticism discussed above
is that of Quine, whose basic complaint is that Carnap does not rest content with a
standpoint internal to science, in mediis rebus (Dreben 1994), but insists on stepping
outside the “conceptual scheme” of working science and appraising it from an external
perspective. Most forms of the second criticism, e.g. from a modernized Kantian
standpoint such as that of Habermas or Rawls, have just the opposite complaint. For
them, Carnap’s external standpoint is too modest. The external perspective Carnap
is willing to countenance, in their view, is insufficiently robust in that it allows too
little scope for practical reason to yield substantive principles of morality, even of the
extremely formal kind that Kant found in the categorical imperative.
The actual Carnap, though, proceeds from a starting point in which our cognition
is just as entangled, just as much in mediis rebus, as Quine or Putnam. Language
users, in Carnap’s view, are also participants in the world that a language is about.
They are not just representers, and consumers of representations, but also agents in
the world so represented. Unlike Quine (or Putnam in some of his moods), however,
Carnap thought that we can also step back from that existential Geworfenheit of having
to act and facing choices. We can abstract from it and isolate certain features from
immediate, deeply entangled experience. We can separate out a purely representational
(or semantic) component of language from its pragmatic embeddedness in a world of
agent-users. (If we were unable to do this, our species would have been trapped from
the outset in Parmenidean paralysis; cognitive life could never have got off the ground.)
For everyday scientific and technological purposes, this semantic abstraction works
smoothly and requires no help from pragmatics. Indeed, the cognitive progress of the
species depends, to a large degree, on our institutionalization of this and similar feats of
abstraction. But when the need for explication or language change arises, semantics is
no longer self-sufficient. The choice among explications (as Thomas Kuhn emphasized
in different words, from a different perspective) is external to the particular language
framework(s) under consideration (Stein 1992, p. 280).27
Carnap’s approach to this predicament differs sharply from Quine’s. Quine excluded
external discourse altogether, and so had to take the mother tongue at face value, i.e. in
a universal sense.28 Carnap too excluded external discourse, including questions about
the truth or the nature or the existence of something without reference to a language. But
for him this exclusion extended only to semantics, where the user-as-agent is abstracted

Footnote 26 continued
vernacular, and the ultimate source of this overlap—empirical evidence—is a more compelling motivation
than the supposed “reasonableness” of the contending “comprehensive doctrines” that drives Rawls’s well-
known version of such an overlapping consensus (Rawls 1996, pp. 58–66). Of course it could perhaps
be argued that all Rawls means by “reasonableness” here is, in fact, a willingness to accept and use the
scientific vernacular.
27 A comparison of Kuhn and Stein in this respect, with particular emphasis on Stein’s Carnapian prag-
matics, is to be found in Carus (2013).
28 This is the upshot of the “reciprocal containment” discussed by Ricketts (2009); on Quine’s “univer-
salism” cf. Hintikka (1990) and Quine’s (1990) thin-skinned but hardly convincing response.

123
 Synthese

away. Faced with a problem that points beyond semantics, a problem that puts the
concepts themselves or the constitutive rules of the language into question, we retreat—
not to a “mother tongue,” as Ricketts (2004, p. 199) maintains—but to pragmatics,
where the user is once again an actor in the world represented, and there are choices
to be made.29 The inescapability of choice does not, however, mean we must retreat
to to the arbitrariness and inconsistency of ordinary language, or to the immediacy
of a Lebenswelt. Of course we live—inescapably—in a Lebenswelt, but pragmatics,
unlike the mother tongue, affords us tools to transcend it when we need to (just as
language, especially written language, gives us tools to transcend the limitations of
immediate subjective consciousness). Descriptive pragmatics draws on our entire fund
of empirical knowledge (specifically on our knowledge about knowledge), and pure
pragmatics enables us, by whatever standards we set ourselves, to understand and act
in conformity with the consequences of our provisionally chosen values (and revise
them if we cannot accept those consequences).
In this limited sense, though, Carnap did, in contrast to Quine, insist on the availabil-
ity of an external perspective, from which criticism of our knowledge and our language
is possible (e.g. George 2012). An example of such a perspective is his posthumously
published discussion of entropy in thermodynamics (Carnap 1977). In the process
of seeking an abstract definition of entropy for purposes of inductive logic, he also
looked at the “nature of the physical concept of entropy in its classical statistical form,
as developed by Boltzmann and Gibbs,”

