MICHAEL MORREAU
Prima f acie and Seeming Duties*
Abstract. Sir David Ross introduced prima facie duties, or acts with a tendency to be duties proper. He also spoke of general prima facie principles, which attribute to acts having some feature the tendency to be a duty proper. Like Utilitarians from Mill to Hare, he saw a role for such principles in the epistemology of duty: in the process by means of which, in any given situation, a moral code can help us to find out what we ought to do. After formalizing general prima facie principles as universally quantified conditionals I will show how seeming duties can be detached from them. There will be examples involving ties, burnt offerings and the question of whether to have a napkin on your lap while eating asparagus. They will illustrate the defensibility of this detachment, how it can lead into dilemmas, and how general prima facie principles are overridden by more specific ones.
Key words: prima facie duties, defensible reasoning, practical reasoning, epistemology.
In this Old Testament episode Jephthah find himself in trouble. The text is Judges 11, 30-35. A n d Jephthah vowed a vow unto the LORD, and said, If thou shalt without fail deliver the children of A m m o n into mine hands, Then it shall be that whatsoever cometh forth of the doors of my house to meet me, when I return in peace from the children of Ammon, shall surely be the LORD's, and I will offer it up for a burnt offering. So Jephthah passed over unto the children of A m m o n to fight against them: and the LORD delivered them into his hands. A n d he smote them from Aroer, even till thou come to Minnith, even twenty cities, and unto the plain of the vineyards, with a very great slaughter. Thus the children of A m m o n were subdued before the children of Israel. A n d Jephthah came to Mizpeh unto his house, and behold, his daughter came out to meet him with timbrels and with dances: and she was his only child; beside her he had neither son nor daughter. *I've been lucky to discuss parts of this project with among others Jeff Horty, Paul McNamara, Alasdair MacIntyre, Wlodek Rabinowicz and Michael Siote. Thanks, too, to Henry Prakken and the reviewers for Studia Logica
Studia Logica 57: 47-71, 1996. 9 1996 Kluwer Academic Publishers. Printed in the Netherlands.
48
M. Morreau And it came to pass, when he saw her, that he rent his clothes, and said, Alas my daughter/ thou hast brought me very low, and thou art one of them that trouble me: for I have opened my mouth unto the LORD, and I cannot go back.
No surprise that Jephthah didn't want to keep his word and didn't want to break it, either. One obvious problem was to W H O M this particular promise had been made; the slaughter of the Ammonites must have seemed a clear sign of what could go wrong, were he to default. On the other h a n d the burnt offering was to be Jephthah's own daughter, and his only child at that. These w e r e fors and againsts that J e p h t h a h was aware of; most likely they were the ones which impressed him most. If so then the dilemma which presented itself to him was not a specifically moral one, since these fors and againsts do not involve a conflict of obligations but would have force even if there were no such thing as duties, and people were motivated only to save their skins and to multiply. There was a moral conflict though, whether or not J e p h t h a h was aware of it. One ought to do what he has promised. And it is wrong to kill. These are principles which almost everybody accepts; supposing J e p h t h a h did too, the moral side of his dilemma can be caricatured like this: whether he should have offered up his daughter as a burnt offering depended on how you looked at it. On the one hand, not to go a h e a d with it would be to break his promise. On the other hand, it would kill her. Considering the first of these things it must have seemed to J e p h t h a h that he ought to go ahead with the burnt offering. Considering the second it must have seemed that he ought to do no such thing. Seemed he ought, and seemed he ought not. It is obvious that in a case like this it can seem that something is right, or a duty, and yet at the same time seem that it is wrong~ or forbidden. There are, so to speak, seeming dilemmas. W h a t is not obvious is whether an act can really be both right and wrong. Are there real dilemmas? 1.
P r i m a f acie d u t i e s
Ross thought that there are none. The conflict in cases like this he saw not in there being severM duties which cannot all be done, but instead in the opposition of different "tendencies" of acts. Ross might have said t h a t in virtue of its being the only way to keep a promise, burning his daughter tended to be Jephthah's duty. But in virtue of its being an act of killing someone this act at the same time tended to be forbidden. Act which tend
Prima fade...
49
to be duties in virtue of one or other of their features he called prima facie duties. Ross explained the tendency of an act to be duty in terms of counterfactuM conditionals: I suggest 'prima facie duty' or 'conditional duty' as a brief way of referring to the characteristic (quite distinct from that of being a d u t y proper) which an act has, in virtue of being of a certain kind (e.g. the keeping of a promise), of being an act which would be a duty proper if it were not at the same time of another kind which is morally significant. (The Right and the Good, p. 19) It is u n f o r t u n a t e that prima facie is the term which has stuck, and not conditional duty, since to qualify an act in this way is not to say anything about its epistemic status. A prima facie duty is not in general an act which seems to be a duty; on the contrary, it can happen that one's actual d u t y is to refrain from a prima facie duty, and that this is quite clear to him. Think of a trivial promise, about to be broken in order to prevent some disaster. According to Ross there is and there remains the prima facie duty to keep the promise: this would have been a duty proper, were it not for the disaster. But it will seem to anyone who is sufficiently well aware of the situation that, under the circumstances, duty calls for the promise to be broken. It is possible to model Ross's prima facie duties using techniques familiar from modal conditional logic. Let L be a language for classical first-order logic. Add a unary modal operator O, which will be used to express Ross's "duties proper" and a binary operator >, which will express tendencies. The formulas of the resulting language LO> are defined by the usual clauses, together with these two: If 9~ is formula, then so is O~; If 9~ and ~b are formulas then ~ > ~b is a formula. The operator O is familiar from modal deontic logic; the intended interpretation of O9~ is that 9~ ought to ho]d. Where 90 is a sentence expressing t h a t some act is done, O9~ express that this act ought to be done; it is a d u t y proper. The operator > comes from the modM conditional logic of Lewis, Stalnaker and Thomason, but has a different interpretation: 9~ > ~b means, informally, t h a t in virtue of ~, ~b tends to hold. Using this language, interpreted appropriately in possible-worlds models, it is possible to express Ross's notion of an act which tends to be a d u t y proper, in virtue of one of its characteristics. For example, let R and D be monadic predicates of acts, with R a expressing that a is an act of r e p a y m e n t
50
M. Morreau
and Da that a is done. And let a be a term standing for some particular act of repayment. That a is a prima facie duty can be expressed: Ra & Ra > ODa Informally, c~ is the repayment of some debt; and in virtue of this a tends to be a duty. To m y mind there are serious problems with the notion of prima facie duties and the moral philosophy in which it originated, some of which are discussed in [17]. Ross introduced prima facie duties as apart of his Intuitionist account of the grounds of duty. His view, in outline, was that what makes an act actually right or wrong is the bMance of its divergent tendencies to be right in virtue of some of its characteristics, wrong in virtue of others. The analogy springs to mind of opposing forces acting on a body, whose vector sum is what determines the direction in which it will move. This Intuitionist metaphysics is not taken very seriously by modern moral philosophers, though formal anMyses of prima facie duties along the above lines, suggested by among others [4], go some way towards rehabilitating it. Such work suggests that Intuitionism m a y at least be viable, from a conceptual point of view. W h e t h e r it is or not does not m a t t e r here. I am not concerned with the reality of dilemmas, or with Intuitionist grounds of duty; and I am not offering another formal reconstruction such as that of [4]. Indeed I introduced these things here only so as to contrast them with my own concern. The question which I will try to answer in the rest of this paper is not, what makes the one act right and the other wrong? Rather the question is, how can I know the one act is right and the other wrong? In the answer I will suggest to this latter epistemological question, prima facie duties have no part whatsoever. Utilitarians have suggested there is a part for what can be called general prima facie principles. I agree with them, b u t argue in the next section that these must be understood properly.
