In Delusions and Other Irrational Beliefs, Lisa Bortolotti presents a carefully argued case against rationality-based objections to the doxastic conception of delusions. The book operates with certain methodological assumptions, which are spelled out clearly in the first chapter. First, Bortolotti’s main concern is with the criteria for the ascription of beliefs, not with the nature of beliefs, and she remains officially neutral on what beliefs are or even whether they exist (p.2) (So, to be precise, the view she defends is that delusion ascription is a species of belief ascription.) Second, Bortolotti adopts an inclusive characterization of beliefs as states that have relations to the subject’s other beliefs and intentional states, are sensitive to evidence, and are manifested in behaviour (p.12). Third, Bortolotti focuses on beliefs that are available to consciousness and can be introspected (p.12).

These assumptions are sensible ones, given the aims of the book. They are widespread in the literature on which Bortolotti comments, and they focus the debate effectively. However, I wish to consider some issues that arise if the assumptions are relaxed. In particular, I want to ask what consequences there may be for our view of delusions if we accept that there are distinct types of belief, which have different profiles and different relationships to consciousness, and which are subject to different ascriptive constraints.

It would be useful to have a comprehensive survey of the various ways in which the term ‘belief’ is used in folk psychology, philosophy, and scientific psychology. I cannot provide this here, but I want to introduce a basic distinction between two types of belief, which is latent in folk psychology. On the one hand, we ascribe beliefs to a wide variety of creatures and artefacts on the basis of their unreflective, nonverbal behaviour and without assuming that the attitudes we ascribe can be introspected, are under personal control, or are functionally discrete. On the other hand, we also use the term ‘belief’ in a more restricted way, for a state that is available to consciousness, is controlled, and can be selectively employed in reasoning and decision making. In addition, we sometimes treat belief as a matter of graded confidence and sometimes as a binary, on/off state.

The best explanation for these facts, I have argued, is that we use the term ‘belief’ to track states of two different types: in one case, multi-track behavioural dispositions, which are non-conscious, passive, graded, and holistic, and, on the other, personal commitments to deliberative policies, which are conscious, controlled, binary, and discrete. I shall call these ‘level 1 belief’ and ‘level 2 belief’ respectively. To have the level 1 belief that p is simply to be disposed to behave in ways that would be rational on the assumption that p is true, given one’s other beliefs and desires (level 1 belief ascription is holistic). This is roughly the conception of belief advocated by Dennett and Ryle [1–3]. By contrast, to have a level 2 belief that p is to be committed to a policy of taking p as a premise in one’s conscious reasoning and decision making. (Committing to a premising policy is what we call ‘making up one’s mind’. We can make such commitments, I argue, because conscious reasoning is to a large extent under our personal control. I defend this view in [6] and [8].) Premising commitments of this kind are often referred to as ‘acceptances’ (e.g., [4, 5]).

Now, not all acceptances are beliefs. One may accept a proposition for pragmatic reasons, such as professional ethics or loyalty to a friend, and the acceptance may be restricted to a certain context [4]. For example, a lawyer may accept that her client is innocent in her professional life, but not in private. We might call this ‘professional belief’, but it would be misleading to describe it as belief tout court (the lawyer may in fact believe that her client is guilty). I shall refer to pragmatic attitudes of this kind as non-doxastic acceptances. Non-doxastic acceptance is not subject to norms of epistemic rationality, and it may be subject to norms of procedural and agential rationality only within certain contexts. (The lawyer is obliged to reason with, and act on, the premise that her client is innocent only when thinking and acting in a professional capacity.) Level 2 belief, by contrast, is a doxastic form of acceptance, which is typically formed for epistemic reasons, is context independent, and is unqualifiedly subject to the norms of epistemic, procedural, and agential rationality.

There is a similar division in our folk ascriptions of desires and other propositional attitudes, and these attitudes too, I have argued, have level 1 and level 2 forms, with characteristics similar to those of the two forms of belief.

