2. Two theses about conditional attitudes

Josephine is considering whether to go to the game tonight. She knows that her favorite team is playing in her city sometime this week, but she can’t remember if it is tonight. Before she can check the schedule, a friend asks if she is going to the game. Josephine explains that she intends to go to the game if her favorite team is in town.

Josephine’s attitude is a conditional intention. It is clearly not an unconditional or simple intention to go to the game. It is less obvious, but still plausible that Josephine’s attitude is not a simple intention of any kind—in particular, not a simple intention towards a conditional content. We can see this by focusing on the situations in which Josephine satisfies her conditional intention. It appears that it is satisfied in cases where her favorite team is in town and she goes to the game, it is frustrated in cases where her favorite team is in town and she doesn’t go, and no question of whether it is satisfied arises in cases where her favorite team is not in town. What is true for intention is true for other sorts of attitudes. The patterns of satisfaction for conditional versions of other attitudes—desires, fears, and beliefs—are similar.Footnote ¹ Even if one disputes these claims for some attitudes, they appear true for many. Regardless, I’m going to state the lesson boldly:

This constrains an account of conditional attitudes.

Attempts to construe Josephine’s conditional intention as a simple intention with a conditional content don’t replicate this pattern of satisfaction, frustration, and the absence of both. Consider the hypothesis that her conditional intention is an intention with a material conditional content—that either she goes to the game or her favorite team is not in town. This can’t be right. She has not carried out her conditional intention in circumstances where her favorite team is not in town, though she would carry out a simple intention in the material conditional in such circumstances (Edgington Reference Edgington and Gobble2001, 409–10; Reference Edgington and Zalta2014; McDaniel and Bradley Reference McDaniel and Bradley2008, 270–71; Ferrero Reference Ferrero2009, 703–5; Ludwig Reference Ludwig2015, 46; and Kleinschmidt (Reference Kleinschmidtn.d.)). Her conditional intention also is not an intention to make the closest possible world in which her favorite team is in town one in which she goes to the game (a simple intention in a Stalnaker [Reference Stalnaker, Harper, Stalnaker and Pearce1981] conditional). Again, she hasn’t carried out her intention in circumstances where her favorite team is not in town—even if she does go to the game in nearby counterfactual possibilities in which her team is in town (McDaniel and Bradley Reference McDaniel and Bradley2008, 270). Similar arguments seem available for other similar accounts of conditional propositions, suggesting that her conditional intention is not a simple intention to make any sort of conditional proposition true.Footnote ²

Perhaps we would make more progress by accepting a theory of conditionals which says that they don’t have propositional contents and their truth does not depend on the possible world of evaluation. An alternative theory of this sort introduces a new parameter, an information state, modelled as a set of possible worlds. According to this theory, a conditional is true if and only if the consequent is true in all worlds of the information state in which the antecedent is true (Yalcin Reference Yalcin2007, 998; Jerzak Reference Jerzak2019, 77).Footnote ³

Can we make sense of conditional attitudes as attitudes toward information-sensitive conditional contents? Let’s think first about belief. Let a person’s belief state be the set of worlds consistent with their beliefs. Yalcin’s (Reference Yalcin2007, 996) account of “believes” does two things: (i) it shifts the information state for evaluation to the subject’s belief state and (ii) it asks whether the complement is true evaluated relative to each world-information state pair for the worlds in that information state. This yields the traditional account for all non-information-sensitive complements. For information-sensitive conditional complements, it implies that for S to believe a conditional is for the consequent to be true in all worlds of S’s belief state in which the antecedent is true. In other words, it is for S to have a belief state which, when updated by the antecedent, accepts the consequent.Footnote ⁴

Unfortunately, this account doesn’t extend in a plausible way to nonbelieflike attitudes. Let’s try something similar for intention, letting “intends” shift the information state parameter to the subject’s intention state—the set of worlds in which their intentions are carried out—and then evaluate the complement relative to each world-information state pair for the worlds in this state. Here’s an example that shows that this doesn’t work. Suppose a candidate intends to win the nomination, but also conditionally intends to vote for their primary opponent if they lose the nomination. The candidate’s intention state will include only worlds where they win the nomination. Since that becomes the information state relative to which the conditional is evaluated, any conditional with the antecedent “the candidate loses the nomination” is trivially true. So, though it is then true that the candidate conditionally intends to vote for their primary opponent if they lose the nomination, it is also true that they conditionally intend to vote for the other party’s nominee (and to stay home, etc.) if they lose the nomination.Footnote ⁵ This problem arises because shifting the information state of evaluation to the subject’s intention state mixes information about the world with how they intend it to be. Insofar as we represent intention states with the same structure as we’ve been representing information states, we seem to encounter this problem.Footnote ⁶

We might, therefore, suggest that “intends” doesn’t shift the information state relative to which we evaluate its complement. Given that the information-sensitive account of conditionals says their truth depends only on the information state, this yields the problematic result that a conditional intention ascription is true in exactly those world-information state pairs in which the unembedded conditional is true. This means that every true conditional is something a subject intends and every conditional intention a subject has is satisfied. Surely, this can’t be right.