2.
Prima facie p r i n c i p l e s a n d t h e e p i s t e m o l o g y
of duty
Unlike prima facie duties, prima facie principles are not acts b u t general claims a b o u t acts, linking the tendency to be a d u t y to some or other characteristic. Ross likened them to laws of nature: W h e n we try to formulate laws of nature, we find that if we are to state them in a universal form which admits of no exceptions, we must state them not as laws of actual operation but as laws of tendency. We cannot say, for instance, that a certMn force impinging on a b o d y of a certain
Prima facie...
51
mass will always cause it to move with a certain velocity in the line of the force; for if the body is acted on by an equal and opposite force, it actually remains at rest. ( . . . ) In the same way if we want to formulate universal moral laws, we can only formulate them as laws of prima facie obligation, laws stating the tendencies of actions to be obligatory in virtue of this characteristic or that. (Foundations of Ethics p. 86) Ross thought that reasoning from general principles like these towards conclusions about what is right is of only minor importance in moral thinking. He thought that the morally mature can mainly just see what is right without resource to prima facie principles, just as those with a training in logic can just see that a proof is valid, without going through the motions of subsuming the particular under the general. However this may be, other accounts of the grounds of duty leave us with the problem of our knowledge of duties. If act utilitarianism is true, for instance, then reasoning "from first principles" about where our obligations lie (by comparing expected utilities, and selecting as our duties those acts which maximize them) is mostly impracticable, because time and information are limited resources. Utilitarians from Mill through to Hare have suggested that this is where "rules of thumb" and prima facie principles come in; their idea is that they can help us to form a reasoned opinion about where our duties lie in cases where we are not in a position to know for sure. There is a problem with this familiar suggestion which has as far I know gone unnoticed. According to Ross, prima facie principles are exceptionless generalizations; they claim that all acts with a given characteristic are prima facie rights or wrongs, as the case may be. Taking the analysis of prima facie duties from the previous section, the logical form of Ross's principles can be brought out as follows. Introducing into LO> a further conditional ::r the principle that debts ought to be repaid is rendered
Vx(Rx =~ (Rx > ODx)). It is the claim that any act of repaying a debt is a prima facie duty; if you are an Intuitionist you can think of prima facie principles in this way. But only Intuitionists believe in prima facie duties; therefore Utilitarians and other nonIntuitionists cannot have Ross's notion in mind, when they speak of prima facie principles. In order to say what they can have in mind instead it helps to distinguish between what I will call the distributed and the undistributed senses of generalizations. An example will make this distinction clear. The generalization that Washington cabs tend to weave in and out of traffic, for instance, is most
52
M. Morreau
naturally understood in a distributed sense, as the claim that each Washington cab taken individually has the tendency to weave in and out of traffic. The generalization that Washington Cabs tend to be Chevrolets by contrast cannot be understood in this way, since it is nonsense to say of any given car that it has the tendency to be a Chevrolet. A car just is a Chevrolet or it is not. This second generalization must be understood as a claim about Washington cabs as a kind and has only what I am calling the undistributed sense. (This distinction is reminiscent of Abelard's distinction between generalizations in sensu diviso and in sensu composito (see for example [9, pp. 64-68]), but it is not the same.) Now Ross m e a n t prima facie principles to be understood distributively, as claims that all acts of a given kind are without exception prima facie rights or wrongs. But they can have an undistributed sense too, and it is only the distributed sense which is unavailable to nonIntuitionists. Thus a Utilitarian can say that acts of lying taken as a group tend to be wrong, while distancing himself from Intuitionist metaphysics by denying that it is the balance of its divergent tendencies which makes an act right or wrong. In exactly the same way, you can say that Washington cabs tend to be Chevrolets while finding very silly the notion that what makes a given car a Chevrolet is the balance of its divergent tendencies to be a Chevrolet, a Ford, a Buick and so on. Thus, whether duties are acts which will bring greatest happiness to be greatest n u m b e r , or acts which accord with a maxim which you could will to be a general law, or acts pronounced duties by some moral oracle; whatever you think, you are free to admit prima facie generalization into your epistemology of d u t y provided you understand them to be undistributed. Concerning the meaning of undistributed prima facie principles there are several possibilities; it is plausible that they express relations between properties of acts, for example. Then to claim that debts ought to be repaid, is to claim that the property of being a d u t y stands in the appropriate relation to the property of being required in order to repay a debt. Here I will not introduce properties into the domain of quantification but will make do with the assumption that however undistributed prima facie principles are represented, they entail conditionMs of a certain kind. SpecificMly, let the conditional ~ > ~ be given an interpretation different from that of the previous section, which was suited to the anMysis of prima facie duties. From now on let it mean, informally speaking, that if ~ then normally r presently > will be interpreted in possible-worlds frames in keeping with this intended meaning. Now letting 0 and D be interpreted as before, but letting R a express that a needs to be done in order to repay some given debt, I suggest that the principle that debts ought to be repMd entails the
P r i m a facie...
