Home → Research → Basing

The Problem of the Basing Relation

by Ian Evans

Very many of our beliefs are based on other beliefs. Some are based on perceptual experiences and some are based on apparent memories. Others are based on desires, fears, vanity, prejudices and other epistemically disreputable states. Perhaps some of our beliefs are baseless.

This relation between a belief and its basis—call it the basing relation—plays an important role in our everyday attempts to understand why we act and think as we do. It also plays an important theoretical role in epistemology: a pervasive thought is that for a belief to be justified by some evidence, it is at least necessary that it be based on that evidence.Where by “justification” I mean doxastic justification (Bergmann 2006) or well-foundedness (Feldman & Conee 1985). Further, it is common to assume that a belief will be unjustified if it is based on bad evidence or something that is not evidence at all, like a desire or a fear. Recently, the basing relation has figured prominently in a theory of the rationality of belief (Kelly 2002).

We would like a compelling and fruitful theory of the basing relation. But what we would like is not on offer. The theories that have been proposed are untowardly complicated and riddled with difficulties. My aim in this paper is to remedy the situation: I will propose a new theory of the basing relation that is both simple and intuitive. It begins with the pedestrian observation that if your belief b is based on some other state s, then you’re disposed to abandon b if you lose s.Cf. Kelly’s remark that “Whether R is a reason on which S’s [belief] is based, or whether R merely plays a role in the history of S’s [belief] will be reflected in the conditions under which S would (or would not) continue to [believe]” (2002: 175). Beliefs stand or fall with their bases. The theory I offer–the dispositional theory–is an elucidation and refinement of this thought.

Before presenting the dispositional theory, I identify six basic data about the basing relation (§2). I then use the data to assess existing analyses of the basing relation. These fall into two classes: causal theories and doxastic theories. Since the causal theories are the most popular, and since I find that the uninitiated tend towards a naive causal theory, I will spend a good deal of time criticizing such views. The literature contains a plethora of such theories and I cannot discuss each in detail and on its own terms. Instead, in §3, I develop and refute a naive causal theory, show that the best and most well-developed refinement of such a theory–that found in Swain’s (1981)–doesn’t fare much better, and raise a general objection showing that no such theory can succeed. In §4, I briefly consider doxastic theories and find the approach fundamentally flawed. In §5 I motivate and present the dispositional theory and discuss some of the new questions that it raises. I conclude by considering some implications this theory has for existing debates. But first, a few methodological preliminaries.

Methodological Preliminaries

Epistemologists have generally assumed that the basing relation explains the important difference between justified and justifiable beliefs, and so it is primarily epistemologists who have theorized about basing.Pollock & Cruz (1999) mark the distinction using the terms “justified” and “justifiable.” Others contrast “doxastically justified” beliefs with “propositionally justified” beliefs (Firth 1978; Foley 1987; Bergmann 2006), Feldman & Conee talk of “justifed” vs. “well-founded” belief (1985), and Goldman distinguishes between “ex-ante” and “ex-post” justification (1979). Suppose S has excellent, undefeated evidence for p, but believes p not on the basis of that evidence but on the basis of a Tarot card reading. Intuitively, S’s belief is not justified, though it is justifiable. Had her belief instead been based on her evidence, it would have been justified (and justifiable).

This is plausible. But we should be careful not to confuse (a) the basing relation with (b) the relation between justified beliefs and the reasons in virtue of which they’re justified. We might call the latter relation “proper basing”. Proper basing holds only between beliefs and good reasons for those beliefs, while basing can hold between beliefs and any mental states, good reasons or not. I think many epistemologists have failed to keep this distinction in mind, but we can charitably read them as having sought theories of proper basing. Still, this is putting the cart before the horse. Basing is necessary for proper basing. It is also necessary for improper basing. It’s best, then, to seek first an account of the basing relation–we can worry about propriety later. That, anyway, is how I shall approach the problem.

Because I’m not interested here in proper basing, I won’t be assuming, as is common, that the basing relation holds exclusively between beliefs and reasons. As I said at the outset, beliefs can be based on a variety of things–experiences, perceptions, desires, memories, prejudices–that we need not assume are reasons. But I do impose at least the following constraint: one’s belief can only be based on one’s mental states. Theories of the basing relation will be instances of the following schema:

BR S’s belief that p is based (at time t ) on her mental state m iff … .

The basing relation is a psychological relation. There are limits to how much we can theorize about it from the armchair–I will indicate some of these limits in discussing my own theory. But there is quite a lot we can say from the armchair.