There is also a more general worry, which applies regardless of our position about shifting the information state. A straightforward version of the information-sensitive theory of conditionals assigns a sentence a truth-value relative to every possible world-information state pair. So, an attitude with a conditional content is always satisfied or unsatisfied. There are no cases where the question of satisfaction doesn’t arise—contrary to Satisfaction of Conditional Attitudes.

We can avoid this problem by saying that conditionals are neither true nor false but undefined when the antecedent is impossible given the information state (Gillies Reference Gillies2009, 347; see also Yalcin Reference Yalcin2007, 998). According to this view, Josephine’s conditional intention is neither satisfied nor unsatisfied when the information state accepts that her favorite team is not in town. That is the kind of result we want. However, the stipulation of undefinedness, not information-sensitivity, is responsible. De Finetti and Angell (Reference De Finetti and Angell1995), Jeffrey (Reference Jeffrey1963), and Belnap (Reference Belnap1970) secure similar results by giving propositional or truth-functional trivalent semantics with an undefined truth-value.Footnote ⁷ I consider this approach the most promising avenue to vindicate Satisfaction of Conditional Attitudes while maintaining that conditional attitudes are attitudes toward conditional contents. Nonetheless, there are choice-points and problems—which I don’t have room to discuss here—that lead away from such a view (Egré and Cozic Reference Egré, Cozic, Aloni and Dekker2016, sec. 7; Bradley Reference Bradley2002, 369).

Our discussion thus far suggests that conditional attitudes are not ordinary attitudes toward conditional contents—whether those contents are world- or information-sensitive. What we’ve seen for intention seems true for other attitudes as well. This suggests that, for instance, Angie’s desire to present an objection if she thinks of a good one is not a simple, unconditional desire toward a conditional content. Likewise, for Jimmy’s thought that it is likely that a Democrat will win if the economy improves and Christopher’s belief that school will be canceled if the buses can’t get over the hill. A natural reaction is that a conditional attitude is a different sort of attitude than a simple one:

In what follows, I’ll show that natural theses about ascribing conditional attitudes and uncertainty are incompatible with Two Arguments in Conditional Attitudes and Satisfaction of Conditional Attitudes. In building a satisfactory view of the structure of conditional attitudes, I will show how to amend our principles.

3. Conditional attitude ascriptions

In the previous section, I made some preliminary conclusions about the nature of attitudes based on considering Josephine’s case. I did so by reflecting on her state of mind and what it would take to satisfy her attitude in the case described. However, in discussing this case, it’s important to ensure that I haven’t been tacitly ignoring an alternative interpretation, which might undermine the considerations in favor of Two Arguments in Conditional Attitudes and Satisfaction of Conditional Attitudes. To see the alternative that I might have missed, it helps to semantically ascend and focus on the language used to describe Josephine’s case. In particular, we should consider:

What does a speaker who utters a sentence like (1) typically do?

We can see the discussion in the previous section as putting forward two possible answers to this question. First, you might think that the logical form of what a person conveys in uttering (1) is roughly the following (where Intend is a two-place function taking an agent and a content and yielding a content and If is a two-place function taking two contents and yielding a content):

Or you might think that the form is rather (where Conditionally Intend is a three-place function that takes an agent and two contents and yields a content):

In section 2, I suggested that a view like (1_sa) is incorrect and that we should instead accept a view like (1_ca). But both take (1) to be used to ascribe an attitude; they simply differ on the attitude ascribed. This ignores a third possibility—that sentences like (1) aren’t used to ascribe attitudes at all. They are merely conditionals whose consequents are sentences that are typically used to ascribe attitudes. As a matter of the scope of the conditional and the attitude verb, it is possible that the form of what is typically expressed by a use of (1) is as follows:

So, in order to be confident that something like Two Arguments in Conditional Attitudes and Satisfaction of Conditional Attitudes are correct, I must explicitly discuss the plausibility of a (1_c)-style reading of (1).

