Lecture Notes, UC Davis Philosophy 102, Theory of Knowledge: Reliabilism

Lecture Notes, UC Davis Philosophy 102, Theory of Knowledge

Reliabilism

G. J. Mattey, Senior Lecturer

As noted at the end of the last lecture, Goldman gave up his causal account of knowledge in favor of reliabilism. There is an important division in the way the role reliable belief-formation is described by reliabilists. Some, like Ramsey, considered reliability as a necessary condition for knowledge. Others, like Goldman, consider it to be a necessary condition for justification.

But there is an obvious objection to Goldman's approach--an objection that would naturally be raised by "internalists." Whether or not one's own belief has been formed in a reliable way is an objective, "external," fact about a person. It is not something that could be appealed to in what Audi called a "dialectical" answer to a skeptical question.

But it does not seem to be a feature of the "structure" of justification either. As Audi described it, the structure is a system of relations among internal states of a person. If there are unjustified justifiers, such as perceptual states, they are still accessible to a person in a way that one's reliability is not.

We have seen that what is internal is the assessment of one's reliability. The need for such an assessment as a necessary condition for justification is one of the key motivations for the adoption of coherentism. Externalists such as Goldman want to hold that one's actual degree of reliability figures into justification as well. In his 1988 paper, Goldman finds an ambiguity in the notion of justification that he thinks can accommodate both internalist and externalist points of view.

In a paper published in 1978 ("Discrimination and Perceptual Knowledge"), Goldman rejected the causal analysis of knowledge he had given in his 1967 paper. He stated that the analysis is subject to a fatal counter-example.

Henry is driving in the countryside with his son, pointing out various objects. He points to a barn, saying, "That's a barn." It is true that the object is a barn, and he believes that it is. Moreover, his belief is the end-point in a causal chain beginning with the fact that the barn is there. So according to the causal analysis, he knows that the object to which he is pointing is a barn.

Now suppose that in the area there are many phony barns which are not distinguishable from the car from real barns. Henry has no knowledge of their presence. According to Goldman, Henry in this case does not know that he sees a barn, even though he continues to satisfy the conditions of the causal analysis.

Goldman's reaction to this kind of example is to drop the causal link to the object as a necessary condition for knowledge of the object. He substitutes for it causal condition which does not mention the object at all. The condition is that the process by which the belief is formed is a reliable process. So his analysis went from being causal to being reliabilist. Goldman could have blocked the example by expanding his causal analysis, adding a fourth condition. He probably did not take this route because he wanted to allow for knowledge in cases where a causal relation to the object is hard to explain or even impossible, as with mathematical knowledge.

Goldman motivates his claim that there is no single univocal conception of justification by the use of an example. Consider a member of a pre-scientific culture that had no notion of probability, statistics, or anything of what could be called an experimental method. Instead, it made claims about what is not observed, including claims about the future, on the basis of such practices as astrology. This person makes a prediction based on the a "culturally approved" method M. Goldman asks whether this belief is justified.

There are contrary tendencies to answer affirmatively and to answer negatively. The negative answer is based on the idea that there is something improper about the method (say, astrology), and that justification demands the use of a proper method for forming beliefs. If the belief turns out to be true, it is not knowledge. We would explain this by saying that it lacks justification, since the person believes it and there is nothing like a Gettier case involved here.

On the other hand, we are inclined to allow that the person has a justified belief because of the context in which it is formed. Because the person has no scientific training, he is in no position to cast doubt on M, while he has every reason (based on the cultural approval of M) to think that it is a proper method. "It is beyond his intellectual scope to find flaws in M." For this reason, we hold him blameless epistemically for using M, "and that seems to explain why we are tempted to call [the belief in question] justified."

So Goldman thinks that there are two distinct conceptions of justification, each of which "captures some chunks of intuition involving the term 'justified' (in its epistemic applications."

