Journal of Philosophy, Inc.

Coherence and Morality

Author(s): Carl G. Hempel
Source: The Journal of Philosophy, Vol. 62, No. 19, American Philosophical Association
Eastern Division Sixty-Second Annual Meeting (Oct. 7, 1965), pp. 539-542
Published by: Journal of Philosophy, Inc.
Stable URL: http://www.jstor.org/stable/2023749
Accessed: 19/07/2010 22:49

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=jphil.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

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.

Journal of Philosophy, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal
of Philosophy.

http://www.jstor.org
 SYMPOSIUM: ETHICS AND DECISION THEORY 539

with the best will in the world, agent 1 will not be able to find a
propositionE that will serve the purposewhen A is the proposition
that agent 1 will die immediately and B is the corresponding
propositionabout agent 2. But, if the mirroringtrick is applicable
to a sufficientlyextensive and varied coherent subrankingof agent
2's preferences,the two rankings may plausibly be correlated by
setting des, (AE) = des2A for each propositionA in the subrank-
ing. The desirabilitiesof the rest of the propositionsin the two
full rankings will then be determined by their relationships to
propositions in the subrankings. (In Ramsey's theory or in the
present theory, once the desirabilitiesof two or of three differently
ranked propositionsare fixed, the desirabilitiesof all other proposi-
tions in the rankingare determined.)
Alternatively, by applying the sorts of considerationthat arise
naturally in discussions of fairness, one might alter the mirroring
so as to extend the domain of comparability by such devices as
permuting appropriateproper names, so that with A and B as in
the precedingparagraphwe have des1 (AB) = des2B and des, (BE)
= des2A for some (now, more plausible) propositionE.
RICHARDC. JEFFREY
THE CITY COLLEGE
CITr UNIVERSITY OF NEW YoRK

COHERENCEAND MORALITY*
AN analysis of moral deliberationin the light of current logical
and mathematicaltheories of decision is, I think, an impor-
tant undertaking, and Professor Jeffrey's paper provides illumi-
nating illustrations of the kinds of insight this approachmay yield
and of the problems it raises.
Jeffrey's decision-theoreticalmodel is applicable especially to
situations where an agent wishes to make a choice among several
courses of action that he believes to be available to him, each with
a set of "possible outcomes." To each of the outcomes,the agent
is assumed to assign, in accordance with his beliefs and prefer-
ences, (i) a probability of its occurrenceas a result of the rele-
vant action, and (ii) a valuation expressed by a numerical de-
sirability. The decision rule Jeffrey proposes for these cases IS
based on Bayes' principle, which assigns to each action a desir-
ability that is defined as the probabilisticallyweighted average of
* Abstract of a paper to be delivered in an APA symposium, commenting
on Richard C. Jeffrey, "Ethics and the Logic of Decision," this JOuRNAL,62,
19 (Oct. 7, 1965): 528-539.
540 THE JOURNAL OF PHILOSOPHY

the utilities of its different possible outcomes. An optimal de-

cision will then select an action for which this desirability is
maximal.
Although the basic conception of the model is Bayesian, the
specific form Jeffrey gives it includes some interesting technical
innovations. Thus, for example, he represents actions by propo-
sitions (to the effect that such and such action is taken), and he
assigns desirabilities not only to propositions that characterize
actions, but to all propositionsexpressiblein that fragment of the
agent's language in which he formulates his deliberations. These
desirabilities are fully determinedby the "expectation principle"
once desirabilities and probabilities have been assigned to the
"basic cases," i.e., to the relatively most specific propositions ex-
pressible in the given language fragment. There follows the in-
teresting result that, even though the desirabilities of the basic
cases may be chosen independently of their probabilities, never-
theless a change in the probabilities alone, reflecting a change in
belief, may entail a change in the desirabilities of some nonbasic
propositions, such as those indicating the possible outcomes of a
contemplated action.
Jeffrey's model offers us just one general rule of deliberation,
namely, to choosea preferenceranking that is coherentin the sense
of being derivable from basic probability and desirability assign-
ments in conformity with the expectation principle. This leads
to his question whether coherence by itself affords "a sufficient
basis for all morality." He does not make clear, however, just
what features he would take to qualify a rule as such a "sufficient
basis." His discussion suggests that he is concerned with some-
thing I will vaguely refer to as the "moral soundness" of the
resulting decisions, and perhaps also with the avoidance of dis-
ruptive conflicts between the preferences of different agents using
the rule. Surely neither of these desiderata is guaranteed by the
formal rule of deliberation alone; and indeed, Jeffrey notes that
a radical optimism concerning human nature is required if co-
herence is to be adopted as a sufficientbasis for moral deliberation.
But, since the qualifying characteristic for such a basis is not
made explicit, it remains unclear what specific assumptions about
human nature such optimism would imply.
Moreover,it seems that, in addition to some form of "moral
optimism" concerning human nature, the use of coherenceas the
basic standard of moral deliberationwould also require a "cogni-
tive optimism" concerning the agents' ways of forming their
empirical beliefs. For grave conflicts could arise over a given
 SYMPOSIUM: ETHICS AND DECISION THEORY 541

