A. Elements of Bratman's Planning Theory

Intentions, like other mental states, can be identified as such by their function in a person's reasoning and behavior. Bratman observes that future-directed intentions are like plans that the agent adopts both to guide her behavior and to coordinate it with others.Footnote ²⁷ Intentions have two characteristic features that allow them to serve those functions. First, and most obviously, A's intention at t⁰ to x at t₁ tends to cause A to x at t₁. Second, A's intention at t₀ to x at t₁ structures A's deliberations and actions between t₀ and t₁. The first feature is perhaps the more familiar one. Exclusive focus on an intention's causal efficacy with respect to the intended act suggests the billiard-ball picture that I describe in the introductory paragraphs. Section III describes how it can also give rise to the illusion that there is no intrinsic difference between background and foreground conditions. Less familiar is the role that future-oriented intentions play in structuring an agent's ongoing deliberation and behavior. This section provides a quick overview of Bratman's theory of this function. I argue in Section III that it is here that the difference between background and foreground conditions plays out.

Practical reasoning is a norm-governed activity. To describe the governing norms is to describe both how rational agents should deliberate and how competent agents do deliberate. That is, the relevant norms not only are the rules of practical reasoning but also depict the reasoning habits of competent agents. Having an intention requires having the capacity to adhere to the relevant norms, which means having the right deliberative habits.

Bratman identifies two norms that govern how A's intention to x should influence A's practical deliberation: means-end coherence and consistency.Footnote ²⁸ For the purposes of the analysis here, I ignore a third constraint, agglomerativity, which requires that an agent's intention to x and her intention to y be such that she could also intend x and y.Footnote ²⁹ The consequences of a conditional intention for agglomerativity are similar to those for consistency.

Means-end coherence requires that if A intends to x, she also eventually adopt intentions to accomplish the necessary means of xing. Stated otherwise, means-end coherence requires, “roughly, that it not be true that one intends E, believes that E requires that one intend means M, and yet not intend M.”Footnote ³⁰ This constraint goes hand in hand with the fact that intentions are typically but partial plans. Jones's intent to expand into the neighboring storefront does not yet decide how the new space will be designed, who will do the construction work, whether her coffee shop will remain open during construction, and so forth. In order to realize that intention, Jones must eventually intend requisite means of expanding. Means-end coherence therefore “provides rational pressure for the addition of further intentions.”Footnote ³¹

The consistency requirement “provides constraints on what further intentions may be introduced,” be they intentions to accomplish the means of xing or other projects.Footnote ³² It requires that “one's intentions, taken together with one's beliefs, fit together into a consistent model of one's future.”Footnote ³³ If Jones intends to expand the coffee shop, she cannot consistently also intend to maintain its interior layout unchanged. Whether two intentions are consistent often depends on the agent's other beliefs. Thus if Jones believes that expanding her business will require hiring more employees, her intention to expand is incompatible with an intention to shrink her workforce. If she believes the expanded business will not require more workers, those intentions are compatible.

These two rational constraints—means-end coherence and consistency—ground Bratman's claim that intentions do not so much provide an agent with first-order reasons to act as frame her ongoing practical deliberation. Settling on an intention to x provides a second-order reason to exclude from one's deliberation options that are inconsistent with xing and creates second-order reasons to adopt further intentions concerning the means of xing. Intentions therefore “provide special kinds of reasons—framework reasons—whose role is to help determine the relevance and admissibility of options. These reasons do not compete with desire-belief reasons, but rather structure the process of weighing such reasons.”Footnote ³⁴

The other necessary piece of Bratman's theory is his technical concept of acceptance. Intentions tie into beliefs or belief-like attitudes at several points. First and perhaps most familiarly, a person's practical reasoning is premised on her beliefs. My reasoning from a desire for a snack to an intent to go to the refrigerator passes through my belief that the refrigerator contains food. Second, having an intention to x requires something like a belief that one is able to x and is closely related to the belief that one will x. Third, the demands of both means-end coherence and consistency depend on the agent's beliefs. I have already mentioned that “consistency” refers not only to logical consistency (A cannot intend both to x and to not-x) but to compossible realizability given the agent's beliefs about the world (Jones's intention to reduce her workforce is inconsistent with an intention to expand her business because she believes expansion will require more employees). Means-end coherence is equally belief-dependent. Jones's intention to hire Nathan to do the construction fills in her intention to expand because she believes that expanding requires renovation of the new space, that she cannot do the work herself, that Nathan can, and so forth.

But belief-type commitments do not function in the same way in our practical and in our theoretical reasoning.Footnote ³⁵ Bratman suggests we reserve the term “belief” for attitudes subject to the requirements of theoretical reasoning and use “acceptance” to denote belief-like attitudes subject only to the requirements of practical reasoning. For the purposes of the argument below, it is enough to observe two differences between belief and acceptance. First, in theoretical reasoning we often assign our beliefs a degree of certainty, though in our practical reasoning we might treat the content of those attitudes as given or simply true. Thus while Jones knows that buildings occasionally catch fire and burn down, in deciding to expand her business into the neighboring storefront she brackets such considerations and assumes the space will be available. There are a number of reasons to accept a proposition in one's practical reasoning that on reflection one would think is subject to some uncertainty. These include simplification, asymmetries in the cost of errors, and the demands of social cooperation.Footnote ³⁶ The net effect is that “practical pressures can make it reasonable of me to accept that p in a certain practical context even if it is not reasonable of me to assign p a probability of 1 in my theoretical reflections.”Footnote ³⁷

Second, whether one accepts a proposition in one's practical reasoning depends on the context. For the purpose of deciding whether to expand her business, Jones reasonably accepts that she will be able to use the space; when thinking about whether to buy insurance, she allows that accidents happen. The practical pressures for accepting a proposition:

Beliefs, in contrast, are not context relative. Theoretical reasoning requires that if I have a belief when reasoning about one matter, I maintain that belief when reasoning about another.

B. Strictly Conditional, Conditioned, and Biconditional Intentions

I turn now to the structure of conditional intentions. It is a familiar fact that the grammar of natural language does not always mirror the logical structure of the propositions we express in it. This is no less true of reports of conditional intentions. Suppose Jones tells the manager of her coffee shop, “I intend to expand into the space next door if I can get a construction loan.” In context, it is reasonable to interpret Jones as saying that she intends to expand if she gets the loan, and that she intends not to expand if she does not. Despite its surface grammar, her statement reports a biconditional intention. She intends that getting a loan shall be a necessary and sufficient condition of expansion. She intends to expand if and only if she gets a loan.

