THIS IS A DRAFT.
DEFINING KNOWLEDGE
Knowledge is not simply justified true belief, but it is justified true belief justifiably arrived at.
Robert Fogelin
Before analyzing knowledge, we need to make a pre-philosophical analysis of the meanings of the concept-word ‘knowledge’. Although one can find a variety of definitions in dictionaries, there are at least three main meanings or concepts of knowledge usually selected by epistemologists.
The first concerns ability knowledge, often called knowing how (to do something). It is the knowledge one has of swimming, bicycling, walking, speaking a language, playing the violin, cradling… Even animals have this kind of knowledge: a bird knows how to make a nest. The form is: “the living being a knows how to perform the activity x.” This procedural kind of knowledge is for us less relevant, not only because it is shared with animals, but because it is usually devoid of conscious awareness. It often appears after a process of automatization: one thinks about the movements one makes in cycling only in the beginning; soon one learns to make them automatically, without any conscious awareness. But in some cases, it is innately established. Consider the sequence of movements a human being makes when crawling. Are you able to specify the exact, true sequence?
The second selected sense is acquaintance knowledge. In order to have it, one needs to be personally acquainted with a person or a thing. Examples are given by sentences like “I know Mary”, “I know Paris”, ”I know Niagara Falls”. However, one cannot have this type of knowledge relative to something one has no personal acquaintance with. I cannot say that I know Moscow or that I know China, because I was never in these places. All I can say is that I know some facts about Moscow and some other facts about China. Of course, I can say that I know Aristotle, but this is an extended sense of the word. I do not mean that I was ever introduced to this philosopher. It is an indirect way to say that I have studied his philosophy. The form of knowledge by acquaintance is “I know x”, where x is a singular term like Mary or China. (French has different verbs for this type of knowledge, namely, ‘connaȋtre’ as opposed to ‘savoir’, which is also the case with other Romance languages, e.g., Spanish: conocer vs. saber. In German: wissen vs. kennen.)
The third and by far most important sense of knowledge is that of propositional knowledge. It is also called knowing-that, because very often (though not always) in its verbal expression, after the verb ‘to know’ comes the preposition “that”, and after the preposition “that” comes a declarative sentence expressing a proposition that can be true or false, which is essential to this kind of knowledge. Examples are “I know that 7 + 5 = 12”, “I know that the Eiffel Tower is located in Paris”, “I know that neutrinos have no electrical charge”.[1] Propositional knowledge is fundamental to us, because most of our sciences, most of our culture, and even our everyday knowledge is of this type. When epistemologists speak of analyzing knowledge they are usually referring to propositional knowledge. It has a complex nature that will be analytically deciphered in a satisfactory way in the course of this chapter.
A well-placed question here would be: is there enough semantic proximity between these three types of knowledge to justify the use of the same word ‘knowledge’ to classify them? I think the answer is ‘yes’. One characteristic of knowing-how is that in humans it is usually learned and that in the process of learning we begin by becoming aware of the sequence of things to be done, for example, the right physical movements made in cycling or swimming. There must be some knowing-that at first, even if afterwards these movements become automatic and we lose any awareness of them. (Another example: When people learn to touch type they at first learn the keyboard, but with experience they are able to find the keys without looking at the keyboard but cannot state the sequence of letters.) Hence, knowing-how is often a result of knowing-that. Concerning knowledge by acquaintance, it is associated with propositional knowledge, since it entails the latter. If I know Mary or if I know Paris, this entails that I know a good number of true propositions or facts about that woman and that city. Finally, it is important to remember Russell’s view, according to which acquaintance knowledge would be the primary form of knowledge (Russell 1912, Ch. V). We must first have some kind of sensory-perceptual acquaintance with empirical things in order to reach a basis for forming propositions detached from the existence of the facts they should represent.
Traditional definition of knowledge
A further point is how to analyze propositional knowledge. Plato seems to have been the first person to make this kind of analysis. According to him (propositional) knowledge is true belief with a logos, a reason[2]. Understanding the ‘reason’ as a justification, we come to the definition of knowledge as justified true belief that has passed through the whole history of philosophy until it was challenged by Edmund Gettier in a famous short article (1963). This traditional, standard or tripartite definition of knowledge is intuitive. You cannot know that the statement p is true without believing in its truth, and if you believe in its truth usually you have some kind of justification: A school-boy who professes to know that Columbus discovered America not only believes in the truth of this claim, but must also be able to tell us something in order to justify this statement. Before going into the details of this definition, I will consider more closely each of these three conditions.
Consider first the condition of truth. We cannot properly say that we know the false. Anna cannot know that the moon is made of Swiss cheese. It is true that we can say that the old Greeks knew that the Gods lived on Mount Olympus or that we once thought we knew things that turned out to be false. But in both cases the verb ‘to know’ is not used in its literal sense. This is shown when we replace the word ‘knew’ in those sentences by the literal expression, which is ‘believed to know’.
Concerning the condition of belief, it is important to see that a belief is an attitude towards a proposition. A proposition (or thought) is what is said by a declarative sentence. The two declarative sentences “Arminius defeated the Romans” and “The Romans were defeated by Arminius” are different, but they say the same, that is, they express the same proposition. Consequently, if we believe in its truth, we believe in the truth of the proposition expressed by them.
One cannot know that something is true without believing that it is true, except in a deranged state of mind in which the degree of belief is distorted by feelings. One case is that of stress: in an oral exam a nervous student does not believe he knows, but he often answers questions correctly. Another case is that of wishful thinking: an elderly English lady believed the impostor Tom Castro was her long lost son, despite all evidence to the contrary.[3]
However, these are not the kind of beliefs meant in the traditional definition of knowledge as justified true belief. Our proper epistemic beliefs must be grounded on some positive probability of truth that we rationally give to a proposition; they must be rational beliefs. In this epistemically relevant sense, beliefs not only come in degrees, but we are even able to find words approximately corresponding to these degrees: if the probability is 1 or above 0.9, we often use words such as ‘certainty’, ‘conviction’, ‘confidence’; if it is around 0.7 or 0.8, we use the words ‘opinion’ or ‘standpoint’; if it is somewhat higher than 0.5 we use words like ‘hunch’, ‘inkling’, ‘suspicion’; if it is around 0.5 we use the word ‘doubt’ or ‘hesitation’ (or ‘suspension of belief’), and if it is below 0.5 we use the word ‘disbelief’. Disbelief is the belief that a proposition is false, which is the same as belief in the negation of the proposition.
When we feel ourselves able to see our degree of belief as proper to knowledge, we say that we are certain of its truth. Certainty is the key-word here. But what is the criterion for certainty? To this, I would suggest that the answer must be pragmatic: it must be a probability high enough to have the expected relevant practical consequences, acknowledging that this relevance might change with the context. If my friend asks me if I closed the door of my apartment as I left it, I answer that I know for certain that I closed the door, since I do this by habit. But if a policeman asks me if I know that I closed my door, I have to be more cautious, answering that I really do not know, though I believe I did.
Another objection to the condition of belief is a well-known counter-example offered by Colin Radford (1966):
Jean is a French-Canadian who claims not to know any English history. He is given a verbal quiz on English history. He answers questions hesitantly, but gets many answers right. One question asks the date of Elisabeth I’s death. Jean says, I’d just be guessing, but, um, lets say, 1603. This is the correct answer. Suppose in fact that Jean did learn this answer, along with many others long ago in school and that his present “guess” is based on a vague memory that in fact traces back to his learning the date of Elisabeth I’s death.
