Inductive Reliabilism

Inductive Reliabilism

By Samuel Bookler, University of Pennsylvania

The Importance of Descriptive Epistemology

Descriptive epistemology should not be completely neglected for its normative counterpart. Alvin I. Goldman points out that normative epistemology should have its roots in “the concepts and practices [and the norms] of the folk,” what he calls our ‘epistemic folkways.’1 To reject these roots would be to reject continuity in epistemology: that is, it would be to reject the importance of understanding how we might move from employing our commonsense concepts to employing the prescriptions of a normative theory. This ascribes to descriptive epistemology a preliminary role: it paves the way for a normative project. Even if one denies the need for continuity, however, descriptive epistemology still has an important role to play. While Goldman admits that it “may well be desirable to reform or transcend our epistemic folkways,” a task for normative epistemology, an elucidation of these epistemic folkways is also necessary (524). For without this description, “how would one know what to criticize, or what needs to be transcended”? In this way, the grounding of normative epistemology is found in descriptive epistemology. Therefore, we require a descriptive account of justification, without which we could not fully consider any epistemological project complete. Let us examine one such account, Goldman’s reliabilist approach to justification.

Goldman and Reliabilism

The core idea of the standard reliabilist approach is to identify the concept of a justified belief with the concept of a belief acquired through ‘virtuous’ psychological processes. Any belief acquired that is even partly due to a ‘cognitive vice’ is unjustified. Goldman holds that it is insufficient to formulate justification in this basic way. He hypothesizes that the folk have “a mentally stored set, or list, of cognitive virtues and vices” (525). To determine the status of a belief, an evaluator considers the processes by which the belief was formed and matches these with either virtues or vices in the mentally stored set. If the processes match with all virtues and no vices, then the belief is justified. If the processes include a match with any vice, on the other hand, the belief is unjustified.

An evaluator matching processes to virtues and vices, what Goldman terms an “epistemic evaluator, the possessor and deployer of [concepts],” may meet with a third alternative. If the processes do not match with any virtue or vice in the mentally stored list, then the belief is classified as “neither justified nor unjustified, but simply nonjustified” (525). This scenario only occurs when an epistemic evaluator judges there to be a complete disconnect between the mentally stored list and the belief forming process. This disconnect is necessary because Goldman asserts that the match between process and virtue or vice need not be exact. A completely novel process can be judged based on its similarity to a virtue or vice on the list and classified accordingly. Since any novel process may fall somewhere on a similarity continuum depending on the already listed virtues and vices, judgments can vary between epistemic evaluators. In addition, Goldman proposes that people exhibit ‘categorical conservatism,’ “[displaying] a preference for ‘entrenched’ categories” (527). So, it may be that a novel process must be sufficiently2 similar to an already entrenched, listed virtue or vice for an epistemic evaluator to successfully judge it so and make the match.

Goldman claims that epistemic evaluators form their mental lists of virtues and vices based on the reliability of processes. Virtuous processes, like vision, hearing, and memory, are virtuous because “they (are deemed to) produce a high ratio of true beliefs” (527). Guesswork, wishful thinking, and ignoring contrary evidence are vices “because they (are deemed to) produce a low ratio of true beliefs”. It does not follow that the list of processes is completely formed by judgments of reliability, however. Members of the same linguistic community can inherit elements of a list from one another. In addition, Goldman claims that the categorical conservatism of epistemic evaluators ensures that new elements are not added to the list on the basis of a single case. Neither do imaginary cases have much effect on the list, since epistemic evaluators “do not lightly supplement or revise their categorical schemes”.

The discussion has so far been purely theoretical. What evidence is there in support of the existence of epistemic evaluators’ lists of virtues and vices? Goldman proposes that virtues and vices can be gathered from “answers to the question ‘How does X know?’”. While we consider it common to reply “‘He saw it,’” “‘He heard it,’” “‘He remembers it,’” and so on, it is quite odd to receive the reply, “‘By guesswork’” (529). Analysis of standards regarding this question suggests lists of virtues and vices.