Data

I begin with six data for which any theory of the basing relation must account. These desiderata are derived from reflection on cases. In his (2002), Thomas Kelly suggests an additional desideratum not on my list: that other psychological states besides beliefs can have bases. For example, it makes sense to ask what my desires are based on. I agree with Kelly here, but this desideratum seems less central to me than the six I discuss and the paper is long enough as is. It seems to me that the theory I eventually offer satisfies Kelly’s desideratum.

  1. The basing relation has at least the following logical property: two beliefs can be (partly) based on each other, whence the relation is not asymmetric. This sort of thing happens when one completes a crossword puzzle. My belief that one-across is “aero” is partly based on my belief that one-down is “acidic”. My reason for thinking one-across is “aero” is not just that it fits its clue and the allotted spaces; it is also that it fits with “acidic”. Correspondingly, my reason for thinking one-down is “acidic” is not just that it fits its clue and allotted space; it is also that it fits with “aero”. I take the two beliefs to support each other. Of course, the basing relation isn’t symmetric either: from the fact that my belief that p is based on my belief that q it doesn’t follow that my belief that q is based on my belief that p. As for other logical properties (transitivity and reflexivity, to name two), I think it is best to let theory decide; at least, I’m at a loss for cases that tell us much about these other logical properties.
  2. The basis of a belief might have been unavailable when the belief was originally formed (the phenomenon of backwards basing). This can be illustrated by a relatively simple case. A bit of wishful thinking causes Green to believe that Mary is at the coffee shop. He walks there to meet her. Fortuitously, Mary is there when Green arrives. As Green greets her, his belief that she is at the coffee shop is based on excellent perceptual evidence.
  3. The basis of a belief might have been acquired at the same time as the belief (the phenomenon of simultaneous basing). A variation on the previous example illustrates the point. Suppose Green sees Mary in a coffee shop working on her book. He then believes that Mary is in the coffee shop and he believes that the person he saw was working on a book. He later realizes that there is some question whether he saw Mary or her twin sister Marie. But Green remembers that Marie isn’t working on a book, and so remains confident that it was Mary he saw. Green’s belief that Mary was in the coffee shop is now partly based on his belief that the person in the coffee shop was working on a book.
  4. The belief can be based on a reason that is a mere (late) preempted cause of the of the belief (the phenomenon of preempted basing). The example used to illustrate backwards basing shows this provided we stipulate that Green’s perceptual evidence would have caused him to believe that Mary was in the coffee shop had he not already believed that.
  5. We often only base a belief on an available reason once we have already formed the belief, since without the belief we would not notice that the reason supports the belief (the phenomenon of biased basing). We might suppose that Plum becomes convinced that Scarlet is the killer. Seeking to confirm his belief, he reviews his evidence and (finally) sees how it supports Scarlet’s guilt. He wouldn’t have noticed the connection to Scarlet if he weren’t already convinced of her guilt.
  6. The basis of a belief can change over time (the phenomenon of basing termination). A twist on John Pollock’s widget factory shows how. Smith walks into a widget factory and sees what appears to be, and is, a red widget. He believes as much on the basis of this experience. The owner of the factory approaches Smith and says, “That’s a fine red widget, eh? I just finished painting it.” “I see that,” Smith says. “No, you have to be careful around here–most of the widgets only look red because of the overhead lights,” says the owner. Smith looks up and sees that the place is lit by red lightbulbs. His belief that he sees a red widget is no longer based on his visual experience; it is now based on the testimony of the factory owner.

Since they’re fictional, the cases generate claims about possible ways for beliefs to be based on reasons. But I take knowledge that the cases are possible to derive from knowledge that similar cases are actual–each is the sort of thing that most of us have encountered (if not in ourselves, then in others). The desiderata then, are empirical truths–at least, I don’t pretend to have any sort of deep apriori insight into their truth. But empirical data is data nonetheless. We will see, however, that many existing analyses cannot explain the data.

Causal Theories

When one first hears about the problem of the basing relation, I think it is very natural to suppose that the relation is somehow causal. If a belief is based on a reason, the thought goes, then the reason must in some sense have caused the belief. If a belief is based on wishful thinking, then a desire must in some sense have caused the belief. Conversely, it seems that if a reason causes a belief, then the belief is based on that reason. If a desire causes a belief, then the belief is based on that desire (wishful thinking). Thinking like this yields the naive causal theory:

nct S’s belief that p is based on her mental state m iff m caused her belief that p.

Probably no one who has thought much about the matter has ever held such a view, but I think many who have not thought much about the matter have. The thought that nct is approximately correct is surely what motivates the more sophisticated causal theories that we will soon discuss. So it will be instructive to see why nct fails.

