Text: Alvin Goldman, "A Causal Theory of Knowing" (1967)
At least since the time of Aristotle (De Anima, Chapters 2 and 3), philosophers have suggested that the human mind knows an objects when the object causes a change of some kind in the mind. This kind of account is most obviously applicable to knowledge of physical objects, since they interact causally with the body. If we suppose that changes in the body, perhaps in the central nervous system, cause changes in the mind, then there is a causal chain from the body to the change in the mind. The existence of such a chain could then explain how it is that knowledge is "tied down," to use Plato's metaphor.
The ancient Stoics made the "cognitive presentation" the standard of truth. Diogenes Laertius described it in this way:
The cognitive [impression], which they say is the criterion of things, is that which arises from what is and is stamped and impressed exactly in accordance with what is. (Lives of the Eminent Philosophers, Book 7, Chapter 46. See Long and Sedley, The Hellenistic Philosophers Volume I, Section 40.)The Stoics here seemed to think of the mind as being able to take on the very shape of the object. If it did, then it would seem reasonable to say that such an impression is the sort of thing that would give us knowledge of the object (perhaps along with other conditions).
The image of stamping and impressing does not hold up very well in the context of modern accounts of human perception. The causal process that begins with physical object ends in the production of some change in the brain or in the mind. John Locke, writing in the seventeenth century, called what is produced in the mind an "idea." He held that we have "sensitive knowledge," which is of "the existence of particular external objects, by that perception and consciousness we have of the actual entrance of ideas from them" (An Essay Concerning Human Understanding, Book IV, Chapter II, Section 15).
A problem with this view is that we do not have any direct way of being conscious of the causal relation between the bodies and whatever they produce in us. We are only directly conscious of the end-product, be it the Lockean "idea" or something else. Many philosophers have thought that we cannot have knowledge of the object unless we have knowledge that there is a causal relation between the object and the brain or mind. So, they would hold, a causal relation to the object is not sufficient for knowledge.
Once you allow this, however, it begins to look as though a causal relation to the object is not a necessary condition for knowledge of the object, either. If all our evidence consists of "ideas," then it can give us knowledge by bringing about a correct representation of objects, without necessarily having been caused by them. The emphasis has generally been on the adequacy of "internal" evidence, so that what is needed for knowledge is "justification."
One of the leading proponents of "externalist" conditions for knowledge is Alvin Goldman, a long-time member of the philosophy department at the University of Arizona who now teaches at Rutgers University in New Jersey. Goldman's views have undergone many changes over the years. We will examine one of his earliest papers in this lecture and one of his later ones in the next.
The Gettier Problem
Goldman's early article was one of many that appeared within a few years of the publication of Edmund Gettier's vastly influential paper, "The Analysis of Knowledge" (1963). Gettier gave a couple of counter-examples to what he called the "traditional" analysis of knowledge. According to this analysis, S knows that q if and only if:
Gettier showed that these conditions can be satisfied in the absence of knowledge. One of his examples involves a person, Smith, who works in the same office as Jones. Smith has excellent evidence that:
q: Jones owns a FordJones has owned a Ford for a long time and has just offered Smith a ride in one. The glitch is that Jones does not now own a Ford, having just sold it. He has offered Smith a ride in a rental car. So q is false, and Smith does not know that q.
Nonetheless, Smith has made a logical inference from q to another proposition which he believes:
p: Jones owns a Ford or Brown is in BarcelonaThis follows logically from q. That is to say, if q is true, then p must be true by the meaning of the logical word "or." Gettier plausibly maintains that if Smith is justified in believing q, then he is justified in believing p.
We suppose further that Smith made his inference by picking a city at random. He had no idea of Brown's whereabouts. But it turns out, by the purest coincidence, that Brown in fact is in Barcelona. This means, again by the logic of "or," that p is true. So by the "traditional" analysis, Smith knows that p. Yet it is implausible that Smith has this knowledge, since his belief is true only by coincidence.