The suggestion is that Jean knows the date of Elisabeth I’s death without believing it. Although this counter-example has embarrassed many, I think that in a real case we would see in Jean’s answers a serious problem of likelihood. As admitted, he does not get all answers right… Thus, in the same way as he said 1603 he could perhaps someday say 1613… He is right in saying that he is just guessing. What he has is a belief with a probability that is somewhat higher than 0.5, enough to be called a good hunch, but not enough to guarantee knowledge. He does not have knowledge, not because he lacks belief, but because he lacks the certain degree of rational belief required for knowledge. In a similar way, if an experienced psychiatrist says she can recognize a schizophrenic patient just by looking at him, she is resorting to hyperbole. She only believes, since she lacks the level of rational belief that this case requires for knowledge.
Finally, there is the condition of justification. Most knowledge-claims clearly require justifying evidence. I know that the Apollo XI mission landed on the Moon, because I saw it in the film they transmitted back to the earth. I know that ‘The Fate knocks at the door symphony’ was composed by Beethoven, because I have read about the composition of this work. I can say that I know that my car is at the university, because I remember leaving it in the parking lot an hour ago. We can all call these epistemically acceptable justifications good or reasonable. However, not all justifications are good in this sense. I cannot justify my statement that a person will die soon, just because I read this in the life line on her palm. I cannot justify my statement that I flew to the moon last night only because I dreamed that I was on the moon. We can define a good or reasonable justification as something that is prima facie acceptable by the majority of people belonging to the epistemic community to which the knowledge-claimer belongs. (One can imagine different societies, where reading the life line in a person’s palm and the dream that one has flown to the moon are very good justifying evidence of the expected life of a person and of a real visit to the moon.) Finally, the fact that a justification is good or reasonable does not mean that it produces true belief: it can be good and produce false belief. The Newtonian gravitational law, for instance, was considered perfectly justified in the nineteenth century, and all the contemporary epistemic community of physicists would have agreed with the truth of this law. However, after Einstein presented his general theory of relativity, Newton’s gravitational law came to be considered strictly speaking false, since this law has been shown to be merely approximately true. Moreover, compared with the justifications given for general relativity theory, the justification for Newtonian gravitational laws has turned out to be deficient as well.
Formulating the standard definition symbolically
Symbolic logic has allowed us to formulate the traditional definition of knowledge symbolically. Calling ‘p’ a proposition in question, ‘a’ a knowledge-claimer, ‘B’ his or her belief in the truth of p, ‘K’ the knowledge-operator, and ‘E’ the justifying evidence for the truth of p given by a, we can say that: (i) that aKp → p (If a knows p, then p is true); (ii) aKp → aBp (if a knows p, then a believes in the truth of p); (iii) aKp → aEBp (if a knows p, then a has justifying evidence for the truth of p). Combining these three conditions, the traditional or standard or tripartite definition of knowledge can easily be formalized as:
(i) (ii) (iii)
aKp = p & aBp & aEBp
We can call (i) the condition of truth, (ii) the condition of belief, and (iii) the condition of justifying evidence or justification. Each of them should be considered necessary, and the conjunction of the three should be seen as a condition sufficient to establish knowledge. It is true that we can simplify the definition by excluding the condition of belief, since it is repeated in the condition of justifying evidence (which it is a justification for a belief in the truth of p) and write: aKp = p & aEBp. This would be more economical, though less transparent.
Important exceptions are so-called basic propositions, which offer justifications like “I have headache” or, maybe, “I am seeing a blue sky” and “~(p & ~p)”. We know the truth of these statements because they impose themselves on us in a non-inferential way, assuming that expected adequate conditions (I am not being induced to believe that I have a headache… I am looking at the sky outdoors on a sunny day…) are all fulfilled. We can formalize the knowledge claim of such basic propositions simply as:
aKp = p & aBp
At least from the usual perspective , knowledge of basic propositions does not need justification, because they are self-justifying in the sense that they have what we see as a non-cognitive, non-doxastic evidential source.
Evidentialist versus reliabilist theories of justification
What is the nature of epistemic justification? There are two main competing theories: reliabilism (Armstrong 1973; Goldman 1979, 1986, 2010, 2015) and evidentialism (Feldman & Conee, 1985, 2004). Reliabilism emerged in the eighties mainly in opposition to reliabilism. In what follows, I will present some arguments favoring evidentialism against reliabilism. But I will first try to make a fair summary of each theory.
Evidentialism is the view endorsed by the philosophical tradition. It is intuitive, since it seems that there is no justification without some kind of evidence. It can be generally defined as follows (Mittag 2017):
[EJ] An epistemic agent a is justified in believing that p in time t iff the evidence E that a has for p supports his belief in p.
Richard Feldman and Earl Conee noted that evidence is necessary to justify belief, suspension of belief, and disbelief (1985). For them, evidence must be any information relevant for the truth or falsity of a proposition. This can be beliefs that are used as evidence for other beliefs or simply experiences that are direct evidence for beliefs, for instance, in perception. Thus, my evidence for the belief that my car is at the university can be my visual memory of having left the car in a parking lot, which is another belief. But my evidence that I am seeing my car in front of me should be simply the experience of perceiving my car in front of me. For these authors, evidence must be mental states in the knower’s mind, even if he is unable to have introspective awareness of them. This means that their theory is internalist, since internalism requires that the justification of a belief is after all internal to the believer. Important is here the support relation. Evidence must be able to support the belief corresponding to it, giving it a probability higher than 0.5 (which as we saw, in the case of epistemic certainty should be near to 1.0). A final point is that we need to select at least three kinds of evidence: (i) evidence that is ‘before our minds’ (I am seeing the blue sky), (ii) evidence that is stored in memory, but that is able to be actualized when necessary as a way to support the belief, (iii) evidence that is stored in memory, but that is unable to be actualized, though still linked to the believer in order to support it.
The competing view is so-called reliabilism. The most successful reliabilist theory to date is Alving Goldman’s process-reliabilism. According to him, a justified true belief must be produced by a reliable causal cognitive process, which can be defined as an empirical mechanism that makes the truth of the belief probable, that is, with a probability greater than 0.5. Considering that empirical justifications can usually be defeated or overridden by other competing reliable cognitive processes, Goldman defines justified belief in a broad way as (1979):
(RG) The belief of a in p in t is justified iff it is (i) causally produced by a reliable cognitive process, and (ii) there is no reliable cognitive process accessible to a so that if applied it would result in the negation of a’s belief in p.
Goldman adds that there are conditional and non-conditional reliabilist processes. Conditional reliabilist processes have beliefs as inputs and beliefs as outputs (these output beliefs are what we usually call non-basic beliefs). Non-conditional reliable processes are those that have experiences as inputs and beliefs as outputs (these output beliefs are what we used to call basic beliefs).
Goldman’s reliabilist theory of justification is externalist, which means that the believer does not need to be able to have any awareness of the causal process that leads to his belief. Some of his examples seem to show the appeal of his theory. Mary reads in a magazine called Gossip that her favorite Hollywood couple is divorcing (2015: 35-36). After one week she still thinks that her favorite couple will divorce, but she has forgotten where she learned this. However, since this process of belief acquiring and belief retention is reliable, Mary is justified in her belief that the couple will divorce. Another example (1968): Goldman knows that Lincoln was born in 1808, but he does not need to be able to justify this knowledge. Indeed, we all know a lot without remembering the evidential sources, things like telephone numbers, passwords, historical dates, equations. The evidential sources can be completely forgotten. Moreover, animals and small children know a lot, but they would not be able to justify, even for themselves, their knowledge (1968, 1978, 1986). It seems that a reliabilist theory of justification has the advantage of demanding only the existence of those reliable cognitive processes that lead to these different instances of knowledge, without demanding any actual or possible introspective evidential access. This is why reliabilism is considered a so-called externalist theory of epistemic justification. Even if internally often accessible, what really matters is the causal process that originated the belief, even if this causal process is only externally accessible. This is the case of Mary. She knows about the impending divorce, which can be acknowledged by a third person. But its evidential origin is completely forgotten by Mary.
