Ramsey’s Reliabilism

2.1. A Human Logic

The section in question is titled “The Logic of Truth.” Ramsey is interested in a human logic, which isn’t “reducible to formal logic” and “is concerned not with what men actually believe, but what they ought to believe, or what it would be reasonable to believe” (TP 89). In other words, his concerns are epistemological in nature.

Ramsey thinks that “we can identify reasonable opinion with the opinion of an ideal person in similar circumstances” (TP 89). By “an ideal person,” he has an mind an ideal human being and not an omniscient agent. As he puts it, “the highest ideal would be always to have a true opinion and be certain of it; but this ideal is more suited to God than to man” (TP 89–90). According to Ramsey, the “human mind works essentially according to general rules or habits” (TP 90).⁶ And the ideal person is one whose mental habits are “in a general sense…best for the human mind to have” (TP 90–91). Ramsey admits that the question of which habits are best is “large and vague,” but he clarifies that whether a habit is any good depends on whether it’s “useful”—on whether it’s “conducive or otherwise to the discovery of truth or to entertaining such degrees of belief as will be most useful” (TP 90).

Set aside degrees of belief for now. Ramsey’s suggestion is that reasonable opinions are produced by truth-conducive mental habits, which an ideal person would have. At this point, we can already see that his account of reasonable belief is similar to Goldman’s (1979) process reliabilist account. Roughly, according to process reliabilists, a belief is justified if and only if it’s produced by a reliable belief-forming process, where such a process tends to produce a high ratio of true to false beliefs. For example, suppose I’ve a visual experience as of a hawk and consequently form the belief that there’s a hawk. According to process reliabilists, my belief is justified if it’s formed via a reliable process—say, via careful observation in good light. But it’s unjustified if it’s formed via an unreliable process, say, that of forming beliefs via wishful thinking. Similarly, according to Ramsey, a belief formed in a certain situation is reasonable if and only if an ideal person in a similar situation would form that belief—that is, if and only if the belief is formed via a truth-conducive mental habit.

Just as for process reliabilists, the justifiedness of a single belief derives from the reliability of the process that produced it, for Ramsey, the reasonableness of a single belief derives from the reliability of the habit that led to it. Ramsey thinks that, fundamentally, a single belief is to be evaluated in terms of its usefulness, where a useful belief is true.⁷ But he thinks that though, “given a single opinion, we can only praise or blame it on the ground of truth or falsity,” we may nonetheless “praise or blame opinions derivatively from our praise or blame of the habits that produce them” (TP 92, my emphasis). In particular, he takes a belief to be reasonable insofar as it’s produced by a truth-conducive habit and not reasonable otherwise. Though Goldman (1979) uses the terms “justified belief” and “belief-forming processes,” whereas Ramsey uses the terms “reasonable belief” and “mental habits,” the similarities between their accounts stand out.

That said, Ramsey’s reliabilism bears even greater affinity with Alston’s account of justified belief. Various features of their respective accounts are strikingly similar, and to show this, I’ll first describe Alston’s account in some detail. We’ll then look at certain passages in “Truth and Probability” that support my claim.

2.2. Ramsey and Alston on Justified Belief

According to Alston (1988; 2005), a belief b is justified if and only if it’s based on a ground g, and the objective probability of b being true given that it’s based on g is sufficiently high, where objective probability is understood in terms of hypothetical relative frequencies.

That’s a lot to digest. Let’s start more slowly and ask: What’s the ground of a belief and what is it for a belief to be based on a ground?

Alston (2005) holds that the ground of a belief is the input to a belief-forming process that produces that belief. Such input comes in the form of “something psychological—some psychological state or process” such as a belief, a memory, or a visual experience (Alston 2005: 83, 122, 133–134). Suppose, for instance, that a belief-forming process takes our visual experience as of a cat on the tree as input and produces a belief that there’s a cat on the tree as its output. Then the visual experience is the ground of the corresponding belief, and such a belief is based on the ground in question.

Further on Alston’s view, our belief that there’s a cat on the tree is justified if and only if the objective probability of our belief’s being true, conditional on its being based on the visual experience in question, is sufficiently high. He proposes that we understand objective probability in terms of hypothetical relative frequency, where the reference class is determined in part by the kind of input or ground that feeds into the belief-forming process.

In our example, to determine the objective probability in question, we first consider a class of cases similar to that in which our belief is formed. Such cases may be ones in which we are standing at a similar distance from the tree, have similarly good eyesight, are looking at the cat under similar lighting conditions, etc. The relevant objective probability is sufficiently high just in case there’s a sufficiently high proportion of such cases in which the output belief is true.⁸

The example above involves non-inferential justification. But we can extend Alston’s account to handle inferential justification. After all, Alston allows the ground of a belief to be other beliefs. Suppose that based on a belief that it’s raining and a belief that if it’s raining, the streets are wet, we form the belief that the streets are wet. Is the inferred belief justified?