An initial concern is how we are to understand nct. It is not clear what it means to say that S’s mental state m caused her belief that p. Plausibly, causation is a relation between events. Plausibly, beliefs (and desires and reasons, &c.) are not events. On the effect side, though, there is an obvious candidate event: the formation of the belief. What about the cause? If we think of mental state m as dispositions, then they can manifest themselves in various ways. Such a manifestation is an event that might cause the formation of a belief.

This gets us the less naive causal theory:

lnct S’s belief that p is based on her mental state m iff a manifestation of m caused the formation of her belief that p.

Let us see how well lnct accounts for the data discussed in the previous section.

lnct mistakenly entails that the basing relation is asymmetric, which contradicts our first datum. For suppose that S’s belief that p is based on her belief that q and that her belief that q is based on her belief that p. Then, by lnct each of the following are true:

  1. A manifestation of her belief that q caused her to form the belief that p.
  2. A manifestation of her belief that p caused her to form the belief that q.

Assuming that a belief cannot manifest itself unless it has already been formed, it is impossible for both 1 and 2 to obtain. This is just a consequence of the fact that the causal relation is asymmetric–causal loops are impossible. But, of course, it’s perfectly possible that one’s belief that p is partly based on one’s belief that q and vice versa.

lnct also cannot explain the possibility of backwards basing: lnct requires that a manifestation of the basis cause the formation of the belief; this couldn’t happen unless the basis were already available. Similar reasoning establishes that lnct fails to explain simultaneous basing, preempted basing, and biased basing.

What about the the phenomenon of basing termination? Pretty clearly, what caused the formation of a belief cannot change over time.Assuming that backwards causation is impossible. Lewis has famously argued that it is possible, but only in bizarre and remote worlds. But for the basis of a belief to change over time is an ordinary and frequent occurrence that shouldn’t require something as controversial as backwards causation. If e causes the formation of someone’s belief that p, then that will be the case for the duration of the belief. It doesn’t look like SSCT can account for this desideratum either. Might the defender of lnct reply that we sometimes temporarily abandon our beliefs only to take them up again in light of new evidence? That could plausibly be described as a case of the basis of a belief changing over time, and it appears to be one that lnct can explain, for when the belief is taken up anew, it must be formed anew. This response is fine so far as it goes, but it doesn’t go far enough. Recall the case used to illustrate the sixth datum. In that case, the belief that the widget is red was never abandoned and so never re-formed. lnct cannot explain a case like this.

lnct fails to account for any of our six data. Such utter failure should give serious pause to the would-be causal theorists–we aren’t discussing recherché counterexamples, but rather the core features of the basing relation. That lnct cannot account for any is telling. But it would be premature to conclude on this basis that no such causal analysis can succeed. I will argue for that conclusion in two steps: we will first consider the most serious causal theory on offer and find it wanting; we will then consider a case that suggests that the basing relation can obtain absent any relevant causal relations. Before the second step, we will have reason to believe that no causal analysis has succeeded. After the second step, we will be able to explain the first step: no causal analysis has succeeded because no causal analysis can succeed.

Swain’s Causal Theory

It will have occurred to some readers that a bit of chisholming will fix many of lnct’s problems. First, we should not privelege the formation of the belief. The formation of a belief is just a privileged instance of a more generic type of event: the having of a belief. We should only require that the basis of one’s belief causes one to have the belief at that time. Second, we might accept that the basis of a belief might not actually have caused the having of the belief: it might be a mere preempted cause (a would-have-caused). If we make both of these modifications, then we have the essentials of Swain’s theory of the basing relation (but not in Swain’s terminology):

st At time t, S’s belief that p is based on her mental state m iff a manifestation of m caused, or is a preempted cause of, her having at t the belief that p.

It is worth remarking that “preempted cause” is syncategorematic: a preempted cause is not a type of cause (in the same way that a decoy duck is not a duck). A preempted cause did not cause the relevant effect, though it would have had it not been preempted.Preempted causation is the bane of many counterfactual analyses of causation–they can’t distinguish a real cause from a merely preempted cause. If we keep this point in mind, allowing the basis of a belief to be a mere preempted cause appears to be an entirely ad hoc maneuver for the causal theorist. To see why, consider the killing relation, which plainly is a causal relation. If my action is a mere preempted cause of Smith’s death then I have not killed Smith.Though this may not exonerate my action. To allow preempted causes as bases of beliefs is to abandon the original motivation for causal theories. Having noted this objection, I pass over it: the present theory faces more serious difficulties.

To its credit, st can plausibly explain the first three data points, if we make some controversial assumptions about the possibility of causal overdetermination. st of course can allow for preempted basing, though it does so by arbitrary fiat.