I’ll discuss few authors as foils, though they merely suggest that some uses of sentences like (1) convey contents the form (1_c) (indeed, the surface forms of almost all of the examples they discuss begin with the “if”-clause, which might privilege such a reading). Because of this, it isn’t clear that their actual positions are in tension with mine, which is a more detailed view in the spirit of (1_ca):

In this section, I’ll show that a reading of (1) like (1_c) is not witnessed in general and that Conditional Attitude Ascriptions is a more plausible generalization.

Jerzak’s (Reference Jerzak2019) account of “wants,” which presupposes an information-sensitive view of conditionals, might suggest rejecting Conditional Attitude Ascriptions. Suppose John prefers to buy the type of wine that will most please his dinner comrades, but he doesn’t know which wine that is. It is appropriate to say:

Jerzak secures the truth of these conditionals by (i) making “wants” information-sensitive—what a person wants is what has the highest expected utility given their preferences and the information state and (ii) letting the conditional take wider scope, so that we evaluate the “wants” consequent relative to an information state updated by the antecedent. Move (ii) makes Jerzak’s account yield readings like the (1_c) reading of (1):

So, if John has a strong preference for honoring his comrades wine tastes, then updating the information state by the antecedents of (2) and (3) about John’s comrades preferences makes the corresponding “wants” claims in the consequents—and, so, the conditionals—true.

Jerzak acknowledges that (2) and (3) are similar to advisory “ought” statements. I claim that they are different from ascriptions of mental states that the subject actually has.Footnote ⁹ Suppose I say before a lecture begins:

I am self-ascribing a current conditional desire even though it may not be clear now whether I will think of a good one. I am not merely asserting a conditional with a “wants” ascription in the consequent. That is, (4_ca) looks like a more accurate paraphrase of what I convey with (4) than (4_c) does:

Furthermore, we cannot plausibly extend Jerzak’s picture to other attitudes. When a speaker utters (1), they aren’t communicating a sort of advisory reading. Such a (1_c)-style reading would allow the speaker to utter (1), find out that Josephine’s favorite team is in town, and, so, conclude that Josephine intends to go to the game. But Josephine does not intend to go to the game in such a scenario. The speaker’s reasoning would be inaccessible to Josephine, herself. And Josephine would not evince the key dispositions of the functional role of that intention, like a disposition to intend the means to her intended end (for example, to intend to take the bus to the stadium). Instead, it seems more plausible that in uttering (1), the speaker ascribes to Josephine a state that is useful in her own contingency planning. The speaker ascribes the sort of state that, if Josephine is in that state, lets her plan about what to do on the contingency that her favorite team is in town (Ferrero Reference Ferrero2009; Ludwig Reference Ludwig2015). For instance, that state might lead Josephine to form an ordinary plan to find out if her team is in town and a contingency plan to take the bus to the stadium if she finds out that they are.

Jerzak notes that belief ascriptions don’t work like (2) and (3) either for similar reasons. If there is a true interpretation of (5), it is not (5_c):

Cases where we’re interested in ascribing attitudes that someone holds are different from the advisory uses of “wants,” and it is on these cases that I focus. In these cases, uses of sentences involving conditionals and verbs like “intends,” “believes,” and even “wants” are not merely assertions of conditionals whose consequents include attitude verbs. They are ascriptions of conditional attitudes.

Other authors claim that a wider class of attitude verbs often get wide-scope conditional readings. Drucker’s (Reference Drucker2019) Policy Externalism suggests this for cases involving attitudes like “hate,” “regret,” and “admire” (see also Drucker [Reference DruckerForthcoming]). He explicitly rules out “believe” from having these features. This is because believing a proposition implies being able to use it in practical and theoretical reasoning but beliefs held simply because conditions are unknowingly met can’t be used in this way (Drucker Reference Drucker2019, 268–71). I suggest that similar concerns apply to credence and, when considering practical reasoning, intention. Drucker is ambivalent about desire (alternately, 271 and 280), but I think there is a case to be made that it too plays a role in reasoning and decision making.

Blumberg and Holguín (Reference Blumberg and Holguín2019, 379) suggest wide-scope conditional readings for “virtually all attitude reports,” though they note significant differences across attitudes. I can’t fully address the nuances of their and Drucker’s views and will merely note a few things. First, Conditional Attitude Ascriptions is a pragmatic thesis. It is compatible with a semantics that allows or privileges wide-scope conditional readings. I’ll show one suggestion for how that might go below. Second, insofar as these readings are not forced, they are not in deep conflict with Conditional Attitude Ascriptions. Two authors concerned with ascriptions of conditional intentions (Ferrero Reference Ferrero2009, 701–2; Ludwig Reference Ludwig2015, 32), distinguish between ascribing conditional intentions and asserting or predicting what one will intend if the condition holds. The former involves a current commitment while the latter doesn’t. Conditional Attitude Ascriptions concern the former readings. Finally, if some attitude verbs—and this may be true of Drucker’s examples of “regret” or “admire”—only yield wide-scope conditional readings and don’t pragmatically convey conditional attitudes of the sort I’m interested in, we could restrict Conditional Attitude Ascriptions to sentences involving other attitude verbs, like “intends,” “believes,” and “wants.”

