The distinction between the subjective and objective ‘should’ is often used by those who work on ethics or theories of rationality, i.e., those who are concerned with what we practically should do. It is roughly the distinction between what a subject should do in light of their epistemic perspective and what a subject should do in light of the facts.1 Imagine that you’re in a hotel room and the hotel is on fire, but you’re not aware of it. In some sense of ‘should’, you should jump out of the window (assuming there is no other safe exit). Since the hotel is on fire, this is the best available action for you. In some other sense of ‘should’, you should not jump. Given that you don’t know that the hotel is on fire, it would be irrational of you to simply jump out of the window.2 We seem to have a puzzle: on one hand, you should jump out of the window. On the other hand, you should not. Distinguishing between the subjective and objective ‘should’ solves the puzzle: subjectively, in light of your evidence, you should not jump out of the window.
1See Ross (1939), Prichard (1932), Ewing (1953), Brandt (1963), Jackson (1986), Parfit
(2011), Kolodny and MacFarlane (2010), and Wedgwood (forthcoming).
Objectively, in light of the fact that the hotel is on fire, you should jump out of the window.
Drawing the same distinction in epistemology is less common, but some authors have done so.3 It is the analogous distinction between what one should believe in light of the facts and what one should believe in light of one’s epistemic perspective.4
The facts that the objective sense of the practical ‘should’ is relativized to are often described asall the facts.5 I assume the same for the objective doxastic ‘should’. Which doxastic attitude one objectively should adopt towards a specific proposition P depends only on one fact in the set of all true propositions. If this set containsP, i.e., ifP is true, one objectively should believeP. If this set contains¬P, i.e., ifP is false, one objectively should believe¬P (or disbelieve
P). We can give the following corresponding definition of the objective doxastic ‘should’:
Objective Doxastic Should For any subjectS, proposition P, and timet, S
should objectively believeP attiffP is true.6
Of course, the ‘should’ in this definition is supposed to be doxastic. That is, it does not state that one should, for practical reasons, objectively believe P iff
P. I claim that the objective doxastic ‘should’ occurs in the formulation of the truth-norm:
3See Brandt (1967), Gibbard (2005), and Wedgwood (2007: ch. 5.2). 4
There is a quite widely discussed distinction between subjective and objectivejustification, which is not to be confused with the distinction that I have in mind here. Roughly, one is objectively justified to have a doxastic attitudeDto a propositionP iff one is actually justified to haveD toP; and one is subjectively justified to haveD toP iff one (justifiably) believes that one is objectively justified to haveD toP. (See Goldman 1986: 73; Pollock 1979: 109f.; Alston 1985: 62; Kornblith 1985: 264; Moser 1985: 62; and Kvanvig 1984: 71.) Sometimes, this distinction is also described as a distinction between a subjective and objective ‘should’ in epistemology, where the first expresses what one should believe in light of the actual evidence or reasons one has, and the second expresses what one should believe in light of what one (justifiably) believes one’s evidence or reasons to be (Pollock 1979: 109f.; Kornblith 1985: 264; and Moser 1985: 62). I do not want to deny that there is such a distinction between senses of ‘justification’ or ‘should’. In fact, I think that this distinction can be accommodated by the model for the information-sensitivity of the doxastic ‘should’ that I have suggested in chapter 2. While I am accordingly not opposed to the distinction between subjective and objective justification, it is not the distinction between subjective and objective ‘should’ that I am after here.
5See Kolodny and MacFarlane (2010: 117). Similarly, Wedgwood (forthcoming: sec. 4)
speaks of the objective ‘should’ as relativized to an “omniscient” information-state.
Truth For any subjectS, propositionP, and time t, S should believeP at t iff
P is true.7
The subjective doxastic ‘should’ expresses, roughly, what one should believe in light of one’s evidence. I suggest the following definition:
Subjective Doxastic Should For any subject S, proposition P, doxastic attitude Dand timet, S should subjectively adopt doxastic attitudeD toP att
iff havingDtoP reflects accurately how muchS’s information-state attsupports
P.
I have chosen the technical term ‘information-state’ instead of evidence on purpose. This allowsSubjective Doxastic Should to be open to different accounts of the subjective doxastic ‘should’. In subsection 3.4.1, I will argue that due to this flexibility many epistemic norms in the literature that I call subjective can be interpreted as being formulated in terms of the doxastic subjective ‘should’. A subjective epistemic norm makes what a subject should believe dependent on the subject’s epistemic perspective. A good example is the following norm:
Evidentialism For any subject S, propositionP, doxastic attitudeDand time
t,S should haveD toP attiff DtoP is supported byS’s evidence att.8 It is easy to see that Evidentialismcomes out true on a subjective reading of ‘should’ if we assume that the subject’s ‘information-state’ that is mentioned in
Subjective Doxastic Should is the subject’s evidence.