In spite of this modest success, st cannot be correct for it fails to allow for basing termination and biased basing. Recall the widget example. Smith walks into a widget factory and a visual appearance as of a red widget causes him to believe that there’s a red widget in front of him. He then continues to believe this; at any given moment, his believing this is caused by his believing it a moment prior. A few moments later, he hears from the factory owner that that particular widget has just been painted red. This, along with his prior believing, overdetermine his believing at the next instant. But the factory owner then says that most of the widgets only look red because they are under red lights. Smith sees that this undercuts the evidential force of his visual experience. Nevertheless, Smith continues to believe that the widget is red because of the owner’s testimony. Intuitively, Smith’s belief is no longer based on his visual experience. The problem is, by the transitivity of causation, that visual experience is a cause of his present belief.

By st then, Smith’s belief is still based on his earlier visual experience. This is the wrong result. Smith himself would presumably deny that his belief is now based on that experience (while of course admitting that it was that experience that first generated the belief). People can make mistakes about the bases of their beliefs, but why should we attribute error to Smith here? He knows what initially caused his belief; he knows that that initial cause no longer supports the belief; and he knows that his belief is now supported on independent grounds. There is nothing about the etiology or epistemology of his belief about which he is deceived.Similar reasoning shows that st cannot explain the possibility of biased basing.

We can also offer a theoretical motivation for the judgment. If a belief is based on grounds for which one has an undercutting defeater, then that belief is epistemically defective. But there is nothing epistemically defective about Smith’s belief. So it isn’t based on a ground for which he has an undercutting defeater. Since the factory owner’s testimony undercuts Smith’s visual experience, it follows that his widget belief is not based on that experience.

Think of it this way. If you find out that something no longer supports your belief, continuing to base your belief on it is unreasonable. According to st, the only way for you to terminate the basing relation is to abandon your belief (and perhaps re-form it later on a new basis). But if you have adequate, independent, and undefeated grounds for the belief, then abandoning it is also unreasonable. If st were right, then, there would be no way for you to fix your epistemic situation, no way to be reasonable.

This is a serious problem for st. The sort of phenomenon described in the widget case is common. We form a belief for one reason, amass evidence for it, and later learn that our initial reason has been defeated. Such cases are pedestrian and don’t seem to involve any irrationality or unreasonableness.

st entails the impossibility of basing termination and biased basing and it gives us no explanation of why they seem possible–indeed, common. Further, there are strong theoretical reasons for thinking basing termination is possible. I take this to refute st. I said that st was the best causal theory on offer; sometimes the best isn’t good enough.

Sophisticated Swampman

That completes the first step in my main argument against causal theories of the basing relation–I argued that even the best on offer doesn’t make the grade. I now turn the second step, arguing that no causal (or pseudo-causal) relations are necessary for the basing relation.

We begin by remembering Davidson’s Swampman (1987). Green is standing by a swamp in a thunderstorm. The swamp is struck by lightning and, quite miraculously, a physical duplicate of Green appears. None of Swampman’s internal states are causally connected. The strategy is to argue that Swampman’s beliefs have all the same bases as Green’s. The problem with Swampman is that his internal states don’t have any content–content requires appropriate causal contact with the world. But we can remove Swampman’s handicap. Let us allow Swampman enough causal contact with the world, and perhaps a linguistic community, for his internal states to acquire content–all of the same content as Green’s. But while we are doing this, let us hold fixed the causal structure of his internal states: he will acquire no new beliefs, lose no existing beliefs, and develop no new causal connection between beliefs. We’ll call him Sophisticated Swampman. If it is possible for us to give Sophisticated Swampman content in this way, then we will have a being whose beliefs are not causally connected. Will some of Sophisticated Swampman’s beliefs have bases? It seems so. In fact, I suspect that, for all p and m, Sophisticated Swampman’s belief that p is based on m iff Green’s belief that p is based on his m. We can imagine Sophisticated Swampman having a conversation and citing a reason for some belief. Citing a reason for a belief might not constitute basing the belief on the reason, but it sure seems like evidence that the belief is based on the reason. We can further imagine someone providing Swampman with undermining evidence for the reason he has cited. Swampman might then abandon the belief that he said the reason supported.