We’d like to explain Conditional Attitude Ascriptions from a combination of the semantics of the terms involved and general pragmatic principles. One option is to give a semantics for these sentences out of which it straightforwardly falls. One way to do so is to give a compositional picture, enriching attitude verbs with argument places for the condition and the content. The antecedent of the conditional supplies the condition. The idea is that a form like (1_ca) is compositionally derivable. And in what appear to be unconditional attitude ascriptions, the condition is a covertly or contextually supplied tautology (this is inspired by Hájek [Reference Hájek2003] on conditional credences and McDaniel and Bradley [Reference McDaniel and Bradley2008] on conditional desires). Developing the details of such a picture in a plausible way is challenging, so this sketch will have to do for now. Alternatively, we could give a noncompositional semantic picture. For instance, Ludwig argues that the way of expressing conditional attitudes described in Conditional Attitude Ascriptions is idiomatic—it doesn’t derive from the compositional account of attitude verbs and conditionals—and “has arisen to fill a semantic gap in the language” (Reference Ludwig2015, 51).

Since Conditional Attitude Ascriptions is a pragmatic thesis, it may be compatible with the semantic pictures of Jerzak, Drucker, or Blumberg and Holguín. According to Blumberg and Holguín (Reference Blumberg and Holguín2019), attitude verbs commonly have restricted readings. For instance, consider (6) uttered by someone who does not know whether it is Chris who is injured:

Blumberg and Holguín explain why (6) seems true but (7) doesn’t, even in cases where Chris did injure himself:

It is because “regret” takes a restricted reading. The speaker regrets that they taught Chris how to play on the assumption that he injured himself horribly on the tennis court, even though the speaker doesn’t regret it independently of this assumption. A natural hypothesis is that in many cases, the antecedent of a conditional provides a restriction for an attitude verb that appears in the consequent (Blumberg and Holguín Reference Blumberg and Holguín2019, 387–88).Footnote ¹⁰ What is important for us is that it is plausible that uses of these sentences with restricted readings also ascribe conditional attitudes, as in Conditional Attitude Ascriptions. Why would we utter conditionals with restricted attitude ascriptions in the consequent? A natural answer is that these restricted attitude readings express something very much like claims about the attitudes that the bearer is disposed to or is committed to having upon learning that the condition is true. So, from this reading, a hearer can infer that the subject has the corresponding conditional attitude.

4. Attitudes about what is likely and probabilistic propositions

Suppose that I say:

A natural first pass at the attitude I ascribe to Fertz says that the type of attitude is belief and the content of the attitude is the proposition that it is likely that it will be sunny tomorrow in Phoenix. However, I will discuss some reasons why this natural first pass has been rejected by many in the literature.Footnote ¹¹

First, it doesn’t seem like there is a notion of probability that can feature in these sorts of thoughts. In the literature on what are called interpretations of probability, none describe the thought I ascribe to Fertz (Christensen Reference Christensen2004, 18–20). His belief is not about some relative frequency of sunny tomorrows to possible tomorrows. Nor is his belief about the objective physical probability or quantum chance of sun. Most plausible is the hypothesis that his belief is about evidential probability. But even this looks problematic. Suppose Fertz knows that his evidence either definitely supports it being sunny tomorrow in Phoenix or supports, to degree 49 percent, it being sunny tomorrow in Phoenix, but suppose that he can’t figure out which. In such a case, he may reasonably think that it is likely that it will be sunny tomorrow in Phoenix. But it looks like Fertz doesn’t believe that the evidential probability of sun tomorrow in Phoenix is merely high; he thinks it is either 100 percent or 49 percent. So, evidential probability seems not to be a constituent of the content of his thought (Konek Reference Konek2016, 514).Footnote ¹²

Second, even if there is a concept of probability that avoids the previous argument, we still can’t make sense of the range of true ascriptions we make with sentences like (8). Price (Reference Price1986, 19) and Frankish (Reference Frankish, Huber and Schmidt-Petri2009) claim that we can truly ascribe a thought that it is likely that it will be sunny tomorrow in Phoenix to a young child that has no concept of probability at all (see also Yalcin’s (Reference Yalcin2007, 997; Reference Yalcin, Egan and Weatherson2011, 308) example of the dog thinking he might get my table scraps). So, some ascriptions of likely-thoughts are not ascriptions of beliefs in probabilistic contents. And, since the thought we ascribe to Fertz is the same sort that we would ascribe to a child, it seems that in uttering (8), we are not ascribing a belief in some probabilistic content.

