Two Senses of ‘Ought’ in Forrester’s Paradox
Joseph Ulatowski∗
September 25, 2013
Abstract
If Smith commits murder, he should do so gently. Suppose then that
Smith commits murder. He should do so gently. From “Smith should
do A” it logically follows that he should do anything logically implied
by A. Smith’s murdering gently entails his committing murder. Thus,
Smith should commit murder. Since a rule stipulates that Smith should
not murder Jones, Smith both should and should not murder Jones. This
paper will attempt to present a viable solution to Forrester’s paradox. I
will argue that the different functions of two deontic operators show that
the contradiction cannot possibly follow. [97 words]
This is a DRAFT. Please do not cite without permission.
Comments, criticisms, and suggestions are welcome. c 2013
Joseph Ulatowski.
1
Introduction
The aim of this paper is to present a viable solution to the paradox of gentle murder, or Forrester’s paradox.1 First, I will summarize the paradox
and explain its logical structure. Next, I will address a a few proposed
solutions to the paradox. The first proposal has us reject one of the deontic principles to lead us out of the paradox. Such a solution overlooks
an important aspect of deontic pragmatics. Then, I will consider Walter
∗ This
paper has benefited from conversations or correspondence with Robert Barnard,
David Beisecker, Donald Bruckner, Robert Colter, James Forrester, Franz-Peter Griesmaier,
Marc Moffett, and Walter Sinnott-Armstrong. I am also indebted to members of the audience at the 2009 MidSouth Philosophy Conference for valuable feedback I received after a
presentation of this paper.
1 James Forrester (1984) is the originator of the paradox. He called it the paradox of
the adverbial samaritan because of its resemblance to the paradox of the good samaritan.
(Åqvist 1967)
1
Sinnott-Armstrong’s (1985) recommendation that limiting the scope of the
deontic operator to range over the manner in which the act is performed
will solve the paradox. His solution has presumed that ‘gently’ may be
treated as a predicate. Since not all adverbs of action may represent a
predicate of action, Sinnott-Armstrong’s proposed solution fails. Finally,
my solution consists in the distinction between two senses of ‘ought’. Under one formulation the obligation may concern how one ought to act and
under the other formulation the obligation may concern how the world
ought to be. The distinction suggests that there are actually two deontic operators at work in the paradox. Since the two deontic operators
function in significantly different ways, I will argue that the paradoxical
conclusion does not follow.
2
The paradox
I will start with a summary of the paradox. Three assumptions and three
rules of standard deontic logic (“SDL”) are at work in the paradox. Two
of the assumptions are rules that govern a subject’s behavior and the third
is an action sentence:
1. It is obligatory that Smith not murder Jones.
2. It is obligatory that if Smith murders Jones, Smith murders
Jones gently.
3. Smith murders Jones.
The three rules of SDL are2 :
R1: (p ⊃ q) ⊢ (Op ⊃ Oq)
R2: O(p ⊃ q) ⊢ (p ⊃ Oq)
R3: O p ⊃ ∼O∼p
The paradox arises when we apply the logical rules to the assumptions.
First, 2 and R2 yields:
2 For the formal rules of SDL, let “O” stand for “it is obligatory that. . . ,” and let the
variables “p” and “q” stand in for action sentences.
2
4. If Smith murders Jones, then it is obligatory that Smith
murder Jones gently.
From 3 and 4, we may derive:
5. It is obligatory that Smith murder Jones gently.
It seems obvious that:
6. If Smith murders Jones gently, then Smith murders Jones.
The application of R1 to 6 entails:
7. If it is obligatory that Smith murders Jones gently, then it
is obligatory that Smith murders Jones.
5 and 7 entail:
8. It is obligatory that Smith murder Jones.
Given R3, 8 yields 9: “it is not obligatory that Smith not murder Jones.”
We cannot consistently hold 9 and 1.