I think Sophisticated Swampman shows that a belief can be based on a reason even when there is no causal connection between the belief and the reason. But maybe you don’t. Maybe you think an upstart like Sophisticated Swampman has nothing to teach us. In that case, I would like to try a more ordinary case. Surely it’s possible for a belief and some other state to bear no causal connection in an agent. Then it’s possible for Green to have both the belief that p and the belief that q, where there is no causal connection between the two. It does not thereby seem impossible for Green’s belief that p to be based on q. In fact, it seems quite possible. Let’s suppose that Green would cite q as a reason for p, is disposed to think of q as his reason for p, and is disposed to abandon the belief that p if his belief q is defeated. Doesn’t it just seem true to say that Green’s belief that p is based on his belief that q? We needn’t claim that any of the facts about Green constitute the basing relation, but their presence makes it eminently reasonable to conclude that Green’s belief that p is based on m. And that’s all I need, for we have stipulated that Green’s belief that p is not causally connected with his reason q.

This completes the second step in my main argument against causal theories of the basing relation–I argued that no causal (or pseudo-causal) relations are necessary for the basing relation. Some readers will object that I have not discussed the notion of causal sustainment invoked by many causal theorists. I find it easier to discuss this notion after I have introduced my own theory and so §6.1 is devoted to the topic. I’ll argue that accounts invoking causal sustainment are best analyzed using my own theory of the basing relation–if causal sustainment is analyzed using causal or counterfactual notions, the resulting account of the basing relation falls victim to the sorts of objections above.

Doxastic Theories

So-called doxastic theories of the basing relation are the only developed alternatives to causal theories. The distinctive feature of doxastic theories is the requirement that an agent believe that the basis of her belief is a good reason for her belief. It is easy to see that these theories won’t do. Beliefs can be based on wishful thinking. When they are, it isn’t because anyone thinks desiring that p is a good reason for believing that p! Similarly with beliefs based on prejudice: imagine that Professor Green’s belief that Professor Smith is unqualified is based on solely his belief that she is a woman. But Professor Smith believes that he harbors no sexist prejudices. He in fact strongly believes that his belief is not based on Professor’s Smith’s gender, but on the quality of her work. There is nothing inconsistent about this case; indeed, there is a robust psychological literature on just this sort of unconscious gender bias [cite]. That shows that meta-beliefs about reasons are not necessary for basing. They’re also not sufficient. Green believes that his belief about Smith is based on the quality of her work, but it isn’t.

Why has anyone held such a theory? It could very well be that meta-beliefs about reasons are required for proper basing and that doxastic theories are intended by their authors as theories of proper basing. But proper basing isn’t basing, and it is a theory of basing that occupies us here.

The Dispositional Theory

I seem to have painted myself into a corner. I have argued that the only existing theories of the basing relation are fundamentally flawed: neither causal relations nor meta-beliefs about reasons are necessary or sufficient for the basing relation. What else is there?

The core of the basing relation is a particular sort of dependence: for one belief to be based on another is for the one to depend on the other in the right way. I think that the sort of dependence in question involves how one would respond were the basis of one’s belief lost–this is why we say that a belief stands or falls with its basis. If one’s belief that p really is based on one’s belief that q, one responds by revising one’s belief that p. One way of understanding this dependence is via counterfactuals. This would yield the following theory:

CT S’s belief that p is based on m iff S would revise her belief that p were she to lose m.

But as is the case with so many other philosophically interesting notions, the counterfactual dependence expressed by CT is much too crude to capture the basing relation. Often, one wouldn’t revise one’s belief that p upon losing one’s belief that q, even though the former is based on the latter; one might, for example, have additional reasons for believing p.

A more promising means of capturing the dependence is via dispositionals. After all, one can be disposed to revise one’s belief that p upon losing one’s belief that q even if one wouldn’t. The above remarks suggest the following theory of the basing relation.

DT S’s belief that p is based on m iff S is disposed to revise her belief that p when she loses m.

I think dt is correct. First and foremost, dt succeeds where others have failed: it accounts for all six data with which we began. This may seem like a basic task, but as the above discussion revealed, it has proved a difficult one. The first datum was the phenomenon of backwards basing. It’s easy to see how dt allows for the phenomena of backwards, simultaneous, preempted, and biased basing. The major stumbling block for causal theories was the phenomenon of basing termination. dt handles this simply: one can, for various reasons, lose the disposition to revise one’s belief that p upon losing m; one’s belief that p will no longer be based on m.

The dispositional theory is also quite intuitive–it is simply a systematization of the idea that a belief stands or falls with its basis. To further explain and defend the theory, I’ll discuss each of the following points below:

I’ll also say a few words about cases that may seem problematic for my account: undercutting defeat, beliefs based on experience, and forgotten evidence.