The subjective/objective distinction can easily be accommodated by the model for the information-sensitivity of the doxastic ‘should’ that I presented in chapter 2:
Doxastic Should For any context of utterance c that provides a doxastic ordering source, ‘S should have doxastic attitudeDto P’ is true at ciff havingD
7
I will discuss the norm in more detail in sections 3.4 and 3.5.
8
See Feldman (2000: 678). For the sake of brevity, I will drop the quantification over propositions, subjects, doxastic attitudes, and times and the reference to times in most of my formulation of norms in the following. Unless otherwise stated, the reader should take them to be implicit.
toP reflects accurately how muchisupports P, whereiis the information-state provided byc.9
The doxastic ‘should’ takes on a subjective sense where the contextually provided information-state i is S’s information-state (whatever that exactly is) at the time t of which the speaker says thatS should have D to P then. It takes on an objective sense wherei is all the facts. As mentioned in subsection 2.5.2, I will call this latter information-state the fully realistic information-state. If we look at the objective doxastic ‘should’ in this way, it is a limiting case of the information-sensitivity of the doxastic ‘should’.10
On this picture, the doxastic ‘should’ is strictly speaking not ambiguous between two senses. Rather, it is akin to an indexical like ‘I’. The sentence “I am hungry” has different truth-conditions in different contexts, i.e., when uttered by different speakers, but we wouldn’t say that ‘I’ has different senses. To put it in Kaplan’s (1989a, 1989b) standard terminology for indexicals, ‘I’ has different
contents in different contexts, i.e., refers to different persons, but only onecharacter. That is, there is a unique rule according to which ‘I’ is mapped in a context to a content. Analogously, ‘should’ has different contents in different contexts and can, for example, mean ‘should in light of the subject’s evidence’ in one context and ‘should in light of the facts’ in another. However, it has one character, i.e., there
is one rule determining which content ‘should’ has in a context.
What unites context-sensitive and ambiguous terms is that both kinds of terms can have multiple meanings. What distinguishes them is that context-sensitivity is characterized “by interaction with (extra-linguistic) context” (Sennet 2016), whereas ambiguity is not. That is, which specific meaning (content, in Kaplan’s terms) a context-sensitive term has is a function of the context in which the term occurs. Strictlyambiguous terms can easily be distinguished from context-sensitive ones since the former have multiple meanings that are unrelated, as in the case of ‘bark’, which is ambiguous between the sound dogs make and the outermost layer of trees.11 It is harder to distinguish context-sensitive and polysemous terms. The
9
For the sake of brevity, I have dropped the quantification over subjects, doxastic attitudes, propositions, and information-states in my formulation of the truth-conditions of doxastic-should sentences here and in the following, except where I deem it helpful to make it explicit.
10As I also mention in subsection 2.5.2, Dowell (2013) construes the objective ‘should’ as
information-insensitive. As I explain there, her position is only superficially at odds with mine.
multiple meanings of polysemous terms are somehow related, for example in the form of a type-token relation, as in the following case:
(1) He left the bank 5 minutes ago. He left the bank 5 years ago.
‘Bank’ in the first sentence refers to a specific building, whereas in the second sentence it refers to an institution that owns this building.
It is clear that the subjective/objective distinction is not a form of strict ambiguity. Furthermore, what speaks in favour of it being context-sensitive rather than polysemous is that we can give a semantic model of the doxastic ‘should’—like the ones discussed in section 2.6—that gives us a unique rule that fixes a contextual parameter, i.e., an information-state, to the subjective and objective meanings of the doxastic ‘should’. Thus, we can give a plausible semantic treatment of the doxastic ‘should’ on which the subjective/objective distinction is the result of interaction with linguistic context.
I could go into a deeper linguistic analysis here to settle the issue whether the subjective/objective distinction is a form of ambiguity or context-sensitivity. I don’t think that this is necessary though. What matters to the discussion in this chapter is that ‘should’ can carry a subjective and objective meaning. Whether this is so in the form of different Kaplanian contents due to context-sensitivity or in the form of different senses due to ambiguity is not important. Furthermore, it is also not decisive for the general project of my thesis. My claim is that the doxastic ‘should’ is relativized to different information-states and that this fact has interesting implications for normative epistemology. Whether the relativizability to information-states is a form of ambiguity or context-sensitivity is, in the light of the purposes of the project, a mere technicality.12