Room for responsibility: Kant on direct doxastic voluntarism

ABSTRACT

Kant's theory of assent seems to combine two incompatible claims that (i) we are responsible for our assent and (ii) we have no direct voluntary control over our assent. But how can we be responsible for something over which we have no direct voluntary control? Scholars have tried to resolve this tension by arguing that, according to Kant, assent is under our indirect voluntary or intellectual control. This paper defends a different solution. It is argued that contrary to first impressions, Kant is actually committed to the view that most assent is under our direct voluntary control, which, in turn, accounts for our doxastic responsibility. From this emerges the outline of a unified Kantian theory of doxastic agency.

KEYWORDS:

• Assent
• knowledge
• persuasion
• opinion
• belief

1. Introduction¹

Kant’s theory of assent (Fürwahrhalten) seems to combine two incompatible claims. On the one hand, Kant claims that we are responsible for our assent. For example, he states that ‘we can, of course, blame someone who has given approval to a false cognition’ (24:160; see also 9:54). On the other hand, Kant also seems to claim that we have no direct voluntary control over our assent: ‘Immediately, the will has no influence on the assent’ (9:73). In other words, we cannot simply assent to p because we choose to assent to p. But how can we be responsible for our assent if we have no direct voluntary control over it?

Different authors have tried different solutions to this problem. Cohen (2013, 2014, 2021) has argued that Kant grounds our doxastic responsibility in our indirect, not direct, voluntary control over assent. While we cannot simply choose to assent to p, we can influence our assent by choosing ‘to perform a number of intermediate actions’ (2013, 34).² Apart from seeking additional evidence, we can influence our assent by choosing ‘epistemic maxims that determine belief-acquisition’ (2021, 687). These epistemic maxims, Cohen claims, are the primary ground of our doxastic responsibility and include principles such as: ‘I should proportion my beliefs according to evidence’ (2021, 687–8).³

By contrast, Kohl (2015) has argued that Kant grounds doxastic responsibility in our intellectual, not voluntary, control over assent. Specifically, Kohl suggests that Kant admits of ‘a non-volitional sense in which we are free and active in our assents’ (309). It is not the freedom of the will, but the freedom of the intellect or understanding that makes us responsible for our assent. Despite their different solutions, Cohen and Kohl agree on the main premise of the problem: Kant was not a direct doxastic voluntarist. Cohen notes that, according to Kant, ‘we have no direct doxastic control over [our assent]’ (2021, 687). And Kohl writes: ‘I agree that Kant rejects doxastic voluntarism’ (2015, 309).

I disagree with both Cohen and Kohl on this main premise. My goal in this paper is to show that, on Kant’s account, we have doxastic responsibility because we have direct voluntary control over most assent. The structure of the paper is as follows: First, I argue that Kant’s claim that the will has no direct influence on our assent should be read as being restricted to knowledge (§ 2). I then suggest that Kant grants direct reflective control over persuasions (§ 3). More importantly, however, I argue that, according to Kant, we have direct voluntary, and indeed intentional, control over all opinion (§ 4) and all Belief (Glaube) (§ 5).⁴ I conclude by explaining how my reading accounts for Kant’s claims about doxastic responsibility (§ 6).⁵

My reading aims to contribute to two debates. Primarily, I hope to advance the scholarship on Kant’s theory of assent by providing the first in-depth analysis of the voluntaristic profile of assent;⁶ and by showing that Kant articulates his doxastic voluntarism in the language of assent. Secondarily, I wish to contribute to discussions on Kant’s theory of doxastic agency, not by proposing my own reading of such a theory, but by introducing constraints on any future reading. Kant’s doxastic voluntarism, I argue, not only grounds our doxastic responsibility, but it also limits its scope; contra Cohen and Kohl, I will claim that Kant limits our doxastic responsibility to only some species of assent.

2. Knowledge

Readings à la Cohen and Kohl can draw on a long list of passages in which Kant seems to claim, without restriction, that the will has no direct influence on our assent. We have already encountered the passage from the Jäsche Logic: ‘Immediately, the will has no influence on the assent’ (9:73). In the Blomberg Logic, Kant similarly claims that ‘our approval […] does not rest on our free will at all’ (24:156). We find further such passages in the Philippi Logic (24:423), Pölitz Logic (24:545), and Vienna Logic (24:859). My aim in this section is to show that, contrary to first impressions, these passages should be read as making a restricted claim about knowledge.

Note first that the cited passages respond to Meier’s Excerpt from the Doctrine of Reason, which was the textbook for Kant’s Logic Lectures. In § 168, Meier states that ‘[w]e give a cognition our approval or we assume it (assentiri, ponere aliquid) when we hold it to be true; we reject it (tollere aliquid) when we hold it to be false; and we suspend our approval (suspendere iudicium) when we do neither of the two’ (my emphasis). Kant’s main concern is that this façon de parler – giving, rejecting, suspending assent – suggests that our assent is a function of our will and not the understanding; that is, we merely ‘hold something to be true because we will it to be true’ (24:545).

In response to Meier, Kant claims that our ‘approval is taken, not given’ (24:423). Or more emphatically, the understanding ‘wrests [abdringen] and forces [abzwingen]’ the assent from us (24:158). It is in this context, then, that Kant claims that our assent ‘does not rest on our free will at all, but is […] necessitated by the laws of our understanding’ (24:156). However, does Kant mean to say here that all assent is forced by the understanding, or does he mean to restrict this claim to only some assent?

To answer this question, let us look at Kant’s explanation for why the understanding forces our assent. In his personal copy of Meier’s Excerpt, Kant notes in the margin of § 168 that ‘[w]e cannot suspend our approval, except in so far as we recognise the possibility of error. (Force by the understanding to approval: mathematics.) We can suspend our approval even for the most evident proofs if we know that in similar cases one has already been deceived’ (R2507, 16:397–8). This reflection motivates the following principle: the understanding forces our assent iff, and because, our grounds of assent are infallible.

This principle makes a plausible philosophical point. If we see that our justification for p is infallible and thus guarantees the truth of p – as Kant thinks is the case, for example, with mathematics and experience: ‘Mathematics and pure, direct experience […] leave us no grounds for the opposite’ (24:160) –, then we are compelled by that insight to accept p as true. Borrowing from Habermas (1992), we may call this the unforced force (zwanglose Zwang) of infallible arguments. Conversely, if we realise that our justification isn’t watertight, we can decide to, at least, suspend judgement on the matter.

Indeed, this principle underpins Kant’s discussion of § 168 in the Logic Lectures. Following his claim that the will has no direct influence on our assent, Kant typically illustrates this point with an example from mathematics. Consider Kant’s example in the Blomberg Logic: a shopkeeper ‘realises from his calculations that he owes much and more than he […] owns’ (24:157). Much as he would want to deny this situation, he cannot will himself to ‘suspend his approval’ (24:157), let alone judge the opposite, that he owns more than he owes. Since ‘he is convinced of the correctness of his arithmetic’, the understanding forces him to doxastically confront his economic plight (24:157).

Following examples like this one, Kant then typically restricts his initial claim that the will has no direct influence on our assent. In the Jäsche Logic, he writes that ‘the will cannot dispute convincing proofs of truths that are contrary to its desires and inclinations’ (9:74; my emphasis). We find similar wording in the Pölitz Logic: ‘[The will] cannot dispute convincing proofs of things that we do not like to wish for’ (24:545; my emphasis).⁷ And in the Philippi Logic, Kant says that ‘[t]here are certain grounds’, namely infallible grounds, ‘that force us to approve. With many, however, […] at least the suspension of approval is under our will’ (24:423; my emphasis). Passages like these strongly suggest that the will has no direct influence on our assent, not just if, but also only if, the understanding forces our assent based on infallible grounds.