Belief Revision

One might worry about the following sort of counterexample:

A large tumor is growing in my brain. It has the following odd effect on me: if I lose my belief that I was born in Colorado, I am disposed to abandon my belief that Lynn Hill was the first to free climb El Capitan’s Nose in 1993. And this because of the way the tumor presses on my brain—nothing in my cognitive system yields any dependence between the two beliefs.Notoriously, causal theories of the basing relation have struggled to solve the problem of deviant causal chains (Pollock & Cruz 1999; Plantinga 1993: 69n8). My dispositional theory of the basing relation faces a similar problem, what I call the problem of deviant dispositions. Thanks to Victor Kumar for forcing me to confront this issue

It’s clear that my Lynn Hill belief is not based on my birthplace belief, though it does, in some sense, depend on it. The dispositional theory can account for this. In the above case, I am disposed to lose my Lynn Hill belief when I lose my birthplace belief, but I am not disposed to revise it. Intuitively, not just any change in belief counts as a revision: if I forget that p, I haven’t revised my belief that p; if a mad scientist cuts out a chunk of my brain resulting in massive belief loss, I haven’t revised all of those beliefs. Belief revision is a special sort of belief loss, and DT makes use of this fact.

Though an intuitive understanding suffices for present purposes, some philosophical elucidation of this distinction may be helpful. When you lose a belief, the rest of your belief system needs to be updated accordingly – other beliefs may need to be deleted or weakened. We can postulate a cognitive process whose function is to carry out this sort of updating. We then say that a belief that p was revised if it was lost as the result of the proper functioning of this process. This definition raises a host of interesting questions about belief revision, but they’re largely orthogonal to the present topic and so I leave them for another day.

Multiple Bases

Much of the talk in this paper makes it sound as though each belief has a single mental state that is its basis. The truth, of course, is that most of our beliefs are based on a whole host of background beliefs, independent reasons, apparent memories, and so forth. Most beliefs have multiple bases. In such cases, will losing one of the bases of a belief result in revision? The answer is in many such cases is “yes.” As I suggested above, revision need not involve wholesale termination of a belief; sometimes revising a belief is just weakening the degree of that belief. When we lose one of a belief’s many bases, we often respond by weakening our degree of belief.

Sometimes, of course, not even this modest sort of revision occurs. The remaining bases are enough to preserve the strength of the belief. But this isn’t a problem for DT, for it doesn’t mean that the agent wasn’t disposed to revise when he lost the basis; it just means that the disposition was masked by the additional, independent, bases. The phenomenon of masking is familiar to philosophers who work on theories of dispositions (Fara 2005; Manley & Wasserman forthcoming). If I package a wine glass in styrofoam, it is still fragile, still disposed to break when dropped. Nevertheless, so long as it is packed in styrofoam, it won’t break when dropped. Its disposition to break when dropped is masked. Basing a belief on multiple independent reasons is a lot like packing a glass in styrofoam.

In many cases, it is difficult to tell whether a disposition to revise is being masked or whether it is absent. Lehrer’s gypsy lawyer is a nice example (Lehrer 1971). On the basis of a card reading, a gypsy lawyer comes to believe that her client, on trial for a grisly murder, is innocent. She then reviews the evidence and discovers a complicated, though quite conclusive, line of reasoning that establishes her client’s innocence. The question is whether her belief is now also based on the evidence. On my account, this reduces to the question whether she is disposed to revise if she loses her evidence (or the complicated line of reasoning).Some of the interest in the question is in whether she now knows that her client is innocent–for further discussion see Kvanvig’s (2003). My own view is that the gypsy lawyer does know; but I think this turns out to be independent of whether her belief is based on the evidence. She knows because she accepts, for epistemic reasons, that her client is innocent. Acceptance and belief are distinct propositional attitudes, and either one can yield knowledge. But that’s a discussion for another day. Given her faith in the cards, it’s pretty clear that the gypsy lawyer would not revise her belief if she discovered that her reasoning were incorrect or that she had been using evidence from a different case (or whatever). But is that because her faith in the cards masks her disposition to revise, or is it because she lacks such a disposition? Of course, in thought experiments like this we can just stipulate an answer to this question, but this won’t help us in understanding real-life cases. I think that to figure out what’s going on in difficult cases like this, we need to know more about the categorical basis of the disposition to revise when the basis is lost.

Categorical Basis

Dispositions have categorical bases. A wine glass is fragile (disposed to break upon impact) because it has certain intrinsic physical properties (certain chemical bonds between its constituent molecules, say). It would be nice to know about the categorical basis of the disposition invoked by dt. This in some sense would yield a deeper explanation of the basing relation. Unfortunately, I don’t think there is much that can be said about this issue from the armchair. Given my objections to causal and doxastic theories, we can say with some confidence that neither causation nor meta-beliefs are required for the agent to have the right disposition. But to say more than this will require the help of empirical psychology.

