The decision theory developed in Decision Theory with a Human Face proposes, as all major decision theories do, that agents should choose in line with their preferences, that is, in the ideal case, choose one of their most preferred options (the “Choice Principle”, p. 25). Moreover—and this is what most of the discussion focuses on—a rational and unbounded agent’s preferences should have a certain kind of structure, whereby they can be thought of as tracking the expectation of a numerical measure of desirability or utility (the “Rationality Hypothesis”, p. 21). Bradley provides a representation theorem that shows that an agent’s preferences can be given such a desirability representation if and only if they abide by certain axioms—some of which Bradley goes on to relax later in the book to make room, e.g., for incomplete preference and imprecision. Preference is a binary relation, and notably, in Bradley’s framework, preference, just like the binary ‘at least as credible as’ relation, can have as its object any ‘prospect’, which could be a state of affairs, an outcome, an action, as well as a conditional possibility. Moreover, in the formal framework proposed, prospects are propositions which form an atomless Boolean algebra. My main concern here is the judgementalist interpretation Bradley gives to his framework. The following subsections will elaborate on each of the core commitments of judgementalism in turn.
Preferences are judgements
That preferences are to be understood as kinds of judgements is a claim Bradley returns to whenever the interpretation of his framework is explicitly discussed. Regarding the question of what these are judgements of, he primarily speaks of judgements of desirability (p. 28), but also of choice-worthiness (p. 47), betterness (p. 47), or benefit. However, in each of the latter three cases, it is made explicit that he has in mind choice-worthiness or value or benefit in light of the agent’s beliefs and desires. And likewise, the intended notion of desirability is presumably desirability of a prospect for the agent as an object of choice. According to judgementalism about preferences, preferences are judgements by an agent that some prospect a is more desirable than another prospect b as a potential object of choice for the agent. Bradley also stresses at various points (e.g. on p. 46) that preferences are meant to be ‘all-things-considered’ judgements that reflect all of an agent’s relevant reasons, agreeing here with Hausman (2012).
Two caveats should be noted. At one point in the book (pp. 46–47), Bradley proposes a ‘hybrid’ account of preference, whereby preferences are not only all-things-considered judgements of desirability, but also need to be instantiated in dispositions to choose. He there notes, “that they are also dispositions to choose both explains the connection between preference and choice and fixes the sense of betterness (namely as choice-worthiness) characteristic of preference judgements.” (p. 47) However, the Choice Principle arguably already explains the connection between preference and choice insofar as agents are rational, and fixes the sense of betterness characteristic of preference judgements—which also distinguishes preferences from other kinds of evaluative judgement. Given that the judgemental element of the hybrid account already removes the main potential advantage of dispositional accounts of preference, namely their parsimony, the addition of a dispositional element in Bradley’s account of preference seems unnecessarily restrictive. I will focus on the purely judgemental interpretation of preference in the following.
Secondly, Bradley defends the judgementalist account of preference primarily when decision theory is applied from the first person perspective, that is, when an agent is deliberating about what to do. While from the first person perspective, preferences are judgements of relative desirability, and, by the second judgementalist claim, utilities are graded judgements of of desirability, Bradley allows that from the third person perspective, preferences and utilities are given a different interpretation. Namely, they could be interpreted as denoting relative desire and degrees of desire respectively, rather than judgements about them (p. 28). He dubs this the ‘mental state interpretation’ of preference, and takes it to be different from judgementalism:
On the judgement interpretation, [the maximisation hypothesis] says that rationality requires of agents that they judge actions to be desirable to the degree that they can be expected to have desirable consequences, given how likely they judge the possible states of the world to be and how desirable they judge possible consequences. Similarly, on the mental state interpretation, the hypothesis says that an agent is rational only if the value she attaches to each action is its expected desirability, relative to her degrees of belief and desire. (pp. 28–29)
The mental state interpretation is more permissive, as we can ascribe desires and beliefs whenever we can identify something that plays the causal role of belief or desire in action explanation. And, to “play this role it is not essential that they be formed as a result of a conscious judgement on the part of the agent.” (p. 28) But as, according to Bradley, the formation of a judgement usually leads to the formation of a corresponding belief and desire, from the third person perspective, we might think of the judgement interpretation as just picking out special cases of the mental state interpretation.