I submit that Kant identifies assent from infallible grounds with knowledge (Wissen). Knowledge, for Kant, is based on sufficient objective grounds: ‘Knowledge [is] the assent from a ground of cognition that is objectively […] sufficient’ (9:70; see also A822/B850). Sufficient objective grounds, in turn, are infallible according to Kant; that is, sufficient objective grounds for p guarantee the truth of p and so make ¬p impossible: ‘the sufficient [objective] ground is that whose opposite cannot possibly be thought […] to be true’ (24:145). Knowledge, and knowledge alone, thus rests on infallible grounds that force our assent.⁸ By implication, then, Kant must be committed to the following principle:

2.1. Principle of voluntary control

(PVC) S has direct voluntary control over the assent that p iff (i) S does not know that p and (ii) S does not know that ¬p.⁹

The if-direction of the principle allows Kant to accommodate popular motivations for doxastic involuntarism. For example, Kant would agree with Williams that we lack direct voluntary control if p is supported by ‘very strong evidence’ (1973, 149); and with Alston that we lack such control if p is ‘clearly true or false’ or even ‘certain’ (1988, 265–6).¹⁰ Kant would also agree with Wiggins that ‘knowledge […] is precisely not up to [us]’, for then we would lose ‘connection with the world’ (1969, 143). In fact, Kant explicitly argues that the will cannot have a global influence on all assent, or else ‘we would constantly create chimeras of a happy state for ourselves, and then always hold them to be true’ (9:74; my emphasis).¹¹

But as I have shown in this section, Kant is also committed to the only-if-direction of PVC. Kant’s response to Meier is not to deny that the will has direct influence on any assent; rather, Kant denies that the will has a direct influence on all assent. While the grounds of knowledge forcibly tie us to the world, assent outside the domain of knowledge is under our direct voluntary control. My aim in the following sections is to fully articulate this view for each of the remaining species of assent, thereby providing independent support for the only-if-direction of PVC. The next section makes the start by showing that the will has a direct influence on persuasion, thus creating chimeras sometimes.

3. Persuasion

Following Meier (§ 184), Kant defines persuasion (Überredung) as the assent by which we ‘take an uncertain cognition for certain’ (24:439).¹² Persuasion is uncertain insofar as it, unlike knowledge, is ‘objectively insufficient assent’ (R2465, 16:382).¹³ Yet we take it to be certain because we mistake subjective grounds of assent for sufficient objective ones: ‘Persuasion is the assent from merely subjective causes that are falsely taken to be objective’ (24:547).¹⁴ In this way, persuasion is a mere prejudice (Vorurteil), which results from the same confusion: ‘Prejudice is the confusion of a […] subjective cause of judgement with the objective ground’ (R2539, 16:409).^15,16

Persuasion is always irrational, for Kant, because we don’t proportion the strength of our assent to the strength of our grounds: ‘Persuasion is always blameworthy [tadelhaft] in that something is held to be certain on insufficient grounds’ (24:218). What goes for persuasion also goes for prejudice: ‘To accept something as true through prejudice is always wrong [unrecht]’ (R2544, 16:410). Note, however, that we can be both overconfident and right, or as Kant puts it: ‘Although every persuasion is false in form (formaliter), insofar as an uncertain cognition seems to be certain, it can still be true in matter (materialiter)’ (9:73).¹⁷ So, we can be persuaded that p, and yet p is true.

Let’s look at some examples. While persuasion and prejudice can arise from different subjective grounds, Kant identifies three main sources: ‘imitation, habit, and inclination’ (9:76).¹⁸ Inclination is particularly interesting because ‘[t]he prejudices that arise from inclination are manifold’ (24:162). Children are the paradigmatic example: ‘Who is most easily persuaded? Children. […] They believe what they wish. […] [I]nexperience makes them assume what they like, and they do not even search for what the understanding requires’ (24:854; my emphasis). Kant also offers the example of a ‘partial judge’, whose verdict depends on his sympathy for the defendant (R2521, 16:403–4).¹⁹

In these cases, where inclination persuades us, the will has a direct influence on our assent. Children hold p to be true because they want p to be true; the inclination is the ground of their assent. Kant expresses this point more generally: ‘The inclination is also the cause and ground of some judgements, whereby one judges […] because it appeals to us, because it is pleasant and suits our taste’ (24:167; my emphasis). This, then, provides independent support for the claim of the previous section that the will does have a direct influence on, at least, some assent.

But these cases also motivate a crucial distinction. While inclination-induced persuasion results from the direct influence of our will, it does not result from our direct voluntary control: ‘In prejudice, reason is only suffering [leidend], for it arises from inclination, which is in itself passive’ (24:641; see also R2531, 16:407–9). When inclination forces our assent, ‘the mind […] [is], as it were, fettered [gefesselt] and restricted [restringiert], so that we […] give our approval to that which is favourable to us’ (24:159). It would seem, therefore, that persuasion is forced by the will, just as knowledge is forced by the understanding.

But that’s not quite right. For while persuasion does not result from our direct voluntary control, it, unlike knowledge, stands under our direct voluntary control. Kant suggests that persuasions are rushed judgements that we make without using our understanding properly: ‘Prejudices are judgements that forestall [zuvorkommen] the understanding and […] [the understanding] comes late afterwards. The understanding is initially passive’ (R2531, 16:407–9; my emphasis). Specifically, we failed to initially reflect on our grounds: ‘to judge without reflection is called prejudice’ (R2519, 16:403).²⁰

Note that Kant here employs a technical notion of reflection (Überlegung). On his account, reflection is the process by which we compare our grounds of assent with the rules of understanding: ‘To reflect is to compare something with the laws of the understanding’ (24:424).²¹ Insofar as the laws of understanding are sensitive only to objective grounds, Kant thinks that, through reflection, we can tell whether or not our assent is based on sufficient objective grounds: ‘to reflect, i.e. […] to examine whether the grounds are sufficient or insufficient with regard to the object’ (9:73).

I propose that persuasion stands under our direct voluntary control insofar as we are able to suspend judgement upon reflecting on our grounds. After all, it was only because we didn’t reflect on our grounds that the inclination was able to force our assent in the first place; however, we are free to reflect on our grounds post factum, in which case we realise that the grounds of our assent are objectively insufficient. Kant is clear on this point: ‘Persuasion is followed by reflection, i.e., […] we see whether the grounds are sufficient or insufficient in view of the object’ (24:559; see also 9:73).

Once we realise that our assent does not rest on sufficient objective grounds, we are free to suspend our judgement. In the Philippi Logic, Kant observes that ‘[t]here are certain grounds on which we find ourselves forced to approve’ – sufficient objective grounds, I suggest –, ‘[b]ut with many we can postpone our approval, […] the suspension of approval is under our will’ (24:423; my emphasis). Suspension of judgement, Kant further states, is a free act: ‘In the suspensio judicii there is some freedom’ (24:736). In fact, the choice of whether to suspend judgement or fall back into persuasion requires an ‘intention [Vorsatz]’, making it an intentional choice (9:73; see also 24:860).