The answer depends on answers to the following two questions. First, is the belief-forming process in question conditionally reliable—does it tend to produce true beliefs conditional on the input beliefs being true? Second, are the input beliefs justified? For any inference, if the input beliefs are justified but the inferential process is not conditionally reliable, the output belief will not count as being justified. And likewise if the inferential process is conditionally reliable, but the input beliefs are unjustified. In our current example, the inferential process is conditionally reliable—given that the inputs are true, the output has to be true. Whether the output is justified then boils down to whether the inputs are justified.

I won’t attempt to spell out further how Alston may account for inferential justification. It suffices to note that he has the resources for a unified account of non-inferential and inferential justification.⁹ As we’ll see, Ramsey gives a similar treatment of both kinds of justification. He first discusses what he calls “habits of inference” before discussing what he calls “habits of observation and memory” (TP 92). But his remarks about the latter make the similarities between his account and Alston’s especially striking. I’ll discuss those remarks first.

According to Ramsey, his account of inferential habits

can be applied … also to habits of observation and memory; when we have a certain feeling in connection with an image we think the image represents something which actually happened to us, but we may not be sure about it; the degree of direct confidence in our memory varies. If we ask what is the best degree of confidence to place in a certain specific memory feeling, the answer must depend on how often when that feeling occurs the event whose image it attaches to has actually taken place. (TP 92)

What does Ramsey mean by placing a degree of belief or confidence in a memory feeling? To have a degree of belief or confidence is to have a degree of belief or confidence in something being true. But a memory feeling is not in itself something that’s truth-apt. Nonetheless, since it’s connected with an image that represents something to be true, I take it that by “best degree of confidence to place in a memory feeling,” Ramsey means the best degree of confidence to place in the corresponding representation being true.

Though Ramsey focuses on degrees of belief, we may extract an account of reasonable all-or-nothing belief from his remarks above. (I’ll return to degrees of belief in §2.3.)

For Ramsey, all-or-nothing beliefs are reasonable insofar as they are produced by a truth-conducive habit. This holds for memory beliefs formed on the basis of memory feelings too. Suppose that we’ve a memory feeling in connection with an image that represents that we had pancakes for breakfast. Suppose further that a certain mental habit produces the belief that we had pancakes on the basis of such a feeling. If we ask whether it’s best for us to form this belief on the basis of the feeling, we might imagine Ramsey holding that the answer must depend on “how often when that feeling occurs the event whose image it attaches to has actually taken place” (TP 92). Consider relevant cases in which the mental habit in question leads us to form a belief on the basis of a corresponding memory feeling. If in a high proportion of such cases the belief formed is true, we may praise the habit for being truth conducive. Derivatively, we may then praise the belief that we had pancakes for breakfast and deem it reasonable.

Ramsey’s very brief remarks on habits of observation and memory show certain parallels between his account of reasonable belief and Alston’s account of justified belief.¹⁰ Alston would hold that to find out whether a memory belief based on a particular ground is justified, we should consider similar cases in which the relevant belief-forming process takes as input a similar ground and produces as output a similar belief.¹¹ He would then hold that the memory belief is justified just in case the proportion of cases in which the relevant output is true is sufficiently high. Now, substitute “reasonable” for “justified,” “habit” for “belief-forming process,” and “memory feeling” for “ground.” The account that emerges is strikingly similar to Ramsey’s.

Let’s now turn to Ramsey on habits of inference. He writes:

Let us take a habit of forming opinion in a certain way; e.g. the habit of proceeding from the opinion that a toadstool is yellow to the opinion that it is unwholesome. Then we can accept the fact that the person has a habit of this sort, and ask merely what degree of opinion that the toadstool is unwholesome it would be best for him to entertain when he sees it; i.e. granting that he is going to think always in the same way about all yellow toadstools, we can ask what degree of confidence it would be best for him to have that they are unwholesome. And the answer is that it will in general be best for his degree of belief that a yellow toadstool is unwholesome to be equal to the proportion of yellow toadstools which are in fact unwholesome. (TP 91)

Again, Ramsey focuses on degrees of belief, though his remarks may also apply to (all-or-nothing) beliefs. Suppose we’ve a habit that proceeds from a belief that a toadstool is yellow to the belief that it’s unwholesome. Extending Ramsey’s remarks to beliefs, one might hold that whether a belief produced by this habit is reasonable depends on whether the proportion of yellow toadstools that are unwholesome is sufficiently high.