In analyzing A's conditional intention to x, it is useful to distinguish A's attitude toward the conditions under which she shall x from her possible attitudes toward xing as such—whether she prefers xing to not-xing, intends to x, intends to try to x, and so forth. To keep with the example, Jones's intention to expand the coffee shop if and only if she gets a construction loan does not of itself tell us whether she wishes or intends to try to expand. In the context of the above exchange, it seems that she does. But we can imagine other scenarios in which Jones does not want to expand but intends to do so if she gets a construction loan—say, where she has entered into a contract to purchase conditioned on obtaining the loan, and she now wants to be relieved of the obligation.

This section focuses on the different attitudes an agent A might have toward the conditions under which she shall x. My claim is that an agent might intend a state of affairs as a sufficient condition on performing an act (A intends to x if C), as a necessary condition of performing it (A intends to x only if C), or as a necessary and sufficient condition (A intends to x if and only if C). For the moment I leave it open whether these connectives—“if,” “only if,” and “if and only if”—should be read to accord with the logic of material implication. I argue in Section III that the truth-functional reading at least captures conditional intentions' role in organizing an agent's practical reasoning.

I will call intentions whose content includes sufficient conditions “strictly conditional,” intentions that include necessary conditions “conditioned,” and intentions that include necessary and sufficient conditions “biconditional.” This typology is not defined with reference to the distinctions between foreground and background or between elective and enabling conditions.Footnote ³⁹ The different possible characteristics of a condition, however, interact and limit one another. In this section I describe how elective and then enabling conditions might or might not fall under one or another of these logical types. The relation to background and foreground conditions is a topic of Section III.

An elective condition on an intention is not imposed by the world but comes from the agent herself. In some instances an elective condition is a matter of a conscious choice: the agent has decided that the relevant state of affairs shall be a condition of performing the intended act. In others, the condition is adopted without deliberation or reflection, as the result of broader policy-based intentions or simply as a matter of habit. In either case, the condition is not the mere corollary of a belief that one will be able to perform the intended act only if it the condition is satisfied. This is what distinguishes elective from enabling conditions.

An elective condition might be located in the antecedent or the consequent of a conditional intention or in both; that is, elective conditions can appear in strictly conditional, conditioned, or biconditional intentions.

Sometimes the surface grammar of “A intends to x if C” matches the speaker's meaning, and the sentence reports that A intends that C shall be a sufficient but not necessary condition of A's xing. This is commonly the case when C is a reason to do the intended act. Jones might intend, for example, to expand the coffee shop if but not only if a planned movie theater opens down the street. In that case, she has made a decision that the movie theater opening shall be a sufficient condition of expanding but she has not yet decided what to do if the movie theater does not open—if the condition is not satisfied. We can formally represent the content of such an intention by using the “→” connective, stipulating for the moment that it should not be read as a material conditional but as a generic implication operator. Where A intends that C shall be a sufficient but not necessary condition of A's xing, the structure of her intention is:

(2)

Footnote ⁴⁰ I suggest reserving the term “strictly conditional” for intentions that have the form (2).

A strictly conditional intention need not be combined with any other attitude to effect the act in question. A strictly conditional intention to x if C suffices to explain the agent's xing when C pertains. That said, an agent who intends to x if C might have other attitudes that further illuminate her disposition toward xing. Thus A might also intend to x if D, whether or not C pertains. Jones might intend to expand her business if the city council passes a measure to create a business improvement district, whether or not the cinema opens. If the agent has control or influence over whether C pertains, her attitude toward C or her other attitudes toward xing can also be highly informative. In the examples, Jones might mean to do everything in her power to ensure satisfaction of the condition—write a letter encouraging the developers to open the cinema, lobby the city council. In other situations, an agent might prefer not-C or not-x, and therefore intend to try to prevent satisfaction of an elective sufficient condition.

Alternatively, an agent A might intend that C be a necessary but not sufficient condition of her xing; that is, she might intend to x only if elective condition C is satisfied without yet deciding whether she shall x if C. Thus Jones might decide that she will perform her promise to purchase only if her architect determines the space is suitable, though even then she might choose not to buy it. Here the structure of Jones's intention is:

(3)

I will use the term “conditioned” for intentions of form (3). The clearest natural-language description of an elective conditioned intention is often its contrapositive: A intends not to x if not-C (Jones intends not to purchase if the architect does not report the space is suitable). Elective conditioned intentions commonly appear in policy-based intentions, such as an intention never to swim on a full stomach or never to drive drunk, which have the forms: A intends (swim → A has an empty stomach) and A intends (drive → A is sober).Footnote ⁴¹

If we read conditional intentions as operating according to the logic of material implication—and I argue in Section III that we should with respect to their consequences for the agent's deliberations—the equivalence of the contrapositive means that every strictly conditional intention is equivalent to a conditioned intention, and vice versa. A's intention (C → x) is functionally equivalent to her intention (not-x → not-C), and A's intention (x → C) is functionally equivalent to her intention (not-C → not-x). I believe this is correct. A formal analysis of the rational constraints that an agent's intentions impose on her deliberations could do without one or the other category. Separating them out, however, will be useful in analyzing the different ways in which conditions figure into practical reasoning. And the opacity of propositional attitudes means that an agent who reports a conditioned intention might not admit to the corresponding strictly conditional intention, or vice versa.

Unlike strictly conditional intentions, a conditioned intention as such does not tell us the agent's attitude toward the act should the condition be satisfied. A's intention (x → C) does not say what A plans to do if C pertains.Footnote ⁴² This is most obviously the case in policy-based intentions. My intention to swim only on an empty stomach, together with the fact that it has been two hours since I last ate, does not determine whether I intend to swim. But the point also applies to one-shot conditioned intentions. Jones intends that a favorable architect's report shall be only a necessary, not a sufficient, condition of purchasing. Consequently, as distinguished from strictly conditional intentions, A's conditioned intention to x only if C and C's pertaining are never a sufficient explanation of A's subsequently xing. For that we must refer to A's other attitudes, like her intention to x or her intention to try to x.