It emerges a two-stage model. Persuasion stands under our direct voluntary control because (i) we are able to reflect freely on our grounds of assent, assessing whether they are objectively sufficient, which, having done so, (ii) allows us to choose freely to suspend our judgement. The second stage always presupposes the first, I suggest, because we cannot will ourselves to suspend judgement as long as we (mistakenly) take our grounds to be objectively sufficient. This model then explains why ‘[u]nderstanding and reason are free: subjective causes affect, but do not determine, understanding’ (R2476, 16:386).²²

This two-stage model invites three objections.²³ The first objection is that we are not always free to reflect. Indeed, Kant acknowledges the ‘stubbornness of prejudices’ (R2546, 16:410) and suggests that it is often ‘difficult to detect persuasions’ (25:1451).²⁴ As a result, ‘[m]any remain with persuasion‘ and only ‘some make it to reflection’ (9:73; see also 24:559). But while I concede that reflection is difficult and that, as a matter of fact, not everyone will reflect on their grounds, I maintain that, on Kant’s account, we are always able, at least in principle, to reflect freely on our grounds. After all, we can always get out of our epistemic comfort zone by testing our grounds with others or by placing a bet (A820–5/B848–53).²⁵

The second objection is that suspension is not a free choice. The idea here is that, once we realise that our assent is not based on grounds that are objectively sufficient but, say, on inclination, we have no choice but to suspend our assent.²⁶ This objection, I think, is too optimistic about our epistemic constitution. Inclination always tempts us back to persuasion, just as it always tempts us to act selfishly in the moral sphere. Kant is clear that ‘the suspension of our approval is a very difficult thing’ that ‘requires a trained power of judgement’ (9:74–5). Suspending judgement thus is a genuine choice, which Kant gets to characterise as voluntary (24:423), free (24:736), and intentional (9:73, 24:840).

The third objection requires special attention. It says that our voluntary control over the suspension of persuasion is only indirect. Even if we have a genuine choice to suspend judgement, we only have that choice once we reflected on our grounds. Refection, therefore, turns out to be an intermediate action necessary for suspension, suggesting that we only have direct control over reflection, not suspension. Cohen (2013) seems to entertain a version of this objection when she claims that suspension only grants ‘[i]ndirect negative doxastic control’ (2013, 37; my emphasis) because we are only able to suspend judgement if we create the ‘epistemic environment necessary’ to reflect on our grounds (41).

The objection, I argue, confuses conditions of choice with grounds of choice. This distinction is best illustrated by an analogy. Picture this: every morning I unlock my bike and cycle to work. Strictly speaking, I only have a genuine choice to cycle once I have unlocked my bike: locked bikes cannot be cycled. Unlocking, then, is an intermediate action to my cycling. Nevertheless, I clearly have direct voluntary control over whether I cycle or not. Unlocking is a condition for my decision to cycle, not a ground; I don’t cycle because of my unlocking. Similarly, reflection is a condition but not a ground for my suspension. While I can only suspend persuasion upon reflecting, I don’t suspend because I’ve reflected; I suspend because I chose to do so.

The analogy also demonstrates that we can have direct voluntary control over the suspension of persuasion, even when such suspension is based on, what Cohen calls, ‘epistemic maxims’ (2021, 687). After all, my decision to cycle to work can be directly voluntary, even if it is based on the maxim ‘whenever I unlock my bike, I cycle to work’. Similarly, my decision to suspend persuasion can be directly voluntary, even if it is based on the epistemic maxim ‘whenever I reflect on the grounds of my assent (and realise they are objectively insufficient), I suspend judgement’.²⁷ The mere fact that a decision is based on a maxim, does not entail that we have only indirect voluntary control over it; maxims may just as well explicate the nature of our direct control, for example, as being principled.²⁸

In conclusion, then, we have direct voluntary control not only over our epistemic maxims and whether we reflect on the grounds of our persuasion – though we have that control, too – but Kant claims that we also have direct voluntary control over whether or not we suspend persuasion, and so over persuasion itself. This is not to say that persuasion results from a voluntary choice, but that it is in a robust sense up to us whether we continue being persuaded or whether we choose suspension. In the next section, I argue that Kant identifies the choice to suspend persuasion with the choice to adopt opinion.

4. Opinion

Borrowing again from Meier (§ 181), Kant defines opinion (Meinung) as ‘an uncertain cognition, insofar as it is taken to be uncertain’ (9:73).²⁹ Thus defined, opinion differs from both knowledge and persuasion. Unlike knowledge, opinion is uncertain in that it rests on insufficient objective grounds that are fallible: ‘Opinion is an […] insufficient assent, which is not sufficient to rule out the opposite’ (24:736).³⁰ Unlike persuasion, opinion is not taken to be certain: ‘Opinion […] is opposed to persuasion. Both involve uncertain cognitions, but […] in the case of opinion they are not held to be certain’ (24:218). In this section, I argue that, for Kant, all opinion results from our direct voluntary control.

The first thing to note is that opinion, for Kant, is a doxastic attitude that is very weak. An opinion is merely a provisional judgement (vorläufiges Urteil): ‘Opinion is a provisional judgement’ (24:541).³¹ Provisional judgements are contrasted with determinate judgements (bestimmte Urteile): ‘Judgements are categorised as provisional […] or determinate’ (24:546).³² While determinate judgements are decided (entschieden), ‘I decide nothing in the provisional judgement’ (24:426); and accordingly, ‘one always remains undecided in opinion’ (24:148).³³ Kant’s idea here seems to be that, when we hold an opinion, we have not yet made up our minds; we are not committed outright.

Kant even suggests that provisional judgements aren’t full judgement but the suspension of (determinate) judgement: ‘suspensio judicii nostri […] is the intention not to allow a provisional judgement to become a determinate one’ (24:545).³⁴ Accordingly, Kant also thinks that opinion requires the suspension of (determinate) judgement: ‘In opinion, [one assumes something] in suspenso’ (R2463, 16:381).³⁵ This point deserves special emphasis: for Kant, assent and suspension are not mutually exclusive; in opinion, we assent that p and, at the same time, intentionally suspend (determinate) judgement about p.

One might object that all opinion is assent, and so should be an assertoric commitment that we are allowed to assert. But this is evidently not the case. Kant identifies opinion, Belief, and knowledge with problematic, assertoric, and apodictic judgements: ‘In opinion, I judge problematically; in Belief, assertorically; in knowledge, apodictically’ (24:638).³⁶ Opinion thus falls short of an assertoric judgement; indeed, Kant identifies provisional judgements with problematic ones: ‘The provisional judgement is […] a merely problematic judgement’ (9:74).³⁷ Moreover, Kant restricts assertion (Behauptung) to assent that qualifies as knowledge (A821–2/B849–50; R2450, 16:373).

Non-assertoric assent also makes philosophical sense. We must not forget that the English term ‘assent’ is a translation of the German ‘Fürwahrhalten’, which more literally means ‘holding-as-true’. While many ways of holding a proposition as true express an outright commitment, I venture that there are also non-committal types of holding; for example, I might merely entertain or assume something as true. Indeed, Kant sometimes uses the term ‘assumption’ (‘Annahme’) to denote this weak holding: ‘The problematic judgement, what I assume’ (R2496, 16:394).³⁸

Let’s look at an example. The paradigmatic case of opinion, for Kant, are hypotheses – ‘Hypotheses are opinion’ (24:733) – or what Meier calls a ‘philosophical […] opinion’ (§ 181). Hypotheses, for Kant, are assumptions we make to explain phenomena that are given to us in experience: ‘Hypothesis is a proposition that one assumes to explain certain phenomena’ (29:918). A doctor, for example, might adopt the working assumption that the patient has the flu to explain the fever; a judge might provisionally assume that Adam is an arsonist to explain why the house caught on fire.