The elements of the list are not assumed to be uniform across all epistemic evaluators. There may be different opinions as to the reliability of processes resulting in differing lists of virtues and vices. Different cultures might hold different processes as reliable as well. Furthermore, reliabilism is relative in that reliable processes are only reliable within a specific domain. For example, “color vision is reliable on the earth but unreliable in the universe at large” (528). Goldman asserts that this domain relativity is explained by the fact that epistemic evaluators form lists of reliable processes on the basis of personally experienced cases. Since these cases are domain relative, so too are the lists formed on the basis of them.

Inductive Reliabilism

I would here like to suggest a refinement of Goldman’s proposed process by which epistemic evaluators come to possess lists of virtues and vices: epistemic evaluators use a system of induction to establish the reliability of any process. So, on my view, a belief is justified if and only if it is formed through a process with appropriate inductive support. (We shall address what ‘appropriate’ means later.) As Goldman notes, epistemic evaluators form the belief that color vision is reliable on the basis of personal experiences. Positive instances give inductive support to the hypothesis that ‘color vision is reliable.’ Accounting for the categorical conservatism mentioned earlier, a novel process does not become a listed virtue or vice based on only a single experience. Given the process is judged as novel, this is sufficient to generate the question “is this process reliable?” A single positive experience, then, is enough to generate the hypothesis “this process is reliable,” making it a candidate for the set of virtues, albeit one requiring further inductive support3.

Different epistemic evaluators no doubt accept a hypothesis as a virtue or vice after varying amounts of instances. Positive instances of the hypothesis ‘color vision is reliable’ are instances when the use of color vision has led to true beliefs. Negative instances are those when the use of color vision has not led to true beliefs. Just as Goldman notes, when a hypothesis produces a high ratio of true beliefs, it becomes a listed virtue (where ‘listed’ simply means that the virtue is placed in the epistemic evaluators’ mentally stored set as Goldman’s account describes). When a hypothesis produces a low ratio of true beliefs, on the other hand, it becomes a listed vice. The ‘ratio’ refers to the number of positive instances (when the hypothesis produces a true belief) over the total number of instances (where the hypothesis produces either a true or false belief). In addition, Goldman’s conception of similarity, in my account, comes down to the transfer of inductive support for one process to a similar, novel process.

The formulation of reliabilism in terms of induction makes clear the reasons behind the domain relativity of lists and the differing processes on such lists between epistemic evaluators. Let ‘x’ be a belief-forming process. For any process ‘x’, the hypothesis, ‘x is reliable,’ requires inductive support of the kind described above. The inductive support consists of positive and negative instances experienced personally by an epistemic evaluator. An epistemic evaluator does not occupy all actual domains. So, positive and negative instances for x are not experienced in all domains. Therefore, the inductive support for x is not domain general; it is necessarily domain specific. Hypotheses become listed virtues or vices after appropriate inductive support. This support is domain specific and so different domains can result in different lists of virtues and vices. The lists of epistemic evaluators, then, are necessarily relative to domain. Furthermore, different cultures provide different domains, so not all epistemic evaluators occupy the same domain.

It is an interesting question whether allowing inductive support by testimony will eliminate individual differences in lists. Others, particularly in the same culture, will report on positive and negative instances of their own vision, audition, et cetera. Would inductive support by testimony be enough to overwhelm the support gained or lost by the total number of personal instances? Such impersonal support might be enough to smooth individual differences in lists, but not to eradicate them. Testimony is not a stable, personal process, but a changing, relative source of information. It may be considered a virtue or a vice depending on an epistemic evaluator’s judgment of the individual or source that is giving the testimony. Relying on testimony from a source that is known to consistently deceive would certainly be a vice. In the absence of information confirming the reliability of a source, testimony may be considered secondary to personal processes. If this is the case, then even a large number of testimonial instances might not overwhelm4 personal instances. In addition, the accepted beliefs of one culture may reduce individual differences in lists relative to other cultures, but will not remove them completely. For example, consider a person who, having had several positive (individual) experiences with astrological predictions, wholeheartedly accepts future predictions as reliable despite cultural, familial, or peer skepticism. So although some differences may be eliminated by testimony and common culture, sets of vices and virtues are still varied and domain relative.