Undercutting Defeat

What about being disposed to revise p when one comes to believe that m is not a reason for p? It is tempting to think that such a disposition is necessary or at least sufficient for the belief that p’s being based on m. But it’s not too difficult to see that such a disposition is not necessary for basing. Self-deceived bigots provide an example: Green, a sexist male, regards himself as free of any sexist prejudice. He believes that the fact that Smith is a woman does not provide a reason for thinking she is unqualified. Nevertheless, it’s easy to imagine that his belief that Smith is unqualified is based on his belief that she’s a woman. In these cases, one doesn’t know what one’s belief is based on, so one isn’t disposed to revise when one comes to believe that the basis (not under that description) is no reason for the belief.Is it necessary provided that one does know that one’s belief is based on m? No. In such a case, one (epistemically) should be disposed to revise, but might not be. The basing relation isn’t a matter of having explicit meta-belief that m is a reason for p and can survive even an explicit meta-belief to the contrary.

The sort of disposition in question also doesn’t seem sufficient for basing. This mainly because a disposition to revise when the basis is lost seems necessary. If am not disposed to revise my belief that p even upon becoming convinced that q is false, then my belief that p is not based on my belief that q. If I am disposed to revise my belief that p upon becoming convinced that q is no reason for p, this indicates that my belief that p is based on my belief that q is a reason for p. It is difficult to be sure because such cases are odd. It is odd to be disposed to revise my belief that p when I learn that q is no reason for p, but not when I learn that q is false.

Experiential Bases

Unlike beliefs perceptual experiences are transient. But, in general, we do not (and should not) revise the beliefs we form on the basis of perception when the perceptual experience ends. I see a red widget before me, form the belief that there is a red widget before me, and then close my eyes. I no longer see a red widget before me, but I still believe it’s there. I still believe this because I remember having seen it (and because I know things don’t just disappear, and that I closed my eyes, etc.). My belief is now based on this apparent memory and the associated background information. Indeed, without the apparent memory I very likely would revise my belief. Similarly, without the associated background information I might very conclude that the widget had, in fact, disappeared. So when we form the belief on the basis of an experience, we are disposed to revise the belief when the experience is lost. It’s just that this disposition gets masked in normal cases because the belief is sustained by memory and complicated background knowledge about external objects.

Forgotten Evidence

We might also ask about forgotten evidence. There are at least two types of forgotten evidence. One sort occurs when you forget what your evidence for some belief is, but you still retain the evidence. Some chemical experiment may have convinced you that broccoli is a complete protein. You may remember the experiment and that broccoli is a complete protein, but forget that the experiment was your reason for your belief. If this seems implausible, one need only recall that beliefs can easily become “compartmentalized” such that we don’t notice their connections. I recently believed both that I was meeting a friend for coffee in the afternoon and that that friend was in a different state for the rest of the week. It wasn’t until I was halfway to the coffee shop that I noticed the contradiction. In such a case, you could very well still be disposed to revise your belief if you lose the evidence. Then your belief is still based on the evidence even though you’ve forgotten. If you aren’t so disposed, then, on my account, your belief is not based on the evidence (though it may have once have been).

The other sort of forgotten evidence occurs when you not only forget what the evidence is for your belief that p, but you forget the evidence itself. This sort of thing happens all the time. I actually do believe that broccoli is a complete protein, but I can’t for the life of me remember any evidence for this. At one time, my belief was based on the evidence that originally led me to form the belief; it no longer is (you can’t base your belief that p on your belief that q if you no longer believe that q!). My belief is now based on my seeming to remember that broccoli is a complete protein (whether that’s an epistemically adequate basis is another question). And, I suppose, this seeming to remember probably explains why I didn’t revise the belief when I forgot the evidence. Seeming to remember p can mask a disposition to revise p. Indeed, masking such dispositions would seem to be the point of apparent memories.

Causal Sustainment

Finally, I should say something about causal sustainment. When confronted with my arguments, I find that many with a sympathetic ear for causal theories want to invoke a notion of causal sustainment.Victor Kumar, in particular, had many helpful things to say on this score. They want to say, for example, that in my widget case the reason that Smith’s belief is no longer based on his perceptual experience is that said experience no longer causally sustains the belief. Here is Dretske discussing a similar case: (1981: 88–91)

[S]uppose K believes that s is F because an ignorant meddler told him s was F. The ignorant meddler knows nothing about the matters whereof he speaks, but K, unaware of this, believes him. After acquiring this belief, K undertakes an examination of s and observes it to be F… In such a case, K knows that s is F…but his belief that s is F was not caused or produced by the information that s is F. Rather, this belief was caused or produced by the ignorant meddler assurances… . The reason we say that K (now) knows is not because his belief was originally caused by the relevant piece of information, but because his belief is (now) supported or sustained by the relevant piece of information.