Opinion, thus weakened, results from two voluntary choices. The first choice is to suspend judgement. Opinion requires the suspension of judgement insofar as we don’t let our provisional judgements become determinate ones. This suspension, Kant says, is an intentional choice: ‘suspensio judicii, the intention not to let a provisional judgement become a determinate one’ (24:860; my emphasis). We are free in choosing to suspend our judgement: ‘With opinion there is suspension. One is still free’ (R2463, 16:381). Moreover, this choice may not be a one-off act, but requires that one ‘maintains oneself in suspensione judicii at will [willkürlich]’ (24:736; my emphasis). Although Kant doesn’t say much about how psychologically demanding this maintaining is, he seems to think that suspension a sustained effort.

I propose that this choice to suspend judgement is the same choice we have in the case of persuasion. Recall that we are able to suspend a persuasion once we are aware that the grounds of our assent are objectively insufficient. In light of the present discussion, this suspension would consist in downgrading a determinate judgement to a provisional one. So by suspending judgement, we move from persuasion to opinion. Conversely, then, our choice to suspend judgement in opinion is the choice not to be persuaded: ‘Provisional judgement is not prejudice, but is a precaution to avoid it’ (R2523, 16:404).³⁹

In addition to this negative choice, I suggest that opinion also rests on a second positive choice. As long as our assent is objectively insufficient, we have several assent-options to choose from. This is most clearly the case with hypotheses: ‘A hypothesis is a proposition which is assumed to explain certain phenomena, but which could also be explained by another hypothesis’ (29:918; my emphasis). To return to my previous examples, the doctor could also explain the patient’s fever by assuming an autoimmune disease; the judge could explain why the house caught fire by assuming that Bill, not Adam, is an arsonist.⁴⁰

While evidence typically recommends one option over another, opinion, for Kant, always requires a voluntary choice to adopt one of the options. Kant puts this point in very strong terms when he discusses problematic judgements in general: ‘The problematic proposition is […] a free choice to allow such a proposition to hold, a merely arbitrary assumption [Aufnehmung] of it in the understanding’ (A75/B101). Echoing these words, Kant regularly emphasises that ‘hypotheses are relative authorisations to assume something at will [willkürlich]’ (R2681, 16:469).^41,42 In Kant’s picture, the doctor and the judge choose which of the hypotheses to entertain or assume (for the time being).⁴³

One might object that my reading makes no sense of high-probability opinion. Sure, if we have two hypotheses, A and B, that have a similar (evidential) probability, we can choose to seriously entertain both A and B as true. But once, say, A has probability 0.99, it would seem that we can no longer suspend judgement on A, and so a fortiori that we can no longer choose to seriously entertain B as true. In fact, Kant seems to raise the very same worry when he notes: ‘the greater the probability of a thing is, the more the understanding is forced to hold it to be true’ (24:229).

While it may become increasingly difficult to suspend judgement as the probability increases, I maintain that, on Kant’s account, we are at least in principle able to suspend high-probability judgements in favour of their low-probability alternatives – which, of course, is not to say that doing so would be rational.⁴⁴ Only in the limit case, when our assent has probability 1, are we left with no choice but to assent. Note that Kant defines probability as the ratio ‘of the grounds that one has […] to the sufficient [objective] ground’ (R2591, 16:432).⁴⁵ This definition entails that assent has probability 1 iff it is based on sufficient objective grounds, which is only the case with knowledge.⁴⁶ So once more we find: knowledge is the only forced assent. Opinion, by contrast, is a matter of choice.

5. Belief

I now turn to Kant’s notion of Belief (Glaube), specifically moral Belief,⁴⁷ which is a special kind of assent. Unlike knowledge, Belief is not based on infallible objective grounds and is thus not logically certain: ‘What I believe, of that I am […] not (logically) certain’ (R2468, 16:383). Unlike persuasion, which is taken to be objectively sufficient, Belief is ‘taken to be objectively insufficient’ (A822/B850). And unlike opinion, which is merely a provisional or problematic judgement and so undecided, Belief is determinate or assertoric and so decided: ‘In opinion, one always remains undecided […]. [In Belief,] I am always decided’ (24:148).⁴⁸

But Belief shouldn’t be confused with a kind of clear-eyed persuasion. The assertoric commitment of Belief is not only practically justified – I will turn to this shortly –, but it is also certain. For while Belief lacks logical certainty, Kant insists that it has ‘moral certainty’ (A829/B857). To be sure, moral certainty does not rest on objective grounds that guarantee truth, but it is nevertheless infallible in that we can never have sufficient grounds to prove its opposite; we ‘cannot prove its impossibility’ (24:855). This implies that, ‘if one does something at the risk of being wrong, one is never morally certain’ (24:734).⁴⁹

Kant is clear that Belief, so understood, results from our free choice. Not only does he call Belief ‘free assent’ (5:472), but he even clarifies that ‘Belief is the assumption of a cognition based on intention [Vorsatz]’ (R2462, 16:380).⁵⁰ My aim in this section is thus not to establish that Belief is voluntary; instead, I address two questions that arise by comparing Belief with opinion. (i) Kant states that ‘[w]ith opinion […] one is still free. With [Belief] in force’ (R2463, 16:381). In what ways is opinion free, but Belief forced? And (ii) opinion can be voluntary because it is a weak doxastic attitude; but how could there ever be something like ‘free assertoric’ assent (6:154n)? I will tackle both questions in turn.

There are two ways in which opinion is free, but Belief is forced. First, opinion involves the freedom to suspend judgement; we have not yet committed ourselves to a proposition. In contrast, Belief is ‘a full commitment [Entschiedenheit]’ (R2462, 16:380). Once a Belief is held, there is no room for changing one’s mind; we are fully bought in. Kant puts this point in very strong terms, writing that ‘[t]he one who opines reserves the right to withdraw his judgement. […] [W]ho believes: not, but immutability’ (R2450, 16:374; my emphasis). In Belief, we thus cannot choose to suspend judgement.

Second, Belief eliminates the freedom to choose between options. Remember that opinion, especially hypotheses, were always just one among multiple possible candidates for assent. Yet Belief is practically justified as the ‘hypothetically necessary’ means to realise a given end of reason (A823/B852). In contrast to the contingent hypotheses discussed before, Kant thus characterises Belief as a necessary hypothesis. Necessary hypotheses leave no room for choosing between options.⁵¹ We cannot, say, hold a Belief either in the existence or non-existence of God. Put succinctly: ‘Theoretical hypotheses are contingent [beliebig]. A practical one is necessary’ (R2692, 16:472).⁵²

But if Belief involves neither the freedom to suspend judgement nor the freedom to choose between options, in what sense is it free? I suggest that Belief, for Kant, is free in so far as it ‘is based on subjective grounds of the will’ that are internal to us (R2470, 16:383). We adopt Belief as a necessary means to realise rational ends we have set for ourselves. By contrast, the objective grounds of knowledge are external to us; they are literally in the ‘constitution of the object’ (A821/B849).⁵³ Accordingly, Kant also defines ‘Belief [as] an assent based on internally sufficient, though not externally sufficient, grounds’ (R2461, 16:380; my emphasis).

At a minimum, then, internal grounds of Belief should be free in the negative sense, meaning that they are ‘independent of […] the influence of external causes’ (4:446; see also 5:33) – they are not heteronomous.⁵⁴ Unlike external grounds of knowledge, internal grounds don’t ‘wrest and force’ the assent from us (24:158). One might further argue that the internal grounds of Belief are also free in the positive sense, meaning that they are self-legislated or autonomous (5:33; 4:446).⁵⁵ Such a reading is not without merit. After all, Belief is justified a as necessary means to realise our rational ends, ends that we have set ourselves autonomously.