Forrester has rejected R1 to solve the paradox because other solutions
are not going to work. Forrester writes:
The most likely candidate for the scrap heap is [R1] - a basic
inference principle in standard deontic logic. If this is indeed
the rotten apple, the entire barrel of standard deontic logic
must be in a bad way.”(Forrester 1984, 197)
Forrester’s rejecting R1 will prevent the paradox from arising, but its
removal will not come without a high cost. Recommending the rejection
of R1 calls for a whole scale revision of SDL, since R1 is a rule derived
from more primitive axioms of deontic logic. This is too large a task for
him to undertake in his (1984) and by my own estimation it is something
we should definitely want to avoid, if we can help it.
3
Rejecting R2
R1 may not be the only deontic principle we should reconsider in light of
Forrester’s paradox. Someone may want to reject R2, given that deontic
3
logic is a branch of alethic modal logic.3
In alethic modal logic, the
principle M2: (p ⊃ q) ⊢ (p ⊃ q) is false. So, by analogy, the deontic
principle R2 must be false. Without R2, the paradox does not arise.
Forrester has defended the truth of R2 because the principle holds in
“certain special cases.” (Forrester 1984, 196) The paradox is one of the
special cases. Suppose I converse with Smith before the killing takes place.
I point to the moral rule that forbids Smith from killing Jones, and he
responds that he is going to kill Jones regardless of what I or anyone else
says. It seems natural for me to say to Smith, “well, then you had better
kill Jones gently.” According to Forrester, I have used R2 to come to this
conclusion; “and my reasoning, as well as my morals, would appear above
reproach.” (ibid.) The principle may be formally inadequate; however,
R2 ought to be upheld since it bears some importance for our moral
deliberation.
4
Manners matter
I will turn now to Sinnott-Armstrong’s radical way of disarming the paradox. Sinnott-Armstrong has used Donald Davidson’s (1967) formal analysis of action sentences to distinguish the occurrence of an act from the
manner of its perpetration. The antecedent of 6, “Smith murders Jones
gently,” in Davidsonian fashion becomes:
6.1. There is an act of murder by Smith of Jones and it is
gentle,
which in turn formalizes as:
6.2. (∃x )(Mxsj & Gx )
Since the act of murder and the gentleness of the act are disambiguated
on Davidson’s account, the obligation operator, O, may then be placed
3I
want to thank redacted for blind review for a discussion about this argument. See
von Wright’s (1951) for discussion of the close connection between deontic and alethic modal
logic. In that paper, von Wright argues that deontic logic is a branch of modal logic.
4
with distinct effect.
This gives rise to a wide and a narrow interpretation of 5. The operator
can express an obligation to murder when its scope includes the whole
proposition (the “wide” view), or the operator can express an obligation
to do so gently when it ranges over Gx (the “narrow” view). So, 5 may
be interpreted as:
5.1. O(∃x )(Mxsj & Gx )
5.2. (∃x )(Mxsj & OGx )
While the act of murder is obligatory in 5.1, the act of murder occurs and
the gentle manner in which it is carried out is obligatory in 5.2.4
The distinction between 5.1 and 5.2 now allows for various interpretations of 4. 4 and 5.1 yields:
4.1. (∃x )Mxsj ⊃ O(∃x )(Mxsj & Gx ).
4 and 5.2 entails:
4.2. (∃x )Mxsj ⊃ (∃x )(Mxsj & OGx ).
4.1 generates the paradox, since Smith is not only obligated to murder
Jones gently but is obligated to murder Jones. If, however, we employ
4.2, only the gentleness of Smith’s act is obligatory. 4.2 disables the
inconsistency because it rules out an obligation of Smith to murder Jones.
No contradiction can be derived from 1, 3, and 4.2 because only 5.2
follows from 3 and 4.2. If 6 is analyzed as:
6.3. (∃x )(Mxsj & Gx ) ⊃ (∃x )Mxsj
then R1 only warrants the conclusion:
7.1. O(∃x )(Mxsj & Gx ) ⊃ O(∃x )Mxsj
4 One
may claim that there is a third interpretation of 5. The third way is:
5.3. (∃x )(OMxsj & Gx ).
This interpretation may be formally correct, but informally unsound. To say that “there is an
act such that it is an obligatory murder by Smith of Jones and it is gentle,” is not an accurate
interpretation of 5, that “it is obligatory that Smith murders Jones gently.”