. . . and I arrived at certain objections against the the customary definitions,

not from a factual-experimental, but from a logical point of view. It seemed
to me that the customary way in which the statistical concept of entropy is
defined or interpreted makes it, perhaps against the intention of the physicists,
a purely logical instead of physical concept; if so, it can no longer be, as it
was intended to be, a counterpart to the classical macro-concept of entropy
introduced by Clausius, which is obviously a physical and not a logical concept.
The same objection holds in my opinion against the recent view that entropy
may be regarded as identical with the negative amount of information. (Carnap
1963, pp. 36–37)

He was unable, to his own disappointment, to get this idea across to the physicists
at Princeton, “chiefly,” he surmised, “because of great differences in point of view and
language.” Regardless of the merits of Carnap’s critique30 in this particular case, it
illustrates the sort of external, critical perspective he regards as legitimate and suggests
how it could be extended to other aspects of scientific practice. Such a perspective is not
available to Quine, who would in any case have rejected Carnap’s central distinction

29 Friedman’s (2001, pp. 47–68, 105–115; 2011, pp. 712–729) argument to the effect that object-level
Kuhnian incommensurability (e.g. during scientific revolutions) can coexist with meta-level rationality at
the level of philosophical or meta-scientific debate (e.g. the debate about geometry between Poincaré and
Helmholtz as a backdrop to Einstein’s introduction of relativity) makes essentially this same point; Carnap
would have classified such meta-scientific debate as belonging to descriptive pragmatics.
30 See Shimony (1977) for a balanced discussion, which however appears somewhat to misconstrue the
aims of Carnap’s critique.

123
Synthese

between “purely logical” and “physical” concepts of entropy. More generally, though,
choices among language forms do not fall within the scope of Quine’s conception of
philosophy as continuous with science.
Along the lines of Carnap’s entropy example, one could imagine critiques showing,
for instance, that certain features of the classical game theory used in economics or
political science are at variance with accepted empirical facts, or with theories in other
disciplines. From here it is a short step to more general systems of practical reason
in which e.g. consistency among various normative principles is itself at issue, or of
course consistency between normative principles and observed practices.
This use of “consistency” (which implies a background logic) may appear to be
in tension with Carnap’s statement in the “value concepts” fragment that “all logic,
including inductive logic, and factual knowledge are irrelevant” to any constraints
on the rationality of comprehensive value functions. But as we saw above, this can
hardly be taken literally, as such constraints are impossible to state in “pure optatives”
alone, i.e. without some factual component, as Carnap recognized in labelling that
component “descriptive pragmatics.”
Nor is there anything circular about the use of logic or facts in their capacity as
descriptive pragmatics. What we use at the pragmatic level, for purposes of choosing
among frameworks or explications, as our working conception of reason, is (just as
in the case of inductive axioms) provisionally fixed. We have at our disposal all the
knowledge available via that provisionally fixed meta-framework, to inform us, as
well as is currently possible, about the consequences of selecting any given frame-
work. This does not mean that either the meta-framework or the knowledge framed
in it is inviolate or frozen in place (as Carnap’s rhetoric sometimes—misleadingly—
suggests). It is, from the viewpoint of the individual or group making choices among
concepts or systems of concepts, the best available right now. Nor is it applied induc-
tively to select frameworks that, in turn, define what is to be counted as knowledge.31
The role of descriptive pragmatics can never be more than a supportive, “heuristic”
one; the logical and empirical knowledge employed as descriptive pragmatics can-
not decide the issue, it can only inform decision-makers of the consequences of their
choices.
So Carnap, in contrast to Quine, left room for an external perspective, even a critical
and valuational perspective, on scientific knowledge. But in contrast to Kant, or most
philosophers from the Kantian or German idealistic or phenomenological traditions,
Carnap did not think this perspective could yield anything more than critique; he did
not think we could actually find out anything, either about facts or about values, from
this external perspective. He did not think that Vernunft could give us a categorical
imperative any more than he thought could it give us synthetic a priori knowledge.
The most it could do—though this was of overriding importance—was to guide Ver-
stand, the search for knowledge, by the wider view it offered.32 And this wider view
afforded by the perspective of Vernunft could in turn be informed and disciplined by

31 Which would invite the kind of circularity charge sometimes levelled at Quine, discussed in depth by
Gregory (2008); cf. also the review of Gregory’s book by Burgess (2009).
32 Michael Friedman provides good examples of this in his cases of philosophical, meta-scientific debate
guiding object-level scientific change (footnote 29 above).