Finally, there are formal triviality results that can be interpreted to suggest that Fertz’s thought is not a belief with a probabilistic content (Russell and Hawthorne Reference Russell and Hawthorne2016; Schroeder Reference Schroeder2018; Goldstein Reference Goldstein2019). While the results merely show that several plausible principles are incompatible, one conclusion we might draw goes as follows: take for granted a notion of being 60 percent confident in a proposition, p. Then there is no other proposition, q, that a person must be completely or very confident in just in case one is 60 percent confident in p. This might be interpreted to show that, for instance, there is no proposition that it is likely that it will be sunny tomorrow in Phoenix, which can be the object of Fertz’s belief.Footnote ¹³

It may seem that the arguments I’ve given cannot be decisive because there is no way to make sense of (8) other than as an ascription of a belief in a probabilistic proposition. This isn’t so. Here are two rough ways of dividing attitude and content:

In (8_cre) the attitude ascribed is not an ordinary belief. It is a belief-that-it-is-likely, high confidence, or credence. The content is an ordinary one—the proposition that it will be sunny tomorrow in Phoenix.Footnote ¹⁴ The reasons discussed above suggest we should make sense of ascription as (8_cre)—high credence in an ordinary proposition—rather than (8_bel)—belief in a probabilistic proposition.Footnote ¹⁵ More generally:

We have talked about ascriptions like (8). What about when we simply say:

It is natural to think that, usually, when we sincerely utter a declarative sentence, we assert a proposition that is the information that sentence conveys. But the reasons given above suggest that there is no proposition that it is likely that it will be sunny tomorrow in Phoenix:

What, then, are we doing when we utter sentence like (9)? A reasonable answer is that we are not asserting a proposition or describing the world. Rather, as many suggest, inspired by work at as far back as Ramsey (Reference Ramsey and Braithwaite1926), we are expressing our high credence that it will be sunny tomorrow in Phoenix.Footnote ¹⁷

5. Conditional attitudes about what is likely

Consider:

Given Conditional Attitude Ascriptions, (12) is used to ascribe a conditional version of the attitude ascribed using (11), with the condition being the proposition that the economy improves. Given Credence Ascriptions, the attitude ascribed to Jimmy by a use of (11) is a high credence in the proposition that a Democrat will win. Putting these conclusions together, we see that in using (12), one ascribes to Jimmy a high conditional credence that a Democrat will win on the condition that the economy improves. More generally, Conditional Attitude Ascriptions and Credence Ascriptions yield:

6. Newfangled Conditional Attitudes

We just saw one way that our theses about ascribing conditional attitudes and ascribing credences interact—in ascribing conditional credences. Another way that they can interact has not been much discussed in the literature. These are cases where “likely” occurs in the antecedent of the conditional.Footnote ¹⁹

Pepe is deliberating about whether to go to the game. She knows that the manager has decided who will pitch but won’t announce his decision publicly. Pepe doesn’t now know how likely it is that Kershaw will pitch, but she is going to watch the manager’s press conference before the game. After that, she expects to have thoughts about how likely it is that Kershaw will pitch, though she doesn’t expect to know whether he will or not. She likes going to games where Kershaw pitches, though she doesn’t care for games where someone else pitches. Before the press conference, a friend asks Pepe if she’s going to the game. She responds:

What attitude does Pepe self-ascribe? Given Conditional Attitude Ascriptions, we should think it is a conditional intention. However, there is something strange about it, since the condition is picked out by:

Given No Likely Propositions, (18) doesn’t express a proposition. So, we have a conditional intention with a condition that isn’t a proposition. That is, we have a counterexample to Two Arguments in Conditional Attitudes. If this is right and propositions are the bearers of truth and falsity, then the satisfaction conditions of the attitude ascribed can’t be those given by Satisfaction of Conditional Attitudes.

Similar examples arise with other types of attitudes. For example, it may be that Lynn does not now want to present her obvious objection, since she doesn’t want to waste the speaker’s time, but she wants to present her obvious objection if it is somewhat likely that the speaker hasn’t heard it. Likewise, Robert does not think that it is likely that a Democrat will win the election, but he thinks it is likely that a Democrat will win the election if it is likely that Elizabeth is nominated. Finally, Len does not believe that school will be canceled but he believes that it will be if it is likely there will be a blizzard. Given Conditional Attitude Ascriptions, these are conditional attitudes, but given No Likely Propositions, the condition of each is not a proposition. So, they also are counterexamples to Two Arguments in Conditional Attitudes and Satisfaction of Conditional Attitudes.