Defending evidentialism
Although I agree that there must be a causal cognitive process generating belief, I do not agree that in itself this is what accounts for justification. To see this, we need only re-examine Goldman’s examples from the evidentialist perspective, showing that for all these examples we can find decisive internal evidence, even if we admit that these internal evidences must in the perceptual case cope with externally given evidences, that is, evidences belonging to the external world and, therefore, following natural laws and able to be interpersonally observed. Thus, I inductively know that the fixed memories of my telephone number or of my password or of an historical date, which have the property of returning to my memory when I try to call them up, are correct. But how? Well, I know these things because I have already checked the truthfulness of my memory numerous times in objective (potentially intersubjective) ways, for example, giving my telephone-number to a friend, storing the password in my computer, or recalling correctly to others the date of the French Revolution, and this is a kind of experience often subject to interpersonal checks. These examples are justifying evidence! Once one begins to fail in such checks, e.g. because of Alzheimer, they cease to be considered evidence. All these things can be presented as evidence for the trustworthiness of memorized information, without the need for recourse to historical sources of justification.
There are also many cases of knowledge in which the agent isn’t able to give any justification at all, but that are behaviorally recognized as cases of knowledge by a third person. For instance, I know that a child is able to identify her mother’s face, because she always smiles as she approaches the child, and I know that a dog senses that his owner is arriving because he hears the sound of the car and he runs to the front door. These behaviors are complex evidence that we arrive at as a third person by means of reliable evidence that gives us real beliefs producing verifiable behaviors by ourselves and consequently by children and animals. Consequently, these behaviors are also complex evidence by analogy that the baby and the dog have mental states that work for them as evidence like ours, which like ours justify their beliefs. The baby recognizes the visual evidence of her mother’s face, and the dog applies the evidence of the sound of the car as the inductive warrant of the arrival of his owner. These are actual reasons for their beliefs. The point buried in the discussion is that it is by means of third person (internal) reconstruction of internalist justifying evidence of others that we conclude that they know, even if they are unable to be reflexively aware of what they are really doing. Internal justifying evidence or evidence of evidence come first.
Consider, finally, the following intriguing case. Henry loves meat and always eats all kinds of meat when he goes to a restaurant. One day by mistake he enters a vegetarian restaurant called ‘Food for Thought’, and when he asks for meat the waiter tells him that the last thing they would do in this restaurant would be to serve meat. Upset, he leaves the restaurant. Many years later he receives a call from a friend inviting him to have dinner in a beautiful restaurant called ‘Food for Thought’. Although he had this incident stored in his long-term memory, he fails to recall the name of the restaurant and, based on his knowledge of the fact that most restaurants serve meat, he immediately agrees. Later, informed that it is a vegetarian restaurant, he remembers the incident and rejects the choice. At first view, it seems that there is a problem for the evidentialist here. Henry should already in the beginning reject the invitation, since he has evidence that it is a vegetarian restaurant in his long-term memory. I think the change of opinion only shows the diversity and limits of the relevant justifying evidence. At first, the knowledge that most restaurants serve meat was evidence for the view that ‘Food for Thought’ would serve meat. Later, the information that ‘Food for Thought’ was a vegetarian restaurant triggered Henry’s memory of the earlier mistake, which now serves as the justifying evidence for the conclusion that this is the wrong restaurant for him. The example speaks for access evidentialism, questioning the possibility (iii). There is, however, some plausibility in (iii), when we consider unconscious justifications. A woman can be unconsciously justified in her belief that she could be loved by someone based on reasons she is unable to retrieve (this is what some call ‘women intuition’).
We can justify our point further, turning on its head an example often used by reliabilists against internalists. The example is the following: suppose I am a brain in a vat with my afferent and efferent nerves linked to a super-computer on the planet Omega, and that the program of this super-computer gives me the impression that I am looking at this screen now. The internalist justification, they say, fails, because the evidence is a kind of forgery: I am not writing a real sentence in a notebook in a room situated in a city on a planet called Earth. The only way to show that this internalist justification fails, they say, is externalist. People from the planet Omega, for example, the programmers of the super-computer, know on external grounds that my justification is based on false evidence, since the evidence is only electronic patterns produced by the program running in the super-computer. The point, however, is the same as above. People from the planet Omega only know that (from their perspective) my justification is inadequate because they have their own internalist justification for their judgments based on their own internal evidence. They know that my evidence is insufficient because they have access to a much larger information set, able to overthrow it. In conclusion: the original mechanism of justification is always internalist. The fact that others can have reasons to reject my justification does not change this state of affairs.
Another controversial suggestion from Goldman is his rejection of foundationalist and coherentist theories as non-historical “current time-slice theories”, which take account only of the actual justification of beliefs (2010: 144; 2015: 26). Moreover, he understands his externalism as including external “justifiers” and external “truth-values” (2010: 145). But if the external justifiers could really be incorporated, justification would not be probabilistic, but infallible.
Reliabilism hides a persistent lack of explanatory power. The role of the input, which we might call justifying evidence, is underscored in this theory. Goldman’s theory focuses on the causal cognitive reliable process but has nothing relevant to tell us about the process. If it is a reasoning process, it can fairly belong to what we call evidence, for instance, when someone makes a mathematical calculation, the evidence for the result extends itself through the whole procedure. Moreover, what we typically call ‘reliable’ is not a psychological process, but own evidence, and it is the evidence that makes the psychological process relevant, insofar as it retains the evidential information (e.g., “I still remember the address”). We say, “Tutankhamun really lived, because we have reliable evidence of his existence”, but we do not say “Tutankhamun existed because there are reliable causal cognitive processes that make probable the belief in his existence”. Even if I do not intend to deny the existence of these causal processes, it is curious that we usually have nothing to say about them when we present a justification. It also seems that it is not by means of a causally reliable process, but rather by means of reliable justifying evidence (that may include a reasoning process) that we measure the probability of a belief. Moreover, there are many causal processes involved, and it is because we are able to detect reliable evidence that we have the thread to find the relevant causal process, and not the other way around.
I conclude that evidentialism is the best way of giving a fundamental role to evidence, and that this evidence must make itself reliable by cognitively causing belief. Evidence and evidential processes work causally, of course. And although they must be internal, in any perceptual case they are only able to be reliable insofar as they are seen as corresponding to external evidence. We can summarize these conclusions in the following definition of justification:
[EJ*] An epistemic agent a is justified in believing that p in t iff p is for a at t causally supported by some reliable evidence E (which can be another belief or an experience), and if there is for a no counter-evidence able to defeat the reliability of E for p in t.
This is an easy way to preserve the many advantages of evidentialism without having to sustain the deficiencies of reliabilism.