53
following thing: that any act is normally a duty which needs to be done in order to repay a debt,
Vx(Rx > ODx). In the rest of this paper I will assume that undistributed prima facie principles can simply be identified with such universMly quantified conditionM obligations. In section 4, 0 will be interpreted in possible-worlds models in a way which is completely standard and > will be interpreted in a way which is quite familiar except for the fact that no modal constraint will be introduced to validate modus ponens across >. Similar proposMs have been a r o u n d for years; the earliest I kI/ow of is in [19]; an elegant technical t r e a t m e n t can be found in [3]. A reviewer complained that if this conditional is as standard as it seems, then it will have some undesirable properties. It is as standard as it seems; I chose it that way so as to get on with the topic of this paper, which is detaching seeming duties from prima facie principles. One who prefers a different analysis of prima facie principles can borrow m y account of moral reasoning, provided his analysis can be given in possible-worlds semantics, in situation theory, or in some related framework in which moral principles are given t r u t h vMues. (To anticipate a little, he can substitute his preferred representation into definition 4, in place of the universally quantified conditional; t h a t is all that needs to be done.) Before going on develop m y account, a few examples will illustrate some general characteristics of reasoning using prima facie principles. In section 5 these and other examples will be anMyzed in detail. 3.
Seeming
duties
Suppose one accepts the prima facie principle that lying is wrong, but at some point considers telling a lie. Suppose further that this lie has no obvious features which the morM code says anything about, apart from its being a lie. All things considered, it seems to him that he ought not to tell the lie. The case of a lie with no other morally significant features m a y seem so simple as to be hardly worth mentioning; in fact it is anything but trivial. One reason for this is that if the relation between moral codes including prima facie principles and seeming duties is some kind of reasoning from the general to the particular, then it is invalid reasoning. This is because the principle that lying is wrong can have "exceptions"; it does not exclude the possibility of lies which are not wrong. Secondly, such reasoning is defeasibIe. In examples like this the function of prima facie principles is to complement more or less firmly held moral
54
M. Morreau
principles and beliefs about the contingencies of the situation at hand, adding to t h e m less firmly held suppositions about what ought to be done. It is in the n a t u r e of this that the principles will never displace firmly held opinions, since it is only where these are lacking that the general principles need be called on. Accordingly, conclusions about seeming duties must be r e t r a c t e d once it turns out that things are not as they seemed to be. A third characteristic of such reasoning is that it is perplexing; one and the same act can seem right and also seem wrong. Cases like J e p h t h a h ' s show that a distinction must be made: in addition to all-things-considered duties we need a some-things-considered notion. For what are J e p h t h a h ' s all-things-considered duties? Since his situation is quite symmetrical we must say, considering the act both as a burning and as a killing, that there is neither an all-things-considered duty to offer his daughter up, nor and all-things considered duty not to. But a seeming dilemma is not a pair of Mternatives, neither of which seems to be a duty. T h e y both seem to be duties. In the analysis given later both are some-things-considered seeming duties. These three characteristics of reasoning about seeming duties are of course not peculiar to moral thinking. They are Mso familiar, to take just two examples, from Gricean pragmatics of natural language and from commonsense reasoning, as studied in the artificial intelligence field of nonmonotonic reasoning. This is not an accident; in all of these cases tentative conclusions are based on generalizations of one kind or another: in moral reasoning the generalizations are prima facie principles; in naturM-language pragmatics they concern the observance of conversational maxims; in nonmonotonic reasoning they are "defaults" of various kinds. Reasoning about seeming duties just is a species of nonmonotonic reasoning. T h a t is why I represented prima facie principles in section 2 not using a special defeasible deontic conditional, but instead using a general purpose defensible conditional together with a standard deontic operator. In the coining sections, seeming duties will be detached from them using a form of nonmonotonic reasoning which has its origins in the notion of "commonsense entailment" of [1]. (An interesting application of these ideas in natural-language pragmatics is [8]). The attraction of embedding moral reasoning into nonmonotonic reasoning is that it becomes clearer how knowledge of things other than morals can inform our moral thinking. In the coming analysis of Jephthah's case, for example, it will be critical t h a t offering someone up as a burnt offering will kill them.
Prima facie... 4.
The interpretation
55 of prima facie principles
In this section LO> will be interpreted in possible- worlds frames. In keeping with the discussion of the previous section I suppose these frames to have the following structure in common:
LO> frame F is a structure ( W , D , o , * / w h e r e : W is a n o n e m p t y set (of possible worlds); D is a n o n e m p t y set (the domain of quantification);
DEFINITION 1. A (i) (ii)
(ii) o C_ W • W is a deontic-alternatives relation;
(iv) , : w • The components W and D are standard: sentences are to be evaluated at possible worlds, and the quantifiers and individual constants range over the things in D. Concerning the acts over which the quantifiers of prima facie principles range it is better to be clear about two things. First, they include hypothetical, not just actual acts. When you consider repaying some particular debt, for example, there seems to be something which you consider doing, and which you might actually do, but which you will perhaps never do. The quantifiers range over such things. Second, it is i m p o r t a n t to be clear that these hypothetical acts are not individual actions but types or kinds of action. This is because in general it is kinds of things which one is obligated to do, not individual actions of those kinds. There are any number of particular things which could be done in order to repay a given debt, for example; some involve handing over cash, others writing a check; some happen today, others tomorrow, and so on. But none of these particular ways of discharging the obligation is itself an obligation, since having repaid a debt in one way but not in another you will have discharged all obligations which stem from it. Obligation attaches only to the totality of different possible ways of repaying the debt; one or other of these had b e t t e r come to pass, but it doesn't m a t t e r which one. Since the acts about which prima facic principles generalize are the sort of thing to which obligation can attach, it is these totalities, or act types, which make up D. The deontic-alternatives relation o, by means of which 0 will be interpreted, is standard. A possible world w relates, through o, to all and only possible worlds where everything holds which, at w, ought to be the case. One familiar constraint which can be placed on o is: OUGHT IMPLIES CAN: For each u E W there is some v E W such that (u, v) E o.