Often more salient than either strictly conditional or conditioned intentions are elective biconditional intentions. A biconditional intention has the structure:

(4)

So interpreted, Jones's statement, “I intend to expand the coffee shop if I can get a loan,” reports that Jones intends to expand if she obtains a loan and not to expand if she does not. That is, Jones intends that getting a loan shall be a necessary and sufficient condition of expanding. A's elective biconditional intention to x if and only if C is sufficient to explain the fact that A xes when C pertains. Like other conditional intentions, however, an agent's other attitudes toward xing might further illuminate the biconditional intention's place in her practical reasoning.

Enabling conditions exhibit less variety in their logical structure. An enabling condition is a condition of the possibility of performing the intended act. By definition, then, enabling conditions are always necessary conditions. It does not follow, however, that they can appear in conditioned intentions. A conditioned intention has form (3) above: A intends (x → C). Substituting an enabling condition for C in (3) produces an ill-formed proposition. What a person intends is within her control and is often a matter of decision. We do not have the ability to control or decide the conditions of the possibility of our acts. Recall Jones's intent to purchase the building only if it is not condemned. Jones does not intend to purchase only if there is no condemnation; the world forces that condition on her. Rather than substituting in (3), we should therefore say:

(5)

A cannot intend that enabling condition C is a necessary condition of her xing, though she can believe or accept that it is.

If an agent has an intention with respect to xing, it will in all likelihood include an intention to x if she can. This might be because A intends to x or intends to try to x.Footnote ⁴³ Thus, absent other conditions or considerations, Jones's intention to try to buy the building, together with her belief that she can do so only if it is not condemned, entails an intent to purchase if it is not condemned. That is, Jones has an attitude of the form:

(6)

In other contexts, an agent might have an attitude of form (6) for reasons other than an intent to x or to try to x. I might intend to attend a faculty meeting if it occurs but hope that it does not happen and even encourage the dean to cancel it. I am not trying to attend the faculty meeting, though I have an intention of form (6). In the analysis that follows, I assume that (6) describes the minimal content of an intention that involves an enabling condition. While not equivalent to a biconditional intention, it is close enough for my purposes. I therefore treat (4), A intends (x ↔ C), as shorthand for (6) as well.

The fact that enabling conditions appear only in something like biconditional intentions marks a significant difference between enabling and elective conditions. An elective condition might be intended as a sufficient, as a necessary, or as a necessary and sufficient condition of acting; that is, an intention subject to an elective condition might be strictly conditional, conditioned, or biconditional. In contrast, enabling conditions, insofar as they figure into the content of an agent's attitudes, are always believed to be necessary and can therefore be intended only as sufficient conditions of acting. Intentions that include enabling conditions in their scope are always biconditional in the sense described above.

A. Consistency

I begin with an obvious case: a foreground biconditional intention, such as Jones's intention to purchase if and only if she obtains a construction loan. On the proposed theory, characterizing the condition as foreground says that Jones accepts neither that she will get a loan nor that she will not get one. This is intuitively plausible. There are two reasons for Jones to mention the condition in conversation or attend to it in her practical reasoning: her belief that she might not get a loan and her decision that if she does not, she will not perform her promise to purchase. That is, Jones is planning for two possible futures, or two scenarios:Footnote ⁴⁸ one in which she obtains a loan and one in which she does not. She intends in the first to perform and in the second not to do so.

The fact that an agent treats the condition in her biconditional intention as an open question has obvious implications for the consistency of her practical attitudes.Footnote ⁴⁹ For example, while Jones's simple intention to expand the coffee shop would exclude a plan to keep the interior unchanged, her foreground biconditional intention does not. She expects that if she does not obtain a loan, she will not expand, in which case she might keep the coffee shop as it is. Her foreground biconditional intention to expand (if and only if she gets a loan) permits her to intend conditionally to keep the layout the same (only if she does not obtain a loan).

This suggests a truth-functional reading of the consistency consequences of foreground biconditional intentions. A's behavior complies with her intent (x ↔_F C) in two possible states of affairs: C pertains and A xes, and C does not pertain and A does not x. Thus while A's unconditional intention to x excludes from A's deliberation plans that entail A not-xing, A's foreground biconditional intention to x if and only if C excludes from her deliberations only plans that entail (a) A xing and not-C, and (b) A not-xing and C. This analysis does not depend on whether the condition is elective or enabling. An intention to purchase the building if and only if the city does not condemn it is of the same structure and has similar consequences for the consistency of Jones's deliberations. (It will be recalled that in these cases, one direction of the biconditional is not an intention but a belief or acceptance regarding the conditions of the possibility of acting.)

The truth-functional reading also captures the consistency constraints of strictly conditional and conditioned intentions. Jones's foreground strictly conditional intention to expand the coffee shop if (but not only if) the movie theater opens excludes from her deliberations plans in which the theater opens and she does not expand but leaves open for her deliberative consideration states of affairs in which (a) it opens and she expands, (b) it does not open and she does not expand, and (c) it does not open and she nonetheless expands. Similarly, Jones's foreground conditioned intention to expand only if (but perhaps not if) her architect confirms the building's suitability excludes from her deliberations only plans that entail expanding against the architect's advice. It permits the consideration of possible futures in which (a) the architect reports that the space is suitable and she expands, (b) he reports that it is suitable and she does not expand, or (c) he reports that it is unsuitable and she does not expand.

My claim here is that a truth-functional reading of the content of foreground-conditional intentions captures the consistency demands they impose. There is an argument that this is too permissive—that this reading does not describe all the ways that conditional intentions constrain an agent's deliberations. The argument turns on situations in which satisfaction of the condition is within the agent's control.Footnote ⁵⁰ Consider a biconditional intention of the form: A intends (C ↔_Fx). On the truth-functional reading of the biconditional, not-C and not-x together establish the truth of “(C ↔_Fx).” If C is within A's control, the truth-functional analysis suggests that A can realize her biconditional intention by preventing the satisfaction of C and then not xing. And this can seem wrong. Suppose Jones intends to purchase if and only if she obtains a construction loan. It seems odd to say that she might realize such an intention by not applying for a loan and then not purchasing. Yet that is what the truth-functional reading entails. Similar intuitions apply to strictly conditional and conditioned intentions.Footnote ⁵¹

These intuitions, however, are context-dependent. Consider the intentions involved in contingency planning. Suppose I fully intend to write a paper for an upcoming conference. Because I recognize that I might not finish it in time, I adopt an intention that if the new paper is not yet finished (an elective condition), I will present an older one. My strictly conditional intention is: Klass intends (I have not finished new paper ↔_F present old paper). In this situation I might do my best to falsify the antecedent without lapsing into irrationality. One can construct similar examples involving enabling conditions. Thus imagine a loving niece who means to do everything in her power to keep her rich uncle alive yet plans that should she inherit from him, she will take the trip to Paris she has always dreamed of. Her biconditional intention with respect to Paris is: Niece intends (uncle dies ↔_F go to Paris). Yet here too we would allow that she might rationally intend to prevent satisfaction of the condition.