5
and the antecedent of 7.1 is 5.1 rather than 5.2. Since 5.1 cannot be
derived from 3 and 4.2, neither can 8. Since 8 generates the inconsistency,
no paradox can possibly arise. Hence, 1, 3, and 4.2 are consistent.
Sinnott-Armstrong’s revolutionary maneuver does not come without
a serious disadvantage. It is not clear what it means for an agent to be
obligated gently to do something without also being obligated to do something. The adverb ‘gently’ is not amenable to treatment as a predicate.
Davidson (2001) claims that ‘slowly’ is such an adverb. If ‘slowly’ is not
representative of a predicate of action, then adverbs like it are not either.
We can now propose the Paradox of Slow Murder. Replace “gently” with
“slowly” in the paradox, and we come to the same paradoxical conclusion
as in the original. But, since Davidson’s account of adverbial modification
does not apply to ‘slowly’, Sinnott-Armstrong’s solution does not apply
to the paradox.5
5
‘Ought to do’ v. ‘Ought to be’
The previous section showed that a theory of adverbial modification will
not likely solve Forrester’s paradox.6 In this section, I will argue that
there are at least two senses of ‘ought’. They have distinct meanings and
their logical behavior differs too. I will show that the paradox employs
both ‘ought to be’ and ‘ought to do’ statements without consideration for
the differences between them. If my analysis is correct, then a solution of
5 Romane Clark (1986) has argued that by defining a special kind of adverbial modifier, M,
where it governs the manner of doing but not the act done, renders the untoward obligation
to murder inoperative. She then says something very curious, “From the general “truth” that
one ought “gently” to do what one does, it does not follow, nor should it, that one ought to
do that which one does do.” (Clark 1986, 55) So, for instance, from “one ought gently to pet
my cat” it does not follow that “one ought to pet my cat.” I would agree if that was what was
occurring in the paradox. But an obligation to do x gently is an obligation to do x because
gently should not be interpreted here as a predicate. I think that Clark has generated an
interesting problem about adverbial modification, but it is unclear to me how it applies to the
paradox. I would like to thank Franz-Peter Greismaier for a discussion on this very point.
6 It should be noted that other people have tried to solve Forrester’s paradox by presenting
alternative theories of adverbial modification. For example, see Hector-Neri Castañeda (1985)
or Dale Jacquette (1986).
6
Forrester’s paradox is possible when the two deontic operators are properly
differentiated.7
‘Ought’ has many uses. Think of how the word comes up in ordinary
conversation. First, a conjectural ‘ought’ may express an expectation.
For example, “the train ought to arrive at 3pm.” According to a conjectural use, ‘ought’ merely indicates that the train will likely arrive at 3pm.
Second, ‘ought’ may report a suggestion. For example, “the truck driver
ought to use North McCarran Boulevard instead of Interstate-80.” An
accident on Interstate-80 may have made the highway impassable, and
the best alternative is for motorists to use North McCarran Blvd. Since
nothing prescribes that the train arrives at 3pm or that the trucker takes
North McCarran Blvd., neither the conjectural nor the suggestive ‘ought’
is a categorical use of the term.
Two ‘oughts’ are often confused with one another in deontic speech.
The first claims that a person should perform some action, so the ‘ought’
statement they contain is an ‘ought to do’. The second ‘ought to be’
statement tells us how the world would be improved if a given state-ofaffairs obtained or failed to obtain. What I will suggest now is not only
that they have two different meanings but that the two senses of ‘ought’
behave in logically different ways.
‘Ought to do’ and ‘ought to be’ statements have different meanings.
While ‘ought to do’ statements are generally directive, judgmental, and
informative, ‘ought to be’ statements are optative. Imagine a case where
Donna Elkin has competed in the new hit reality television contest American Rocker. Suppose that Donna’s performance is clearly superior in
comparison with the other contestants, but that the judges are biased,
and it is likely that the next American Rocker goes to someone else. If
one then said, “Donna Elkin ought to win American Rocker,” this kind
of statement is about the situation, not about Donna. A legitimate para7 Jim Forrester (1996) distinguishes between these two forms of obligation, but he does not
attempt to apply the distinction to his own paradox in that book.