Before turning to his elucidation of these two notions, Goldman makes a further distinction between to kinds of "justifiedness." These correspond to a distinction between "processes" and "methods" of belief-formation. A process is a basic, "wired-in" means of forming beliefs, while a method is an explicitly inferential means, such as the use of statistical analysis or of the readings of instruments.

A belief is fully justified (at least strongly) only if it is formed by both adequate processes and adequate methods, if methods are used. Primary justifiedness results from the use of adequate processes and secondary justifiedness from the use of adequate methods. "A complete account of justifiedness must present conditions for both these levels."

A belief is strongly justified only if it makes use of adequate methods. A good candidate for explaining adequacy is reliability. This allows us to explain why the person in the original example lacked strong justification: the methods of his society which he used in forming his beliefs are unreliable. It may be that what is needed for strong secondary justification is maximal reliability, where the method used is no less reliable than any other, but Goldman will not consider this stricter requirement here.

Reliability of the belief-formation method is not sufficient for justification, though. The method itself must have been acquired by other methods or processes that are themselves reliable or meta-reliable (productive of reliable methods). Secondly, there is a "no-defeater" clause: that the person's cognitive state contain nothing that would undermine the adequacy or correctness of the method. The believer should not think or be justified in believing that the method is unreliable.

Chisholm has claimed that reliability is too weak a condition because it is too easy to obtain. A method might be reliable because it can only be employed once, and it works at that time, say finding out the number of planets by reading tea-leaves at 2:17 on a Friday afternoon. Goldman denies that one uses this method, because to do so, one must represent it in one's head, and nobody fills in the details of the time as in this case. The method should be described simply as reading the tea-leaves.

Even if a person does have a method of this sort that is "reliable" in the sense that it works whenever it is used, it would not confer secondary justifiedness. The reason is that this method could not be suitably acquired. If the method were suitably required, Goldman concedes, then one would have strong justification.

Goldman digresses to consider the idea that justifiedness comes in degrees, rather than being a catergorical concept, according to which in every case, one is either justified or not. The degree of reliability determines the degree of justifiedness, all else being equal. To motivate this claim, Goldman considers an example of someone adding numbers in the normal way from top to bottom. He forms a belief about the sum using this method. Someone else does a second addition in another way and forms a belief only if the two answers are the same. We say that the second person is more justified by the method he uses (which includes the first method) because it is more reliable.

There is an initial problem in discussing weak justification in the context of reliability. If a person uses a reliable method which is properly acquired and not undermined is blameless. So by the definition of weak justification, the person is weakly justified. Goldman wants to consider a contrastive notion, so he turns to that of being blameless in the formation of a belief that is ill-formed, i.e., not produced by the kind of method that confers strong justification.

Some sufficient conditions for weak justification of the ill-formed belief that p for S are now proposed.

An objection could be that our subject could have checked on the reliability of astrological predictions simply by observing their success-ratio. Goldman covers this objection by noting that the astrological establishment might have a way of "covering" or explaining away all their failures by appealing to methods our subject is not in a position to expose as fraudulent.

A second objection is that reliable methods of telling that M is unreliable may not always be possessed by a person, but are always "available" to him, no matter how scientifically backward is the culture in which he finds himself. Goldman responds by saying that "possesses," as well as "available," are a vague terms, which are open to many interpretations, which are applicable in different contexts.

We can see this in the case of the possession of evidence. The most conservative approach would be to say that evidence is possessed only when it is something of which the subject is consciously thinking. The most liberal view would allow that evidence is possessed when stored in memory. Intermediate between them is the claim that evidence is possessed when it is either the object of conscious thinking or easily retrieved from memory.

Similar remarks apply to the "availability" of evidence, though the scope of that term includes not only what is in the head, but also the social world in which one finds one's self. Much depends on how hard it is to get evidence that is at least theoretically "available" in one's social sphere. Standards vary, and because of this, Goldman will concede that the notion of "availability," and therefore of weak justification, is vague. Still, under normal conceptions of availability, there was no method available to the person in the example that would have exposed the unreliability of the use of astrology.