Inductive reliabilism explains why lists of virtues and vices are domain relative and why uniformity is not guaranteed or expected among epistemic evaluators. Let us now consider some problem cases for standard reliabilism and how Goldman’s account fares against the inductive reliability formulation.

The Problem Cases for Goldman:

Reevaluating the Unsolved with Inductive Reliabilism

Justification and the Evil Demon

Goldman examines how his improved account of reliabilism copes with problem cases. If the Cartesian evil demon “gives people deceptive visual experiences, which systematically lead to false beliefs,” are these beliefs based on vision justified? Goldman claims that intuitively, we count the beliefs in this case to be justified (526). He argues that a victim of the demon’s deception has visual experiences identical in kind to those that we experience. The victim would also use the same processes to form beliefs. So, this is enough, Goldman asserts, to label the beliefs justified. An epistemic evaluator judges them to be justified because he matches the vision-based process of the victim with a listed virtue.

Goldman asserts that the victim’s beliefs are justified without further qualification, but I think this is wrong. It seems that the beliefs formed from the demon’s deception are justified relative to the victim of the demon and unjustified relative to an outside evaluator who is aware of the victim’s situation. The beliefs are not justified from our perspective and this points to a limitation in Goldman’s view that he himself should want to overcome. Despite his own account of reliabilism suggesting that justification is relative to perspective, Goldman nonetheless misses the importance of perspective in this case. Goldman notes that the reliability of a process is established on the basis of personal experience (528). Personal experience is relative to a particular evaluator. To be justified is to utilize a reliable belief-forming process. Since the establishment of reliability depends on personal experience, which is relative to a particular evaluator, it is relative to perspective as well. Goldman’s own view suggests this conclusion by noting that individuals of different cultures may have different processes on their lists of virtues and vices (527). Since judgments of reliability are relative to perspective, justification, which is tied to this concept, is also relative to perspective.

Inductive reliabilism refines Goldman’s account and produces exactly the result we need. Goldman’s reliabilism predicts that an outside epistemic evaluator, who already has vision as a listed virtue, will judge the victim of the demon as justified. This, however, is not quite correct. It is stipulated that the victim is led to false beliefs by vision. With inductive reliabilism, we must consider that from the outside perspective, the victim is not establishing the appropriate inductive support for the belief that ‘vision is reliable.’ From the perspective of the outside epistemic evaluator, the victim is facing many negative instances and thus is unjustified. The victim, however, lacks information concerning the falsity of his evidence and establishes apparent inductive reliability based on the same kinds of visual experiences and belief-forming processes as we do. In fact, this is much how our own situation would be if a Cartesian demon did exist! Thus, from the inside perspective, from the perspective of the victim, he is justified. Since justification is a function of inductive support, and the degree of inductive support for the belief is based on perspective, justification is relative to perspective as well. More generally formulated, given any belief, justification is dependent on inductive support and relative to any further perspective-based information regarding the truth or falsity of the evidence and beliefs used in the induction. So, inductive reliabilism refines Goldman’s hypothesis that an epistemic evaluator has a mentally stored set of virtues and vices; it hypothesizes that this set of virtues and vices is based on inductive support that is domain relative.

The Case of Clairvoyance

Now consider a case like those proposed by Laurence BonJour “in which hypothetical agents are assumed to possess a perfectly reliable clairvoyant faculty”.5 Imagine that Samantha alone possesses this perfectly reliable clairvoyant faculty and through it comes to believe that the president is in New York. She also has a large body of independent evidence in favor of the belief that the president is in Washington, but nonetheless she believes he is in New York. In addition, imagine Casper, another reliable clairvoyant, who has “large amounts of ostensibly cogent evidence that [he] has no reliable clairvoyant power,” but still uses the power to form beliefs. BonJour asserts that the beliefs in these cases are not justified, although they are formed by a reliable process.

Goldman argues that these clairvoyants are indeed unjustified in their beliefs. In each case, the agent has a large amount of evidence against the belief that he/she holds. In order to hold the belief, they are ignoring contrary evidence. ‘Ignoring contrary evidence,’ of course, is a vice according to Goldman. So, the epistemic evaluator will match the belief-forming process of these agents with this vice and judge the agents unjustified.