Gettier’s problem
Now we will consider the pain in the neck of contemporary epistemology: the so-called Gettier problem. In 1963 Gettier presented two cases in which the three conditions of the tripartite definition of knowledge seem to be satisfied, although there is in fact no knowledge, rendering this condition insufficient. The result was a flood of articles and books, either trying to add a fourth condition to the tripartite definition of knowledge, or trying to offer a substitute for it, or even claiming that a definition of knowledge is impossible. All this work created a new sub-field of epistemology called ‘analysis of knowledge’. I will first explain Gettier’s problem, and then I will show how it can in my view be successfully answered.
There are many counter-examples of Gettier’s type. I will choose one of them. Suppose that yesterday Professor Stone said to Mary that he would be at the University this morning in order to hold a doctoral examination. Now, since Professor Stone is an extremely correct professional (hard as stone), Mary has a very good justification to believe that he is at the University now (10 a.m.). Moreover, when Mary says that she knows that Professor Stone is at the University, he is indeed at the University, which makes her statement true. Hence, we see a justified true belief. Nevertheless, in fact Mary does not know it! And, the reason is that Professor Stone’s three teenage children were involved in a severe car accident last night and he has cancelled all his appointments today in order to stay with his children in the hospital. However, by mere coincidence, Professor Stone briefly returned to his office at the University to fetch some documents and then hurry back to the hospital. Since the justification given by Mary does not have anything to do with what makes the proposition true, we reject her knowledge claim. However, even though the three conditions of the traditional definition are satisfied, they do not constitute taken together a sufficient condition of knowledge, which means that there is something wrong with the traditional definition as we have understood it.
Path to a solution
It is often said that the paths to the false are many, while the path to the truth is only one. The first time I read about Gettier’s problem the real answer seemed obvious: the justifications given in these counter-examples to the classical definition, though good, were not adequate, because none of them was able to make the proposition true. Since I felt that this answer was too intuitive not to be noticed, I went through the literature on Gettier’s problem, searching for someone who had said something similar. And indeed, I found what I was looking for.
As far as I know, the first attempt to develop this insight appears in Robert Almeder’s papers, written in the Seventies. And a later attempt appeared in a book written by Richard Fogelin, published in 1994. The answer was there, though incomplete and more complicated than it seemed at first view. I was not the only author to see things in this way. Here is a passage in an old introductory text from Brian Carr and D. J. O’Connor with which I am in full agreement (1984: 81):
For a justified belief to constitute knowledge it would appear that there should exist a connection between the truth of the proposition believed and the grounds on which it is believed. The reason why the proposition is true must not be independent of the facts asserted in the propositions constituting the grounds for the belief. Or to put it in different terminology, those justified true beliefs which constitute knowledge are those in which it is not just a coincidence that the believer is right but where the belief has been arrived at on the basis of facts which are relevant to the truth of the belief.
Further, they notice that it is a flaw of traditional analysis that it allows the three conditions to be independently satisfied. This flaw in the standard analysis can be eliminated, they write, “not by adding some fourth condition to the other three, but by insisting that these three previously recognized conditions should not be independently satisfied.” (1984: 82). And they conclude, though still in need of clarification, this straightforward effective way to solve Gettier’s problem should be further developed: “It is somewhat surprising, therefore,” they comment, “that it is not a response to the Gettier problem which has found much support in the considerable literature on the subject.” (1984: 82)
Now, the next pages are dedicated to the development of this program. I will first discuss Robert Almeder’s and Richard Fogelin’s solutions, which have different focuses. Then I will develop what seems to me a sufficiently complete conservative analysis of the idea of knowledge as justified true belief, able to answer Gettier’s problem without leaving unsolved difficulties behind.
Almeder’s and Fogelin’s attempts
As I noted, a first step in the right direction was made earlier by Robert Almeder (1974). His solution emerged from the perception that in Gettier’s examples the justification given by a has nothing to do with what makes the proposition p true. Consequently, what the traditional definition needs is to show the right relationship between the condition (iii) of justification and the condition (i) of truth. According to Almeder, this relationship should be one of entailment. The justification must entail the truth of p. Using => to symbolize entailment, we can formulate Almeder’s version of the traditional definition as:
aKp = p & aBp & aEBp & (E => p).
There is, however, a serious problem with Almeder’s solution. The requirement of entailment is too strong. The solution works well for formal knowledge, when the justification is deductive. In this case, the justificational evidence allows us to make the proposition true by means of something like entailment. But it does not work with empirical justification, since this justification has an inductive form and cannot have the strength of entailment. We do not wish to have a solution that precludes empirical knowledge.
A more hopeful solution is that of Richard Fogelin. This author avoids the attempt to establish a precise logical relation between conditions (iii) and (i). All he demands is that justification E makes proposition p true for us. Consequently, his version of the traditional definition of knowledge can be informally stated as:
a knows p =
(i) p is true
(ii) a believes that p is true.
(iii) a has a justification E for her belief that p is true.
(iv) a’s justification E makes p true.
This would not be a great contribution if Fogelin had not considered a more important point. As he writes, person a has a certain body of information by means of which she comes to her justification E for p. We, however, have more information than a possess, and based on a wider informational set, we see that the grounds given by a do not justify p. Then he concludes (1994: 23):
I think that this double informational setting – this informational mismatch between the evidence possessed by a and the evidence we are given – lies at the heart of Gettier’s problem.
Indeed, we know that Mary in the above example does not know that Professor Stone is at the University now because we are aware of information that she lacks, namely, that there was a car accident the night before and he cancelled his appointments at the University in order to be at the hospital.
Almeder suggested the necessity of establishing a relation between the condition of justification and the condition of truth, even if he does not give us the right logical relation. Fogelin introduced a third person, whom we could call the knowledge-evaluator s, who will judge whether the justification given by knowledge-claimer a makes the proposition p true or not, in the first case deciding that a knows p and in the second case denying a’s knowledge of p. But Fogelin does not explain how this conclusion is arrived at. That is: in this aspect Almeder’s solution is too stringent, while Fogelin’s solution is too loose. The establishment of a more precise and therefore more adequate relationship between the condition of justification and the condition of truth is the problem that will occupy us now.
Perspectival definition of knowledge
We can summarize Fogelin’s view as follows. In Gettier’s examples there is a mismatch between what we could call the informational background of knowledge-claimer a and the informational background of knowledge-evaluator s (who often represents a community of ideas, but can instead be the same a at a later time). Knowledge-evaluator s is better informed. Because of this difference, knowledge-evaluator s does not accept the justification given by a for the truth of p, even if the knowledge evaluator has his own sufficient reasons to assume the truth of p.
For instance: Carl (knowledge-evaluator s), who is speaking with Mary, met Professor Stone some minutes ago, and he told him about the accident. He knows that by chance Professor Stone is at the University now, but he is also informed not only about the accident, but also about Professor Stone’s decision to cancel his activities at the University today. As Carl hears Mary’s statement p = “Professor Stone is at the University now”, and he hears as justification the information that yesterday Professor Stone told her that he would hold a Ph.D. exam in this morning, Carl refuses to accept Mary’s justification, because he knows that it is completely inadequate as a way to make the proposition p (that Professor Stone is at the University now) true, and consequently he rejects Mary’s knowledge claim. He would accept Mary’s justification and the consequent knowledge-claim if she said she had met Professor Stone in the corridor of the University, or if she said she had seem Professor Stone’s car parked where he always left it, since these justifications would be consistent with Carl’s informational set.
This insight can be made more precise in the form of what could be called a perspectival definition of knowledge[4] (Costa: 2011, 2014). This definition requires a revision of the condition of truth (i) and of the condition of justification (iii) of the traditional definition of knowledge.