I have elsewhere made a detailed case for the existence of this division in folk practice and for this interpretation of it [6]. The distinction between level 1 and level 2 belief corresponds roughly to Dennett’s distinction between belief and ‘opinion’ [7] and also to that between tacit and explicit belief often made by psychologists (for a survey of relevant psychological theories and discussion of their relation to the folk-psychological distinction, see [8–10]).

On this view, there is a constitutive dependency between the two types of belief, which is what makes it appropriate to talk of levels. Level 2 beliefs are constituted by policy commitments which are themselves motivated and sustained by level 1 beliefs about the utility of the policies in question (see [6], ch.4). In addition, there may be mutual influence between contents at the two levels. In particular, adopting an explicit, level 2 belief may gradually cause one to acquire a tacit, level 1 belief with the same content. (That is, endorsing a proposition for use in conscious reasoning may gradually cause one to acquire an habitual, dispositional belief in the proposition, manifested in one’s unreflective behaviour.) However, the contents of the two levels can differ significantly, and a proposition may be believed at one level and its contradictory at the other. This may be what happens in cases of so-called attitude/behaviour inconsistency—for example, where a person avows egalitarian attitudes but continues to display prejudiced behaviour (see, e.g., [11]). Such cases, I would argue, typically involve conflict between a level 2 belief manifested in the subject’s avowals, and a contrasting level 1 belief manifested in their unreflective behaviour.

Bortolotti mentions the distinction between level 1 and level 2 belief ([12], pp. 167–71) and seems not unsympathetic to it, but she argues that it is an idealization and asks why there should be a necessary connection between the listed features of the two kinds of belief—non-conscious, passive, graded, and holistic, on the one hand, and conscious, controlled, binary, and discrete, on the other (p. 169). Bortolotti also questions the claim that level 2 beliefs are binary, arguing that beliefs typically come in degrees (p. 12). Finally, she assumes that, on the proposed view, level 2 beliefs will be behaviourally inert, and she objects—rightly—that this is implausible (pp. 169–70).

These considerations (coupled, I assume, with the methodological assumptions mentioned earlier), lead Bortolotti to set aside the two-level framework. But I think this is too swift. I shall briefly address Bortolotti’s objections and then outline some of the questions about delusions that arise if we accept the framework. There is no space here to deal with the issues fully, but I shall highlight the main points (for more detail, see my [6], [8], and [13]).

On the question of why the features should go together in the way described, my reply is that there is no necessary connection between them. Rather, my claim is that the best explanation for the division in folk practice is that the term ‘belief’ is used to refer to two different types of state (dispositions and premising policies) in which the features do in fact go together. Behavioural dispositions are not introspectable, and they are passively formed, graded, and holistic, whereas premising policies are conscious, controlled, binary, and discrete.

On the issue of whether level 2 belief is graded, I would distinguish two senses of ‘graded’. There is a sense in which level 2 beliefs are graded. They can have grades attached to them, reflecting how committed the subject is to them and corresponding to the strength of the subject’s level 1 beliefs in the relevant contents. This seems to be the sense Bortolotti has in mind.Footnote 1 This is compatible with these beliefs also being binary, in the sense that they are either categorically present or absent. For any given proposition, one either has or does not have a level 2 belief in it (that is, one either has or has not endorsed it as a premise). But there is also a more radical sense of ‘graded’ in which level 2 belief is not graded. In this sense, a belief is graded if it consists in a graded state, which can be present in any degree from zero to full. Degrees of confidence are graded in this way, and a subject can be interpreted as manifesting some, perhaps very low, degree of confidence in any given proposition. It is in this sense that level 1 belief is graded and level 2 belief is not.

As to the putative behavioural inertness of level 2 belief, I deny it. It is true that I have argued that level 2 belief has a much less tight connection to action than level 1 belief, and suggested that this may help to explain how delusions can fail to influence action [13]. I shall say more about this in a moment, when I discuss the role of norms of agential rationality in constraining belief ascription. But I want to stress that the proposed two-level framework does not entail that level 2 beliefs are typically inert or that they influence only verbal behaviour. What it does entail is that their direct influence is confined to reflective behaviour (that is, behaviour that is the product of conscious reasoning) and that they can sometimes fail to influence behaviour in the way they should.