In the 1967 paper, Goldman proposed that the problem here is that "what makes p true is the fact that Brown is in Barcelona, but that this fact has nothing to do with Smith's believing that p." Smith's belief that p was arrived at from q. He could have substituted any truth in the place of the truth that Brown is in Barcelona and arrived at a justified belief about that.
So Goldman offered as a solution a new condition on knowledge, not found in "traditional" analysis: that what makes p true must be the cause of S's believing that p. (Note that this is offered only as a condition for empirical knowledge.) What Goldman must do to make his account plausible is to show that it can be generalized to all cases of empirical knowledge. He does this by providing a survey of types of causal connections.
Perception is probably the simplest type of causal connection leading to knowledge. Goldman adopts the analysis of Paul Grice, according to which a necessary condition for S's seeing that there is a vase in front of him is that there is a causal connection of a certain sort between S's belief that he sees a vase and "the presence of the vase." Goldman does not try to explain the specific causal process involved. But he notes that we would not say that someone sees something unless the relevant causal process has occurred.
To drive this point home, Goldman suggests a thought-experiment. Suppose someone interposed a hologram of a vase between S and the vase, thus cutting off the causal connection between S and the vase. In that case, we would not say that he sees the vase, though we could allow that he knows (by some other means) that there is a vase before him. So "seeing that" is treated as a special case of "knowing that." It requires a causal connection.
For the moment, we will regard the process leading to perceptual knowledge as not involving inference. It does not seem plausible, from our experience, that we do make inferences from "ideas" or "sense data" or brain states when we know something perceptually. (Though if we do make such inferences, the more complex process can still be described as causal, as Goldman will try to show later. He also thinks he can account for knowing that the painting before one is a Picasso, which seems to involve inference.)
Some of our knowledge is based at least in part on memory, and memory itself is a causal process. A belief at the present can be traced back to an earlier belief as a cause. As before, Goldman does not try to give a detailed account of how the process works.
The need for a causal connection is shown by considering two types of cases. In the first, one believes that p at a time and then later believes p again. This may occur without being a case of remembering. In the second, one knows that p at one time and knows that p later. This second type of case could happen without memory, as there could be a different source of the belief or knowledge at the two times, such as perception at the first time and testimony at the second.
A further consideration is that the mere having of a "memory impression" is not sufficient for remembering. A memory impression may be stimulated by an external source, and in such a case is not caused by the original belief. Goldman holds that this is not a case of remembering.
Finally, knowledge may be transmitted by a combination of perception and memory. One sees that p at a time, forms a belief that p, and later comes to believe it anew on the basis of memory. A single causal process would have united the fact with the later belief.
The cases of direct perception and remembering are fairly straightforward, because they involve processes that are clearly causal. The process by which someone reasons explicitly is also be pretty clearly a causal process. For example, one may carrying out the reasoning process by a kind of interior dialogue "Since such-and-such is true, p must also be true."
But Goldman wants to construe inference more broadly. One may have knowledge based on inference without having gone through the process of inferring explicitly. Goldman's example is the belief that a nearby mountain has erupted hundreds of years ago. The basis for this belief is the observation of solidified lava lying about, along with "various 'background' beliefs about the production of lava." It important to note that these "background" beliefs are not generally to be found in the chain linking the object to the belief.
Goldman thinks that knowledge in this case requires a causal process as well. He describes a variant similar to the hologram variant of the vase case. Someone long ago removed the lava that was blown off the mountain, then later someone else scattered some lava round. The causal chain would have originated from the new lava, and the person would not know that the mountain erupted. "A necessary condition of S's knowing that p is that his believing p be connected with p by a causal chain.
The example does not show, however, that S knows by inference that the mountain erupted. For this to hold, given Goldman's condition, the inferential process linking the perception of the lava to the conclusion that the mountain erupted. Here, Goldman says that he is inclined to construe the process of inference as a causal process, since the new belief is based on the belief from which it was inferred.
One need not do so, however, to reconcile the causal condition with inferential knowledge. All that need be done is to be careful about how "causal" is understood in the condition. Thus the "causal chain" required for any knowledge remains "causal" when inferential links are added to it. But this does not imply that there are "causal connections" as a result.