I begin by reconstructing the condition of truth. It is common to consider the condition of truth as the truth-value of the proposition independently of any epistemic agent. This is, however, an illusory fata morgana. It is the illusion that we can give any actual use to the absolute truth-value of a proposition. Only God, the infallible knower, would be able to tell us the ultimate truth value of any or almost any proposition. But since our communication with the infallible knower is as unverifiable as his own actual existence, we are fated to remain in the dark about this. Indeed, if the absolute truth-value of p were demanded, we would not be able to know anything, since the ultimate truth-value of our propositions would always remain beyond our reach. Of course, we might have the normative concept (a Kantian idea) of absolute truth, and we pragmatically proceed as if we had reached the final truth when we accept something as true, but we are painfully aware that this final truth can always vanish in thin air when it bumps into some obstacle along the way.
For such reasons, the only way to understand the condition of truth is to relativize what real epistemic agents are be able to determine as truth. In order to get this in the traditional definition, we demand that the knowledge-evaluator s must offer a set of reasons (equivalent to justifying evidence) for truth, each element of the set being evidence sufficient for the acceptance or non-acceptance of p’s truth, a set that we might call the justifiability body of evidence E* for p or E*p. Assuming that the knowledge-evaluator is rational, either he has a body of evidence in which each piece of evidence is considered sufficient to make the proposition true, or he has a body of evidence in which each piece of evidence is considered sufficient to make the proposition false (it would be unreasonable to have evidence sufficient for the truth and also for the falsity of p in the same set). An instance of the first case (the only one that interests us) makes the point clear. If p is the statement “The earth is round”, s can have accepted this as evidence for p: E1 = “We have authentic photos of the earth taken from telescopes in outer space”, E2 = “Ships sailing away from us always seem to eventually disappear below the horizon, beginning with the hull”, E3 = “There are many stories of circumnavigation of the earth”. Each of these justifications is for s sufficient to warrant the truth of p. If p = “The earth is round” and E*p is the body of evidence for p, this set is made up of{E1, E2, E3}. Calling the sign ‘~>’ an attribution of probability able to give certainty (that is, of 1, for the cases of formal evidence, or at least sufficiently near to 1, for cases of empirical evidence), which could be called the probability of epistemic acceptance[5], we can rewrite the condition of truth p (i) as E*p & (E*p ~> p) or (i’). Indeed, if s has an E*p and from any evidence belonging to E*p (assuming his rationality) he is able to derive the certainty of p, then he must accept the truth of p. In other words:
(i) E*p & (E*p ~> p)
After making explicit what was hidden in the condition of truth, we move to condition (iii), the condition of justification, which we hope to be able to link with the new formulation of the first condition. The condition of justification must be written so:
(iii) aBEp & (E ∈ E*p).
This condition of justification requires that the evidence given by any knowledge-claimer a either belongs to a pre-existent E*p accepted by the knowledge-evaluator s at t or can be accepted by s at t as belonging to an extended form of E*p, which includes E. With the help of these few formal devices, and adding to s the time of evaluation ‘t’, we get the following epistemic equivalence, establishing the conditions that must be fulfilled for s’s attribution of knowledge to the knowledge-claimer a;
(1) stK [aKp] = stK [E*p & (E*p ~> p)] & aBp & [aBEp & (E*p ∈ E*p)].
Here we can clearly see how the condition of justification is related to the condition of truth. If an s has an E*p that gives him certainty of the truth of p, and the justifying evidence E given by a is such that it belongs or is able to belong to the s’s E*p, then the justification is not only good but also epistemically adequate. Gettier’s cases are based on good justifications that are not epistemically adequate, since they lack the expected relation to E*p.
It is important to see the role of ‘t’, which is the time of the evaluation. It is essential because the E*p that s gives to a belief can vary from time to time. For instance: When Columbus discovered the New World, he claimed p = “I discovered the sea route to India”. Most evaluators accepted this sentence as true in 1492. But ten years later, the relevant informational set of people had changed. Columbus continued to believe he had discovered the sea route to India until his death in 1506, although around this time most evaluators would have judged his claim false. They would not have accepted his justifications as sufficient to make the proposition true, based on the increasing amount of information showing that he had in fact discovered a new continent.
We see that the time of evaluation is essential for the acceptance of epistemic equivalence from the evaluator’s perspective. The next step is to place epistemic equivalence at the level of an assumption. Since stK is present on both sides of the equivalence, we can bring it to the background and formulate the following definition:
(2) aKp (for s in t) = [E*p & (E* ~> p)] & aBp & [aBEp & (E ∈ E*)].
Finally, if you wish, since the condition of belief is repeated in the condition of justification, we can elide it and get the shorter formulation:
(3) aKp (for s in t) = E*p & (E*p ~> p) & aBEp & (E ∈ E*).
The point of any of these formulations is to link the condition of justification to the condition of truth in the appropriate way. In the case of the statement “The earth is round”, if someone, as a knowledge-claimer, says that the earth is round because of the many artificial satellites orbiting the earth, we, as the knowledge-evaluators s, will accept this, even if we have not thoughts about it, since we know that our knowledge allows its inclusion as an element of the justifying corpus E*. Some examples will show that if this response is well-understood it is seemingly flawless.
Consider now the Gettierian cases under the light of the perspectival definition. Mary claims to know that Professor Stone is at the University now. Since her justifying evidence E cannot belong to Carl’s body of evidence for the truth of this claim, not even to its possible extensions, as the evaluator of Mary’s knowledge claim, Carl rejects Mary’s knowledge-claim.
Another example is that of Bertrand Russell’s stopped watch. Suppose that at time t1 you look at your watch. It shows 11:15 a.m. Then you look the church clock on the other side of the square: it also shows 11:15 a.m. It seems clear that that you are well-justified. But then, in the following moment you remember that your watch was not working properly yesterday. You look at the watch again and see that it has stopped. Probably it stopped last night at 11:15 p.m. Now, at first you had good justifying evidence for a true belief, since the hands of our watches are normally reliable. But after you noticed that something was very wrong with your watch, you conclude that your evidence for the time was flawed and you didn’t really know the time.
In this case the knowledge-evaluator is yourself at a later moment. At time t1 you accept the usual justifying evidence E you have given to yourself. But at time t2 you have the information that the watch is not working and you come to the conclusion that E cannot belong to your present E*, according to which only the time shown by the church clock gives the right justification for your present knowledge that now it is 11:15 a.m. Your first evidence was good, since our watches are normally reliable, but it was not adequate for knowledge, since it was unable to make p true, making your first knowledge-claim a Gettier case.
A good perceptual example is the following. Carol is visiting a region of the country she does not know. The driver of the car, Mr. Smart, knows the region from living there a long time. After crossing a bridge, Carol, glancing out of the car window, comments, “What a beautiful red barn we see in this field!” This exclamation includes the knowledge-claim of p: “There is a red barn in this field”. However, it is only by chance that what she sees is really a red barn – for with the exception of this one, all the red barns in the vicinity are really only barn façades, which were built for a film, although they are convincing enough to fool even the most observant traveller. Although a satisfies the conditions of justified true belief as stated in the traditional definition, for Smart, the knowledge evaluator, a does not satisfy these conditions as demanded by the perspectival form. For in this form, the knowledge-evaluator Smart needs to consider the reasons for belief in the truth of p, which always arises from a knowledge evaluator’s point of view. Now, since Smart lives in the region and knows that Carol is not aware of the story of the fake barns, he knows that Carol has identified the only true Barn by chance and that her justifying evidence isn’t sufficient. In order to have justifying evidence that could be incorporated into Smart’s body of evidence E* for p, Carol should give evidence like the examination of all sides of the barn, or, for instance, telling Smart that she already knew about the barn façades and that the only real barn would be this one after the bridge.