But there is a puzzle I can only mention here, not solve. Autonomous grounds would need to be universally valid. Yet Kant insists that internal grounds of Belief are merely subjective, making them incommunicable: ‘Belief gives a conviction that is not communicable. (because of the subjective grounds)’ (R2489, 16:391).⁵⁶ By contrast, the external grounds of knowledge are objective, and so ‘can be universally communicated’ (R2459, 16:378).⁵⁷ The main difference here seems to be that Belief rests on an end that is ‘directed at [an] interest’,⁵⁸ whereas objective grounds are entirely ‘without interest’; and even if Belief is directed at ‘the interest […] of everyone’ (R2489, 16:391), my Belief is always based on my interest.

The point here is not only that grounds of Belief are first-personal, but also that grounds of Belief have merely private validity.⁵⁹ For example, Kant writes: ‘Belief is a private assent [Privatfürwahrhalten], only for me sufficiently certain’ (24:732; my emphasis).⁶⁰ Moreover, he contrasts his notion of private Belief with a notion of ‘people’s Belief [Volksglaube]’, which is ‘[a] Belief sanctioned by public authority’ (23:431).⁶¹ In keeping with the spirit of the Enlightenment, Kant wants to remove Belief – including religious Belief – from the public sphere and make it a matter of private life. He even allows us to fend off any attack on our Beliefs by engaging in defensive polemics (A754/B782).

What follows from this? While it is clear that Belief is negatively free – enough to ground our direct voluntary control –, it remains unclear whether, and if so how, it can also be positively free. It seems, for example, that Belief cannot be autonomous in Cohen’s sense of epistemic autonomy. Cohen states that ‘we believe autonomously if we believe in accordance with universalisable epistemic maxims’ (2014, 690; my emphasis). But arguably, private Belief cannot be based on universal maxims since it would then be subject to the public’s epistemic authority.⁶² Perhaps it’s because of these struggles with Belief that Cohen limits her account to empirical assent (2013, 40). More on this in the next section.

Having discussed the sense(s) in which Belief is and isn’t free, we can turn to the second question of this section: how could we ever freely choose an assertoric assent, i.e. an assent that is decided? Kant seems to think that the will compels our assent in the case of Belief. Take, for example, the moral Belief in our immortality that we must adopt in order to realise the highest good (5:122–4). Kant suggests that ‘this assumption is a full commitment if it must be assumed according to practical forcing [nötingenden] laws’ (R2462, 16:380). It would seem, then, that once we are clear about the conditions under which we are able to realise our moral ends, we are compelled by our own will to adopt the assertoric assent.

However, Kant recognises obstacles to the free adoption of a Belief. For example, he observes that many people struggle with their (im)mortality: ‘We realise that death is not to be feared, and yet we fear it’ (R6749, 19:148). This struggle can have different reason. A common reason for Unbelief (Unglaube) is ‘a lack of moral interest’ (6:70). If we aren’t fully committed to the moral end itself, then we won’t be committed to the conditions under which we are able to realise that end. But Kant also suggests that Unbelief can occur when the reasons for our assent are not apparent to us; we don’t quite see how we can be immortal: ‘The soul is immortal: that is quite certain, but the grounds of philosophy are not so apparent that one should be completely convinced by them’ (24:204).⁶³

But in those cases where we can’t bring ourselves to freely form an assertoric assent, we can still choose to act as if we had assented. Again, in relation to immortality, Kant states: ‘But if I cannot believe it (e.g. the future life), then I have reason enough to act as if such a life were coming’ (R2503, 16:395). Indeed, Kant seems to suggest that if we act as if we were immortal, e.g. by subordinating our own happiness to moral demands, we can be said to possess moral Belief or conviction in some (presumably weaker) sense: ‘He who lets certain advantages of life pass by and lives holy […] has a strong Belief’ (24:853).

6. Responsibility

I have argued that, according to Kant, we have direct voluntary control over all assent that is not based on sufficient objective grounds, that is, all assent other than knowledge. Kant should thus be read as endorsing PVC. In particular, I have suggested that (i) persuasion is under our direct voluntary control insofar as we are able to deliberately suspend our judgement once we reflected on our grounds; (ii) opinion results from a dual choice to suspend judgement and to choose one of several available options; and (iii) Belief is voluntary insofar as it is a necessary assumption to our willed ends an assumption that is not forced by external grounds. In concluding this paper, I show that this reading accounts for Kant’s claims about our doxastic responsibility.

While there must be a minimal sense in which all assent is imputable to us – the understanding’s spontaneity may explain this imputation (Kohl 2015; Stevenson 2003)⁶⁴ –, it is important to note that Kant never asserts a global responsibility for our assent. Instead, he makes several more local claims. Specifically, he states: ‘I am not responsible for my opinion […]. But I must take responsibility for what I believe’ (R2462, 16:381); ‘Persuasion is always blameworthy’ (24:218; see also 24:165); and ‘we can, of course, blame someone who has given approval to a false cognition’ (24:160; see also 9:54). Insofar as blame entails responsibility, these passages entail the following four local commitments about our doxastic responsibility:

1. We are not responsible for any opinion.

2. We are responsible for all Belief.⁶⁵

3. We are responsible for all persuasion.

4. We are responsible for all false cognition.

Let’s start with (1): why are we not responsible for opinion? Kant suggests that ‘I am not responsible for my opinion […] because I have assumed it’ (R2462, 16:381). As noted above, Kant sometimes uses the term ‘assumption’ to indicate the very weak holding-as-true that is characteristic of opinion. So the idea is simple: if I have not yet assertorically committed to a proposition, then I cannot possibly be held responsible for it. Conversely, however, I suggest that we must take responsibility for all determinate judgements as long as we have direct control over them. I thus claim that Kant endorses the following principle:

6.1. Principle of doxastic responsibility

(PDR) S is responsible for the assent that p iff, and because, (i) S has direct voluntary control over the assent that p (as determined by PVC), and (ii) S’s assent that p is assertoric.⁶⁶

PDR easily explains (2) and (3). We are responsible for all Belief and persuasion because Belief and persuasion are assertoric judgements that are under our direct voluntary control.⁶⁷ But what about (4)? To avoid contradiction with (1), we must restrict (4) to assertoric approval of false cognition: we are responsible for all false assertoric cognition.⁶⁸ Since Kant is an infallibilist about sufficient objective grounds, these false judgements must rest on insufficient objective grounds; and since Belief does not admit of alethic evaluation, the judgements must be persuasions. Indeed, Kant argues that ‘all error arises only because we confuse the causes of approval, which lie in the object itself, with the grounds which lie in the subject’ (24:146; my emphasis) – that’s just persuasion. Since we are responsible for all persuasions, we are a fortiori also responsible for all false cognition.⁶⁹

What’s more, PDR has an important implication for knowledge. While, as noted, there must be a weak sense in which knowledge is imputable to us, I suggest that there is a stronger sense in which we are not responsible for our knowledge. After all, we have no direct control over knowledge. Sufficient objective grounds are external to us and ‘wrest and force’ the assent from us (24:158). Note, however, that this implication is perfectly consistent with the four claims about our doxastic responsibility noted above. For knowledge is neither Belief, nor persuasion, nor false cognition. And Kant also never claims that knowledge deserves praise.⁷⁰

Although we’re not responsible for knowledge, we are, of course, responsible for our knowledge-directed inquiry. Kant thinks our will has an indirect, or ‘mediate’ (24:434), influence over knowledge by guiding our inquiry.⁷¹ This guidance is rational when we will to extend our knowledge by seeking further evidence, as we do, for example, when we conduct experiments. The guidance is irrational, however, when we will to limit our knowledge by refusing to seek further evidence. Kant describes the case of a ‘professor in Padua’, an Aristotelian astronomer, who refused to look through the telescope so that he wouldn’t see the moons of Saturn (24:424).