7
phrase of the statement is “It ought to be that Donna wins American
Rocker.” Saying that Donna should win Rocker is not to judge her or
blame her if she loses. The sentence tells us how the world ought to be,
but it does not prescribe actions Donna should do to bring about her
winning the competition.
In Forrester’s paradox, 1. governs Smith’s behavior by indicating how
the world ought not to be. The rule states that, “It is obligatory that
Smith not murder Jones.” The rule is logically equivalent to “It ought
not to be that Smith murder Jones.” Under this reading, the rule refers
to a hypothetical state of the world that should not be brought about.
Since the rule refers to a hypothetical situation, rule 1 is an ‘ought to be’
statement.
‘Ought to do’ statements prescribe how one ought to act. Suppose
that Donna has not kept up with her voice training. So, one says, “Donna
ought to practice harder.” It tells Donna what she ought to do. The
statement implies that Donna is able to practice harder, and places blame
on her for not doing so. 7, which is logically derived from the assumption
6, says “if it is obligatory that Smith murders Jones gently, then it is
obligatory that Smith murder Jones.” Since in this case the antecedent
and consequent prescribe how Smith ought to act, “it is obligatory that
Smith murders Jones gently” is logically equivalent to “Smith ought to
murder Jones gently.” A paraphrase of the modified antecedent is “Smith
ought to do x.” Therefore, “Smith ought to murder Jones gently” is an
‘ought to do’ statement and he should do so if he can.
Whereas ‘ought to do’ statements imply can, ‘ought to be’ statements
do not. It is a commonly accepted principle that ‘ought implies can.’
Statements of the form ‘x ought to y’ imply, entail or presuppose statements of the form ‘x can y’.8 “It ought not to be that Smith murder
Jones” does not imply that Smith cannot murder Jones. In fact, Smith
8 Similarly, ‘x ought not to y’ implies, entails, or presupposes that x can refrain from, avoid,
or stop y-ing.
8
can murder Jones under the appropriate circumstances. But, when we
say, “Smith ought not to murder Jones,” it implies that he can refrain
from doing so. If Smith murders Jones, then he may be held morally
responsible for his action. The predicate of an ‘ought to do’ statement
allocates responsibility to the agent.
The two meanings of ‘ought’ seem to imply a difference between two
deontic operators. ‘Ought to do’ and ‘ought to be’ have different logical
behaviors. So, we must employ distinct logical operators for each.
The de re/de dicto distinction is familiar from modal semantics, and
I believe that deontic operators can also be either de dicto or de re. The
sentence “Sam can move the car” presents an apparent de re possibility.
A certain possibility holds of Sam, and Sam must exist for the sentence
to be true. Similarly, “Smith ought to murder Jones” can only be a de
re deontic modality. The proposition can be true only if there is such
a person as Smith. The sentence predicates a certain ‘ought’ of Smith.
The ‘ought to do’ operator does not operate on the entire proposition
because the subject term is outside the scope of the operator. It ranges
over predicates only.
An ‘ought to be’ statement is a de dicto modality. Consider “It is
possible that Sam moves the car.” This represents a de dicto possibility,
and Sam’s existence is not required for its truth. Correspondingly, ‘ought
to be’ statements are not committed to the existence of any of its components in stipulating how the world should be. The rule, “It ought to
be that Smith not murder Jones,” would be true and would apply to the
world, even if there is not and never will be such a person Smith. Unlike
the ‘ought to do’ deontic operator, the ‘ought to be’ operator ranges over
the entire proposition, not merely predicates.
The deontic operators for ‘ought to do’ and ‘ought to be’ statements
J
may now be applied in Forrester’s paradox with great effect. Let “ ”
stand in for ‘ought to be’ statements, let “O” stand in for ‘ought to do’
9
statements, M is “Smith murders Jones,” and G is “Smith murders Jones
gently.” Modified versions of the three rules, R1, R2, and R3 may be
J
used with the ‘ought to be’ operator, but O should be replaced with
in
J
each of them. The deontic operators in 1 and 2 should be replaced by
because they refer to states of a hypothetical world. The argument now
reads:
1*.