M. Morreau
56
This constraint reflects a substantial view in ethics: t h a t of necessity all duties can be discharged. On some views of the grounds of d u t y this is so, on others it is not. In order to remain neutral on this point it would be better not to place this constraint on the frames; still I will adopt it. (The coming analysis of dilemmas as conflicts between seeming obligations is more interesting if it is assumed that obligations cannot actually conflict; this said, what I say later still holds if this assumption is not made.) T h e only novelty is ,, the worlds-selection function by means of which > will be given its interpretation. Informally, , ( w , p) is the set of worlds where p holds Mong with everything which (relative to w) normally holds if p. The following is now a very basic constraint to place on ,: FACTICITY:
*(W,p) C p.
If p holds the norma~y p holds. Another less immediately obvious constraint is: DISJUNCTION:
*(w,pUq) C_*(w,p) U*(w,q).
A n y t h i n g which is normally the case where p holds, and where q holds, is normally the case where either p or q holds. Later, this constraint will validate the scheme (~2 > ~ ~ r > ~) -+ (c2 V ~p > ~), which m a n y find desirable on intuitive grounds. It will also play a crucial and I find very surprising part in the means by which, in this account, more specific generalizations override less specific ones. Look ahead to example 3. T h e following two observations provide examples of frames satisfying these two constraints: A. OBSERVATION: Given any W, let 9 be identity: for each x C W, 9 (x,p) = p. Then 9 satisfies facticity and disjunction. B. OBSERVATION: Let W contain just two possible worlds, v and w. Let 9 be such t h a t for each x E W, either of the following holds: (i) *(x, {}) = {), ,(x, {v}) = {v}, and , ( x , {v, w)) = , ( x , {w}) = { ~ ) ; or (ii) , ( x , p ) = p, for each p C_ W. T h e n 9 satisfies facticity and disjunction. Now for the models which interpret LO>. DEFINITION 2. A
where
model M for LO>
is a
structure
(W,D,o,.,H),
P r i m a facie...
57
(i) < W, D, o , , > is an LO> frame; and (ii) [[] is a function m a p p i n g nonlogical constants of L to appropriate intensions. T h e t r u t h definition for LO> relative to variable assignments v is familiar: DEFINITION 3. For any model M and any possible world w: (i) M,w I= (p[v] is as usual for atomic formulas ~ of LO> ; (ii) M, w ]= O~[v] if, and only if, for all z such that (w,x) E o,M,x I= ~2[v]; (hi) T h e usual clauses for V, &, ~ , -~, and the quantifiers, together with: M, w ]= ~ > ~b[v] if, and only if, *(w, II~llM,v) c IIr I have chosen the familiar possible-worlds interpretation of O so as to be able to concentrate on the main point, which is the semantics and pragmatics of generalizations about obligations. In fact not much depends on this choice. Other formally specified accounts compatible with my account of prima facie principles are provided by Sven Ove Hansson [5] and Richmond T h o m a s o n [21], to take just two examples. Suppose there is a prima facie duty to refrain from lying; by analogy with the earlier obligation to repay debts this can be expressed
Vx(Lz > O-~Dx). It is worth checking what it takes for this sentence to be true at possible world; for any act-type a it has to be so that if a is an act of lying, then normally c~ ought not to be done. It can be shown using one of the frames provided by observation B that prima facie principles can have exceptions: models can easily be constructed which show that L a and Vx(Lx > O~Dx) are jointly satisfiable not only with O~Dx, but also with -~O-~Dc~. W h a t we have so far is this: an extremely weak conditional language in which to express prima facie principles; it is so weak that modus ponens does not even hold across >. This is as it should be since acts can be exceptions to the rule; but it is clear from this failure of modus ponens that there m u s t be more to be said about prima facie principles, and their relation to particular duties, t h a n can be explained with reference to their truthconditional meanings. How can the detachment of particular duties from general principles be justified, if not by logic? Which particulars can under which conditions be detached from which principles? T h e idea I will develop in the next section in answer to these questions is t h a t the logic of prima facie principles can be strengthened by adding,
M. Morreau
58
to the logical consequences of the moral code of a moral agent and what he knows about the situation in which he finds himself, such assumptions as can be m a d e by supposing exceptions, like the lie which ought to be told, to be as few as possible. 5.
F r o m prima facie p r i n c i p l e s t o s e e m i n g
duties
I begin with some definitions which make this idea precise. First, there is the notion of an exception of some kind, or an individual which is not typical of t h a t kind. DEFINITION 4. Let M be a model and u E WM; also, let 7x be a monadic predicate expression with free variable x. The exceptional 7's (at u in M ) , written Ex~(M, u), are just these: {d: for some ~,M,u [= Vx(7 > () but M,u [= 7d &= -~(d}. The exceptional 7's are the ones which lack some property which 7's normally have. Where there is the previously formalized prima facie obligation not to lie, for example, an act is an exceptional lie if it is a lie, but there is no d u t y not to tell it. This notion generalizes in the obvious way from monadic predicates to arbitrary relations 7. Leading towards the notion of exceptions being at a minimum I now define a relation < on the possible worlds of (possibly different) models. One world stands in this relation to another if it has no exceptions of any kind which the other doesn't have. DEFINITION 5. Let M and N be two models; let u E WM, and v E WN. Define M, u _< N, v just in case for every 7, Ex.y(M, u) C Ex~(N, v). Where M, u _< N, v and also N, v _< M, u, write M, u ~ N, v. To move to the left along _< is to lose an exception of some sort. And to move as far to the left along _< as you can go, while remaining within the class of worlds where some set F of sentences hold, is to move to a place where F is true and exceptions are at a minimum. These worlds, where things are as unexceptional as is consistent with F, are the subject of the next definition. DEFINITION 6. Let F be a set of sentences, v E (in N ) , in symbols N, v I=, F, just in case
WN minimally satisfies F
(i) N, v I= F; and (ii) M, u l = F a n d M ,
u_
Prima facie...