Using a simplified format for the moment, Goldman holds that S is strongly justified at the primary level when and only when his belief is produced or sustained by a cognitive process that is sufficiently reliable and is not undermined by S's cognitive state. The conditions for weak primary justifiedness also parallel those for secondary justifiedness. So just substitute "process" for "method" in the analysis of weak secondary justifiedness.

A criticism of reliabilism is that a victim of massive deception forms beliefs in an unreliable way but is nonetheless justified in holding them. In such a scenario, the victim has the same range of experience as someone who is in contact with the physical world, but the beliefs he forms are false as the result of the deception. The process of forming beliefs from experience would be unreliable if it failed in enough cases. This can be accommodated by there being only person using the belief-forming process or many cases in which people using the process are deceived. (If our world is not a demon world, then our use of the processes presumably would be reliable.)

The criticism is handled by the claim that the victim is weakly, but not strongly, justified. Although his cognitive processes are not reliable, he does not believe that they are unreliable, he has no reliable way of telling that they are unreliable, and there is no method or process he believes reliable that would tell him this.

A rather implausible objection to this result is that there is a method which is reliable and would tell the victim that his cognitive processes are unreliable. On the "single-output" method, the victims beliefs serve as premises, and the conclusion is that "My perceptual processes are unreliable." It produces a single belief, which is true, and so it might be said to be a reliable method.

Goldman raises two objections to this ploy. First, he doubts that a "single-output" procedure even qualifies as a method. Second, it is not "available" in the sense that there is no natural way to produce it "by some processes and/or methods he already employs, operating on his actual beliefs and experiences." In the present case, everything leads away from this alleged "method." Thus, the condition for weak justification is satisfied.

A different way of handling the demon scenario was given in Goldman's earlier book, Epistemology and Cognition. In that book, Goldman proposed a more complicated version of the reliability needed for justification. It is the rightness of rules that govern the processes that produce belief, rather than of the processes themselves, which is the necessary condition. Rightness of a rule system is in turn explained by its reliability. And the reliability of rules systems is understood in the sense that they produce a high ratio of true beliefs when complied with.

Although Goldman does not mention it in the present paper, a main reason he formulated a "rule framework" for reliabilism in the book was to capture the normative element that he now encapsulates in his notion of weak justification.

We can still ask what it is that makes a set of rules reliable. Ordinarily, we think of reliability as determined by behavior in a range of situations, which we could call "possible worlds." Suppose W is the world inhabited by the believer. We want to say whether the rules that govern the processes and methods that produce belief in W are reliable. Should we confine our attention to the results of the use of the rules in W itself, or perhaps in some other way.

Recall that we are looking for a notion of reliability that will support the notion of a right rule system. Goldman holds that a system must be right in all worlds if it is to yield justification. To be right is to be reliable, but this does not imply that the system must be reliable in all worlds. Instead, Goldman requires only that the system be reliable in all "normal" worlds.

A normal world is one which is consistent with our general beliefs about the actual world. More specifically, these are beliefs about "the sorts of objects, events, and changes that occur in the actual world." So the demon world is not a normal world, in which case the reliability of the rules used by people in non-demon worlds is not affected by the failure of the rules in the demon world. It could even be that our world is a demon world and non-normal at the same time, in which case we would remain justified in our beliefs about physical objects.

Goldman discusses some of the advantages of this approach. One is that it is consistent with the intuitive view that a demon victim remains justified in his beliefs about the physical world. A second is that it seems natural to assess reliability in normal situations. We do not generally condemn a process as unreliable when it breaks down in extreme conditions. But this consideration really does not provide much motivation, because here "normal" seems to mean something like "typical," as opposed to "as is the norm." It this latter sense that is at work when Goldman delineates "normal" worlds.