2*.
3*.
4*.
5*.
6*.
7*.
J
J ∼M (premise)
(M ⊃ G) (premise)
M (premise)
J
M
J ⊃ G (2, modified R2)
G (3, 4)
G ⊃ M (assumption)
OM ⊃ OG (6, original R1)
Something like 8 does not follow because 5* and 7* cannot possibly yield
anything like “It is obligatory that Smith murder Jones” without conflating the two senses of ‘ought’ that were clearly distinguished above. A
contradiction does not follow, thus circumventing the paradox.
6
Why two deontic operators?
One may object by arguing that the introduction of two different deontic
operators is too cumbersome for deontic logic. The introduction of the new
deontic operators would result in the development of whole new systems of
deontic logic. That would merely muddy matters already quite obscure in
SDL. As I have mentioned, SDL is not well off now by not distinguishing
between two deontic operators. If we use an operator that stands in for
‘ought’ in a generic way, we run the risk of making matters far worse
by continuing to employ just one operator. Two operators permit us to
circumvent paradoxes, such as Forrester’s paradox, and they may shed
some light on other pressing matters such as conditional obligation.
Finally, it seems possible to argue that the employment of two operators generates two paradoxes rather than just one. If 7 were derived
10
from the modified R1, then it seems like we could derive a contradiction
because the operators would be consistent all the way down. Either the
paradox goes through using the ‘ought to be’ operator or it goes through
using the ‘ought to do’ operator. But, if it goes through with the ‘ought
to be’ operator, then it has nothing whatsoever to do with how we ought
to act. The paradox is impotent at best. If the paradox goes through with
the ‘ought to do’ operator, the function of moral rules in deontic logic is
unclear. Rules would apply to certain activities without governing our
behavior generally. Thus, it is highly unlikely that the paradox obtains
when we apply just one of the deontic operators to Forrester’s paradox.
7
Conclusion
Forrester’s and Sinnott-Armstrong’s proposed solutions to the paradox of
gentle murder have failed. In Forrester’s case, we have no reason to reject
a principle of SDL if in rejecting that principle we must revise all of SDL.
In Sinnott-Armstrong’s case, since ‘gently’ is not an adverb of action, it
does not act like a predicate. So, his Davidsonian analysis of propositions
in Forrester’s paradox does not lead us out of the paradox. I have argued
that two distinct deontic operators are at work in the paradox. Since
‘ought to be’ and ‘ought to do’ have two distinct meanings and behave in
logically different ways, some of the propositions refer to how the world
ought to be and other propositions prescribe how Smith ought to act.
Given their different employments, the contradiction does not follow, thus
circumventing the paradox.
11
References
Åqvist, L.: 1967, Good samaritans, contrary-to-duty imperatives, and
epistemic obligations, Nous 1(4), 361–379.
Castañeda, H.-N.: 1985, Aspectual actions and davidson’s theory of
events, in E. LePore and B. McLaughlin (eds), Aspects and Events,
Blackwell, Oxford, pp. 294–310.
Clark, R.: 1986, Murderers are not obliged to murder; another solution
to forrester’s paradox, Philosophical Papers 15, 51–57.
Davidson, D.: 1967, Logical form of action sentences, in N. Rescher (ed.),
The Logic of Decision and Action, University of Pittsburgh Press,
Pittsburgh, PA, pp. 81–95.
Davidson, D.: 2001, Adverbs of action, Essays on Actions and Events,
Claredon Press, Oxford, pp. 293–304.
Forrester, J.: 1984, Gentle murder, or the adverbial samaritan, Journal
of Philosophy 81, 193–197.
Forrester, J.: 1996, Being Good and Being Logical: Philosophical Groundwork for a New Deontic Logic, M.E. Sharpe, Armonk, N.Y.
Jacquette, D.: 1986, Forrester’s paradox, Dialogue 35, 761–763.
Sinnott-Armstrong, W.: 1985, A solution to forrester’s paradox of gentle
murder, Journal of Philosophy 82, 162–168.
von Wright, G.: 1951, Deontic logic, Mind 60, 1–15.
12