Compare this account to Alston’s. Ramsey’s habits of inference and Alston’s inferential processes both take as input beliefs and deliver as output inferred beliefs. Now, consider a belief-forming process that takes as input a belief that a toadstool is yellow and produces as output a belief that the toadstool is unwholesome. What makes the output belief justified? According to Alston, the input belief must be justified. Further, the process in question has to be conditionally reliable. So a sufficiently high proportion of cases in which we base our belief that a toadstool is unwholesome on the true belief that it’s yellow must be ones in which it’s indeed unwholesome.

On the face of it, there’s an important difference between Alston’s account and Ramsey’s (as it’s presented so far). Given the latter, so long as the relevant proportion of yellow toadstools that are unwholesome is high enough, the inferred belief counts as reasonable—never mind whether the input belief is reasonable. On this score, Alston’s account seems to fare better than Ramsey’s. Nonetheless, to be charitable, one might take Ramsey to have made the simplifying assumption that the relevant input beliefs are all reasonable.¹² This would bring his account of inferential justification closer to Alston’s.

There’s another difference. On Ramsey’s account, whether the inferred belief is reasonable depends on the proportion of yellow toadstools that are unwholesome. But on Alston’s account the relevant proportion is the proportion of yellow toadstools that we believe to be yellow and that are indeed unwholesome. The two can come apart. The first proportion might be high while the second is low—perhaps, for whatever reason, the many yellow toadstools that are unwholesome are invisible to us, and we see only the relatively few yellow toadstools that are wholesome. Yet, the two accounts of inferential justification remain close cousins. We can take a small, gentle step from Ramsey’s account towards Alston’s by praising the inferential habit in the toadstool example if and only if a high proportion of yellow toadstools—that we believe to be yellow—are unwholesome.¹³

2.3. Ramsey and Justified Degree of Belief

Contemporary reliabilists have focused on justified all-or-nothing belief, with relatively little said about justified degree of belief. Only fairly recently have there been more attempts to extend reliabilism to account for the latter. I’ll now argue that Ramsey’s account of reasonable degree of belief anticipates some of these contemporary accounts.

On the face of it, reliabilism as usually formulated is not particularly well suited to accounting for degrees of belief. Degrees of belief are not in general true or false, and it doesn’t seem promising to hold that a justified degree of belief of 0.6, say, is produced by a process that yields a high ratio of true beliefs. Perhaps one might suggest that such a degree of belief is justified if and only if it’s produced by a process that yields true beliefs 60% of the time. But problems remain. First, the suggestion does not deal with cases in which a degree of belief is produced by a process that produces only degrees of belief. Second, there’s something odd about holding that a process produces justified degrees of belief only if it also produces all-or-nothing beliefs that are unjustified. (A process that produces true beliefs merely 60% of the time is presumably unreliable.)

In response, one might suggest that we focus on whether processes that produce degrees of belief are well calibrated. Here’s what van Fraassen (1984: 245) says about the notion of calibration:

[C]onsider a weather forecaster who says in the morning that the probability of rain equals 0.8. That day it either rains or does not. How good a forecaster is he? Clearly to evaluate him we must look at his performance over a longer period of time. Calibration is a measure of agreement between judgments and actual frequencies…This forecaster was perfectly calibrated over the past year, for example, if, for every number r, the proportion of rainy days among those days on which he announced probability r for rain, equalled r.

There’s a good sense in which a calibrated process is reliable. Reliabilists might thus appeal to the notion of calibration to account for justified degree of belief. Say that a process is calibrated just in case, for any y, 100y% of the propositions in which the process causes us to have a degree of belief of y are true. Reliabilists might hold that a degree of belief in p is justified if and only if it’s produced by such a process. Call this account the Simple Calibration Account.

Goldman (1986: 113–115) was perhaps the first contemporary reliabilist to have considered such an account of justified degree of belief.¹⁴ But as it stands, the Simple Calibration Account faces a problem.¹⁵ Consider a weather forecaster who knows that the long-term relative frequency of rain in Bali is, say, 0.7. To ensure that she’s perfectly calibrated, she reports a 0.7 probability of rain every day, disregarding whether it’s the monsoon season or whether the skies are cloudy, etc. Such a weather forecaster is well calibrated, but there’s a sense in which she’s not very good at her job—she’s not as informative as we would like our weather forecasters to be.¹⁶

Now consider a process that produces degrees of belief that match the corresponding relative frequencies in a similarly non-discriminating way. Such a process might be calibrated. But suppose the process ignores relevant evidence that comes one’s way. Say, we receive evidence that suggests that it’s almost certain to rain, but the process continues to produce a degree of belief of 0.7 in rain. The process may remain calibrated (given that the relative frequency of rain is 0.7) even if, in light of the new evidence, we might be hesitant to say that a degree of belief of 0.7 is justified. But the Simple Calibration Account says that it is. Thus, the account does not seem to be able to handle the phenomenon of defeat.