A number of problems have been raised for the normal-worlds approach to reliability. Some of them are as follows:

So, Goldman turns his back on the normal-worlds approach. The main thing it had going for it in the first place was that it allowed justification for the beliefs of the demon-victim. This can be accommodated with the notion of weakly justified beliefs. It remains to show how strong justification is determined. Goldman proposes to allow "rightness" of rule systems to float from world to world. So a rule system may be right at one world (underpinned by reliability, understood in terms of high truth-ratio at that world) but wrong at another world.

There remain problems with this approach, however. Allowing reliability to be determined by truth-ratios in the actual world means allows for favorable ratios to be determined by a few cases. So if a rule is little-used, and if it is successful when it is used, then it will confer justification. To overcome this objection, Goldman proposes that "close" worlds be taken into account in fixing the truth-ratios. This means that the ratios that determine success are "modalized," taken across worlds, rather than being actual frequencies.

A second problem lies in the fact that the rules may be complied with in many different ways. They are "permission" rules, which say which methods or procedures are permitted to be used. So which ones are actually used will vary greatly: there will be many "compliance profiles." We might require that all compliance profiles generate a truth-ratio that is "tolerably high," or we might require that some average profile, or profile across worlds, meet the threshold.

The final "technical problem" Goldman discusses here he calls the "epistemic free rider problem. What confers justification is not a single rule, but a system of rules. And it is the reliability of the system which determines rightness. Systematic reliability may not be compromised by the addition to system R of a single unreliable rule, resulting in a new system R*. "But if R* contains a poor rule, we don't want to count it as right." The poor rule gets a "free ride" by virtue of the strength of the other rules. This problem is solved by requiring that no subset of R have an insufficiently high truth-ratio.

By relativizing rightness of rule systems to worlds, Goldman opens his reliabilism to the charge that one could be strongly justified as a matter of luck. Whether we are in a normal world or a demon world seems a matter of chance. This would be a serious objection, as indeed the whole point of strong justification is to capture the idea that a justified belief is supported by "proper, suitable, or adequate methods, procedures, or processes."

Surprisingly, Goldman defends the possibility of lucky strong justification. His first point is that luck is acknowledged in knowledge. A belief which is justified for a person may be false, and its being true could then be thought a matter of luck. Even whether a justified true belief is knowledge could be viewed as a matter of luck, as the Gettier cases show. "Since luck is a component in knowledge, why should it be shocking to find it in justifiedness?"

In both these cases, the luck involved concerns a factual component of knowing: whether the belief is true or whether some element in one's justification is true. But justification itself is an evaluative notion, having to do with what it is proper to believe. Luck should not play a role in this, it might be charged.

Goldman's response is that luck seems to play a role in other evaluative notions. He gives as an example two truck-drivers who drive recklessly. The first hits a child and the second doesn't. This is a matter of luck, but in practice we hold the first to be more accountable than the second. Or, we may call someone a good artist partly on grounds of his talent or training, both of which might be said to be a matter of luck.

The final point is that there is an anti-luck element in justification, only it is to be found in weak justification. Because of this, Goldman says that "the present theory also makes room for anti-luck cravings."

It might seem as if in endorsing weak justification as a condition of knowledge, Goldman has abandoned externalism in favor of internalism. In response, he notes that the question is hard to answer given the vagueness of the terms "internalism" and "externalism." On one conception of internalism, his theory is not internalist, because one may not have the right kind of internal access as to whether one has or has available a way of telling whether his rule system is reliable.

Another point is that weak justification may be too weak a notion for many internalists. Note that the kind of "weak justification" Goldman has been considering involves only ill-formed beliefs. The concept he spells out says nothing about well-formed beliefs. It does not do so because, he thinks, one cannot be blamed when one holds a well-formed belief. Another way of putting the issue is that Goldman's formulation of weak justification is purely negative. The positive work is done by strong justification. An internalist presumably would spell out the positive side of justification in terms of what is internally accessible, rather than in terms of what is not internally accessible.