59
Later I will formalize Jephthah's obligations to keep his promise and not to kill. There it will turn out that the worlds which minimally satisfy his moral code and the details of his situation fall into two equivalence classes. In the first the act of burning his daughter is exceptional as a killing goes but not as the keeping of his promise, and it is right; in the second equivalence class things are just the other way around, and this act is wrong. This suggests understanding some-things-considered support in terms of minimal satisfaction as follows: DEFINITION 7. I~ provides some-things-considered support for ~, written F 1=3 ~, if for some N and v such that N, v ~=, F and for all M and u, if M,u~N, vthen M,u]=~. If F provides some-things-considered support for a sentence of form OT, where ~ expresses that an act of some particular kind is done, then ~ is on the basis of F a some-things-considered seeming obligation. As will be clear from the analysis of Jephthah's dilemma in example 2, it can happen that there is such support for two incompatible things: where there is such support for 0 ~ , but also for 0 - ~ (and thus for its logical consequence -~0~) we have a seeming moral dilemma. In this way this notion of support reflects the perplexing nature of reasoning from prima facie principles to seeming duties. The case of dilemmas is a special, if particularly interesting one. But so is the case in which, on the basis of a moral code, it unambiguously seems that some act ought to be done. We will see later that the example about lying is a case where there is all-things-considered support in the following sense: DEFINITION 8. F provides all-things-considered support for ~, written F I=v ~, just in case for every M and u, if M, u I=, F then M, u I= ~. Before looking at worked examples of the defensible detachment of seeming obligations from prima facie principles, it is a good idea to generalize these two notions of support. Both notions of support take all features of acts into account when judging them, since exceptions of all kinds are supposed to be at a minimum. There are however good reasons for thinking that the features which are taken into account are a contextual parameter, which can vary from reasoning situation to situation. For one thing, this is suggested by Joseph Raz's [13] notion of secondary reasons. An exclusionary reason, for example, is a reason not to take some other reason into consideration in practical
M. Morreau
60
reasoning. K the features of acts are treated as a contextual parameter, exclusionary reasons can be thought of as reasons not to take certain features of acts into account. Adding such a parameter Mso makes the analysis of examples technically simpler, so I now generalize the notions of prima facie support and minimal consequence, adding as a parameter a set f of predicate expressions, the moral reasoner's "focus of attention". From now on only those kinds of exceptions will count which the moral reasoner has in focus. There is. just one definition which needs to be generalized: I define M , u _ g , v just in case for every 7 E f, Ex~(M,u) C_ Ex~(N,v). The rest of the definitions can be left exactly as they were, but with the extra parameter f. The following points may give a sense of the way in which ~]v generalizes I=v: where f is empty <1 holds between any pair of possible worlds in any models. Consequently they are all minima, and I=v] collapses into logical consequence. Taking none of the features of acts into account thus amounts to what, disgustingly, has been called a "morM holiday", since it completely blocks the detachment of obligations from prima facie principles. Where on the other hand f is chosen as the set of all predicates of the formal language, l=v] coincides with I=v. With these concepts in place I can now return with a formal treatment of the earlier examples. EXAMPLE 1:
THE SIMPLE LIE
As before, let the prima facie principle that it is wrong to lie be represented,
Vx(Lx > O-~Dx). Now let F consist of this principle together with La, which expresses that is a lie. Then provided L is in f, (1)
F I=v] O-~Da. In particular, r I=v O-~Da.
Moral thinking reaches the conclusion that all things considered, the lie in question ought not to be told.
Demonstration of (1):
First, I define a structure M which has a single possible world u, of which the domain of quantification D contains just the individual constant a, of which the deontic-alternatives relation contains only (u, u), and which has the trivial worlds-selection function: 9 (u,p) = p (for both propositions p of the model). By observation A this specifies a frame. Finally, [] is chosen such that a is interpreted as
Prima fade...
61
itself, and such t h a t La and -~Da are both true at u. M looks like this: Ou
La -~Dc~ It is straightforward to check that M, u I= F. Furthermore, for every 7, Ex.~(M, u) = {}. I show this just for monadic "~ (the general case is just the same). Suppose first M, u ~= 7 a . T h e n clearly a • Ex~(M, u). If however M, u I= 7 a then i]Tall M = {u). It follows from the choice of 9 t h a t u E . ( u , ]ITaNM). From this follows, for any Cx, that if M , u I= Vx(7 > ~) then M, u I= ~a. So in this case too, a r Ex~(M,u). Since D contains only a, I am done. Now, to see t h a t F ~]v O-~Da, suppose N, v I= ], F. Since Ex~(M, u) = {} for every 7, M , u O-~Dx) & La, ExL(N,v) ~ {}. This completes the d e m o n s t r a t i o n of (1). 9 Now suppose it turns out that in fact there is no duty not to tell this lie (or even t h a t it ought to be told). Accordingly, to the premises of the moral reasoner is added ~ O ~ D a (or O D a ) . Now the support for O ~ D a should be withdrawn, and it is:
(2) r,- O- Da V:v s O- Da (and r, ODa V=v s O- Da). T h e following d e m o n s t r a t i o n shows that there is no longer even some-thingsconsidered support for the conclusion that the lie ought not to be told. T h e d e m o n s t r a t i o n covers only the special case where f contains just L, but given the extreme simplicity of the moral code we are looking at here - - it contains just the single prima facie principle - - it is hard to think what else should be added.