My suggestion is that our intuitions in all of these cases are driven not by the conditional intention per se but by the agent's reasons for adopting it, by her other intentions, and by the broader plans or policies in which the conditional intention fits. Consider again Jones's intention to purchase if and only if she obtains a construction loan. If she wants to expand, she might adopt the biconditional intention because she wants to purchase the neighboring storefront and judges that it will not be worth anything to her if she cannot pay to renovate it. But suppose the sales contract contains a contingency clause providing that if Jones cannot get a loan she is not required to purchase, and she wants to get out of the deal because a Starbucks is opening across the street. In these circumstances, Jones might reason that the only way to get out of the contract without incurring liability for breach is to do her best not to obtain a loan.Footnote ⁵² Jones's intent to purchase is still biconditional, but now she is attempting to defeat the condition.

I am arguing here for a thin conception of intentions, according to which intentions are individuated by the consistency, means-end-coherence, and other rational demands they impose on an agent's deliberations, but not by the agent's reasons for adopting them or the broader plans or projects in which they figure.Footnote ⁵³ On this thin conception, Jones's foreground-conditional intention to purchase if and only if she obtains a loan is the same whether she adopts it because she intends to try to expand and believes she needs the loan to do so or because she wants to get out of the sales contract and thinks being denied a loan is her best exit strategy. While the different reasons for adopting the biconditional intention impose different constraints on Jones's deliberations, those constraints are not part of the intention itself.Footnote ⁵⁴

This is not to say that a thicker conception of intentions, according to which Jones's two conditional intentions are not functionally equivalent, is impossible or incoherent. But to individuate intentions with reference to the agent's reasons for holding them would be to rob the concept of much of its analytic value. By distinguishing intentions from those reasons, we identify an attitude that plays a distinctive role in our practical reasoning. Such intentions constrain an agent's deliberation and influence her action in a way that is rational, without requiring reasons to enter into her decisions every step along the way.Footnote ⁵⁵ And even if we chose the thicker conception and decided to individuate intentions in light of agent's reasons for adopting them, we would still want a way of identifying what is common between attitudes that the thinner conception treats as the same; that is, we would still want something like the thin conception of intentions with its truth-functional reading of a foreground condition's consistency consequences. Finally, as becomes clear in Section IV, the legal importance of a promisor's foreground-conditional intent to perform turns in part on cases in which the promisor intends to realize that intention by preventing satisfaction of the condition, as permitted by the truth-functional reading.

The last several paragraphs have taken us fairly deep into the analytic weeds. While the details differ when satisfaction of the condition is under an agent's control, on both the thin and thick conceptions, foreground conditions on an agent's intention have the same sorts of consistency consequences: they relax the consistency demands the intention would otherwise impose, permitting the agent to consider plans that under certain circumstances entail not performing the act in question. With respect to consistency, this is the core difference between a foreground-conditional and an unconditional intention.

What of the consistency implications of background conditions? On the proposed theory, an intention is background-conditional if it is conditional and the agent accepts the condition's satisfaction or its nonsatisfaction. Acceptance differs from belief in that consistency does not require an agent who accepts S in one context to accept S in others. In considering whether to purchase the building next door, it is rational for Jones to accept that it will serve her purposes; in deciding whether to buy fire insurance, it is rational for her to recognize that accidents happen. That said, consistency would seem to require at least that an agent who accepts S in one context not accept not-S in others. It would be irrational for Jones to accept, when deciding whether to buy, that she will be able to expand into the building, and then decide to insure it because she expects it to burn down. It is one thing to recognize, on reflection or in other practical deliberations, that one's practical assumptions might be wrong; it is another to believe or accept in other contexts that they are wrong. The former is rational; the latter, irrational.Footnote ⁵⁶

If this is right, background conditions do not relax consistency constraints in the same way that foreground conditions do. Suppose A has a (strictly conditional) intention to x if C—A intends (C →_Bx)—or a (biconditional) intention to x if and only if C—A intends (C ↔_Bx). In either case, A's acceptance of C excludes from her deliberation any course of action that entails not-xing. Any state of affairs in which A does not x will entail not-C, which ex hypothesi is incompatible with A's acceptance of C. In contrast to similar foreground conditions, the consistency constraints imposed by a background strictly conditional or background biconditional intention to x are no less stringent than those imposed by an unconditional intention to x. Acceptance of the condition has different implications for conditioned intentions. A conditioned intention as such does not yet tell us the agent's attitude with respect to the act upon satisfaction of the condition. If A intends to x only if C—A intends (x →_B C)—her acceptance of C renders that intention compatible with both xing and not-xing. A background conditioned intention by itself imposes no consistency constraints on the agent's further deliberations with respect to the act in question.

This is not to say that background conditions play no role in consistency evaluations. As Ferrero observes, A's background belief that C is a necessary enabling condition of her xing—in my notation, her belief (x →_B C)—partly determines which other intentions are consistent with her intention to x. Suppose Jones believes that she will have the money to purchase the building next door only if the coffee shop remains in business. If Jones accepts that the coffee shop will stay in business and intends to make the purchase, the background condition renders her intent to do so incompatible with a new plan to close the coffee shop. “[T]o determine whether two intentions can be agglomerated, we have to see whether there is any incompatibility in their preconditions. If there is an incompatibility, the pursuit of one plan sooner or later is going to make the pursuit of the other impossible.”Footnote ⁵⁷

Ferrero makes this point for enabling conditions, but it applies equally to elective ones. Thus suppose Jones could pay for the purchase out of her own pocket but has a policy of never using personal funds to pay for business expenses. This policy, together with the above belief, entails an electively conditional intention to buy the building only if the coffee shop remains open. Adherence to that intention would again render selling the coffee shop incompatible with Jones's intent to purchase. The consistency implications of the condition depend not on whether it is enabling or elective but on whether the agent is practically committed to its being a necessary condition of performing the act in question. That practical commitment might be the result of a belief, an acceptance, or an intention.