The last counter-examples to be examined – admitting that I am already testing your patience – are those of Gettier’s own article. He gives two counter-examples to the traditional definition of knowledge. In the first one, two persons, Smith and Jones, have applied for a certain job. Since the president of the company has assured Smith that Jones would be accepted, and since Smith knows that Jones has ten coins in his pocket, Smith has the best evidence for the knowledge claim (a): “Jones will gain the job and Jones has ten coins in his pocket”. Moreover, from (a) Smith infers (b): “The man who will gain the job has ten coins in his pocket”. However, against all expectations, Smith and not Jones gets the job. Furthermore, by pure coincidence Smith also has ten coins in his pocket. According to Gettier, Smith has a justified true belief that sentence (b) is true, satisfying the traditional definition. But at the same time, it is clear that he does not know the truth of (b), since he misleadingly infers it from the false sentence (a).
Now, applying our perspectival definition of knowledge to Smith’s claim, we can say the following. There must be a person s to evaluate Smith’s knowledge claim of (b). This person, say, Meg, knows that Smith got the job, because, e.g. she has seen the document of his approval E1, and also knows that Smith has ten coins in his pocket, since she has counted the coins (E2).
Now, the E*p that s is disposed to accept as sufficient to make the conjunctive statement (a) true is the justification {E3} constituted by the conjunction E1 & E2. Meg is disposed to extend her set, as far the justification given by Smith is consistent with {E3}. For instance, Smith justifies (b) by saying that he was informed that he has got the job and he has counted ten coins in his pocket. But to her dismay, the justification Smith gives to (b) is completely different; he says that (b) is true because (a) is true, using as justification the knowledge claim that Jones has got the job and that Jones has ten coins in his pocket. But this justification neither belongs to the body of justifications acceptable by Meg as belonging to E*p nor as belonging to a reasonable extension of E*p to be made by Meg. Meg would give the same negative evaluation to Smith’s justification of (a) by reference to Jones’ justification that the president of the company has told him Jones would get the job, even if he is right in saying that Jones has ten coins in his pocket because he has counted them: in order to be true, the conjunctive sentence (a) must have both component sentences adequately justified.
A second and last counter-example given by Gettier is more complicated, but it also exhibits no real difficulty. In case Smith has strong evidence for the truth of (a) “Jones has a Ford,” since he has always met Smith with this car, given lifts, etc. But about Brown, Smith knows nothing. Then Smith constructs the three following sentences:
(b1) “Either Jones owns a Ford, or Brown is in Boston”.
(b2) “Either Jones owns a Ford, or Brown is in Barcelona”.
(b3) “Either Jones owns a Ford, or Brown is in Brest-Litovsk”.
Smith is sure that these three disjunctive statements are true, even if he has no idea about where Brown is, since he uses the disjunctive syllogism to infer their truth from the truth of (a).
But then Gettier adds the following. In fact, Smith is now driving a rented car and by coincidence Brown is in Barcelona. In this case (b2) is true. Smith has a justified true belief regarding (b2), but he does not know (b2).
Our perspectival answer follows the same path. In any real situation there must be an evaluator s who has more information than the knowledge-claimer a. This evaluator, Julia, knows that Jones’ Ford is rented and that Brown is in Barcelona. She knows that (a) is false because she is the person who rented the Ford to Jones, since he does not have a car. This information serves as justification for her knowledge that (a) is false, which constitutes this E*p and would not accept Smith’s justification.
Regarding (b2), it is true because it is a disjunctive sentence in which the first disjunct is false, but the second true. And Julia knows that Brown is in Barcelona, say, because she went to the airport with Brown (E1), and she later received a call from him (E2), so we can say that Julia’s set of evidential justifications E* is made up of sufficient conditions {E1, E2}. Now, in order to evaluate Smith’s knowledge claim of (a), Julia asks Smith his justification for (b2). To her dismay he says that he derived his knowledge from his knowledge that Smith owns a Ford. She cannot accept Smith’s justification as belonging to her corpus of justifications E* {E1, E2} for (b2) or even as able to be included in it, simply because she does not accept his justification for his last statement, neither as belonging to her E* nor able to be assimilated into it as its extension.
Objection of relativism
At this point the following objection could be made: “Your perspectival analysis of knowledge embraces epistemic relativism. The justification given by a knowledge-claimer will be considered adequate for knowledge or not according to the informational set of a knowledge-evaluator, which can always vary. Hence, if the knowledge-evaluator changes, the evaluation of a knowledge-claim can vary. Since there is no infallible knowledge-evaluator, knowledge is relative to the knowledge-evaluator we arbitrarily chose.”
In order to answer this objection, we need first remark that we should not transform an often-present difficulty into an impossibility. Restricting ourselves to Gettierian cases, it is easy to agree that the knowledge-claimer will be convinced when acquainted with the more complete information available to the knowledge-evaluator. But regarding a comparison between knowledge-evaluators – who can treat one another as knowledge-claimers – things can turn out to be less obvious. We can defend the view that there are indeed more privileged knowledge-claimers (or knowledge-evaluators) and that the criterion to find them is to submit these knowledge-claimers to a critical dialogical situation, similar to what Habermas has called an ideal speech situation (ideale Sprachsituation) (1976). This means that knowledge-claimers must be located in an interactive speech situation in which the following conditions must be sufficiently satisfied:
1. the participants must have a truth-searching commitment,
2. they must have similar rights of informational exchange and questioning,
3. they must have similar competence and capacity to evaluate information,
4. they should be subject to no pressure, neither external nor internal, except the pressure of the best argument (…)[6]
Assuming that speakers satisfy this ideal to a sufficient degree, it is reasonable to conclude that the balance will tend to fall upon the most reasonable side.
For instance, according to anthropologists, North American Natives colonized the region around 10,000 years ago, originally coming from Siberia. There is much evidence that they originally came from Siberia: at that time there was a land bridge across the Bering sea; there is no evidence of people living in America at a much earlier time; moreover, DNA evidence has shown that Native Americans are genetically related to populations that lived in Siberia. The explanation of the origins of the Native Americans given by the Natives themselves is, however, very different: in ancestral times supernatural spirits prepared the world for humans to live there. Then the earth opened and their ancestors emerged from the subterranean world of spirits. The anthropologists and natives can play the role of knowledge-claimers or knowledge-evaluators, and (assuming that they are not cultural relativists or social constructivists (see Boghossian 2006)), they will inevitably disagree: the first believe to know p: “The natives originated from earlier Siberian populations”, while the second believe to know q: “The natives originated from the subterranean world of spirits”. Nevertheless, the situation is very asymmetric. If a native comes to Harvard University and studies anthropology, we can bet that – assuming that she has accepted the conditions of a rational dialogical situation – in the end she will agree with the anthropologist, coming to consider the story she learned as a child as nothing beyond beautiful ancient mythology. The informational set of the anthropologist, under the assumption of the best of our scientific and humanist culture, can explain the informational set of the tribes, while the opposite is not the case. Hence, they are not relative. Hence, we can consider that factors such as a larger well-confirmed informational set, containing a higher amount of more precise and varied scientific information, will be seen as advantageous when examined by both sides, insofar as both sides sufficiently satisfy the conditions of a critical dialogical situation.