As we will also see more in the following, for Bradley it is the first person perspective that is crucial for normative decision theory as a theory of rationality: Requirements of rationality are meant to be helpful to agents deliberating about what to do. And in the case of first person deliberation, Bradley takes the mental state interpretation to be inadequate. This is because an agent who is deliberating about what to do is not, or at least not only, looking inward to determine what her desires are, but rather is trying to figure out what options are actually valuable or desirable. In doing so, she makes judgements about the value or desirability of options. This value could be objective—Bradley is explicitly not committed to subjectivism about value. (pp. 29–30)
Aside from this distinction from the mental state interpretation, most of Bradley’s defence of judgementalism focuses on its advantages over two more radically different rivals: Interpretations that propose an objective or empirical interpretation of utility as, e.g., well-being or some other value; and choice-theoretic or dispositional accounts of preference and utility. Bradley’s case against objectivist interpretations of utility is one of the key instances where he appeals to what we will call internalism about the requirements of rationality.
On an objectivist interpretation of utility (and probability), standard decision theory would require agents to choose what in fact maximises expected well-being or value. But, according to Bradley, rationality cannot require an agent to do what is in fact expected to maximise well-being or value if she has false beliefs or makes mistaken judgements of (objective) value. For Bradley, rationality concerns only the questions of what it means for an agent to make consistent judgements, and to make the best choices given the judgements she has made. This is clearly compatible with making mistaken value judgements. All we can ask of agents is that they make the best choices given their judgements of credibility and desirability: “We are where we are, with the judgements that we have arrived at, and at the moment when the decision must be made the best that we can do is act consistently on the basis of those judgements.” (p. 27) Bradley’s remarks here exhibit an internalist conception of the requirements of rationality, which takes them to supervene on agents’ mental states. Rationality cannot require different things of two agents who have the same state of mind. More specifically, for Bradley, the requirements of rationality supervene on the agent’s judgements of desirability and credibility.Footnote 2
Bradley’s core defence against the proponent of choice-theoretic interpretations of preference as disposition to choose is that only mental states and judgements, not choices, are “the sorts of things that are susceptible to rationality conditions.” (p. 46) Given the Choice Principle does rationally constrain choice, Bradley must have in mind here the rational consistency conditions of the Rationality Hypothesis. In line with the third defining claim of judgementalism, the idea is that choice is only rationally constrained by preference judgements, and not subject to its own consistency conditions.
Preferences are prior
The second defining claim of judgementalism is that preferences and judgements of relative credibility have conceptual, explanatory and methodological priority over any numerical notions featuring in decision theory, such as utilities, or, in Bradley’s framework, desirabilities and probabilities. They are thus the fundamental relations decision theory is based on. Bradley calls this view ‘pragmatism’ (p. 43). Methodological priority is given to preferences as they typically form the empirical basis for ascription of the quantitative notions. And conceptual priority is given to preference because the properties of rational preference are supposed to explain the properties of the quantitative notions. Against the background of one’s favoured representation theorem, the plausibility of the axioms on preference is supposed to explain the properties of the desirability or utility functions that capture the preferences. This is a natural view to take when, as Bradley does, one takes the rationality conditions on relational attitudes like preference to have more intuitive appeal than the properties of quantitative notions like utility, probability and desirability.