I suggest, then, that Bortolotti’s worries about the two-level framework are misplaced, and I turn now to some questions that would arise for her project if she were to accept it. I shall outline three of these, and in the process suggest an alternative view of the nature of delusions. I do not think any of the points I shall make poses a serious problem for Bortolotti, though they might require some modest qualifications or extensions to her position.

First, if delusions are beliefs, which type are they, level 1 or level 2? The obvious answer is that they are level 2, at least initially—a view that accords with Bortolotti’s focus on conscious belief. Delusions are typically ascribed on the basis of the patient’s avowals, rather than inference from their unreflective, nonverbal behaviour. Indeed, it is hard to see what sort of nonverbal behaviour could warrant the ascription of some highly detailed or bizarre delusions, such as the delusion that one is dead (as opposed to, say, the belief that life isn’t worth living). But even if delusions begin as level 2 beliefs, they may later generate corresponding level 1 beliefs, perhaps with vaguer, more generalized contents. (In discussing a two-level view of belief, Bortolotti herself suggests an explanation of ‘double-bookkeeping’ in delusions on which delusions are behavioural dispositions as opposed to policy commitments, which suggests that she might be willing to locate some delusions at level 1; see [12], p. 170.) At any rate, if we endorse a two-level theory of belief, it will be important to explore this issue further and to identify the relative contributions of the two levels to the total delusional state.

Second, if there are different types of belief, do rationality constraints operate in the same way in relation to each, and what implications does this have for the doxastic view of delusions? I shall assume for the sake of argument that the two types of belief are subject to the same norms, and that these take the tripartite form Bortolotti describes: epistemic, procedural, and agential. (This may not be right, especially if one form of belief is radically graded and the other binary, but it does not affect the core issues here.) The question is to what extent, if any, adherence to these norms serves as a constraint on the ascription of beliefs of each type.

In the case of level 1 beliefs, adherence to the norms of agential rationality acts as a major constraint, since the multi-track behavioural dispositions that constitute these beliefs are defined by reference to those norms. To have a level 1 belief with content p is to have the behavioural dispositions a rational agent would have on the assumption that p is true (given their other beliefs and desires). If these dispositions include dispositions to form new beliefs, then adherence to the norms of procedural rationality will also play a constraining role, for the same reason. Of course, this defines an ideal dispositional profile, and there will be room for some deviation from it; but the more deviation there is, the weaker and more approximate the belief ascription becomes. In such cases it may be appropriate to speak of ‘in-between’ level 1 belief [14].Footnote 2 The key point is that where the dispositional profile for a particular belief is only partially manifested, there is no further fact that could settle the question of whether the belief is present.

In the case of level 2 belief, on the other hand, adherence to norms of rationality plays a far smaller constraining role. To have the level 2 belief that p is to be committed to a policy of treating p as true. A sincere and comprehending avowal of belief in p normally manifests such a commitment, so it is normally sufficient to warrant the ascription of the belief. This is so whether or not the commitment was made in response to evidence, so norms of epistemic rationality have no significant role in constraining attributions of level belief 2. Norms of procedural and agential rationality, on the other hand, do play a role, though it is limited. In adopting a policy of using p as a premise in reasoning and decision making, one commits oneself to adhering to norms of procedural and agential rationality. This is definitional; the policy is defined by reference to the norms. And if a person egregiously fails to do this, then we shall doubt that their commitment to the policy was sincere and comprehending, and withdraw the belief ascription. However, strict adherence is by no means assumed. Many lapses can be understood as performance errors, or ascribed to lack of motivation or competence rather than lack of commitment.