moral issue even among agents who assign the same desirabilities
to all basic cases, if they differ strongly in their probability as-
signments. In his smoking example, Jeffrey mentions statistical
evidence as a possible guide for the choice of subjective probabili-
ties: but what if some agents assigned their probabilities in a
counterstatistical way? Incidentally, some experimental studies
suggest that the subjective probabilities representing the expecta-
tions of even quite intelligent agents may be distinctly at variance
with relevant statistical information available to them. Thus,
I wonder whether some additionialrequirementof rationality must
not be imposed upon the choice of probabilities and, if so, what
form it might take.
Jeffrey offers his model as a "framework within which any
correct deliberationcan be presented." Some moral decisions are
usually viewed as being dictated by just one more or less specific
qualitative standard, such as the golden rule, or the principle that
killing-with specified exceptions-is evil. Can the model accom-
modatesuch cases? A moral principle of the kind here considered
would have to be construed as assigning a strong negative desir-
ability to the disjunction of those basic cases which involve a vio-
lation of the principle. But this will normally be quite insufficient
to assign definite desirabilities to all the propositions relevant to
the deliberation, and the decision supposedly reached by means
of the principle cannot, therefore, be obtained witlhin the Jeffrey
model. But, rather than pointing out a weakness of the model,
this considerationshould be taken to show, I think, that the con-
strual of a moral decision as determined by one principle of the
kind here referred to involves a considerable oversimplification;
for an action is always performedin some particular setting many
of whose features may be relevant to its moral evaluation.
Finally, two basic questions-one of them theoretical,the other
moral-concerning the justifiability of choosing the Bayesian
maxim as a basic rule of moral deliberation.
Even within the mathematicaltheory of decision, the Bayesian
approachis rejected by a group of theoreticianswho question the
possibility of assigning probabilitiesto propositionsexcept in those
special cases where the probabilitiescan be interpreted in terms of
statistical frequencies. In other cases, decisions have to be made
"under uncertainty," i.e., without the benefit of probabilities;and,
interestingly, the pattern those theoreticianspropose for such de-
cisions is basically the same as that of the argumentwhich Jeffrey
cites and rejects as an example of "fallacious" deliberation.
542 THE JOURNAL OF PHILOSOPHY

Secondly, adherence to the Bayesian rule might be questioned

also on the moral ground that it makes the decision in a moral
conflict strongly dependent upon the probabilities of success of
the alternative actions under consideration. The rule may, there-
fore, direct us to refrain from an action that could bring about
a very high good and to act instead in an alternative way, which-
even at best-will yield a much lesser good, but which has a greater
probability of success.
CARL G. HEMPEL
PRINCETONUNIVERSITY

COMMENTS ON RICHARD C. JEFFREY'S "ETHICS AND

THE LOGIC OF DECISION" *
In this interesting paper, Professor Jeffrey suggests the thesis
that some of the formal models developed in statistical decision
theory provide a useful and constructive device for understanding
certain theoretical problems in ethics. Two such problems, con-
sidered as examples, are: (1) the normative character of coherent
preference rankings, and (2) the interpersonal comparison of
utilities. Parts of the development of his position are contained
in his forthcoming book, The Logic of Decision; presumably, the
questions raised here are answered there. In any case, the fol-
lowing is intended to bring up issues essential to the evaluation
of the thesis rather than to make any closed judgment. Indeed,
it is clear that Jeffrey's paper should be understood as a proposal.

I. EXPERIMENTAL CRITERIA

To the naive reader, many of the concepts developed formally

seem to suffer from the absence of appropriate experimental iden-
tification. For example, in the discussion regarding value and
belief, it is asserted that "we must think of propositions as dis-
junctions of basic cases (themselves propositions)." However, it
isn't stated how one determines whether or not a given proposi-
tion is a basic case. If the rule is, as would suit the requirements
of the formal model, that A is a basic case if and only if des A
is constant for the agent (over all other cases?, over time?, etc.),
then an element of circularity appears. If, however, the rule is a
more fundamental one from which such constancy may be deduced,
it would be extremely useful to have it stated. Some rule seems
*
Abstract of a paper to be presented in an APA symposium,commenting
on Richard C. Jeffrey, "Ethics and the Logic of Decision," this JOuRNAL,62,
19 (Oct. 7, 1965): 528-539.