The main conclusion of this paper is that Kant endorses both PVC and PDR. This conclusion stands in sharp contrast to the readings defended by Cohen (2013, 2014, 2021) and Kohl (2015). For not only does PDR locate the ground of our doxastic responsibility in our direct voluntary control over (assertoric) assent – as opposed to our indirect voluntary or intellectual control –, but together with PVC, PDR also entails different verdicts about our doxastic responsibility. Let me illustrate this point by focusing on Cohen’s account of epistemic maxims, which purports to ‘delineate the domain of epistemic responsibility’ (2013, 42).

On Cohen’s account, we have ‘epistemic responsibility for all our [empirical] judgments’ (2013, 40). I contend that this both over- and understates our doxastic responsibilities. Contrary to Cohen, I have argued that we are responsible neither for knowledge nor for opinion. And while Cohen could just dig in her heels in the case of knowledge – after all, Kant is silent on whether or not we are responsible for knowledge –, she explicitly contradicts Kant’s claim that we are not responsible for opinion. But to be fair to her, I don’t see this as a deep disagreement since she could simply add the second condition of PDR to her account.

The main disagreement concerns Belief. Drawing on my discussion in the previous section, I maintain that Cohen’s account of epistemic maxims cannot explain why we are responsible for Belief: maxims must be universalisable, yet Belief is private. Indeed, Cohen explicitly limits the scope of her account to ‘empirical beliefs’ – in the contemporary sense of ‘belief’ – ‘beliefs that are always, at least in principle, susceptible of evidential support’ (2013, 34). This scope has a prominent predecessor in Williams, who likewise limits his seminal discussion of doxastic voluntarism to empirical, or what he calls, factual belief: ‘I am not going to take religious and moral beliefs, but cases of more straightforward factual belief’ (1973, 136).⁷²

Yet this contemporary scope is all too narrow and fails to recognise the underlying unity of Kant’s theory of doxastic agency. For while Cohen seems to agree that we are responsible for Belief, she cannot explain this responsibility with reference to our epistemic maxims, but must instead admit that Belief is ‘voluntarily acquired’ (2013, 34n). Her theory of doxastic responsibility thus ends up being disjunctive: epistemic maxims explain why we are responsible for empirical assent; direct voluntary control explains why we are responsible for Belief. My reading, by contrast, suggests that Kant held a unified theory of doxastic responsibility: all doxastic responsibility is grounded in our direct voluntary control over (assertoric) assent.

Acknowledgments

I thank Alix Cohen, Hatice Kaya, and the anonymous reviewer for the many helpful comments on earlier drafts of this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 References to the Kritik der reinen Vernunft (1998) are to the standard (A) and (B) pagination of the first and second editions. All other citations of Kant are to the volume and page of Kants gesammelte Schriften (1900–). Reflections are further cited by their R-number. References to Meier are to the paragraph of his Auszug aus der Vernunftlehre (1752). Translations are my own.

2 For a further discussion of the difference between direct and indirect voluntary control, see Alston (1988).

3 Cohen (2013) suggests that our doxastic responsibility is grounded not only in our choice of epistemic maxims, but also in our choice to suspend or ‘withhold judgment’ (40). The former, she suggests, gives us ‘[i]ndirect positive doxastic control’ (41), the latter ‘[i]ndirect negative doxastic control’ (35). I agree with Cohen that suspension is a form of negative voluntary control, but I will argue later that Kant characterises this control as direct rather than indirect. Moreover, I will suggest that Kant identifies suspended judgements with opinion.

4 I capitalise ‘Belief’ to indicate that ‘Glaube’ is a technical term for Kant, to be distunguished from our contemporary notion of small-b ‘belief’.

5 My reading draws extensively on Kant’s unpublished material, especially the notes on logic in Volumes 16 and 24 of the Academy Edition. In view of this, I was encouraged by the anonymous reviewer to indicate which parts of my reading are additionally supported by published writings. The answer to this question depends on whether we count the Jäsche Logic among the published writings. If we do, then all major claims of this paper are supported by published writings. If we don’t, then I can still support my discussion of opinion and Belief in Sections 4 and 5. For not only does Kant align opinion with problematic judgement (A822/B851) and state that we have direct voluntary control over problematic judgement (A75/B101), but he also specifies that we have direct voluntary control over Belief (5:144–6; 5:469n; 6:70; 7:20). That said, my claims about knowledge and persuasion in Sections 2 and 3 do largely depend on the logic notes. But this shouldn’t put my reading at a disadvantage. The logic notes are also the textual basis for other readings, because they offer a more nuanced account of persuasion, and because they are the only place where Kant denies the will’s influence on our assent (9:73; 24:156; 24:545; 24:859). In fact, it is this denial that fuels readings that reject direct doxastic voluntarism; and so, it would be these readings, if any, that are at a disadvantage by basing their main claim on unpublished materials. More generally, however, we can say that no discussion of Kant’s doxastic (in)voluntarism will be able to sidestep the unpublished material altogether.

6 Chignell (2007a, 2007b) sketches a first outline of Kant’s doxastic voluntarism. I agree with Chignell that Kant denies voluntary control over knowledge and grants it over Belief. But I move beyond his reading by arguing that, for Kant, all persuasion and all opinion are also under our voluntary control.

7 I read ‘convincing proofs’ as ‘proofs that bring about conviction’. Note that Kant has a technical notion of conviction, or more precisely ‘logical conviction’ (9:72), as denoting assent from sufficient objective grounds: ‘objectively sufficient […] assent is called conviction‘ (A820/B848). I will argue shortly that sufficient objective grounds are infallible. Therefore, when Kant speaks of ‘convincing proofs’, he must mean those proofs that provide infallible grounds of assent.

8 Following Chignell (2007a, 2007b, 2021), Cohen (2021) has argued that, in Kant’s view, some knowledge rests on fallible grounds. While I have tried to show in my Benzenberg (ms) that such a fallibilist reading faces severe textual challenges, the reading also doesn’t help in the present context. For if some knowledge were based on fallible grounds, then some knowledge would (if I am otherwise right) be subject to the direct influence of our will. Fallibilist readings would, therefore, extend the direct influence of our will. In the same vein, one might also try to extend the direct influence of our will by arguing that infallible grounds for assent are not sufficient to force assent. In addition to being infallible, our grounds only force assent if they are ‘apparent [augenscheinlich]’ and if we ‘reflect upon them’ (24:157). See also (24:160) and (24:546). I won’t pursue these further restrictions in this paper; but see Kohl (2015) for an illuminating discussion of the reflection required for empirical assent.

9 The second condition is needed because, if S knew that ¬p, then S would be forced to assent to ¬p, and so could neither choose to assent to p nor suspend judgement about whether p. In this case, then, S would also lack direct voluntary control over the assent that p.

10 In fact, Kant specifies that the grounds of knowledge guarantee ‘certainty (for everyone)’ (A822/B850). This brings me to another disagreement with Cohen. Cohen (2013) can be read as saying that, according to Kant, ‘judgement can always be suspended’ (40), even if our grounds guarantee ‘objective certainty’ (38). But this is evidently not the case. The passages cited throughout this section demonstrate that, for Kant, we cannot suspend judgement if our assent rests on sufficient objective grounds that guarantee certainty.