123
 Synthese

the progressively better and more comprehensive picture of the world given us by
(Vernunft-guided) Verstand. This is certainly not the “criterial” conception of rational-
ity Putnam (1983) saw and criticized in Carnap, but the modesty of this picture also
sets it apart from the great majority of ethical projects in any tradition.
One reason it does not emerge clearly from Carnap’s writings is, of course, that
he had not fully worked it out and was otherwise preoccupied. But another reason,
it has to be admitted, is Carnap’s own rhetorical clumsiness. The language he used
regarding the “a priori” and the “analytic” (not to mention “intuition”) was, as we
saw above, highly misleading as a way of explaining what he was actually trying to
get across. Likewise, to articulate his idea of holding higher-level inductive principles
provisionally fixed while choosing from the range of possibilities they still left open,
he often conveyed a rigidity that was, in fact, quite alien to the conception he actually
had in mind.
These rhetorical failings, though exasperating, need not cloud our appraisal of the
architectonic of rationality Carnap was feeling his way toward in his later years. Once
we unpack his scattered remarks, we can see that there is actually quite a powerful
vision behind them, an overall perspective from which scientific rationality and ethical
values fit into a single coherent conception without sacrificing the pluralism, tolerance,
and engineering constructivism that characterize Carnap’s ideal of explication. It no
longer looks impoverished or defective in the ways it is often portrayed to be; in fact
it is coherent at quite a deep level, lends itself to ampliative development as suggested
by Stein and Uebel, and could well repay further elaboration.

6 Unpublished sources

Documents cited are from two different collections, the Carnap papers at the Archives
of Scientific Philosophy at the Hillman Library, University of Pittsburgh (abbrevi-
ated ASP/RC) and the Carnap papers (manuscript collection no. 1029) at the Young
Research Library, University of California at Los Angeles (UCLA). Citations are by
collection, followed by location, thus ASP/RC followed by three numbers separated
by hyphens (ASP-RC XX-YY-ZZ, where XX is the box number, YY the folder num-
ber, and ZZ the item number) or UCLA followed by box, folder, and sometimes page
numbers.

Acknowledgment The unpublished texts listed above are quoted by permission of the University of
Pittsburgh and the Carnap heirs, respectively, which is gratefully acknowledged. I am also grateful to Georg
Schiemer for organizing the conference at the Munich Center for Mathematical Philosophy in July 2013
where an earlier version of this paper was presented, as well as to Thomas Uebel, Florian Steinberger,
Pierre Wagner and two anonymous referees for perceptive comments that greatly improved the paper. The
influence of many conversations with Howard Stein and Michael Friedman on the overall viewpoint of the
paper is pervasive.

References

Burgess, J. P. (2009). Review of Gregory 2008. Notre Dame Philosophical Reviews.

Burks, A. (1953). The presupposition theory of induction. Philosophy of Science, 20, 177–197.