Bradley provides two further considerations in favour of pragmatism. One is that “agents are not typically in a position to make precise numerical cognitive or evaluative judgements. Nor does rationality require that they do so.” (p. 72) If agents are more often in a position to make relational judgements of relative desirability or credibility, this seems to count in favour of pragmatism. Secondly, Bradley argues that models formulated in terms of quantitative notions like desirability and probability will arise as a special case of relational ones: If preferences are complete and consistent, then a quantitative model representing them can be generated. If they are not, there will be no precise quantitative model. Given this latter possibility, relational models also have greater generality. (pp. 72–73)
While Bradley is quite clear about his judgementalism about preference, and about taking preferences to be conceptually and explanatorily prior to the quantitative notion of desirability, I’d like to point out here an important ambiguity regarding how quantitative desirabilities are to be interpreted. Quantitative desirabilities are measures that make it meaningful to compare the size differences between options. Yet, they are meant to be conceptually derivative of binary judgements. Despite the name, on the judgementalist picture, we can’t think of these desirabilities as a measure of actual desirability, that is, of the thing preferences are meant to be judgements of. If they were, then preference, which is a judgement of desirability, would not be conceptually and explanatorily prior to it. Unless, that is, judgements of desirability are what determines desirability. However, as we have just seen, Bradley explicitly rejects such subjectivism about value.Footnote 3
But it is not entirely clear how else we should think of quantitative desirabilities within a judgementalist framework. Elsewhere (Bradley 2004), Bradley suggests that desirabilities track degrees of preference, and that thus preference has more than a binary structure after all. But it is not clear how we should think of degrees of preference when preferences are judgements. Two possibilities are the following: First, we could think of them as a measure of degrees of desirability, as judged by the agent, that is, as expressing the agent’s judgements about degrees of desirability rather than just about orders of desirability. Second, we could think of them as expressing degrees of judgement of desirability. In the first case, it is the desirability (as judged by the agent) that is graded. In the second case, it is the judgement itself that is graded.
An interpretation along the second lines would require further explication, as it is not immediately obvious what it would mean to make a judgement of degree x that some option a is desirable, and whether judgements come in degrees at all. It would be implausible for such degrees to track the confidence with which a judgement is made, since agents might be equally confident in all their evaluative judgements, even if some of them are represented by lower utility assignments. At least an agent can be equally confident in all her relational judgements, i.e. preferences, and then it is unclear again how these relational judgements should be explanatorily and conceptually prior to ascriptions of varying degrees of confidence in one’s evaluative judgements.
The first judgementalist interpretation of utility/desirability is somewhat easier to make sense of, and it seems plausible that relational judgements about orders of desirability should be explanatorily and conceptually prior to judgements of degrees of desirability. But the more fundamental problem with both ways of thinking of desirabilities or utilities as capturing some kind of graded judgement is that ascribing utilities and desirabilities to an agent now means ascribing additional judgements to her, different from the relational judgements we started out with. We can either think of judgements as necessarily conscious, or as potentially unconscious, implicit in other judgements we have made. If they are necessarily conscious, it’s implausible that the desirability and utility assignments yielded by the representation theorems should always represent quantitative judgements. After all, an agent who makes consistent relational judgements can conceivably not make any conscious quantitative judgements at all. In fact, as we have just seen, one of the reasons Bradley endorses pragmatism to begin with is that agents are often in a position to make relational judgements when they can’t make precise numerical judgements, at least not consciously.
This problem would be avoided if we thought of judgements as not necessarily conscious. At one point (p. 65), Bradley indeed suggests that judgements need not be conscious. But in many parts of the book, Bradley relies on judgements being necessarily conscious. As noted above, he appeals to conscious awareness of judgements to distinguish judgementalism from mental state accounts. Moreover, much of the book is concerned with the representation of incomplete preferences, where preferences are incomplete when agents have not made up their mind about an issue, for instance because they have not thought about it due to cognitive limitations (pp. 225–226). The standard for preference ascription Bradley applies here thus seems to presuppose a conscious judgement. Lastly, and most importantly for our discussion, for the most part, Bradley considers normative decision theory from the first person perspective, as a tool for decision makers to figure out what to do. And decision makers can only reason from the judgements they are conscious of. In fact, we will see in the next section that appealing to the connection of rationality to reasoning and action-guidance is the most plausible case for internalism about rational requirements, which serves as a justification for other judgementalist claims.
Thus, some of Bradley’s central claims rely on judgements being necessarily conscious, but in that case, the claim that desirability and utility ascriptions denote additional kinds of judgements seems undermined. A final possibility is that quantitative desirabilities are not a measure of anything graded at all, but rather merely a convenient way of representing relational judgements of desirability. This corresponds to what is arguably the standard interpretation of utility in the economics literature, which sees utility as a mere convenient representation of preference.Footnote 4 On this view, there would be no implication at all that desirabilities track degrees of judgement, or degrees of desirability as judged by the agent. The quantitative measure would express only facts about the structure of the agent’s relational judgements. Pragmatism not only makes sense, but indeed is implicit in this interpretation of quantitative desirabilities and utilities. I thus take this to be the best way forward for the judgementalist. However, this third possibility appears to be rejected by Bradley: He thinks that desirabilities are a quantitative measure of some graded attitude.