Demonstration of (2): It is sufficient to find a model M and possible world v such t h a t M,v I=]~F, ~O~Da (and M,v I=1 F, ODa). W i t h observation B, consider any frame (W, D,o, ,} of which D contains just a; of which o is {(v,v),(w,w)); and such that . ( v , { v } ) = {v) and 9 (v, {v, w}) = {w). Turn this frame into a model, M, by choosing [[] so t h a t a is interpreted as itself, and so that M ends up looking like this. Ov
Dw
La Da
La -Da
M. Morreau
62
It is routine to check that M, v [= F. Also M, v [= -~O--Da, so it is clear from definition 4 that ExL(M, v) = {a}. Since M, v t= F, -~O-,Da (and M,v 1= F, ODa), and since f contains just L, to see that M,v I=]~F, -~O--Da (and M, v I=~ F, O D a ) , it is therefore necessary only to note that there is no N , u ~- F, --O-,Da (no N , u I= F, O D a ) such that ExL(N,u) = {}. But this is obvious from F and definition 4. This completes the demonstration of (2). 9 EXAMPLE 2:
JEPHTHAH'SDILEMMA
Let the principle that promises ought to be kept be formalized by setting aside a binary relation constant K. Intuitively, x bears K to a type of act y if x is a promise which has actually been made, and to keep x it is necessary to do y. This principle becomes
Vxy(Kxy > ODy). As for the duty not to kill, let there be a binary relation S, so that, for any given type y of act and person x, Sxy expresses that doing y would kill x. (S for slay - - K has already been taken.) Now this second obligation can be formalized
Vxy(Sxy > O~Dy). Suppose Jephthah's moral code includes these two principles. Also, let his understanding of the physical world extend to the fact that burning someone tends to kill them, which can be represented
Vxy(Bxy > Sxy). Here Bxy means that doing y would burn up x. Let p, d and a be individual constants (standing for Jephthah's promise, his daughter, and the act of offering her up as a burnt offering). Finally, let P include just the three generalizations above, together with Kpa and Bda: the only way to keep promise p is to do a; and this, were it to be done, would be an act of burning d. Jephthah's dilemma is this: it seems he has the duty, expressed O D a , to offer up his daughter for a burnt offering. And it seems he has the duty, expressed O-~Da, to do nothing of the sort. This analysis reflects the situation, since both of these turn out to be some-things-considered obligations: F I=~ ODa and
r ]-/ --B O-~Da. Here f is the obvious set of predicates to focus on in this case: {B, 1(I S}. Little changes if more predicates are included, but the dilemma depends
Pdma fade...
63
on there being all three. It follows immediately that F offers all-thingsconsidered s u p p o r t for neither alternative. I show just t h a t F 1=]30Da; that F 1=3/ O-,Da can be shown with an analogous proof.
Demonstration:
W h a t is needed is a model M and possible world v such t h a t M , v I=~ F, and such that for any N and u, if N, u ~2 M, v t h e n N, u I= ODa. T h e construction of such M and v is very similar to the d e m o n s t r a t i o n of (2) in Example 1. Using observation B, consider any frame (W, D , o , .) of which D contains just the terms p, d and a; of which o is {(v,v),(w,w)}; and of w h i c h , is such that , ( v , { v , w ) ) = {w). A n d t u r n this frame into a model, M , choosing [[] so that a, d and p are interpreted as themselves, and so that M ends up looking like this. (All other atomic sentences involving B, K or S are put false at b o t h possible worlds.) Ov
Ow
Sda
Sda
B da Kpa
-, B da ~Kpa -.Da
Da
It is routine to check t h a t M, v [= F. And obviously M, v ~ Sda and M, v [= ~O~Da. So E x s ( M , v) = {(d, a ) ) and, arguing as in (1) above, E x K ( M , v ) = ExB(M,v) = {}. In order that M , v [=~1 F, all t h a t is needed is t h a t there is no N, u [= F such that E x s ( N , u) = ExB(N, u) = ExK(N, u) = {). There are no such N and u, however, since as can easily be seen, wherever F holds at least one of these sets m u s t be nonempty. So M, v ]=~ F. Now suppose N , u ~1 M,v. Then ExK(N,u) = E x g ( M , v ) = {}, and so N, u ]= ODa. 9 6.
Normal
in this way, not in that
T h e conditional > was introduced for expressing the moral tendencies of acts, b u t usefully abused in the previous example to express their natural properties as well: burning somebody up tends to kill them. In order to bring out more clearly the obvious role of natural generalizations like this one in moral thinking it is better to embed the theory given up until now into a theory of generalizations, moral and otherwise. This is the benefit to be had from treating moral reasoning as one species of defeasible reasoning a m o n g others: it becomes clear how our moral thinking can be informed by
M. Morreau
64
our expectations about the physical world and others' behavior, or by other beliefs and expectations which are not in themselves a part of morality. One way to do this is to distinguish different regards in which things can be normal. Add a set of constant symbols | to the formal language. A regard r in | stands for an abstract object: a regard in which things can be normal. These constants will be appended to the conditional >, so that we know what kind of normality it is which is being expressed: moral, natural, statistical, whatever. I set aside a constant m which stands for the moral features of acts, and which will be appended to all prima facie principles. T h a t it is wrong to lie will for example now be expressed,
Vx(Lx >m O-~Dx). The fact - - a natural, not a moral principle - - that burning someone up will normally kill t h e m can be expressed,
Vxy(Bxy >~ Sxy). where n stands for the natural properties of acts, and so on. All t h a t has happened here is that the conditional has been duplicated: we now have a different one for each constant in | so the more general language can be interpreted simply by duplicating the selection functions of frames. Technically, to each frame is added a set R of regards in which things can be normal, which are given to 9 as an extra argument. The following generalizes the earlier observation B, providing frames satisfying versions of facticity and disjunction which are relativized to regards: B. OBSERVATION: Let W contain just two possible worlds, v and w. Given any set R of regards, let 9 be such that for each r E R and x E W, either of the following holds:
(i) ,(x, {}, r) = {}, ,(x, {v}, r ) = {v}, and *(x, {v, w}, r) = ,(x, {w}, r) = {w}; or (ii)
* ( x , p , r ) = p,
for each for each p _C W.
In turning frames into models, interpretation functions are given the extra task of mapping the constants in | onto regards in R. The t r u t h conditions are just the same, with due attention paid to the duplication. The clause for > becomes: M, w
I:
~ >r r if, and only if, . ( w , II~lIM,~, ffrD ~ IIr
P r i m a facie...