It is worth saying a little more about the influence of level 2 belief on action. On the view proposed, our level 1 attitudes play a crucial role in making our level 2 beliefs effective. We act on our level 2 beliefs because we have a strong level 1 desire to execute the reasoning policies we have adopted ([6], ch.4). But a level 2 belief may sometimes dictate an action that conflicts with other strong level 1 desires we have, and in such a case these other desires may override our desire to act on the belief. If delusions are level 2 beliefs, then this offers a possible explanation for cases where delusions fail to influence action. The subject is strongly motivated to accept and act on a delusional claim (say, that their partner has been replaced by an imposter), but is also strongly motivated to refrain from performing certain actions which it dictates, such as leaving their partner [8, 13].

To sum up, then, on this view, the failures of rationality evident in delusions are no obstacle to delusions being classified as beliefs of the level 2 kind. This is, I take it, broadly compatible with Bortolotti’s position. Indeed, the two-level view actually makes it easier to argue for a version of the doxastic conception, since arguments for strong rationality constraints on belief ascription can be dismissed as applying only to level 1 belief.

Third, might delusions be non-doxastic acceptances rather than irrational beliefs? The functional profile of a non-doxastic acceptance may be indistinguishable from that of an irrational doxastic one. Both may be insensitive to norms of evidential rationality and only partially sensitive to those of procedural and agential rationality. The difference lies in the subject’s motives for holding them—the intentions with which they were formed and are sustained. (These intentions will, I assume, typically be non-conscious, level 1 states.) The formation of doxastic acceptances (level 2 beliefs) is typically motivated by epistemic goals. The subject endorses a proposition with the aim of thereby acquiring a true belief, and thus indirectly furthering their theoretical enquiries and practical deliberations. A non-doxastic acceptance, on the other hand, is formed for practical reasons. The subject endorses a proposition because they have a reason for treating it as true regardless of whether or not it actually is true.

So it may be that delusions, rather than being beliefs that fail to conform to the standards of rationality for belief, are non-doxastic acceptances that were never meant to conform to them. Perhaps patients adopt delusions because they answer some emotional or other psychological need, rather than because they are probable. (Of course, patients may not think of their delusions as being motivated in this way, but this would not be surprising if the motivating attitudes are typically non-conscious ones, as I have suggested.) This view should be attractive to those who advocate psychodynamic theories of delusions, though it does not presuppose any specific theory of that type.

It may be useful here to compare delusions with religious faith. Faith is usually regarded as a form of belief. Indeed the term ‘belief’ is often used precisely for this kind of emotionally charged attitude, as opposed to mundane knowledge. However, some instances of faith may be better classified as non-doxastic acceptances. Religious people sometimes speak as if there is a virtue in holding onto their faith in the face of countervailing evidence, as though maintaining it were a moral duty rather than an epistemic one. This suggests that they do not regard faith as subject to the norms of epistemic rationality, and that they maintain the attitude for non-epistemic reasons. Of course, faith is not delusion, since it is culturally sanctioned and usually benign. But nonetheless it might be the closest analogue for delusion among non-pathological mental states. Delusion might be a sort of pathological faith.

The suggestion I want to make, then, is that delusions may be non-doxastic acceptances rather than irrational beliefs. It would not, of course, follow that delusions are rational. The pragmatic reasons for accepting a proposition may be bad ones. But if delusions are non-doxastic, then they are not irrational simply in virtue of not conforming to the standards of rationality for beliefs. This suggestion is not an objection to Bortolotti’s view, of course, merely an alternative to it. Moreover, it could be that both views are partially true; some delusions may be doxastic acceptances and some non-doxastic ones. And there may be a large of class of acceptances that are made for a mixture of epistemic and non-epistemic reasons, and whose doxastic status is indeterminate. In the end, nothing much hangs on whether a delusion gets the folk title of ‘belief’. The important issue, I suggest, is whether delusions are (primarily) acceptances. If they are (and I would be very interested to know Bortolotti’s view), then it will be important to examine the profile of individual delusions and to assess the relative roles of epistemic and pragmatic factors in each. This will be crucial to understanding and interacting with deluded people, and to helping alleviate their distress.