Beside this, our truths, as well as our knowledge of truths, are always relative to the best or privileged knowledge-evaluator of a dialogical situation. We cannot, in this or any other way sustain the ideal of finding indisputable absolute truth, proper to absolute knowledge. The best we can do in this direction is to sustain absolute truth and knowledge as a normative ideal, something similar to what Kant called an ideal of reason, useful to make comparisons and to measure the growth of our knowledge in a non-relativist way (Popper 1963: Ch. 10).
Comparing with some other attempts
Assuming the perspectival account of knowledge, I will now explain and criticize a few interesting attempts to answer Gettier’s problem.
Among the first ones, there was the attempt to solve the problem by rejecting false justificatory evidence (Clark 1963). It is false that Professor Stone would be at the University today to give an examination. However, this answer does not work very satisfactorily. A well-known example of true justifying evidence is the following. Mr. Nogot tells Smith that he owns a Ford and even shows him a deed to that effect. Since Nogot was always reliable and honest, Smith concludes p: “Someone in my office has a Ford”. However, it is false that Nogot has a Ford. He is a compulsive liar and is driving his sister’s Ford. Nevertheless, the conclusion p is true, since there is another person in Smith’s office, Mr. Jones, who really does own a Ford, and Smith does not know that. In this case the evidence is false. However, one needs only to change the evidence a bit, applying an existential generalization, in order to get true evidence, Thus, suppose that Smith uses as justifying evidence for p the statement q: “Someone in my office told me that he has a Ford, showing me a deed to that effect, and up to now has always been reliable and honest with me”. Although the evidence given by statement q is true, this is a Gettier’s case in which Smith has a justified true belief without knowing that someone in his office owns a Ford. In a similar way, it is true that Professor Stone told Mary he would be at the University to give a doctoral exam.
One could try to refine the no-falsity answer by considering that non-important falsity could be involved in the justifying evidence, though it is difficult to see how we get this. Nonetheless, the real shortcoming of the non-falsity solution is that it is too coarse-grained. It does not contribute to explaining the difference between wrong justifying evidence (“You are wrong in believing you saw a sheep on the mountain”) and the justification that provides us with a Gettier’s case (“There is indeed a sheep on the mountain behind a stone, but what you in fact saw was only a large furry dog”). Our proposed solution shows the difference: when someone believes he can see a sheep on the mountain, but there is actually no sheep there, the false justification cannot be accepted in E*p, because there is no E*p. But if someone believes he can see a sheep when he is really only seeing a furry dog, even though there is indeed a sheep there, the false justification cannot be accepted in E*p in a case where there is E*p.
A more interesting solution is that we need to add a fourth condition, namely, that the justification must have no defeater (Lehrer 1965). The defeater of the justification in our first given example was the fact that the children of Professor Stone are in the hospital in a critical state and he decided to cancel his activities at the university in order to be there. However, the non-defeater condition is also insufficient, since any defeater can also be defeated. For instance, suppose that the information regarding Professor Stone is mistaken. Suppose it really applies, but to another professor also called Stone, a botany professor from the department of biology in the next building, who also is scheduled to give a Ph.D. examination today and has in fact cancelled his activities in order to remain in the hospital. Concerning the Professor Stone meant by Mary, he is actually in the department giving a Ph.D. exam. In this case, the knowledge-evaluator will accept Mary’s claim of knowing that Professor Stone is at the University now. This defeating of defeaters by new defeaters can in principle continue indefinitely. In conclusion: Mary would have to know all the truth in order to neutralize any possible defeater. Even in a case where there were no defeater, she could only neutralize the possibility of a defeater if she knew all truths.
Keith Lehrer and Thomas Paxon (1969) tried to emend the no defeater definition by defining knowledge as completely justified undefeated true belief. The only way to explain a completely justified true belief, however, is to see it as a belief that is ultimately undefeated relative to the set of all truths (Pollock 1986). Yet, this means that in order to know p, the knower must know all truths! Omniscience is not, however, a human attribute. And this means that by taking this approach we cannot reach a plausible solution to Gettier’s problem. The non-defeater solution solves Gettier’s problem only by creating a greater one.
A very different attempt to solve Gettier’s problem was to replace the condition of justification with the condition of appropriate causal connection suggested by Alvin Goldman (1967). The intuition was clear: I know, for instance, that Emperor Nero killed his mother because there is some appropriate causal chain that begins with a fact and ends in my writing this sentence. I know there must be a causal chain because I know that the facts of the world and our consciousness of these facts must be causally related, even if I am only aware of some few links of those causal chains. To see how it works against Gettier’s counter-examples, consider again Mr. Nogot’s counter-example. He says he is the owner of a Ford. But since he has no Ford, this cannot be the cause of the knowledge that he is the owner of a Ford. The real cause of this knowledge should be the fact that Mr. Jones, also employed in Smith’s firm, is the owner of a Ford. If Smith has made the existential generalization based on the fact that Jones has told him that he has a Ford, we would agree that he knows.
In this article Goldman manages to show that some kind of causal connection is present in all cases of knowledge. This seems plausible. But even if we admit its existence, there are serious problems with his solution: there seems to be something wrong in divorcing knowledge from the cognitive procedure of justification. More specifically, it seems that we cannot find the links belonging to the appropriate causal chain without first knowing the justifying procedure. Because Smith knows that the justification for his knowledge that an employee in his firm owns a Ford is based upon the fact that Mr. Jones told him he has a Ford, and we know that there is a correct causal connection between Mr. Jones’ claim of knowledge and the fact that Mr. Jones has a Ford and not the other way around. In other words, putting the causal process before the justification is putting the cart before the horse, since it is through the procedure of justification that we can find the corresponding causal process. This is also valid for supposed external justifications: it is because we know that there are many possible justifications for my knowing that George Washington was the first president of USA. Although I cannot remember when and where I learned this, I know that there is an appropriate causal chain between this fact and my knowing and not the other way around. Even if Goldman has shown us that there is no knowledge without appropriate causal connections, and that the appropriate causal connections are pointed out by the adequate justification, to put causal connections in the place of justification is to fall into a petitio principii.
Goldman rejected his causal theory ten years later, influenced by the perceptual counter-example of Gettier’s type of the barn façades (Ginet, 1975) that we have already considered, in which Carol really sees a barn that luckily is the only real barn in a region of seemingly real barn façades, a reason why her justification cannot be accepted. This example seems to run against Goldman’s causal theory. Carol’s belief that the barn-like structure is a real barn seems to be normally caused by the presence of the barn in a normal perceptual process. Consequently, according to the causal theory Carol should know that she was seeing a real barn.
Goldman’s response was to develop the new theory of justification in terms of reliability that we have already discussed, a theory that requires that a justified true belief must be produced by a reliable causal cognitive process defined as an empirical mechanism that makes the truth probable. Goldman also expects in this way to answer Gettier’s problem. To the reliabilist understanding of justification, the barn-facades are located in an unreliable environment regarding the distinction between real barns and mere barn façades. This Gettier’s case is not one of knowledge because Carol’s belief that she is seeing a real barn can be demonstrated as unreliable. To be reliable, the barn-case demands the exclusion of relevant alternatives. One of them is that it is not a barn façade, which is left unconsidered. Hence, Carol’s justification is not knowledge-producing, because it is unreliable (Goldman, 1988: 63).
To this response, we can object that changing our justifying evidence by means of process reliability does not seem to bring any real improvement, since in any case (also in the causal theory) one could say that because of the special environment the knowledge evaluator should demand a careful examination of all sides of the barn, even its interior… in order accept Carol’s knowledge-claim that it is a real barn to Smart’s E*p, excluding the alternative of a barn façade.