65
Concerning the notions of all- and some-things-considered consequence, there is just one definition which needs to be generalized. Things can now be exceptional in one regard but not in another; accordingly we define the exceptional 7's in regard of r (at u in M), written Ex~,r(M, u) by appending r to the conditional in definition 4. The order on possible worlds < can then be redefined in the obvious way, and the predicate expressions in focus are now paired with regards. The rest of the definitions can left as they were. It is not surprising that since the earlier models without regard constants are just cross sections through the new ones, the analyses of the lie and the dilemma can be repeated with regard constants added to the conditionals in the way I just indicated. Now another example brings out an advantage of distinguishing between different regards. EXAMPLE 3:
A QUESTION ABOUT TABLE MANNERS
John Horty [7] discussed a problem with a theory of conditional obligation. P u t quite generally it is this: how can a code containing prima facie principles allow for cases in which one general principle is overridden by a more specific principle which states an exception, while another general principle, to which no exception is stated, is not overridden? The following example, which derives from Horty's, will make clear what I mean. It involves two principles of etiquette: when at table, you shouldn't eat with your fingers; and when at table~ you should have a napkin on your lap. Besides these two there is a third principle, more specific t h a n both of them: when eating asparagus you should use your fingers. I show how, where asparagus is served, the third principle can override the one about not using your fingers, while not overriding the principle that a napkin should be in place. In such cases, the simple etiquette modelled below prescribes eating with the fingers. It does not prescribe not eating with the fingers; in this way the first general principle is truly overridden, and not merely contradicted. But it still prescribes a napkin; the principle requiring this is not overridden. This is intuitively correct and it is not explained by other formalisms, including the one discussed by Horty in that paper. The principle at table you ought not to eat with your fingers I formalize with the predicate letter T true of situations in which you are at Table; the letter F true of situations in which you eat with your Fingers; and a regard d - - the one in which a mealtime is normal as regards the use of fingers. The principle becomes
Yx(Tx >d O-~Fx). The second principle is, at table, you ought to have a napkin on your lap. Letting e be another respect - - the one in which mealtimes are normal as
M. Morreau
66
regards the precautions against messes - - and letting the predicate letter N be true of situations in which you have a napkin on your lap, this principle becomes
Vx(Tx >~ ONz). Finally, when served asparagus at table you ought to use your fingers. Letting A express that asparagus is served, and with d the same respect as above, the one in which mealtimes are normal as regards the use of fingers, this principle can be written,
Vx(Tx & Ax >d OFx). Your etiquette includes the above three principles. Now you are at table and are served asparagus. W h a t should you do? Let F consist of the above three principles together with premises Ts and As, expressing t h a t s is a situation in which you are at the table and are served asparagus. P u t
f = {(T,d),(T & d , d ) , ( T , e ) , ( T & A,e)) Now it can be shown that F I=v/ OFs, that F ~:v] O~Fs and that F I=fv ONs. It will be clear from the following demonstration t h a t more pairs can be added to f if this is thought reasonable.
Demonstration:
The proof style is the same as in the simpler examples above. Choose a model M using Observation B with just the two possible worlds u and v; of which the domain of quantification D contains just the term s; of which o is {(u, u), (v, v)}; of which the set of respects is just | and such that for each possible world x we have:
9( x , { ) , d ) = { ) ,
.(x,{u),d)={u),
and
9 ( x , { u , v ) , d ) = .(x{v),d) = {v); and , ( x , p, e) = p, for each for each p C W. The interpretation function is chosen so that s and each r E | interpreted as themselves, and so that M looks like this:
Now two observations:
Ou
Ov
T(s)
T(s)
A(s)
~A(s)
F(s)
F(s)
are
Prima facie...
67
(1) It is straightforward to verify that M, u [= F. And repeating exactly the same a r g u m e n t s as in the previous examples it can be shown t h a t EX(T,d)(M,u) = {s}, while for each other pair ~r e f , Ex~(M,u)= {}. (2) T h e above model has s an exceptional T in respect d. Now it turns out t h a t in every model of F this is so. To see this, note first t h a t the scheme
vx(
>,
& w(w
>,
Vx( >,
is valid. (This is a consequence of the modal constraints of facticity and disjunction.) Further, since by the ought implies can constraint OF(x) ~ ~O~F(x),
Vx(Tx & Ax >d OFx) ]= Vx(Tx & Ax >d -~O-~Fx). Instantiating the above scheme - - ~2 is Tx, 7 is O-Fx, ~b is Ax and r is d, F entails Yx(Tx •d -,Ax). But F entails Ts and As, so that if N,w ]= F, then SN,~ E
EX(T,d)(N , w). Now I can finally turn to what I am supposed to be showing. T h a t F ~IV O-,Fs already follows from the above two observations, since in virtue of t h e m M, u [=1~ F, while obviously M, u ~= O-~F(s). It remains to be shown that F I=IVOFs, and that F ]=I v ONs. So now consider any N, w I=~ F. It is sufficient to show that N, w ~ OFs and t h a t N , w [= ONs. By the second of the above observations SN,~ E EX(T,d)(N ,w), and without loss of generality sy,~ is just the t e r m s itself. Then, by the first of the above observations, M, u _~ ONx) are true at w) and it also follows t h a t N, w I= OFs (since Ts ~z As is true, as is Vx(Tx ~ Ax >d OFx). This completes the analysis of Example 3. I T h e m a n n e r in which in this analysis the more specific principle overrides the moral general one is as far as I know unique within the literature of defeasible or "nonmonotonic" reasoning. (But compare [1].) This behaviour arises from the basic definitions without any need for the "prioritization" of principles or the doctoring of defaults rules, as described and criticized by Horty in [7]. T h e modal constraint disjunction is crucial to this feature of the formalism. It is interesting that it was however not originally adopted in order to produce this overriding behaviour; rather, it was adopted in order
M. Morreau
68
to validate reasoning by cases, which some find intuitive on independent grounds. It is simply not possible to validate reasoning by cases without at the same time producing this overriding behaviour, since (as can be seen from the above demonstration) it is a direct consequence of facility and
disjunction. Horty argued that it is desirable that overriding behaviour is in this way inherent in the formalism: for one thing, it makes learning new moral principles simpler. It this analysis, to learn a new moral principle is simply to add it to your moral code. Where the formulation of moral principles has to be carefully chosen so as to enable more specific rules to override less specific ones, however, new principles cannot be added without reformulating older ones.
This having been said in favor of disjunction and the notion t h a t the more specific overrides the more general principle, there is more to overriding that I have said here. For one thing, in legal reasoning we want to allow for some principles to override others than which they are not more specific. For another, as Henry Prakken pointed out, which principles override which can depend on the context in which the reasoning takes place. Furthermore, discussions with Wlodek Rabinowicz have made me more wary t h a n I was of the modal constraint disjunction. It is hard to think why our conception of normality should be as this constraint requires it to be.