To get around the problem, I’ve suggested elsewhere that a justified degree of belief is produced by a process that is calibrated relative to the evidence or grounds on which the relevant degrees of belief are based (Tang 2016). My account is inspired by Alston’s and extends the latter to handle degrees of belief. In Tang (2016: 80–81), I write:

According to Alston, a belief is justified if and only if it is based on a certain ground, and the objective probability of the belief’s being true given that ground is very high. Now suppose that belief comes in degrees. It is natural to think that the stronger (or weaker) the belief, the higher (or lower) the objective probability in question needs to be for the belief to be justified.

To extend Alston’s account to handle degrees of belief and to deal with the phenomenon of defeat, I go on to propose the following:

For any x, any proposition p, and any subject a, a’s credence of x in p is justified if and only if

(1) it is based on some ground g, where the objective probability of the credence having a true content given (that it is based on) g approximates or equals x, and

(2) there is no more inclusive ground g’ had by a such that the objective probability of the credence having a true content given (that it is based on) g’ neither approximates nor equals x. (Tang 2016: 88)

Suppose that we have a visual experience as of rain and consequently form a degree of belief of 0.9 in rain. On my account, this degree of belief is justified if and only if the following two conditions are met. First, the objective probability that it’s raining conditional on the degree of belief being based upon our visual experience as of rain, should equal or approximate 0.9. Further, there should be no defeater—there should be no more inclusive ground g’ such that the objective probability of rain given that our degree of belief in rain is based on this more inclusive ground neither approximates nor equals 0.9. (Here’s an example of a more inclusive ground: a perceptual experience as of rain and a belief that our eyesight is not reliable.)

My aim is not to evaluate the Simple Calibration Account or to defend my account of justified degree of belief. My aim in describing them is to show that Ramsey’s treatment of reasonable degree of belief anticipates such accounts.

Recall: Ramsey holds that “given a habit of a certain form, we can praise or blame it accordingly as the degree of belief it produces is near or far from the actual proportion in which the habit leads to truth” (TP 92). He adds, “[w]e can then praise or blame [degrees of belief] derivatively from our praise or blame of the habits that produce them” (TP 92).¹⁷

For example, suppose we’ve a memory image as of having had pancakes for breakfast and a reasonably strong feeling that the image represents something that happened. Suppose also that we’ve a habit of memory that, in response to such a memory feeling, produces a degree of belief of, say, 0.9 in our having had pancakes for breakfast. On Ramsey’s view (recall the quote concerning memory feeling), this degree of belief is reasonable—and the habit from which it stems deserves praise—if and only if 90% of the relevant cases in which the habit produces a degree of belief of 0.9 in a similar proposition in response to a similar memory feeling are cases in which the proposition in question is true.

More generally, on Ramsey’s view, a degree belief of x in p that’s based on a memory feeling is reasonable if and only if 100x% of the cases in which the relevant habit takes as input a similar memory feeling and produces a degree of belief of x in a similar proposition are cases in which the proposition in question is true. Though Ramsey focuses on memory feelings, he also talks about habits of observation. Presumably he’ll give a similar account involving perceptual experiences.

Ramsey is essentially suggesting that reasonable degrees of belief are produced by calibrated processes, thus anticipating the Simple Calibration Account. Granted, Ramsey’s account holds that the processes in question should be calibrated with respect to the relevant memory images or perceptual experiences—or to what, following Alston (2005), I’ve referred to as the grounds of one’s degrees of belief. Thus Ramsey also anticipates the account of justified degree of belief in Tang (2016). Though he does not talk about defeat, one may extend his account to handle defeat in the way discussed above.

It’s worth stressing that Ramsey’s account shares an important feature with typical reliabilist accounts in that reasonable belief is spelt out in terms of reliability rather than known reliability. The entry on Ramsey in the Stanford Encyclopedia of Philosophy maintains that in “Truth and Probability,” “Ramsey concludes…with a lengthy discussion on what (in addition) might make a set of partial beliefs reasonable” and “proposes a condition of calibration, or fit with known frequencies”(§6) (MacBride et al. 2023). It then quotes the toadstool example from the same paper to support the preceding claim. I agree that Ramsey proposes a condition of calibration. But I disagree that this condition is fit with known frequencies.