11 See also (24:157), (24:545), (24:859), as well as Kohl (2015, 308).

12 See also (R2452, 16:375), (24:218), and (24:437).

13 See also (R2488, 16:391), (24:559), (24:639), and (24:850).

14 See also (A820/B848), (5:461), (9:73), (R2459, 16:379), (R2465, 16:382), (R2486, 16:389), (R2488, 16:391), (20:297), (24:542), (24:547), (24:559), (24:647), (24:747), (24:849), (24:889–90), and (25:1451). For a generalised account of persuasion in Kant, see Gava (2024).

15 See also (R2515, 16:399), (R2528, 16:406), (R2530, 16:407), (R2533, 16:408), (R2539, 16:409), (R2550, 16:411), (24:162), and (24:863).

16 Strictly speaking, prejudices aren't persuasions but the propensity (Hang) to persuasion: ‘A prejudice […] [is] the propensity to persuasion’ (24:547). See also (9:75), (R2529, 16:406), (R2530, 16:407), (R2541, 16:410), (R2550, 16:411), and (R2647, 16:411). Since this distinction makes no difference to my argument, I will identify prejudice and persuasion to reduce complexity.

17 See also (24:226) and (24:890).

18 For a further discussion of the sources of persuasion and prejudice, see (R2519, 16:403), (R2524, 16:404), (R2531, 16:407–9), (R2540, 16:410), (R2550, 16:411), (R2571, 16:424), (24:165–6), (24:227), (24:426), (24:547–8), (24:641), (25:46) as well as Cohen (2014, 323), (2021, 690).

19 For other examples, see (R2562, 16:417), (24:436), (24:747), (25:42), and (25:904).

20 On the lack of reflection in prejudice, see also (R2520, 16:403), (R2521, 16:403–4), (R2524, 16:404), (R2697, 16:474), (24:162), (24:165–7), (24:547), (24:747), and (24:865).

21 See also (R2519, 16:403), (R2536, 16:408), (24:161), (24:165–7), (24:641), (24:896), and (27:131).

22 For a similar model of reflective control, see Alston (1988, 262), Cohen (2013, 38), and Kohl (2015, 311).

23 Thanks to Alix Cohen for pushing me on the first two objections, and thanks the anonymous reviewer for pushing me on the third objection.

24 A special difficulty arises because inclination, which is the ground of some persuasion, also hinders reflection: ‘[I]nclination also causes us to always examine and investigate from […] the side where we wish it to be so […]; there is no own reflection’ (24:167). On the difficulties of reflection, see also (24:146), (24:186), (24:226), (24:747), (25:268), and (28:380).

25 On the touchstones that distinguish persuasion from conviction, see also (6:305), (9:73), (R2450, 16:373), (R2451, 16:374), (R2493, 16:393), (R2756, 16:499), (24:149), (24:242), (24:734), and (24:850–3).

26 Kohl puts the objection succinctly: ‘when people find that they hold beliefs they cannot support with reasons, they are typically alienated from (and not free with regard to) their doxastic states’ (2015, 311).

27 One might further object that there is a key difference between the cycling case and suspension. While I know that I can choose to cycle once the bike is unlocked, the same isn’t true of suspension. I don’t know in advance whether I will be able to choose suspension; after all, it may turn out that my assent rests on sufficient objective grounds, and so cannot be suspended. However, this complication doesn’t undermine our direct voluntary control. Consider a revised analogy, where the unlocked bike moves only as long as its chain is not jammed (which I don’t know in advance). Yet on the condition that the chain is not jammed, it would seem to be under my direct voluntary control to cycle once I unlock the bike. I contend the same goes for the suspension. If my assent is a persuasion based on insufficient objective grounds, it is under my direct voluntary control to suspend once I reflect.

28 This is true even if we have always followed the maxim in question; and even if the action requires training. After all, I still have direct voluntary control over whether I cycle to work, even if I have done so every time I have unlocked my bike; and even if cycling requires training. Similarly, I have direct voluntary control over the suspension of my persuasion, even if I have always suspended my persuasions upon realising that its grounds are objectively insufficient; and even if, as Kant notes, suspension ‘requires a trained power of judgement’ (9:74–5).

29 See also (A822/B850), (9:66–7), (R2459, 16:378), (24:227), (24:541), (24:637), (24:736), and (24:850).

30 ‘An opinion is never a certainty, the opposite can always take place’ (24:219). See also (24:160) and (24:736).

31 See also (9:66–7), (R2459, 16:378–9), and (R2501, 16:395).

32 On the nature of provisional judgements and how they differ from determinate judgements, see also (9:66–7), (9:73–4), (R2444, 16:370), (R2459, 16:378–9), (R2505, 16:396), (R2510, 16:399), (R2531, 16:407), (24:545), (24:555), (24:640), and (24:860–2).

33 See also (R2462, 16:380), (R2464, 16:382), (24:161), (24:228), (24:425–6), (24:545), and (24:736).

34 For the same account of suspension, see (9:74), (24:736), and (24:860).

35 On the relation between opinion and suspension, see also (9:66–7), (R2450, 16:374), (R2459, 16:378–9), (R2511, 16:399), as well as my Benzenberg (forthcoming).

36 See also (A74–6/B100–1), (6:153–4n), (9:66), (R2449, 16:372–3), (R2474, 16:385), (R2475, 16:386), (R2501, 16:395), (24:224), (24:541), (24:637–8), (24:732), and (24:850).

37 See also (R2510, 16:399) and (24:545).

38 For other non-assertoric uses of ‘assumption’, see (6:153–4n), (R2458, 16:378), (R2462, 16:380), (R2505, 16:396), (24:146), (24:160), (24:219), (24:544), (24:557–9), (24:639), and (24:735). Following Meier (§ 168), Kant, at times, also identifies ‘assumption’ with ‘assent’. See (24:202), (24:213), and (24:242–3). Both usages advise against Chignell’s (2007a, 2007b) translation of ‘annehmen’/‘Annahme’ as ‘accept’/‘acceptance’, which to my (non-native) ears would signal an assertoric commitment.

39 On the relation between provisional judgement and prejudice, see also (9:75), (R2523, 16:404), (R2540, 16:410), (24:161), (24:425–6), (24:547), (24:640), and (24:737).

40 In the case of non-hypothesis opinions, we always have the choice between p and ¬p.

41 On this point, see also (R2511, 16:399), (R2681, 16:469), (R2692, 16:472), (R4383, 17:528), (24:219–20), (24:230), (24:748), (24:850), (24:885–9), (28:697), and (28:793).

42 Responding to Meier (§ 190), Kant contrasts the sense in which hypotheses are at will or arbitrary (form the Latin ‘arbitrium') with the sense in which mathematical propositions are arbitrary. The latter ‘makes that something is true […], e.g., in the case of the circle, that one divides it into 360 degrees’ (24:230; my emphasis). See also (24:390), (24:442), (24:560), and (24:892). Hypotheses are not arbitrary in this strong sense.

43 Alston (1988) acknowledges that hypotheses can be under our direct voluntary control, but denies that they are instances of belief (in the contemporary sense of the term). For example, he claims that ‘a scientist can adopt “as a working hypothesis” the proposition that the atomic nucleus is positively charged’ without needing to ‘form the [corresponding] belief’ (267). One might take this to show that Kant’s notion of assent simply differs from our contemporary notion of belief, in part because of its different voluntaristic profile (cf. Chignell 2007a, 341–4, 2007b, 56). I think this conclusion is premature. In my Benzenberg (forthcoming), I defend the view that Kant’s notion of assent is to be identified with our contemporary notion of belief.