One might worry that in each of these cases, the subject really has a related conditional attitude—one where the condition is the proposition expressed by what comes after “likely.” For instance, Len believes that school will be cancelled if there is a blizzard (with no mention of the likelihood of a blizzard). While there may be cases like this, there are also cases where Len believes that school will be canceled if it is likely that there will be a blizzard, even if in the end there isn’t one. I, at least, have been in this very scenario. What was most important to me as a child was that a blizzard was likely, since this, and not the actual appearance of a blizzard, would ground the decision to cancel school. And this conditional belief would be useful to Len when he does not expect to know soon whether there will be a blizzard, but he does expect to soon have an informed credence about whether there will be (for instance, if tomorrow’s forecast will be broadcast soon).Footnote ²⁰

When we ordinarily think of conditional attitudes, we think of a condition that describes the state of the world, a proposition. Call these old-fashioned conditional attitudes. But, if No Likely Propositions is true, when we talk about it being likely that Kershaw will pitch, we aren’t asserting a proposition. So, some conditions aren’t propositions, and the conditional attitudes discussed in this section are not old-fashioned. Call them newfangled conditional attitudes. As we have seen, Two Arguments in Conditional Attitudes and Satisfaction of Conditional Attitudes don’t apply to newfangled conditional attitudes. In the next section, I will sketch a general picture of conditional attitudes along with alternative principles that account for the newfangled variety.

7. A general picture of conditional attitudes

We will take two steps toward our new picture of conditional attitudes. The first is a subject of debate independent of any probabilistic component. The original statement of Two Arguments in Conditional Attitudes says that the condition of the attitude is a proposition, and the original statement of Satisfaction of Conditional Attitudes says that the question of the satisfaction of the attitude arises only when that proposition is true. But we might think that what matters is not simply whether the antecedent proposition is true but also whether the attitude bearer takes it to be true. As Ferrero says, “if the agent intends to ‘φ if C,’ she is under a rational demand to φ only when she takes C to obtain that is, only when she either believes or accepts that C is the case” (Reference Ferrero2009, 702). And, since we might think that the rational demand is to do what satisfies the attitude, it would follow that the agent’s grasp of the condition is relevant to the satisfaction of a conditional attitude.

We can see why this is attractive by reflecting on the functional role of conditional attitudes. In the case of conditional intentions, this means reflecting on their role in contingency planning (Ludwig Reference Ludwig2015). Suppose that Josephine intends to go to the game if her favorite team is in town. That intention doesn’t aid in planning and practical reasoning even if her favorite team is in town until she comes to have a view on the matter. If she comes to believe that her favorite team is in town, this conditional intention serves the same functional role as an ordinary intention to go to the game. For instance, it is governed by characteristic rational norms on intentions—for example, to intend the means to satisfying the intention and to not intend what she believes to be impossible (Bratman Reference Bratman and Robertson2009; Ferrero Reference Ferrero2009). Similar things seem true of the relationship between other conditional attitudes and their simple versions. Conditional belief plays the role of representing how the world is in a person’s theoretical and practical reasoning, once the believer comes to believe the antecedent; conditional desire plays the role of inclining one toward some state of affairs, once the desirer comes to believe the antecedent. In general, the idea is that if one has an old-fashioned conditional attitude and one comes to believe the antecedent, then one’s attitude serves much the same role as an ordinary attitude with the same content.

Furthermore, it doesn’t seem that the truth of the condition of a conditional attitude is enough to, by itself, raise the question of satisfaction of that attitude. There must be some condition about the agent’s grasp. For example, consider a case where Josephine’s favorite team is in town but she can’t figure out whether or not they are and, so, does not go to the game. Do we say that she failed to carry out her conditional intention? Did she fail to do what she intended? The natural answer, I think, is no. She wasn’t in a position to either do or fail to do what she intended because she didn’t grasp that the condition held.Footnote ²¹

On the other hand, the agent’s belief about the condition isn’t all that matters for raising the question of satisfaction. It seems that the condition must also be true. For instance, suppose that Josephine has a conditional intention to buy tickets to the game if her favorite team is playing. Further, suppose that she thinks her favorite team is playing and she buys tickets to the game but, in fact, they are not playing. We don’t want to say that she has succeeded in carrying out her conditional intention in such a case. After all, her team isn’t even playing. But a purely doxastic condition risks saying just this.Footnote ²² So it looks like both whether the condition holds and the agent’s grasp of it matter for whether the question of the satisfaction of a conditional attitude arises. A natural hypothesis, then, is that the question of satisfaction arises when the agent is in the right state of knowledge.