After all, the very notion of calibration concerns fit with frequencies, not fit with known frequencies. Further, in both the toadstool example and the memory example (both discussed in §2.2), Ramsey does not appeal to known frequencies. More generally, Ramsey consistently says that a subject’s degrees of belief should match the relevant frequencies without saying or suggesting that the frequencies should also be known. Granted, given that knowledge is factive, someone who thinks that reasonable degrees of belief should match the relevant known frequencies will also think that they should match the relevant frequencies. But we would also expect Ramsey to state the stronger condition explicitly if he thinks that it’s a necessary condition for reasonable belief.¹⁸

In sum, Ramsey anticipates not just reliabilist accounts of knowledge but also—in some detail—reliabilist accounts of justified belief and degree of belief. In fact, since his 1926 account of reasonable degree of belief precedes his 1929 account of knowledge, one of the earliest versions of reliabilism concerns degrees of belief first and foremost.¹⁹ Remarkably, Ramsey also offers an argument for reliabilism, to which we now turn.²⁰

3.1. The Pragmatist View of Belief

About midway through §5 of “Truth and Probability,” just before Ramsey begins his discussion of reasonable beliefs, he mentions in a footnote that “what follows to the end of the section is almost entirely based on the writings of C. S. Peirce” (TP 90). Ramsey is referring, in particular, to Peirce’s Chance, Love and Logic (henceforth Chance), originally published as “Illustrations in the Logic of Science.”²¹

A key ingredient in Ramsey’s argument for reliabilism is the pragmatist view of belief, which he shares with Peirce, and which Peirce attributes to Alexander Bain (1859) (CP 5.12). Spelling out the view will help us understand why Ramsey thinks that a belief is ultimately to be praised or blamed depending on whether it’s true. That will in turn help us see why he professes to subscribe to “a kind of pragmatism” where “we judge mental habits by whether they work” (TP 93). Focus first on all-or-nothing belief. In his 1927 “Facts and Propositions,” Ramsey writes:

[I]t is possible to say that a chicken believes a certain sort of caterpillar to be poisonous, and mean by that merely that it abstains from eating such caterpillars on account of unpleasant experiences connected with them…[I]t might well be held that in regard to this kind of belief the pragmatist view was correct, i.e. that the relation between the chicken’s behaviour and the objective factors was that the actions were such as to be useful if, and only if, the caterpillars were actually poisonous. (FP 40)²²

In the quote above, Ramsey posits an intimate connection between what a chicken believes, what it desires (to eat caterpillars but not those connected with unpleasant experiences), and what it does due to having such beliefs and desires. Along similar lines, he maintains in “Truth and Probability” that a person will “always choose the course of action which will lead in his opinion to the greatest sum of good,” where “goods” are those “things a person ultimately desires” (TP 69-70). Elsewhere, he holds that a person’s “actions result from his desires and the whole system of his beliefs, roughly according to the rule that he performs those actions which, if his beliefs were true, would have the most satisfactory consequences” (OT 45).²³

Ramsey calls his view of belief “pragmatist.” (I’ll discuss whether the label is apt in §3.3.) In “Facts and Propositions,” he attributes the view to Bertrand Russell, but while it’s debatable whether Russell holds such a view, it’s fairly clear that Peirce does.²⁴ In Chance, Peirce holds that “our beliefs guide our desires and shape our actions” and that when we have a belief, it’s “established in our nature some habit which will determine our actions” (CLL 14–15). Further, he thinks that it’s “certainly best for us that our beliefs should be such as may truly guide our actions so as to satisfy our desires” (CLL 16, my emphasis). To clarify, Peirce does not think that belief will “make us act at once” (CLL 15). Rather, it “puts us into such a condition that we shall behave in some certain way, when the occasion arises” (CLL 15). Likewise, Ramsey writes, “it is not asserted that a belief is an idea which does actually lead to action, but one which would lead to action in suitable circumstances” (TP 66). Ramsey does not specify what he means by suitable. But presumably, the thought is that for a circumstance to be suitable for action, the belief should at least be accompanied by a relevant desire. Suppose we believe that shouting at the top of our lungs will draw attention to ourselves. Without desiring to draw attention to ourselves, we’ll not proceed to shout at the top of our lungs. But should such a desire arise (and we have no countervailing desire), we’d have a circumstance ripe for the relevant action.

3.2. The Argument

What does the pragmatist view of belief have to do with reliabilism? Ramsey does not spell out his argument in full explicit detail. But I’ll argue that Ramsey’s pragmatist view of belief, as well as his view that the ultimate value of belief lies in its connection to action, leads him to maintain that a single belief is ultimately to be praised or blamed depending on whether it’s true. This, coupled with his view that beliefs are produced by mental habits, then leads him to think that insofar as we appraise certain individual beliefs as being reasonable or otherwise, such appraisal is derivative upon our appraisal of the relevant habits, which in turn depends on whether those habits are truth-conducive.

As we’ve seen, Ramsey holds that our desires and beliefs cause us to act in a way such that if our beliefs were true, our actions “would have the most satisfactory consequences” (OT 45). Or as he’d sometimes put it, a true belief is useful in suitable circumstances. But the chicken and caterpillars passage suggests he’d also maintain that a belief is useful in suitable circumstances only if it’s true. In his own words, “[i]t is useful to believe aRb would mean that it is useful to do things which are useful if, and only if, aRb” (FP 40). To illustrate, suppose that in accordance with the pragmatist view of belief, our desire for chocolate and our belief that going to the shop will satisfy this desire lead to us going to the shop. Ceteris paribus, our belief is useful—it’ll lead to useful, desire-satisfying conduct—if and only if true. Or so Ramsey would maintain.