44 Kant maintains that it is only rational to assume the most probable opinion or hypothesis. See (R2450, 16:373), (R2480, 16:388), (24:219), (24:227–8), (24:241–2), and (24:860). But this seems overly restrictive. It may be good for an epistemic community that at least some people explore low-probability hypotheses. After all, we learn more when a low-probability hypothesis turns out to be true than when a high-probability hypothesis turns out to be true.

45 See also (A293/B350), (5:465), (9:82), (R2452, 16:375), (R2583, 16:427), (R2595, 16:434), (R2602, 16:436), (24:145), (24:194–6), (24:433), (24:436), (24:880), and (24:884).

46 For a full defence of the view that all knowledge, for Kant, requires probability 1, see my Benzenberg (ms).

47 The anonymous reviewer has raised the excellent question of whether my discussion in this section extends to all species of Belief in Kant. Although the text remains silent on this point, I suggest that pragmatic Belief (pragmatischer Glaube) and doctrinal Belief (doktrinaler Glaube) should also be under our direct voluntary control because they share the features that make moral Belief directly voluntary: like moral Belief, they do not rest on sufficient objective grounds, which would force our assent; and like moral Belief, they are assumed as means to realise rational ends we have set for ourselves (A823–7/B851–5). However, my discussion does not extend to historical Belief (historischer Glaube), which, depending on the strength of its objective grounds, is either a species of opinion (R2470, 16:383) or a species of knowledge (24:896). As such, historical Belief should share the voluntary profile of either opinion or knowledge as I have developed it in Sections 2 and 4.

48 See also (R2451, 16:374), (R2459, 16:378–9), (R2462, 16:380), (24:150), (24:227–8), (24:543), (24:732–4), (24:851), and compare footnote 33.

49 Moral certainty, then, is certainty in the broader sense of not being able to assume the opposite: ‘With certainty, it is not at all possible to assume the opposite’ (R2468, 16:383). Indeed, moral certainty is yet stronger than logical certainty because we don’t even listen to opposing grounds: ‘Practical Belief is often stronger than all knowledge. With the latter, you can still hear grounds to the contrary, but not with practical Belief’ (24:543). See also (24:148), (24:421), and (24:732).

50 Belief, therefore, is not just a decided judgement, but I ‘decided myself to it’ (24:228). See also (5:144–6), (5:469n), (6:70), (7:20), (9:67), (9:68n), (R2480, 16:388), (R2487, 16:390), (24:148), (24:228), and (24:734–6).

51 Kant’s account of Belief, therefore, presents a counterexample to Alston’s principle that voluntary control requires that we could have chosen otherwise. Alston claims that, for a belief A (in the contemporary sense of ‘belief’), ‘the power to choose A at will is the power to determine at will whether it shall be A or […] not-A’ (1988, 261; my emphasis). Kantian Belief, however, is under our voluntary control even in the absence of alternative possibilities.

52 On the difference between contingent and necessary hypotheses, see also (5:143), (R2692, 16:472), (R5624, 18:260), (R6236, 18:520), (24:750–1), and (28:793).

53 See also (A820/B848), (R2450, 16:374), (R2459, 16:378), (24:198), as well as Chignell (2007a, 326, 2007b, 39).

54 Heteronomy has its source in an external object: ‘If the will seeks the law that is to determine it anywhere other than in […] its own legislation, in the nature of any of its objects, heteronomy always results’ (4:441; my emphasis).

55 Kant identifies positive freedom and autonomy: ‘Freedom and self-legislation of the will are both autonomy, and, therefore, interchangeable concepts’ (4:450).

56 See also (A828–9/B856–7), (9:70), (R2492, 16:393), (R2498, 16:394) (R2627, 16:442), (24:732), as well as Gava (2024).

57 See also (A820–1/B848–9), (R680, 15:302), (R714, 15:317), (R2450, 16:373), (2466, 16:382), (R2489, 16:391), and (24:150).

58 See also (5:144–6), (6:70), (24:542–3), and (24:638).

59 I am grateful to the anonymous reviewer for pushing me on this point.

60 Also consider this Reflexion from the ψ-period: ‘The assent from a ground of cognition that is subjectively but not objectively sufficient is Belief. (^s is not communicable demonstrable.) (^s private judgement.)’ (R2477, 16:387; my emphasis). The wording of this Reflexion parallels Kant’s wording in the first Critique when he discusses persuasion: in persuasion ‘the ground of the judgement […] lies only in the subject’, which implies that ‘such a judgement has only private validity, and the assent cannot be communicated’ (A820/B848; my emphasis). The parallel wording might be taken to suggest that grounds of Belief are privately valid in the same sense that grounds of persuasion are privately valid. Indeed, Kant states more generally that subjective grounds have only private validity in (A820/B848), (A822/B850), (R680, 15:302), (R714, 15:317), (R1488, 15:727), and (R6935, 19:210).

61 Kant more generally contrasts ‘private’ and ‘public’ in (5:214), (R1513, 15:839), and (R2519, 16:403). For the contrast between ‘private’ and ‘universal’, see (7:329), (R541, 15:237), (R876, 15:384), (R1511 15:836), (R1821, 16:128), (R1829, 16:130), (R1994, 16:187), (R1995, 16:187), and (R6935, 19:210).

62 If nothing else, it is not at all obvious from which epistemic maxim private Belief should follow. For example, Belief cannot follow from the maxim Cohen claims is universally valid, namely that ‘the degree of certainty of my belief ought to be proportionate to the evidence I possess’ (2014, 326). Belief involves the highest degree of certainty regardless of the evidence available (A823–4/B851–3).

63 Compare footnote 8.

64 Kohl claims that assent ‘result[s] from imputable actions of thinkers. These actions and their effects are imputable because thinkers have freedom of […] thought that consists, negatively speaking, in the absence of the determination of acts of thought by sensible […] causes’ (2015, 307; my emphasis).

65 This commitment should exclude historical Belief. See footnote 47.

66 Apodictic assent is also assertoric because the third modality of judgement entails the second.

67 This points to a fundamental difference between assent and action. In the case of action, we have only two options: commission or omission; and we are responsible for both. In the case of assent, however, we have three opinions: assent, dissent, and suspension (cf. Meier § 168). While we are responsible for (assertoric) assent and dissent, Kant seems to think that suspension stands outside the scope of our doxastic responsibility. Cohen’s global account of doxastic responsibility will have difficulty explaining this responsibility-free zone.

68 This fits with Kant’s claim that ‘a proposition is 1: true & 2 false 3 at will [willkürlich]’ (24:236). If propositions at will denote the free provisional judgements of opinion, then Kant must think that opinion, qua opinion, cannot even be false; it is not a bearer of truth or falsity.

69 Fallibilist readings will have a hard time explaining why all false judgements are epistemically blameworthy. For false judgements that are based on sufficient objective grounds seem to be epistemically blameless. The epistemic agent was just unlucky that the proposition turned out to be false.

70 The involuntariness of knowledge has an important implication for the norms of assent. For even if we bracket non-evidentially justified assent, like Belief, there cannot, on Kant’s view, be a norm that tells us to proportion the strength of our assent to the evidence. Such a norm would state, for all propositions p and degrees d: if p is supported by evidence of strength d, then we ought to assent to p with strength d. Yet on Kant’s account, ‘ought’ not only implies ‘can’ but also ‘might not’ (4:413). If, however, knowledge is involuntary, then there is one degree, d, for which we cannot but proportion the strength of our assent to the evidence, namely when we have sufficient objective grounds and so our evidence is maximally strong.

71 See also (9:74), (24:158), (24:545), and (24:859).

72 Kohl likewise limits his reading of Kant to ‘doxastic free agency, and the spontaneity in empirical thought’ (2015, 301; my emphasis).