In light of this, we might want to make a change from Two Arguments in Conditional Attitudes and Satisfaction of Conditional Attitudes. This change says, roughly, that the condition of a conditional attitude is a state of belief and the question of satisfaction arising depends on whether the subject is in that state and that state constitutes knowledge. More precisely:

Remember that a belief state is a set of possible worlds left open by a set of beliefs. Now let a knowledge state be a set of possible worlds left open by a set of pieces of knowledge. In Satisfaction of Conditional Attitudes', I talk about a person’s knowledge state being a member of the condition, which a set of belief states. Given what knowledge states are, this is a more formal way of expressing that the person has the knowledge expressed in the condition. Consider, for example, Josephine’s intention to go to the game if her favorite team is in town. The condition that Josephine’s favorite team is in town corresponds to the set of belief states that include the belief that Josephine’s favorite team is in town (i.e., the set of belief states each of which only includes worlds in which Josephine’s favorite team is in town). The question of whether the conditional attitude is satisfied arises when her knowledge state is a member of that set—i.e., when she knows that her favorite team is in town. One advantage of talking in terms of belief and knowledge states rather than individual beliefs and knowledge is that it allows us to make sense of conditions that a person might meet in more specific ways. For instance, suppose Josephine has an intention to go to dinner if it is at her favorite restaurant or she is the guest of honor. And suppose she comes to know that she is the guest of honor. Then the question of whether her intention is satisfied arises regardless of whether or not she comes to have any thoughts about the disjunction that dinner is at her favorite restaurant or she is the guest of honor. This is captured by our representation.

In addition to the reasons presented in the previous pages for moving to Satisfaction of Conditional Attitudes', the ease with which it allows us to generalize to the probabilistic case is a mark in its favor. There needn’t be anything special about a belief that constitutes knowledge in the picture sketched in the first step. Another similar attitude could serve the role. Suppose that in addition to beliefs constituting knowledge, it is possible for credences to constitute knowledge (Moss Reference Moss2013; Konek Reference Konek2016). Then, a person could have a conditional attitude which, instead of having belief as its condition has a credence serving that role. And the question of satisfaction of this conditional attitude would arise in cases where the bearer of the attitude had that credence and it constituted knowledge. This would make sense of our target thought—an intention to go to the game if it is likely that Kershaw will pitch. It is a state of mind that, when the person comes to have a high credence that Kershaw will pitch, has the functional role of an intention to go to the game. And when that high credence that Kershaw will pitch constitutes knowledge, the question of satisfaction of the conditional intention arises. It is satisfied in cases where the bearer goes to the game and fails to be satisfied in cases where they don’t. In general, the functional role of conditional attitudes is largely the same for old-fashioned or newfangled ones, and the satisfaction conditions are similar as well. The major difference is that beliefs play a role in old-fashioned ones while credences play that role in newfangled ones.

I’ve been talking about satisfaction in the newfangled case by talking about credences which constitute knowledge. But traditional pictures think of knowledge as true belief together with some other justificationlike conditions (Plato Reference Levett1992; Gettier Reference Gettier1963; Nozick Reference Nozick1981). And popular views that reject this structure usually still claim that knowing entails believing (Williamson Reference Williamson2000). Realizing that these claims about belief are intended to be about ordinary or qualitative belief, in what sense can I maintain that credences can be the doxastic attitude in an instance of knowledge? Sarah Moss makes a compelling case that credences can constitute knowledge (Reference Moss2013; see also Konek Reference Konek2016). She argues that many of the central hallmarks of beliefs that constitute knowledge can hold of credences as well. She shows that a notion of factivity applies to credences that constitute knowledge. For instance, she argues that knowing that a proposition is likely—having a high credence in that proposition, which constitutes knowledge—entails that that proposition is likely. She also shows how the notions of safety (Williamson Reference Williamson2000; Pritchard Reference Pritchard2005) and sensitivity (Nozick Reference Nozick1981) can apply to credences that constitute knowledge, just as they do in the case of beliefs. I won’t defend the claim that safety and/or sensitivity are necessary for knowledge, but they do apply in many cases. And this is true with both beliefs and credences. For instance, just as, in most cases, knowing some proposition, p, implies that the nearby scenarios where I continue to believe p are p-scenarios, knowing that p is likely implies that the nearby scenarios where I continue to have high credence in p are scenarios where it is still likely (safety). And just as, in most cases, knowing p implies that in nearby scenarios where not-p, I don’t believe p, knowing that p is likely implies that in nearby scenarios where p is not likely, I don’t have high credence in p (sensitivity). Additionally, Moss shows that there are Gettierlike cases (Gettier Reference Gettier1963) involving credences that don’t quite constitute knowledge, just as there are for beliefs. I don’t have space to fully corroborate Moss’s position here, but I think this makes it plausible enough to rely on it in my account of conditional attitudes.