Moreover, Goldman’s process-reliability explanation of justification, as much as his causal theory, is open to the same objections presented against the non-defeasibility view of justification: in order to know p, one needs to know that the reliable process cannot be defeated by another reliable process, and in order to know that it cannot be defeated, one would need to exclude all possible defeaters and defeaters of defeaters, that is… one needs to have omniscience. Our answer to Gettier’s problem solves this problem neatly: the required extended knowledge remains within the extension of the informational set of the knowledge-evaluator which constitutes his body of acceptable evidence E*p.
An attempt to define knowledge that is similar to Goldman’s is Robert Nozick’s tracking theory. According to Nozick, if someone is able to track correctly the truth of a proposition p, this person knows that p is true. The way to find the right track is the satisfaction of two subjunctive conditionals:
(i) if p were not true, a would not believe in p;
(ii) if p were true, a would believe in p.
In fact, under the circumstances of our Gettierian case, if Professor Stone were not at the University, Mary would still believe he was there, conflating against the subjunctive conditional (i).
However, considering (i), how do we know that if Professor Stone were not at the University, Mary would still believe he was there? The reason is given by the perspectival definition: Assuming that Carl (s) accepts the evidence given by Mary (a) as making p true, namely, that for Carl Mary knows p, then if p were not accepted by Carl as true, Mary would not believe in p in a way that makes her know p. Moreover, assuming that for Mary knowing p, p being true for Carl, it is to be expected that Mary would believe in p.
Internalism and externalism again
An epistemic justification is said to be internal when the epistemic agent is able to have cognitive access to the justifying evidence or reasons for the belief in the truth of the proposition. Ideally, the agent must be able to make discursively explicit his justifying evidence. This is the rich standard case around which the more limited cases are aggregated, going further until reaching those borderline cases in which one does not know if the word ‘justification’ is still appropriate. Vagueness inevitably belongs to the semantics of the word.
This is the case of the Gettierian examples we gave and many others and the justifying evidence was all internal. In what follows however, I will consider some borderline cases, showing that their justification is internal and evidential in the proper sense of these words and, furthermore, that they can serve as justifying evidence for our perspectival definition of knowledge.
We begin with cases of memories in which the original evidential link has been lost. For instance, how can I justify my belief that my phone number is 035-216? I have a bad memory for numbers. All I can say is that after I repeated this number many times it was finally anchored in my long-term memory. If pressed to produce a justification, I could check my memory, looking again at the note pad where I wrote it down. What about my knowledge of my old telephone number 225-00-16? I have had this number imprinted in my memory for many years. My old telephone is gone now and it is impossible to check it. Nevertheless, I feel myself justified to affirm that this was my telephone many years ago. However, I know by induction that memories that always repeat themselves are usually correct. Moreover, in order to test my memory I can write it down and later recall it and look to see if the remembered number is the same one I wrote down. Furthermore, the telephone number is associated with surroundings like the old apartment where I lived in Rio de Janeiro, which agrees with my life history. The inductive justifying evidence of such memories is sometimes confirmed. Some days ago there were a quiz on television, where it was asked who was the president who forbade women to wear bikinis on the beach in Brazil? I knew the answer: J. Q. This reinforces my inductive belief in my old long-term memories. Repetitively confirmed induction is the evidence that justifies our belief in our long-term memories. But these are internal reasons, even if they are not the result of direct introspection. Direct introspection is the case when I justify that my car is at the university because I remember leaving it in the parking lot: the memory of a perceptual experience. My conclusion is that cases in which I remember my password or my telephone number or some historical facts are not really different from the case where I know that my car is at the university. The only difference is that in this last case the introspection of a perceptual memory is what serves as evidence. The fact that once in my childhood I had an introspective memory of the source of my knowledge that America was discovered in 1492 does not change anything. It only serves to confuse our minds by focusing on one kind of evidence, which is causally but not factually necessary.
There are more difficult cases. It is said that there are persons who can know with relatively great precision the sex of a chick simply by feeling the animal. Closer consideration shows that there are simple physical techniques that can be used to recognize the sex of a chick by feeling it. Even if this requires practice, it already has the character of cognitive (or pre-cognitive) internal justifying evidence, which, like most such evidence, must indicate an external fact.
Another case is the knowledge we attribute to animals. A dog hears the sound of its owner’s car and runs to the door, where it stands barking. In fact, the dog knows its owner is there, but it does not know reflexively, although we know that it knows. And the justification for our knowing that it knows is through its behavior, added to induction by analogy, considering that the dog is sufficiently similar in behavior to humans able to have cognitions and feelings. This can lead to the mistaken conclusion that the justification is external. Nonetheless, though made from a third person perspective – our own – the justification remains internal, since we are assuming that the dog runs to the door because it has taken as evidence of its owner arriving the sound of his car. Another example of this kind is that of a child who knows that she is in the presence of her mother. We know this by her smile, by behavioral reactions. We justify this in the third person, knowing by her behavior that she is justified in believing that she is in the presence of her mother. She evidentially re-identifies her mother’s face and behavior. Moreover, in both cases, as knowledge-evaluators, we know that if the dog and the baby were able to have reflexive access to their cognitive processes, and could linguistically express what is going on, they would say: “I know that my owner is coming because I hear the sound of his car” and “I know that my mother is with me because I recognize her face”. These justifying pieces of evidence are ones we would immediately accept as able to be included in our justifying body of evidence. We know that they have evidence able to make these propositions true. Still a case to consider is that of generalizations. Scientific laws are the inductive (mainly abductive) results of cumulative experience. But these experiences, though having external counterparts, must as justifying evidence be internally accessible.
A last but also important case is that of testimony. When we are informed by reasonable and trustworthy people, even if indirectly by means of radio, television, internet, books, or other media, we accept information about things we are unable to actually experience. Nevertheless, the testimonial origin has its own evidential grounds, which are cognitively, that is internally, accessed. We only borrow the results, which still makes justification an essentially cognitive phenomenon. These non-actual third person pieces of evidence can also be accepted or rejected as belonging to our body of evidence as knowledge-evaluators in conformity with the perspectival definition of knowledge.
[1] One can find propositional knowledge statements using words like ‘when’ or ‘whether’ in the place of ‘that’. But the sentences can be paraphrased in ways that these words are replaced by ‘that’. For instance: “Hank knows whether the bull is dangerous” can be replaced by “Either Hank knows that the bull is dangerous or Hank does not know that the bull is dangerous”. (See Feldman 2003: 9-10)
[2] As Socrates says in the dialogue Meno: “true beliefs… are not worth much until one ties them down on account of the reason why they are tied down… After they are tied down, in the first place they become knowledge, and then they remain in place. (1997: 895) See also Theatetus (1997: 223)
[3] See J. L. Borges tale, ‘Tom Castro, the implausible impostor’, in his book, A Universal History of Infamy. The tale is based on a real occurrence.
[4] The term was suggested to me by John Cottingham.
[5] A probability of acceptance can vary in accordance with the context. In the exceptional context of a lottery, for instance, the “I know that I will not win” remains below acceptance, even when the probability of not winning is extremely high. The reason is in my view that the probability required by this closed system must be 1.
[6] Complete failures in the satisfaction of such dialogical conditions are catastrophic to science: examples are Catholic dogmas that led to the condemnation of Galileo, the Marxist genetics of Lysenko in the Stalinist USSR, and Nazi Aryan science.