7.
Putting
everything
together
According to this analysis, the obligations which a moral agent seems to have in any given situation can be derived from the prima facie principles which make up his moral code, together with details of the situation in which he finds himself. The reasoning involved, I said, is defensible, and the analysis I have suggested reflects this: as new facts about one's situation become known, or as new principles are added to his moral code, it can happen that he no longer seems to have some of the duties which he previously seemed to have. Now the examples were all rather simplistic and unrealistic. And since the d e t a c h m e n t of obligations is defeasible it is not obvious t h a t they can be m a d e more realistic, by taking more interesting situations and more interesting moral codes into consideration. In example 1, say, I showed that from the principle that it is wrong to lie, and from the fact that to do a given thing would be to lie, follows defensibly that you ought not to do it. But would this still follow if to the premises of this argument were added other principles which we can expect to find in any real moral code, like some of
Prima f a d e . . .
69
those discussed in other examples? The answer is that it would. Technically, it has been shown that Vx(Lx >m O-~Dx), La I=v] O-~Da, provided ( L , m ) is in f . What, now, if to the premises of this argument are added other prima facie principles? Let us give the moral agent a more realistic code, adding other principles from some of the other examples. One, relativized with a suitable regard constant, was
Yxy(Kxy >,~ ODy). This was the
prima facie duty to keep promises. There was also Vxy(Sxy >m O-~Dy).
which expresses the duty not to kill, and
Vxy(Bxy >n Sxy) expressing the natural generalization that burning someone tends to kill them. Now to give him good table manners from example 3 we add
Vx(Tx >d O-~Fx) Vx(Tx >~ ONx) Vx(Tx & Ax >d OFx) These are the principles about not eating with the fingers and wearing napkins. Now the question is, supposing all of these principles are added to the premises of defeasible argument, is there still all-things-considered support for the conclusion O-~Da? The answer is that there is. In fact all of these prima facie principles can be added to each of the moral codes P in each of the examples; the analyses remain intact. The reason is that each proof consists of a demonstration that a particular set of sentences, which includes the prima facie principles of the argument in question, is satisfiable. And each of these demonstrations shows that the set is satisfied by the possible world v of a model based on a frame from observation B. (I did not do this in the case of example 1 since a simpler demonstration was available. But by now it should be clear how to prove the result of example 1 using the more complicated frame.) Now these model constructions were set up so that each can, as it were~ be superimposed on any of the others. (A slight complication is that the frames have different domains of quantification. In superimposition these are simply run together.) The effect is that any or all of the prima facie principles just listed can be added to any set of premises in any of the worked examples.
70
M. Morreau
So the examples are not quite trivial as they might seem. I have shown how a moral agent who knows about lying and promises, who has some simple knowledge of natural science and good table manners, will draw intuitively correct conclusions about what he ought to do in a variety of situations.
References [1] ASHER, N., and M. MOItREAU, 1991, 'Common sense entailment: a modal theory of nonmonotonic reasoning', Proceedings of the 12th IJCAI, Morgan Kaufmann, Palo Alto. [2] BARCAN, M. R., 1980, 'Moral dilemmas and consistency', The Journal of Philosophy 77, 121-136. [3] CHELLAS,B., 1980, Modallogic, an introduction, Cambridge University Press, Cambridge.
[4] CHISHOLM, R., 1978, 'Practical reason and the logic of requirement', In Raz (ed.) Practical reasoning, Oxford Readings in Philosophy, Oxford University Press, Oxford.
[5] HANSSON, S. O., 1990, 'Preference-based deontic logic (PDL)', Journal of Philosophical Logic, 19, 75-93. [6] HARE, R. M., 1981, Moral thinking, its levels, method, andpoint~ Clarendon Press, Oxford. [7] HORTY, J., 1994, 'Moral dilemmas and nonmonotonic logic', Journal of Philosophical Logic. [8] LASCARIDES,A., and N. ASHER, 1993, 'Temporalinterpretation, discourse relations and commonsense entailment', Linguistics and Philosophy, 16, 437-494. [9] LEWIS, D., 1969, Conventions, Harvard University Press, Cambridge,Massachusetts. [10] LEWIS, D., 1973, Counter]actuals, Basil Blackwell, Oxford. [11] MCCARTHY, J., 1980, 'Circumscription - - a form of nonmonotonic reasoning', Artificial Intelligence 13, 27-39. [12] MCCARTHY, J., 1986, 'Applications of circumscription to formahzing commonsense knowledge', Artificial Intelligence 28~ 89-116. [13] RAZ, J., 1990, Practical reason aud norms, 2rid edition, Princeton University Press. [14] REITER, R.,1980, 'A logic for default reasoning', Artificial Intelligence 13~ 81-132. [15] ROSS, D., 1963, Foundations o] ethics, Clarendon Press, Oxford. [16] Ross, D., 1930, The right and the good, Oxford University Press, Oxford. [17] SEARLE, J., 1978, 'Prima facie obhgations', in Raz (ed.) Practical reasoning, Oxford Readings in Philosophy, Oxford University Press, Oxford.
Prima fade...
71
[18] STALNAKER,R., 1968, 'A theory of conditionals', In Rescher (ed.) Studies in logical theory, American Philosophical Quarterly Monograph Series 2, Basil Blackwell, Oxford.
[19] STALNAKER,R., and R. H. T/:IOMASON, 1970, 'A semantic analysis of conditional logic', Theoria 36, 23-42. [20] STALNAKER,R., 1981, 'Indicative conditionals', In Harper, Stalnaker and Pearce (ed.) I]s, D. Reidel Publishing Company, Dordreeht. [21] TI/OMASON, R., 1981, 'Deontic logic as founded on tense logic', In Hilpinen (ed.) New studies in deontic logic, D. Reidel, Dordrecht, 141-152.
PHILOSOPHY DEPARTMENT ~ UMIACS UNIVERSITY OF MARYLAND AT COLLEGE PARK COLLEGE PARK, MD 20742-7615 USA
mimo~umiacs.umd.edu
Studia Logica 57, 1/2 (1996)