Now, one might point out that even if our belief is false, our desire can still be satisfied by luck—say, there’s no chocolate in the shop but just as we leave the shop a friend comes along and hands us some. But I think Ramsey will deny that our belief is useful in this case. He might grant that our going to the shop has led to satisfactory consequences, but he’d presumably deny that the desire is satisfied via a belief or conduct that’s useful—there’s nothing about the belief itself that helps satisfy the desire in question.

One might also bring up standard examples in which we maintain false beliefs because of the consolation they offer or because they enhance our self-image, thus resulting in us performing better at our tasks. One might think that such beliefs, despite being false, are useful. But Ramsey will deny that these are counterexamples to the equivalence between true and useful belief. In holding that a belief is true if and only if it’s useful, Ramsey is concerned with “primary utility,” where a belief is useful by way of its being true (OT 91). We’ll see more of what he has to say about the notion of primary utility in §3.3. For now, note that given this notion, the right-to-left direction of the equivalence is trivial. But the converse is non-trivial. For instance, if you deny that beliefs interact with desires to bring about action or you think it’s possible to have beliefs without having any desires—see Eriksson and Hájek’s (2007) example of the Zen monk—you’d think that a true belief needn’t be useful.

To clarify, in appealing to the aforementioned connection between truth and usefulness, Ramsey is not appealing to a pragmatist theory of truth, insofar as such a theory takes such a connection to give us a definition or analysis of truth.²⁵ Ramsey maintains that he has “a clear definition of truth” that is found in the “platitude that a belief that p is true if and only if p” (OT 11; 13). This suggests that insofar as he maintains an equivalence between true belief and useful belief, such equivalence is to be understood as extensional, not intensional, in nature. Or as Misak (2016: 224) puts the point, Ramsey’s “pragmatist move…links truth to action, without defining it in terms of action.” The fact that Ramsey understands usefulness as primary utility also shows that he is not defining truth in terms of usefulness, for such a definition would be blatantly circular.

Further, Ramsey explicitly holds that the claim that “a belief that A is B [is] roughly a belief leading to such actions as will be useful if A is B, but not otherwise” is “an absurd answer” to the question “Given propositional reference, what determines truth?” (OT 90–91). He maintains, however, that the view is a “sensible sort of answer” to the question “What constitutes propositional reference?”—that is, to the question “what is meant by saying a belief is a belief that so-and-so” (OT 89–91).²⁶ Here, some commentators have interpreted Ramsey to be putting forward a theory of mental content subsequently known as success semantics—see, for example, Dokic & Engel (2001). But for our purposes, note that one might agree with Ramsey that a belief is useful in suitable circumstances if and only if it’s true without subscribing to such a theory of mental content—one might agree with Ramsey about the connection between truth and usefulness without thereby thinking that the connection on its own suffices to fix the content of belief. (You might think, for instance, that to know the conditions under which our actions are useful, we’d also need to know the contents of our desires.)

We’ve finally clarified the first step in Ramsey’s argument for reliabilism—his claim that a belief is true if and only if it’s useful (in the primary sense and in suitable circumstances—I’ll drop the qualifications in what follows). The discussion of the next step will be relatively brief. Ramsey holds that the ultimate value of a belief lies in its usefulness—as he puts it, “the ultimate purpose of thought is to guide our actions” (GPC 153). Granted, Ramsey’s claim about the ultimate purpose of thought occurs in a context in which he’s considering whether the claim helps support the view that, essentially, all beliefs are beliefs in singular propositions and beliefs in general propositions or variable hypotheticals are “superfluous” (GPC 153). He answers the question in the negative but not because he denies the claim itself. In fact, he thinks that beliefs in general propositions or variable hypotheticals are essential precisely because they allow us to “praise and blame” people by considering “what would have happened if they had acted otherwise” (GPC 153–154).²⁷

We can now put the first two steps together. Given that a belief is useful if and only if it’s true, and given that a belief’s ultimate value lies in its usefulness, it is to be ultimately valued if and only if it’s true. This explains why Ramsey thinks that a single belief is ultimately to be praised or blamed depending on whether it’s true. And since he maintains that beliefs are produced by mental habits, he’d also hold that such habits are to be valued—and deemed “useful”—according to whether they are truth conducive (TP 90–93).