Now Dretske is advancing a causal theory of knowledge, not of the basing relation, but we can overlook that difference for present purposes. The sort of thinking behind the above passage motivates the following theory of the basing relation:

pcst At time t, S’s belief that p is based on her mental state m iff a manifestation of m caused (or causally sustains at t) the belief that p.

The immediate problem with such a theory is that it still fails to explain my widget case. If we grant the notion of causal sustainment and say that the testimony of the factory owner sustains Smith’s belief, then we do get to say that Smith’s belief is partly based on this testimony (as it intuitively is). But, since Smith’s belief was caused by his experience, we will also still have to say that it is partly based on his experience, which, intuitively (and for theoretical reasons) it isn’t. The remedy to this problem is to drop the “caused” disjunct:

cst At time t, S’s belief that p is based on her mental state m iff a manifestation of m causally sustains, at t, the belief that p.

At this point, the notion of causal sustainment is being asked to do important work, so it’s fair enough for us to ask for some identification: just what is it for a belief to be causally sustained by another mental state? An example from Dretske adds some intuitive content to the notion: imagine an object being held up by a piece of string–the object’s hanging in the air is being sustained by the string (1981: 89). It doesn’t matter whether the object was initially put to hanging by something else (a hand, say) or whether there is anything else supporting the object (another piece of string, say). All that matters is that the object is now supported by the string, that the string is now sufficient to keep the object suspended in air. Such talk is just as applicable to belief: the basis of a belief supports the belief (in psychological, non-evidential sense of “supports”) and the basis of a belief is sufficient to keep the belief in existence. Now Dretske attempts a theoretical characterization of causal sustainment on the basis of the above intuitive gloss. He says that for something to causally sustain a belief, it must be such that it is sufficient to cause the belief–that it would cause the belief, were the belief not already formed (1981: 89).

I think the notion of sustainment brought out by the piece of string example is the right one for thinking about the basing relation. But I don’t think there’s anything particularly “causal” about this notion nor do I think that Dretske’s theoretical characterization gets it right. Why is the object’s hanging in the air sustained by the string? Because the object is disposed to fall if the string is removed (or cut or whatever). Why is my belief sustained by the evidence on which it is based? Because I am disposed to revise it if I lose the evidence. Such dispositions can exist in a causal vacuum, as my Swampman case illustrated. (Consider SwampObject: lightning strikes a swamp and forms an object that is hanging in the air by a string. Right from the start, SwampObject is supported by the string.)

Often, I suppose, the basis of a belief is sufficient to cause that belief. But sometimes not. Recall my fifth datum, Biased Basing. Sometimes, it is only after we form a belief that we will notice that some evidence we have supports that belief. After all, it’s often a non-trivial task to discover that our evidence supports p and we’re not particularly likely to work it out unless we have some interest in p. In such cases, the evidence on which our belief is based will not be sufficient to cause that belief and it certainly would not have caused us to believe if we didn’t already. Nevertheless, once one does appreciate that some evidence supports one’s belief, it often happens that that evidence becomes part of the basis for one’s belief.

So while I do think that there is something to the idea of the basis of a belief sustaining that belief, I don’t view this as a threat to my theory of the basing relation. A belief is sustained by its basis insofar as the right dispositions are in place and not in virtue of the basis bearing any sort of causal relation to the belief (counterfactual or otherwise). As I’ve suggested before, such causal relations will often effect such dispositions but not always.

Conclusion

To recap, I argued that we were sorely in want of a new theory of the basing relation: the best of the existing theories don’t pass muster. To respond to this want, I offered the dispositional theory of the basing relation and tried to show that it succeeds where others have failed. The theory I have offered is intuitive and accounts for the data. The real test, however, will be its theoretical fruitfulness: it will be confirmed to the degree that it can be play a role in the theories of rational belief and action, and in understanding the similarities and differences between epistemic and practical rationality.

It is worth remarking that the theory posits real psychological kinds – dispositions to revise – whose existence cannot be established from the armchair. The dispositional theory is held hostage to questions in empirical psychology. Supposing such dispositions don’t exist, we will confront an interesting question: does basing not exist, or is it simply that the dispositional theory is incorrect? The answer to that question would have serious ramifications for the theories of rational belief and action; no doubt our answer to the question would turn on how we think the empirical data impact the possibility of rationality.

References