These observations are of a piece with the fact that the consistency constraint requires not only logical consistency but consistency with the agent's practical commitments as a whole, including her other intentions, beliefs, and acceptances. The consistency of A's intention to x with her other possible intentions depends on A's commitments as to what xing entails. These include both the conditions of the possibility of her xing (the enabling conditions Ferrero emphasizes) and the conditions under which she intends to x (elective conditions), not to mention the natural consequences of xing, whether intended or not.Footnote ⁵⁸

So far I have discussed only background necessary conditions, in which A believes or intends that she shall x only if C and accepts C. The results carry over a fortiori to background necessary and sufficient conditions, where A intends to x if and only if C and accepts that C. Here, too, the condition figures into the consistency demands of A's intention: a commitment to xing precludes courses of action that entail not-C. The consistency implications of background sufficient conditions are subtler. If A accepts C, her intention to x if C would seem to impose the same consistency constraints as an unconditional intention to x would. The background strictly conditional intention does, however, constrain A's reconsideration of xing: she may consistently decide to not-x only if she no longer accepts C or gives up her strictly conditional intention to x if C.

With respect to the consistency of an agent's deliberations, then, background and foreground conditions have something like the opposite effects. A foreground condition relaxes the consistency constraints the intention would otherwise impose, allowing the agent to plan for some (but not all) states of affairs in which she does not perform the act in question. Background conditions, on the contrary, figure into the articulation of consistency constraints, further limiting what other intentions the agent may consider. While foreground conditions permit the agent to entertain a wider range of possible plans, background conditions restrict the plans she may contemplate.

B. Means-End Coherence

I turn now to the consequences of background and foreground conditions for means-end coherence. My thesis is that a foreground necessary or necessary and sufficient condition on an intention reduces the rational pressure to intend requisite means of performing the act, that a foreground sufficient condition has no immediate implications for means-end coherence, and that background conditions generally impose no special pressure to intend requisite means of acting.

Ferrero begins to get at the different means-end consequences of background and foreground conditions when he observes that preparing for not-C might be more or less reasonable depending in part “on the agent's expectation about the obtaining of C”:

Ferrero's first case, in which the agent accepts that C will be satisfied, describes a background-conditional intention. Here Ferrero's observation seems basically correct, though it skips over some details. If A intends C to be a sufficient or a necessary and sufficient condition of her xing and A accepts C—in my notation, if A intends (C →_Bx) or A intends (x ↔_B C)—then means-end coherence requires of A exactly what it would if A intended to x simpliciter. A should invest no more or less thought and energy in realizing the means of xing than if she had an unconditional intent to x. If Jones intends to expand her business if (or if and only if) the movie theater opens and then learns that it will open, means-end coherence requires her to take precisely those steps it would require if she unconditionally intended to expand.

Accepting the satisfaction of a necessary but not sufficient condition has somewhat different means-end consequences. If A intends or believes that she shall x only if C—if A intends or believes (x → C)—then her acceptance of C renders that attitude inert with respect to the requirements of means-end coherence. If I have a policy-based intention to swim only on an empty stomach, my belief or acceptance that I have not eaten in the past two hours does not yet say whether I intend to swim. If Jones intends to purchase the building only if, but not necessarily if, her architect reports that the space is suitable, and the architect's report is positive, Jones's subsequent actions will accord with her conditioned intention whether she purchases or not.Footnote ⁶⁰

Ferrero's second case, in which the agent treats satisfaction of C as an open question, describes a foreground-conditional intention. Here things are a bit more complex.

First, notice that Ferrero's observations have more force when applied to conditioned or biconditional intentions. If A has a foreground strictly conditional intention that C shall be a merely sufficient condition of her xing—A intends (C →_Fx)—her practical belief that C is unlikely to be satisfied might have few consequences for her assessment of whether she will x. Should it turn out that not-C, her strictly conditional intention permits her to choose to x anyway. Thus if Jones intends to expand if but not only if the movie theater opens and then finds out that it is unlikely to open, she might still want to plan in a way that makes it possible to expand, as she has not yet decided whether to do so in those circumstances. If, on the contrary, A's intention is conditioned or biconditional—A intends (x →_F C) or (x ↔_F C)—not-C requires that she not x. If Jones intends to expand if and only if she obtains a construction loan, learning that she is unlikely to be approved for a loan means that she should plan on not expanding. The fact that an intention is foreground conditioned or biconditional has more immediate consequences for means-end coherence than does the fact that it is foreground strictly conditional.

Second, so long as A treats the satisfaction of C as an open question, her foreground-conditional intention will put her under some rational pressure to intend necessary means of xing. A person's attitude toward a proposition that exerts no rational pressure on her to realize it in possible future states of the world might qualify as a wish or a desire but is not an intention.

Third, it does not follow that a foreground-conditional intention to x exerts the same degree of rational pressure as an unconditional intention to x would. If A intends (x →_F C) or if she intends (x ↔_F C) and if she considers it possible that not-C, her investment in preparing to x might well be tied to her estimation of the probability of C.Footnote ⁶¹ The less likely she considers C, the less rationality is likely to require or even permit her to invest in requisite means of xing.Footnote ⁶² Reduced investment is rational because planning resources are finite. Planning for possible futures in which C is satisfied requires time and attention that might otherwise be spent planning for possible futures in which C is not satisfied, not to mention planning the rest of the agent's life or engaging in other worthwhile pursuits.

This is not to say that the agent's assessment of the probability of C alone determines how much it is rational to invest in the means of xing. As Ferrero observes, other relevant facts include the relative costs of preparing for C and not-C, the relative benefits A expects from realizing her intentions conditioned on C and from realizing her intentions conditioned on not-C, and A's attitude toward risk.Footnote ⁶³ That said, a person who has a foreground-conditional intention to x only if C or to x if and only if C should invest less in achieving the means of xing the less likely she considers C to be. A foreground condition on her intention is likely to reduce the “rational pressure for the addition of further intentions.”Footnote ⁶⁴

Finally, “planning and preparing for C” will often include adopting foreground-conditional intentions of the form:

(7)

or:

(8)

These intentions, too, will be subject to pressures of means-end coherence, the force of which will, again, partly depend on the agent's assumptions about the probability of C.Footnote ⁶⁵ Foreground necessary conditions on the agent's intention to x can in this way take her in new directions, creating a rational pressure to adopt intentions that are not immediately related to and might be incompatible with xing.