Nonetheless, even if a single belief is to be valued—and thus praised or blamed—according to whether it’s true, we do in practice evaluate beliefs by deeming them reasonable or unreasonable. Given the preceding, such evaluation can only be understood as being secondary in nature. It’s then natural to hold, as Ramsey explicitly does, that such praise or blame derives from our praising or blaming the corresponding belief-forming habits. We praise a belief in the derivative sense—we deem it reasonable—if and only if it stems from a useful, truth-conducive habit.

We’ve seen that Ramsey’s reliabilist account of reasonable belief is based on his pragmatist view of belief, his view about the ultimate value of belief, and his view that beliefs are produced by belief-forming habits. But what about reasonable degrees of belief?

Ramsey holds that we may criticize a mental habit “as being conducive or otherwise…to entertaining such degrees of belief as will be most useful” (TP 90). He also thinks that it is “in general …best” for a person’s degree of belief to match certain relative frequencies—recall his examples involving yellow toadstools and memory feelings (TP 91). If he’s right, then correspondingly, we should praise or blame a habit according to whether it tends to produce degrees of belief that match certain relative frequencies. But why should useful degrees of belief—ones leading to useful conduct—be aligned with frequencies?

In the case of all-or-nothing beliefs, there’s a straightforward sense in which Ramsey may maintain that it’s best for us that such beliefs be true: when true, they help satisfy the relevant desires. But a degree of belief of x in p that matches the relevant relative frequency might lead to us acting in a way that doesn’t satisfy our desires even if we act on it accordingly. For example, suppose our degree of belief in rain is 0.6, the relevant relative frequency is 0.6, and it in fact rains. Still, we might act in a way that benefits us if and only if it doesn’t rain: perhaps we’re out and about without an umbrella and desire not to get wet, but the cost of a new umbrella is exorbitant, and we decide to go without it. This decision may well be rational (in that calculated risks may be rational). In such a case, in what sense is a degree of belief of 0.6 in rain best?

Ramsey would grant that in a single case, acting on a degree of belief that matches the relevant relative frequency might not satisfy your desires. But he holds that “supposing goods to be additive, belief of degree $\frac{m}{n}$ is the sort of belief which leads to the action which would be best if repeated n times in m of which the proposition is true” (TP 84). Why does he think that?

Ramsey doesn’t give an explicit answer. But here’s one on his behalf. Our degrees of belief, in combination with our desires, lead to us performing certain actions. The principle of maximizing subjective expected utility tells us that we ought to perform those actions that maximize subjective expected utility. If we constantly have to choose between two actions A and B, and doing A has greater subjective expected utility than doing B, then from our subjective perspective, doing A will in the long run tend to lead to greater utility than doing B.

Now, suppose that this difference in expected utility is due to our assigning a higher degree of belief to A leading to a desired outcome than to B leading to that same outcome. But suppose that A leads to the desired outcome with a lower frequency than does B. In that case, there’s a mismatch between degree of belief and relative frequency. Then even though from our subjective perspective, doing A maximizes expected utility, it doesn’t do so objectively speaking. In the long run, if we keep choosing A over B, it’s overwhelmingly likely that we’ll in fact be worse off. To avoid this, we should ensure that our degrees of belief match the relevant relative frequencies, in which case our subjective expected utilities will match the objective ones.

3.3. Two Worries

Let me now turn to a couple of worries.

First, one might wonder if Ramsey’s view of belief is accurately labelled “pragmatist.” For instance, Mellor (1990: xix) refers to it as what Ramsey “calls a ‘pragmatist’ (i.e. functionalist) account.” One might read Mellor as suggesting that the view is at bottom functionalist in nature and only pragmatist in name. I agree that a functionalist or proto-functionalist might well accept the view. Nonetheless, the label “pragmatist” remains apt. As mentioned, Ramsey and Peirce share the pragmatist view of belief, which Peirce attributes to Alexander Bain. Peirce also thinks that “pragmatism is scarce more than a corollary” of Bain’s account of belief (CP 5.12).²⁸ Given that in Chance, the pragmatist view of belief is central to Peirce’s pragmatism and given that Peirce is considered to be one of the main founders of pragmatism, the label “pragmatist” is not misplaced.

To be sure, there are different varieties of pragmatism. Talisse & Aikin (2008: 8) point that that A. O. Lovejoy (1908) identified 13 distinct versions before saying that it’s “reasonable to suspect that the number of pragmatisms in currency today far exceeds 13.” As they then go on to note, even the classical pragmatists—Peirce, James, and Dewey—“were often at odds with each other concerning central philosophical questions” (Talisse & Aikin 2008: 8). However, even if your preferred version of pragmatism clashes with what I say above, the following point remains. Ramsey has an early argument for reliabilism that’s based on his view about the value of belief, his view that beliefs are produced by belief-forming habits, and his view on how beliefs and desires are intimately related to action—whether we call this last view pragmatist or not.