This solution to the clairvoyant case is quite puzzling. Goldman does not clarify exactly what counts as ‘contrary evidence.’ Take the case of Samantha above. Samantha has large amounts of evidence in support of the belief that the president is in Washington. All evidence for a belief, however, is connected to some belief-forming process. So, we are assuming that the evidence in question is gathered from some reliable-belief forming process. Yet, Samantha forms her belief, that the president is in New York, based on her clairvoyance rather than this other evidence. While it is true that she is ignoring contrary evidence in this case, it is not possible for her to avoid this vice. If she forms the belief that the president is in Washington on the basis of the other evidence, she is then ignoring the evidence she has received via her clairvoyant faculty. So, Samantha’s unjustified belief cannot be merely explained by the fact that Samantha is ignoring contrary evidence, but rather it must be due to the kind and quantity of evidence that she is ignoring.

It could perhaps be objected that there is no evidence coming from the clairvoyant faculty. However, it is a process that results in a certain belief and that belief can function as evidence. Likewise, vision is a certain process that yields certain beliefs, some of which are counted as evidence. For example, looking at a reputable newspaper and reading that the President is scheduled to be in Washington utilizes vision. This leads to beliefs about the newspaper, what it says, et cetera. A person cannot somehow consider evidence outside of his own mind; evidence is inevitably tied to a belief-state. Thus, to deny that a certain belief (e.g. one through clairvoyance) can be counted as evidence already presupposes some account of why certain kinds of beliefs can be counted as evidence and others cannot. This is exactly what Goldman is missing, but inductive reliabilism, as I shall argue, can clearly account for this.

Furthermore, ‘ignoring contrary evidence’ does not belong on the list of vices. Although vices are unreliable processes leading to a low ratio of true beliefs, they nonetheless have a basis in evidence and the forming of beliefs, like virtues. In order for any evidence to be ‘contrary,’ it must stand in opposition to other evidence. ‘Ignoring contrary evidence’ as opposed to ‘ignoring all evidence’ seems to imply that some evidence is already accepted, some evidence to which the contrary evidence stands in opposition. So, ‘ignoring contrary evidence’ refers to the exclusion of some evidence in favor of other evidence. The act of ‘ignoring contrary evidence’ does not itself have any basis in evidence or belief formation. Rather, one can only ignore contrary evidence when there is already evidence related to other virtues or vices; it is a relation between evidence. It cannot be individuated as a vice due to its dependence on other vices and virtues. Inductive reliabilism, I shall argue, shows that ‘ignoring contrary evidence’ leads to an unjustified belief without labeling it as a vice while Goldman’s view seems unable to account for this. Indeed, if ‘ignoring contrary evidence’ is not a vice, Goldman’s solution to the clairvoyance cases disappears.

Let us consider the hypothetical clairvoyance cases with inductive reliabilism. In the case of Samantha, she has large amounts of evidence connected to other reliable belief forming processes that she ignores. An epistemic evaluator judges her unjustified because it is assumed that her clairvoyant faculty does not have a higher degree of inductive support than the other processes on which the contrary evidence relies. If an epistemic evaluator knew that Samantha had a high degree of inductive support for her clairvoyant faculty, then the intuition would be different. Suppose Samantha ignored her clairvoyant faculty because of its low inductive support and formed a belief based on a more reliable process. Then, it turns out that her other reliable process led to a false belief, which she later discovers. Memory connects revised true belief to the clairvoyant belief-process she earlier ignored. Imagine that this happens many times. Then, the epistemic evaluator will probably judge Samantha correct in ignoring the other evidence, since the clairvoyant faculty would have surpassed the processes related to that other evidence in inductive support. In the case of Casper, his faculty of clairvoyance is stipulated as having low inductive support. It also seems likely that Casper has enough evidence to deem clairvoyance a vice, in which case its use in belief formation makes the consequent belief unjustified. Even if it is not considered a vice, when Casper relies on clairvoyance instead of a process with greater inductive support, he is ignoring contrary evidence and thus led to an unjustified belief.