C. Summary

The main components of the planning theory of conditional intentions can be summarized as follows. In our everyday talk we mark a difference between conditions whose nonoccurrence causes the agent to change her mind and those whose nonoccurrence does not. This intuitive distinction between background and foreground conditions can be explained by the effect of accepting satisfaction or nonsatisfaction of the condition, as distinguished from treating it as an open question, where “accepting” means taking something as given in one's practical deliberations. The fact that C is a foreground condition of A's intention to x has on this theory two important consequences for A's ongoing deliberations and other undertakings. First, while consistency with an unconditional intention to x restricts A's deliberations to plans co-realizable with xing, consistency with her foreground-conditional intention permits A to consider not-xing in the case of not-C and, depending on the whether her intention is strictly conditional, conditioned, or biconditional, perhaps other states of affairs as well. Second, if A's intention is foreground-conditional—and especially if it is foreground conditioned or biconditional—the lower A's evaluation is of the probability of C, the less pressure means-end coherence creates to intend requisite means of xing. A background condition C on A's intention to x, on the contrary, has no such consequences for A's practical deliberations. It will, however, figure into the consistency constraint A's intention imposes on those deliberations.

This account explains both the intuitive plausibility of the awareness theory and where that theory goes wrong. When an agent treats the satisfaction of a condition on her intention as an open question, she is more likely to attend to it. Foreground conditions pose planning problems—what to do if the condition is not satisfied—that conscious deliberation can solve. But an agent might attend to a conditional intention even though she accepts that the condition is satisfied or nonsatisfied. Suppose Jones intends to expand the coffee shop and perform under the sales contract. If asked or if she is in a particularly reflective mood, Jones might allow that she will not perform if the space is condemned or if a Starbucks opens up across the street. She would then be aware of these background conditions on her intention. But that theoretical awareness is unlikely to influence Jones's practical deliberations or actions. Because she accepts their satisfaction, it would be a waste of Jones's limited cognitive resources to plan for either contingency. Such reflective awareness of background conditions belongs to theoretical, not practical reasoning.Footnote ⁶⁶

A. Materiality: Foreground versus Background Conditions

Recall Ayres's and my suggestion that a “promisor's intention to perform is material only because it entails her being likely to perform, only because it provides the promisee crucial information for evaluating [the probability of performance].”Footnote ⁶⁷ On this picture, an undisclosed condition on the promisor's intent to perform is material when it reduces the likelihood of performance to a degree that, if it were known to the promisee, would affect the promisee's decision to enter into or otherwise invest in the transaction. Let P_i stand for the mutually understood probability that the promisor will realize an unconditional intent to do the act promised, and α for the probability that an undisclosed intended necessary and sufficient condition for her performance will be satisfied. While the implicit representation that attaches to A's promise of an intent to perform says that the probability of performance is P_i, in fact the probability is α(P_i). At some point the difference between P_i and α(P_i) will be so great that knowledge of the condition would affect the promisee's decision-making—he would reduce his investment in the deal, take additional precautions against nonperformance, or not enter into the transaction at all. This is the point at which the implicit representation of an unconditional intent to perform materially misrepresents the probability of performance and when an insincere promisor might be held legally liable for promissory fraud.

While this is correct as far as it goes, it is not the whole picture. As I argue in Section III, the above analysis cannot account for our intuitions about the difference between foreground and background conditions. The planning theory can. On the planning theory, background conditions on an agent's intention are those she accepts as satisfied, while foreground conditions are those whose satisfaction she treats as an open question. There are three reasons on this theory to expect a foreground condition on a promisor's intent to perform to be material, though a background condition with the same content would not be: the foreground condition is likely to affect the promisor's preperformance deliberations and behavior in a way that further reduces the likelihood of performance; the fact that the promisor has not accepted satisfaction of the condition for the purposes of her own practical deliberations is evidence that the promisee should also to take the possibility into account; and a foreground condition is more likely to prevent the parties from filling contract gaps in a mutually satisfactory way.

At this point, let me drop Jones's hypothetical coffee shop and use as an example the facts described by the Hawaii Supreme Court in Chung v. Kaonohi Center Co. Footnote ⁶⁸ In January 1972, the Chung plaintiffs signed with the defendants a ten-year lease for a food concession space in a mall that the defendants were building. After the parties signed the lease, the “plaintiffs arranged for financing, ordered equipment and furnishings, hired chefs and workers, advertised in the yellow pages of the telephone book for the to-be-built kitchen, and incurred other expenses.”Footnote ⁶⁹ Though the defendants were aware of the plaintiffs' investments, they continued to offer the space to other potential tenants. Eventually the defendants signed another, presumably more profitable, lease for the space and in June 1973 informed the plaintiffs that it was no longer available.

The Chung defendants' postformation behavior is strong evidence that their intent to perform was foreground conditioned or biconditional—that the defendants intended to perform their promise to lease only if, or if and only if, they did not find another, more profitable tenant. But the defendants' continued negotiations with other tenants were not only evidence of an undisclosed condition on their intent to perform; from the plaintiffs' perspective, those negotiations were also part of the problem with the undisclosed condition. The foreground condition did not decrease the probability of performance by some fixed probability, as “α(P_i)” might suggest. It also affected the defendants' attitudes and actions prior to the time of performance in a way that further reduced the chance they would perform.Footnote ⁷⁰ Most obviously on the facts of the case, the foreground condition relaxed the consistency constraint that an unconditional intent to perform or a background conditioned intent would otherwise impose, permitting the defendants rationally to pursue plans, like leasing the space to someone else, incompatible with performance. Such alternative planning worked to decrease the chances that the condition would be satisfied—that is, it reduced α. Here we see an advantage of the truth-functional reading of conditional intentions. The Chung defendants worked to realize their foreground biconditional intention by preventing satisfaction of the condition.

Chung is perhaps unusual, as the satisfaction of the condition was partly in the defendant's control. But even where this is not the case, or where the defendant has independent reasons to prefer the condition's satisfaction, a foreground necessary condition is still likely to affect her planning and the resources she invests in performing. As I argue in Section III, the fact that a person's intent to x is foreground conditioned or biconditional reduces the rational pressure to adopt necessary means of xing. Recall the Hillcrest case, in which the defendant did not disclose that she intended to purchase a neighboring property for parking only if she found a tenant for it.Footnote ⁷¹ Even if we assume that the defendant made every effort to find a tenant for the property, the undisclosed condition reduced the rational pressure on her to take other steps necessary to purchasing it—to contact the owners, to secure funds for the purchase, to put down a nonrefundable deposit, and so forth. Such decreased investment in performance further reduces the probability of performance—now not by reducing α but by reducing the chances that the promisor will be able to perform if the condition is satisfied, or reducing P_i.