All in all, then, there is tension between Bradley’s pragmatism, the commitment that judgements are conscious implicit in some central arguments in the book, and his insistence that desirabilities and utilities represent some graded attitude. Given I am treating pragmatism as a central claim of judgementalism, and the central role of internalism about rationality in justifying judgementalism, I think it is the latter that should be given up by the judgementalist. I will set these complications aside for now, however. What is crucial about pragmatism from a judgementalist perspective is that quantitative notions such as utility or desirability do not rationally constrain preference, because preference is explanatorily prior to these. According to the third judgementalist claim, discussed in the next section, they also do not constrain choice in their own right—only preference does so.
Only preferences constrain choice
The final defining claim of the judgementalist position is that only preferences, conceived of as relational judgements, directly constrain choices. Bradley proposes the Choice Principle, and only the Choice Principle as a rational requirement on an unbounded agent’s choices, and the Choice Principle asks agents to choose one of their most preferred options. The Choice Principle has to be relaxed in cases of incomplete preferences. But here, too, the recommendations of the potential choice rules discussed by Bradley supervene on the imprecise utility representation of the agent, which in turn, given pragmatism, is simply derived from the preferences the agent does have. Preferences themselves are only constrained by other preferences and relational judgements of credibility. This is the view Bradley calls, following Broome (1991), ‘moderate Humeanism’.Footnote 5 Despite the initial appearance that the Rationality Hypothesis constrains preferences by requiring them to conform to the expectation of benefit, Bradley conceives of the Rationality Hypothesis as merely a requirement of consistency on preferences. In line with the representation theorems, agents should have preferences that are representable as expected desirability/utility maximising.
Why think that only preferences in the judgementalist sense constrain choices? The lazy answer is that this just follows from the first two judgementalist claims in combination with what the formal framework of normative decision theory gives us—one which has proven to be very fruitful, as Bradley’s book amply demonstrates. But to say more, we can appeal again to internalism about the requirements of rationality. As we already saw above, Bradley appears to be committed specifically to the view that rational requirements supervene on judgements, from which it follows that only judgements rationally constrain choice. Preferences being kinds of judgements according to the first judgementalist claim, this might be seen to support this third judgementalist claim.
Let me sum up the judgementalist interpretation of decision theory and Bradley’s motivations for endorsing it. According to judgementalism, preferences are all-things-considered relational judgements of desirability, they are conceptually and explanatorily prior to, and not rationally constrained by any quantitative notions like utility and desirability, and only preferences thus interpreted rationally constrain choices. That preferences are judgements rather than desires is motivated by a rejection of subjectivism about value, as deliberating agents are attempting to figure out, not determine what is in fact desirable. And that they are judgements rather than tracking objective value directly is motivated by internalism about rational requirements, according to which rational requirements supervene on mental states, and in particular judgements. The judgementalist claim that preferences are conceptually and explanatorily prior to quantitative desirabilities and utilities is motivated by the claims that rationality conditions on relational judgements have greater intuitive appeal, that we more easily make relational than quantitative judgements, and that starting from relational judgements yields the more general theory. And the judgementalist claim that only preferences constrain choices is motivated again by internalism about the requirements of rationality.
On this picture, normative decision theory is meant to govern decision-making from the first person perspective, which is assumed to proceed roughly as follows: To choose rationally, I must choose in line with my all-things-considered judgements about what it would be most desirable to do. When making these judgements, my aim is to correctly capture what is in fact desirable, not, or at least not exclusively to determine what is desirable. But when figuring out what all-things-considered judgement of desirability to make, all I can subjectively appeal to are other judgements I have made—of relative desirability and of relative credibility. If rationality only imposes subjective norms, then these can only be norms that appeal to consistency among my judgements, and consistency of my choices with my judgements. The best I can do is make a judgement that is consistent with my other judgements, and then act accordingly. And, crucially, requirements on the preferences and credibility judgements as they feature in the formal framework of normative decision theory are supposed to fully capture this picture of rational choice.