Detailing the Differences: Goldman versus Inductive Reliabilism

Inductive reliabilism avoids both of the difficulties that Goldman’s account faces. So far I have granted Goldman’s distinction between virtues and vices. In fact, however, inductive reliabilism eliminates this distinction: we should instead say just that virtues have a high degree of inductive support while vices have a low degree of inductive support. Obviously, the difference between virtues and vices is only a matter of degree and not kind. It is quite possible for the list of an epistemic evaluator to be a continuum of processes with greater or lesser degrees of inductive support. In modeling this continuum, there must be more than a list of ratios (positive instances over total instances). We must take into account the difference between novel processes with few total instances and little or no inductive support and processes that have time and again proven unreliable. Perhaps we could model this difference by considering the negative instances over the total instances as a negative ratio and then summing this with the positive ratio. In this way, processes that have proven unreliable will be modeled by negative numbers while novel processes will remain close to zero.

To understand this continuum, imagine a situation in which no process with high inductive support can be used to form a belief (due to lack of evidence). If all of these processes were truly unavailable, would an epistemic evaluator be unjustified in using the greatest inductively supported process available? For example, grant that vision has greater inductive support than hearing. If we form a belief based on sound, we are not unjustified simply because we have no visual evidence. Likewise, if evidence is absent for all processes down to belief-formation through guesswork, then an epistemic evaluator is justified in forming a belief based on this available process. It must be noted, however, that epistemic evaluators can revise their beliefs if processes with greater inductive support become available (through new evidence). To be justified, an epistemic evaluator must rely on processes with the greatest inductive support for which there is available evidence. If the process with the absolute highest degree of inductive support on the continuum is not available, then the epistemic evaluator slides down the continuum of processes until an available one is found.

‘Ignoring contrary evidence,’ for inductive reliabilism, comes down to an epistemic evaluator ignoring evidence connected to a process with greater inductive support than the one on which the evaluator is relying. Since in this case there is evidence with greater support, it more clearly comes down to favoring a process with less inductive support over a process with greater inductive support. This is, by definition, to be unjustified. Samantha and Casper both do this and thus are unjustified in their beliefs. This conception of ‘ignoring contrary evidence,’ not only shows that the problems that Goldman’s account faces are actually unified, it also completely avoids them. It explains why some evidence is ignored and why other evidence is not. It describes the kind of evidence that epistemic evaluators value: evidence connected to a process with high inductive support. Furthermore, it is clear why ‘ignoring contrary evidence’ is not on an epistemic evaluators’ list. It is defined based on the relation of inductive support between processes in the virtue/vice continuum on which an epistemic evaluator relies in a particular case.

Inductive reliabilism has the further advantage of being able to clearly explain why one can be completely justified in a belief and still fail to have knowledge. Let y be any belief-forming process. On the basis of ‘y was reliable in the past,’ we conclude that ‘y is reliable in this new case.’ Even if it turns out that y leads to a false belief in the new case, the belief remains justified since the belief-forming process had high inductive support. Of course, because it is false, it is not knowledge; but it is justified false belief.

Conclusion

Descriptive epistemology, describing the concepts and norms of the folk, is a necessary enterprise. It can provide the basis for continuity or the basis for criticism in a normative account. If the discussion of inductive reliabilism, at times, seems normative, then it is because we are addressing the norms of the folk. Furthermore, it also may suggest that normative epistemology is, in fact, continuous with descriptive epistemology. Inductive reliabilism would urge us, then, not only to fully model a system of induction that explains our commonsense concepts and norms, but also one that can be included in normative epistemology. Goldman’s attempt to improve upon standard reliabilism meets with several problems and his solutions to problem cases were inadequate. Through the examination of his view and its problems, I have advanced a refined form of reliabilism: inductive reliabilism. It logically extends Goldman’s discussion of ratios of true and false beliefs, personal experience, and mentally stored lists of virtues and vices. The refined hypothesis, that the set of virtues and vices is based on domain relative inductive support, overcomes the shortcomings of Goldman’s account and provides greater depth to the concept of justification in descriptive epistemology.□