Here’s another worry. Given that true beliefs or degrees of belief that match the relevant frequencies help satisfy our desires in the long run, a truth-conducive or well-calibrated habit may well be evaluated positively. But one might think that this shows at best that there’s something pragmatically good—and not necessarily epistemically good—about the habit. One might thus worry that a belief produced by the habit may be praised only for pragmatic rather than epistemic qualities—and consequently, that Ramsey is talking about pragmatic rather than epistemic reasonableness.

But Ramsey is talking about reasonableness in an epistemic sense, even though on his view, the notion can be spelt out in pragmatic terms. To hold that the epistemic is to be spelt out in terms of the pragmatic isn’t to deny that there are epistemic qualities, just as to hold that consciousness is to be spelt out in physical terms isn’t to deny that consciousness exists.

It’d also help to see that Ramsey is not saying that it’s reasonable to believe p just in case believing p leads to the best practical consequences overall. We’ve already seen that by “usefulness,” Ramsey has primary utility in mind. Here’s more of what he has to say about the notion:

To say a man believes in hell means, according to the pragmatists that he avoids doing those things which would result in his being thrown into hell, and which he would not avoid on account of any other consequences they may have. Such conduct will be useful to the man if it really saves him from hell, but if there is no such place it will be a mere waste of opportunities for enjoyment. But besides this primary utility there are other ways in which such conduct may or may not be useful to the man or others; the actions from which a belief in hell would cause him to abstain might bring disasters in their train either for him or for others even in this present life. But these other consequences of the belief, whether useful or not, are clearly not relevant to its propositional reference or truth, and if the pragmatist definition makes them relevant this only shows that the definition must be corrected by qualifying clauses or even abandoned altogether. (OT 91–92, emphases mine)

In the passage, Ramsey distinguishes the primary utility and the non-primary utility of a belief, where the former but not the latter is tightly connected to the belief’s truth.²⁹ While the passage is concerned with truth rather than reasonableness, Ramsey’s maintaining a tight connection between reasonableness and truth conduciveness suggests that he’d also maintain a tight connection between reasonableness and practical consequences in a primary sense.

In particular, Ramsey would hold that a belief’s reasonableness has to do with whether the mental habit from which the belief stems helps lead to positive practical consequences by way of its being truth conducive. On such an account, reasonableness is epistemic in nature. True, though wishful thinking is unreliable, it might sometimes bring one happiness. But such utility, Ramsey would maintain, is non-primary and has nothing to do with whether the beliefs formed via wishful thinking are reasonable. As Misak (2016: 210) puts it, for Ramsey, “the dimension of usefulness relative to epistemic assessment is such that the belief that p will be useful only if p” (Misak’s emphasis).

Ramsey’s view thus accommodates the thought that reasonable beliefs needn’t always be best for one all things considered—reasonable beliefs, like true beliefs, can sometimes bring about negative practical consequences. And unreasonable beliefs might sometimes make one very happy. But why should we be interested in a notion of reasonableness spelt out in terms of primary usefulness instead of overall usefulness? Ramsey doesn’t address this question explicitly, but the following offers an answer coherent with his view.

Ramsey holds that we want habits that “work” (TP 93). Granted, epistemically reasonable beliefs or truth-conducive habits may not always yield the best outcomes. But the concept of reasonable belief understood along Ramseyan lines may still earn its keep. For when judging whether a belief is reasonable along such lines, we’re also judging whether it’s produced by a truth-conducive habit. And it’s helpful to identify such habits as long as there are significantly enough cases in which these habits help serve our desires. A truth-conducive habit might sometimes bring about bad consequences. But it’d still be helpful to identify such habits if, by and large, having truth-conducive habits serves our desires better than not having them. Also, while certain false beliefs might make us feel better or perform certain tasks better, one might think that most workaday beliefs are not like that.³⁰

Contemporary epistemologists have likewise expressed the thought that true beliefs or truth-conducive habits tend to help serve our desires. Consider Alston’s (2005: 31) claim that “where we seek to produce or influence one outcome rather than another, we are much more likely to succeed if we are guided by true rather than false beliefs about the likely consequences of one or another course of action.” Similarly, Kornblith (1993: 373) holds that “it is for pragmatic reasons that truth takes on the importance that it does in epistemic evaluation.” According to him, it’s “of the first importance that our cognitive systems remain suitable for the purpose of performing the relevant cost-benefit calculations” and “what this requires is that our cognitive systems be accurate, that is, that they reliably get us at the truth”; further he holds that “to evaluate our cognitive systems by their conduciveness to truth…is precisely what epistemic evaluation is all about” (372).³¹

I’m not, of course, aiming to provide a full-fledged defence of Ramsey’s argument for reliabilism or of the claim that overall, having truth-conducive habits is better than not having such habits. But the claim has enough initial plausibility, I think, that Ramsey’s argument for reliabilism, as presented, deserves airing.