These consequences would not have attached if the Chung or Hillcrest defendants' intent to perform had been background biconditional. If the Chung defendants had accepted that they would not find a more profitable tenant—if they had taken it as given in their practical deliberations that the condition on their intention to perform would be satisfied—consistency would have prevented them from seeking out other tenants, and means-end coherence would have required them to invest in planning for performance as if their intent to perform were unconditional. If the Hillcrest defendant had taken it as given that she would find a tenant for the neighboring property, means-end coherence would have required the same investments in preparing to purchase as an unconditional intent to purchase would have.

A second reason foreground conditions on a promisor's intent to perform are much more likely to be material follows from the fact that the promisor has not accepted the condition in her own deliberations. A promisor's acceptance of a condition C on her intent to perform means that she takes C as given in her practical deliberations, though on reflection or when considering other practical problems, she might recognize the possibility of not-C. An agent's acceptance of a condition is often based on an informal and unreflective cost-benefit analysis. It is rational to accept C for the purposes of one's practical deliberations when, given what one knows, C is so likely true that the risk-adjusted benefits of planning for not-C are not worth its cognitive-resource costs.Footnote ⁷² When considering whether to ride my bike to campus, it is rational for me to assume it will get me there. The chance that it will not is so low that it is not worth the time or effort to consider. But if I know that my chain has been slipping and the derailleur sticking, it might be worth taking that possibility into account. That an agent treats C as an open question suggests a (perhaps reflexive) judgment on her part that the chances of not-C are so high that it is worth the time and effort to plan for that contingency.

Such a judgment by a promisor with respect to a condition on her intent to perform would be of obvious interest to the promisee.Footnote ⁷³ This is again illustrated by the facts in Chung. If the Chung defendants' intention to lease the space to the plaintiffs was foreground-conditional on not finding a better tenant, that was because they considered the probability of finding another tenant high enough that it was worth taking that possibility into account in their practical deliberations. That fact alone—apart from the condition's effect on the defendants' preperformance deliberations and actions—suggests that the plaintiffs would also have wanted to take it into account. Knowledge of the condition would also have affected the plaintiffs' deliberations about how much to invest in the transaction. This logic would not apply if the defendants' intent to lease had been merely background conditioned on there being no better offer for the space. In that case, the defendants themselves would have made a judgment that the probability of a better deal was so low that it was not worth planning for that contingency, suggesting that the plaintiffs need not worry about it either.

A final reason we should expect foreground conditions to be material involves Bratman's concept of shared intentions. My discussion of Bratman's planning theory has so far emphasized the intrapersonal function of future-oriented intentions: intentions organize an individual's deliberations and actions over time, allowing her to settle on and effectively pursue objectives that require temporally extended planning and execution. Intentions also enable complex interpersonal coordination. One way is obvious: B's knowledge that A intends to x often supports a prediction that A will x, which permits B to coordinate his behavior with A's.Footnote ⁷⁴ But that is not all. Bratman's theory of shared intentions extends his planning theory to describe the distinctive attitudes of persons engaged in shared projects, such as painting a house together or singing a duet.Footnote ⁷⁵ The technical details of the theory are relatively complex, and its application to contractual agreements raises issues beyond the central themes of this article. But I hope at least to indicate how the role of shared intentions in many contractual agreements provides a third reason to care especially about foreground conditions on a promisor's intent to perform.

It is common wisdom among contract theorists today that parties' contractual obligations—their legal duties—often do not map onto their reasonable expectations of one another.Footnote ⁷⁶ In Karl Llewellyn's words, there is a “difference in content between the running, flexible obligation understood in fact by the parties and the rigid, stereotyped obligation which is all the law will recognize.”Footnote ⁷⁷ For example, while a commercial lease might impose a variety of detailed obligations on the parties—permitted use of the space, how it will be fitted out, maintenance, security, insurance, and so forth—other aspects of their relationship will be left to future agreement, negotiation, or bargaining. One reason parties can rationally leave such matters undecided is that each typically has evidence that the other is committed to the transaction. To take a trivial example, suppose the Chung lease required the defendant-owners to paint the space but did not specify a color.Footnote ⁷⁸ Suppose further (perhaps contrary to fact) that a court would read the lease as permitting the owners to use any color they chose. The tenants might sign such a lease with a reasonable expectation that the owners would consult with them on the color despite not having a legal obligation to do so. Assuming that the owners intended to go through with the transaction, the owners would also be interested in the success of the tenants' venture, which would have positive spillover effects for the mall as a whole. And because the owners are committed to the transaction, they have a general interest in the tenants' continued participation in it and so in the tenants' satisfaction.

We can generalize from this example and translate it into Bratman's theory of shared intentions as follows. Relational contracts are not simply exchanges of unilateral promises but involve agreements to engage in joint intentional or shared cooperative activities—“shared projects” for short.Footnote ⁷⁹ These can be as simple as having someone fix a bicycle or as complex as a long-term management agreement. As with other intentions, each party's intent to take part in the shared project creates a rational pressure further to intend and undertake the necessary means of realizing that project. But unlike nonshared intentions, because each intends the activity as a shared project, each is further committed to intending means that are consistent with the means intended by the other. As Bratman puts it, they are committed to meshing subplans. When the means of realizing the joint project are not specified in advance (as in a contract), the commitment to engage in the shared project imposes a rational pressure to reach agreement on them. Each party's commitment to the shared project therefore underwrites a prediction that they will be able to fill gaps in their original plan in a way generally satisfactory to both.