With this in hand we can see a sense in which the conditions of old-fashioned conditional attitudes are similar to those of newfangled ones; both conditions are doxastic attitudes. But there is also a sense in which the conditions of old-fashioned conditional attitudes are different from those of newfangled ones; the former are ordinary beliefs while the latter are credences. Nonetheless, we can represent them jointly using what I’ll stipulatively call a doxastic state. We already know that a person’s belief state can be modeled as the set of possible worlds left open by what they believe. A person’s credences are often modelled by a probability function. But I’m going to take a more general approach, since a person might reasonably be undecided about all sorts of matters that would be required for a probability function to represent them. I’ll call this a partial probability function, which I will suppose is still a function from propositions to real numbers.Footnote ²³ Their doxastic state is then an ordered pair of their belief state and partial probability function.Footnote ²⁴ From this, we can give a stipulative definition of an epistemic state. An epistemic state is a doxastic state for which all of the beliefs and credences that determine the state constitute knowledge.

As we did with a set of belief states, we can use a set of doxastic states to catalogue all of the ways of having a thought. I want to introduce notation for a set of doxastic states, |S|, corresponding to a sentence, S. This is the set doxastic states that a person could have while accepting S. Footnote ²⁵ For instance, consider:

|(19)| only places substantive constraints on the set of worlds of the doxastic state; it is partial probability function invariant.Footnote ²⁶ |(19)| contains all pairs in which the first member is a set that contains only worlds in which the store is open, regardless of the partial probability function of that pair. A person thinks that the store is open when their doxastic state is a member of |(19)|. |(20)| only places substantive constraints on the partial probability function of the doxastic state; it is world invariant. |(20)| contains all pairs in which the second member is a partial probability function that assigns a high value to the proposition that the manager is there, regardless of the set of worlds of that pair. A person thinks that the manager is likely to be there when their doxastic state is a member of |(20)|. |(21)| is a mixed constraint. It contains all pairs in which either the first member is a set that contains only worlds in which the store is open or the second member is a partial probability function that assigns a high value to the proposition that the manager is there. That is, the disjunctive sentence corresponds to the expected constraints on sets of doxastic states—one of the disjuncts must hold. Expected strategies work for negation and conjunction, as well.Footnote ²⁷ So, a person thinks that the store is open or the manager is likely to be there when their doxastic state is a member of |(21)|.

This representation helps us understand conditional attitudes with mixed conditions. Consider the ascription:

As we just saw, (22)’s antecedent, (21), corresponds to a mixed set of doxastic states, |(21)|. A conditional intention to go to the store if it is either open or the manager is likely to be there is an attitude that has the functional role of an intention to go to the store in cases where the bearer’s doxastic state is a member of |(21)|—that is, where the bearer either believes that the store is open or has high credence that the manager is there.Footnote ²⁸ And the question of whether the conditional intention ascribed in (22) is satisfied only arises when the condition is known—the bearer either believes that the store is open and that belief constitutes knowledge or has high credence that the manager is there and that credence constitutes knowledge.

I’ve been talking about conditional attitudes in a somewhat roundabout way—a state that has the functional role of the ordinary attitude given that the bearer is in particular sort of doxastic state and for which satisfaction arises when the bearer is in that doxastic state in a way that constitutes knowledge. I’ve done this because I want to be ecumenical about what the attitude involved fundamentally is. For example, a conditional attitude, x, in p if q (where q may or may not be probabilistic) may be a disposition to form an unconditional x toward p if one’s doxastic state is a member of |q|.Footnote ²⁹ Or it may be a commitment to form an unconditional x toward p if one’s doxastic state is a member of |q|.Footnote ³⁰ Or it may not possess these dispositional or normative connections with simple attitudes, but might simply be an x-like attitude that acts like a simple x toward p if one’s doxastic state is a member of |q|.Footnote ³¹ This last option does not attempt to reduce conditional attitudes to other mental states like dispositions or commitments.

Whichever view we take, we are in a position to record the second, generalizing step in the form of the following principles:

These principles allow for a unified account of old-fashioned and newfangled conditional attitudes.