123
Synthese

Burks, A. (1963). On the significance of Carnap’s system of inductive logic for the philosophy of induction.
In P. A. Schilpp (Ed.), The philosophy of Rudolf Carnap (pp. 739–759). LaSalle, IL: Open Court.
Carnap, R. (1950a). Logical foundations of probability. Chicago: University of Chicago Press.
Carnap, R. (1950b). Empiricism, semantics, and ontology (Reprinted in his Meaning and Necessity,
pp. 205–221, 2nd Ed.), Chicago: University of Chicago Press 1956.
Carnap, R. (1952). The continuum of inductive methods. Chicago: University of Chicago Press.
Carnap, R. (1962). The aim of inductive logic. In E. Nagel, P. Suppes, & A. Tarski (Eds.), Logic, methodology,
and philosophy of science: Proceedings of the 1960 International Congress Stanford (pp. 303–318).
CA: Stanford University Press.
Carnap, R. (1963). Autobiography and Replies and systematic expositions. In P. A. Schilpp (Ed.), The
philosophy of Rudolf Carnap (pp. 3–84 and 859–1013). LaSalle, IL: Open Court.
Carnap, R. (1968). Inductive logic and inductive intuition. In I. Lakatos (Ed.), The problem of inductive
logic: Proceedings of the International Colloquium in the Philosophy of Science, London 1965 (Vol.
2, pp. 258–267). Amsterdam: North-Holland.
Carnap, R. (1977). Two essays on entropy. Oakland, CA: University of California Press.
Carus, A. W. (2007). Carnap and twentieth-century thought: Explication as enlightenment. Cambridge:
Cambridge University Press.
Carus, A. W. (2010). The pragmatics of scientific knowledge: Howard Stein’s reshaping of logical empiri-
cism. Monist, 93, 618–639.
Carus, A. W. (2013). History and the future of logical empiricism. In E. Reck (Ed.), The historical turn in
analytic philosophy (pp. 261–293). Basingstoke: Palgrave Macmillan.
Dreben, B. (1990). Quine. In R. Barrett & R. Gibson (Eds.), Perspectives on Quine (pp. 81–95). Oxford:
Blackwell.
Dreben, B. (1994). In Mediis Rebus. Inquiry, 37, 441–7.
Earman, J. (1993). Carnap, Kuhn, and the philosophy of scientific methodology. In P. Horwich (Ed.), World
changes: Thomas Kuhn and the nature of science (pp. 9–36). Cambridge, MA: MIT Press.
Feigl, H. (1954). Scientific method without metaphysical presuppositions. Philosophical Studies, 5,
17–32.
Frankfurt, H. G. (1971). Freedom of the will and the concept of a person. Journal of Philosophy,
68, (Reprinted in H.G. Frankfurt, The importance of what we care about: Philosophical essays)
(pp. 11–25). Cambridge: Cambridge University Press.
Friedman, M. (2001). Dynamics of reason: The 1999 Kant Lectures at Stanford University. Stanford, CA:
CSLI Publications.
Friedman, M. (2011). Synthetic history reconsidered. In M. Domski & M. Dickson (Eds.), Discourse on a
new method: Reinvigorating the marriage of history and philosophy of science (pp. 571–813). LaSalle,
IL: Open Court.
George, A. (2012). Opening the door to cloud-cukoo-land: Hempel and Kuhn on rationality. Journal for
the History of Analytical Philosophy, 1, 1–17.
Gregory, P. A. (2008). Quine’s naturalism: Language, theory, and the knowing subject. London: Continuum.
Hintikka, J. (1990). Quine as a member of the tradition of the Universality of Language. In R. Barrett & R.
Gibson (Eds.), Perspectives on Quine (pp. 159–175). Oxford: Blackwell.
Irzik, G. (2003). Changing conceptions of rationality: From logical empiricism to postpositivism. In P.
Parrini, W. C. Salmon, & M. H. Salmon (Eds.), Logical empiricism: Historical and contemporary
perspectives (pp. 325–346). Pittsburgh: University of Pittsburgh Press.
Lenz, J. W. (1956). Carnap on defining “degree of confirmation.” Philosophy of Science, 23, 230–236.
Nozick, R. (1993). The nature of rationality. Princeton, NJ: Princeton University Press.
Putnam, H. (1983). Philosophers and human understanding. In H. Putnam (Ed.), Realism and reason:
Philosophical papers (Vol. 3, pp. 184–204). Cambridge: Cambridge University Press.
Quine, W. V. O. (1969). Ontological relativity (Reprinted in his Ontological relativity and other essays,
pp. 26–68). New York: Columbia University Press.
Quine, W. V. O. (1990). Comment on Hintikka. In R. Barrett & R. Gibson (Eds.), Perspectives on Quine
(p. 176). Oxford: Blackwell.
Rawls, J. (1996). Political liberalism (2nd ed.). New York: Columbia University Press.
Ricketts, T. (2004). Frege, Carnap, Quine: Continuities and discontinuities. In C. Klein & S. Awodey (Eds.),
Carnap brought home: The view from Jena (pp. 181–202). LaSalle, IL: Open Court.
Ricketts, T. (2009). From tolerance to reciprocal containment. In P. Wagner (Ed.), Carnap’s logical syntax
of language (pp. 217–235). Basingstoke: Palgrave Macmillan.

123
 Synthese

Shimony, A. (1977). Introduction, In Carnap (1977) (pp. vii–xxii).

Stein, H. (1992). Was Carnap entirely wrong, after all? Synthese, 93, 275–295.
Stein, H. (1998). How does physics bear upon metaphysics; and why did Plato hold that philosophy cannot
be written down?, Colloquium Talk at the University of Chicago. http://awcarus.net/A.W._Carus/
Home_files/Physics%20and%20metaphysics.pdf.
Steinberger, F. (2014). How tolerant can you be? Carnap on the normativity of logic. (Typescript of January
2014)
Strawson, P. F. (1963). Carnap’s views on constructed systems versus natural languages in analytic phi-
losophy. In P. Schilpp (Ed.), The philosophy of Rudolf Carnap (pp. 503–518). LaSalle, IL: Open
Court.
Uebel, T. (2004). Carnap, the left Vienna circle, and neopositivist antimetaphysics. In C. Klein &
S. Awodey (Eds.), Carnap brought home: The view from Jena (pp. 247–278). LaSalle, IL: Open
Court.
Uebel, T. (2007). Logical empiricism at the crossroads: The Vienna Circle’s protocol-sentence debate.
LaSalle, IL: Open Court.
Wood, A. W. (2006). The supreme principle of morality. In P. Guyer (Ed.), The Cambridge companion to
Kant and modern philosophy (pp. 342–380). Cambridge: Cambridge University Press.

123