While there is more to be said about the details of this picture,Footnote ⁸⁰ this is enough to identify the negative relational consequences of an undisclosed foreground condition on one party's intent to complete the transaction. Just as a conditional intent to perform reduces the rational pressure to intend means of performing and permits the promisor to plan rationally for her breach, so in a relational contract a foreground condition will reduce the pressure to arrive at mutually acceptable approaches to as-yet-unspecified details of the transaction. The Chung defendants' foreground-conditional intent to perform made it all the more likely that they would conform to the letter of the lease (in the imagined case, painting the space) without attending to its spirit (choosing a color that worked with the plaintiffs' planned use). In extreme cases of unilateral, self-interested exercises of contractual discretion, a promisor risks liability for breach of the duty of good faith. Between those extreme cases and collaborative gap-filling lies a range of uncooperative behavior that denies the other party some reasonably expected benefit without breaching the contract. An undisclosed foreground condition (but not a background condition) to perform makes it all the more likely that the promisor's behavior will tend toward the unilateral and uncooperative end of that range, depriving the promisee of benefits she reasonably expects even if the promisor ultimately performs.

In sum, the planning theory of conditional intentions explains as follows the legal relevance of foreground and background necessary conditions on a promisor's intent to perform. Both foreground and background conditions on that intention impact the probability of performance. An intent to x only if, or if and only if, C means that the probability that the agent will x depends in part on the likelihood of C, whether or not the promisor accepts C in her practical deliberations. The promisor's nonacceptance of C, however, has several important consequences. For one thing, it means that she perceives the chances of not-C as high enough to warrant taking that possibility into account in her own practical deliberations. This suggests that the promisee might also want to take into account the possibility of not-C and its impact on the promisor's performance. In addition, the promisor's nonacceptance of a condition on her performance can affect her deliberations and preperformance behavior in ways that harm the promisee. She is more likely to continue planning for possible futures in which she does not perform and less likely to invest in securing the means of performance. And a foreground condition on a promisor's intent to perform reduces the rational pressure on her to fill contract gaps in ways that mesh with the promisee's plans and preferences. None of these consequence attaches to background conditions on a promisor's intent to perform.

B. What a Promise Says

Both Richard Craswell and Aditi Bagchi suggest that on Ayres's and my theory, the law might dispense with promissory representations of intent and focus solely on what a promise says about the probability of performance. Craswell advocates such a rule. He worries that a focus on the promisor's intent “too easily slide[s] into a binary inquiry: Either the defendant intended to perform, or she didn't.”Footnote ⁸¹ Starting from our suggestion that a promisor's intentions are material only because of what they tell the promisee about the probability of performance, Craswell argues:

Bagchi reaches the same conclusion about the implications of our analysis, though she is more critical of them.Footnote ⁸³ On her view, our approach neglects the special moral problems with lying promises and therefore is incomplete as an account of the law of promissory fraud. A concept of promisor intent that is no more than a placeholder for the probability of performance “is a hollow one, and it would empty the doctrine of promissory fraud of much of its moral import—or, it would empty it of all the moral force associated with voluntary commitment and its breach.”Footnote ⁸⁴

I think neither of these criticisms hits the mark, even assuming the relatively impoverished concept of intent that Ayres and I employ. From a strictly consequentialist perspective, there would be reasons for the law to focus on promisor intent even if all that mattered were the probability of performance. The representation of an intent to perform is an integral part of the extralegal practice of entering into an agreement—not because it provides information about the probability of performance, but because an intent to perform is, to use John Searle's phrase, among the normal “felicity conditions” of undertaking a voluntary obligation.Footnote ⁸⁵ That representation of intent, while not as informative or fine-grained as a representation of the probability of performance, is cheap to include, easy to understand, and usually gives the promisee all the information he needs as to the probability of performance. By the same token, framing the question in terms of the promisor's intent serves a valuable heuristic function in an ex post evaluation of whether there was a material misrepresentation. While courts should be sensitive to the fact that intent is not all that is material, it would be wrong to jettison it from the legal inquiry. I would also take issue with Bagchi's characterization of our attitude as purely consequentialist and her assumption that we have an “objection” to moral theories of the law of promissory misrepresentation.Footnote ⁸⁶ Our argument focuses on consequentialist legal arguments because in our view the results of that analysis are the more surprising ones, not because we believe deontological arguments had no place in legal reasoning.Footnote ⁸⁷

That said, Craswell's and Bagchi's readings make a certain amount of sense given the anemic concept of intention we employ in the book. The above, more robust concept suggests deeper, though still consequentialist, reasons why the law should attend to the promisor's intent to perform. Contrary to our assumption in Insincere Promises, promissory representations of intent are material not only because they say something about the probability of performance but because they provide important information about the promisor's likely attitudes and actions in the time leading up to performance. Especially in relational contracts in which the parties rely on nonlegal norms and other incentives, a promissory representation of intent says as much about the value of the transaction as a whole as about the probability of performance. In such transactions, the promisor's lack of an intent to perform can prevent her from acting in ways that are not required by the contract but would benefit the promisee. The richer conception of intent also figures into the moral reasons for holding insincere promisors legally liable that concern Bagchi.

The planning theory also suggests refinements to Ayres's and my analysis of the representational content of contractual promises. As noted above, promissory representations of intent often occur against a shared understanding of conditions under which the promisor might choose not to perform or be prevented from doing so. That “background” understanding should not be confused with the background conditions on the promisor's intent to perform. Background conditions qualify as such by virtue of their role in the promisor's practical reasoning, not the promisee's knowledge of them. And while the parties might have a shared understanding of what those background conditions are likely to be, we should not be too quick to assume that the act of promising says anything about them.

This is so because the implicit content of a speech act depends in part on the objective importance of the information in question.Footnote ⁸⁸ For all the reasons discussed above, the background conditions on a promisor's intent to perform are normally not material. They are relatively unimportant. While the details are complex, this suggests that an unconditional promise represents an unconditional intent to perform, not an intent to perform in certain mutually understood and accepted circumstances.Footnote ⁸⁹ This is so despite the fact that both parties should understand that an intent to perform is not an intent to perform come what may; that is, that the promisor's intent to perform will naturally be background-conditional. Because a foreground condition on the promisor's intent to perform is much more likely to be of importance to the promisee, conversational norms require that a promisor with a foreground-conditional intent qualify the otherwise implicit promissory representation.Footnote ⁹⁰

Another way of putting the above point is to say that a contractual promise does not, as such, represent anything about the promisor's character—about her dispositions to perform or breach in circumstances that while they may be theoretically possible, she does not consider practically relevant. What a promise normally does represent is the promisor's present practical attitude toward performance: that she intends to perform and, by implication, that she presently accepts the satisfaction of any conditions on that intent. The promise represents that the promisor's intent to perform is not foreground-conditional, although, like all intentions, it will be subject to an indefinite number of background conditions.