The fourth type of knowledge that might be given a causal account is knowledge from the testimony of another person. This type is somewhat more complicated than the others. Say that a person T comes to believe that p through a causal chain ending in perception. Then the belief causes an assertion that p (the testimony) as an audible sound. This sound causes S to believe that T is asserting that p, and that belief causes S to believe that T believes that p. Finally, this belief about the testifier's belief causes S to believe that p.
In this five-step process, background beliefs play a role again. Goldman calls attention to the fact that the belief that T believes that p may require, in addition to belief that T asserted that p, beliefs about T's sincerity. Perhaps even another belief is needed, to the effect that T understands what he is asserting.
The passage from the belief that T believes that p to S's own belief that p requires more risky background beliefs. At the minimum, S must believe that T is a competent judge with respect to the kind of information expressed by p. S should also believe that S is in a position to believe correctly that p in the present instance, say by seeing that p.
Reconstructing the Causal Chain
All knowledge based on inference requires that the chain of inference mirroring the part of the causal chain that led up to the inference. In the case just discussed, S's belief that T asserted that p mirrors the fact that T did assert that p. From this, S inferred that S believes that p, and the inference mirrors the fact that S's belief that p caused S to assert that p. Finally, the inference that p from the belief that S believes that p mirrors the fact that p caused S to believe that p.
To motivate the claim that correct reconstruction is a necessary condition for knowledge, Goldman has us imagine a case where a newspaper reporter observes p and reports it to the paper. But there is an error, and the paper reports that not-p instead. A reader then overlooks the word 'not' and believes p. Although the reader gets the end-points of the causal chain right, his reconstruction fails to reflect the actual process. Since the reader lacks knowledge, correct reconstruction is necessary for knowledge.
In general, the reconstruction must contain no errors. It may contain omissions, however. Goldman only held that "all the important links" be reconstructed. A further requirement is that the reconstruction of the causal chain be no accident. The inferences made in the reconstruction must be good inferences, which "genuinely confirm p very highly, whether deductively or inductively." This can be understood as a way of understanding the justification condition found in the traditional analysis.
As it stands, Goldman's analysis does not seem to be able to handle knowledge of the future. If some future fact p must be the cause of my belief that p, then causation must go backwards, which Goldman does not accept. To allow knowledge of the future, he again notes that he requires only a causal connection, and not a cause. The causal connection that could allow for knowledge of the future would be this: that some fact that is causally connected to the belief is also causally connected to the future fact. That is, p and the belief that p have a common cause.
The example Goldman uses to motivate this claim is as follows. T tells S that he intends to go downtown on Monday. The intention that leads to the telling is the beginning of a causal chain that ends in S's belief that T will go downtown on Monday. And in fact that intention is the cause of T's going downtown on Monday. Goldman calls this a "Pattern 2," in contrast to the "Pattern 1" cases discussed earlier.
As with Pattern 1 cases, Goldman notes that a breach in the causal connection prevents knowledge. If T decided later not to go downtown on Monday but ended up there because he was kidnapped and taken there by criminals, then S does not know that T will go downtown on Monday, given that the other features of the example remain the same.
Pattern 2 applies not only to knowledge of the future, but also knowledge of contemporary events. I know that there is smoke going out the chimney because I see the fire in my fireplace. The fire is the cause of the smoke and the end-point in the causal connection leading to my belief that smoke is going up the chimney. The patterns can be combined in various ways, as in the case where I am told by a friend that he saw the smoke last night. His testimony is both a cause of my later belief as well as an effect of the fire in the fireplace.
The causal analysis of knowledge differs from the traditional analysis in that it places further requirements beyond truth, belief, and justification. Belief that p must be the end-point of a causal chain leading from p. If inference is part of the causal chain, then it must mirror the chain up to the point where inference began.
A different analysis was proposed as a solution to the Gettier problem by Michael Clark, who focuses on the "internal" grounds S has for believing p. He suggests that in order to known that p, none of S's grounds for believing p, be they immediate or remote, may contain a falsehood. This would block the original case, because Smith falsely believed that q, Jones owns a Ford, which was essential to his justification that p, Jones owns a Ford or Brown is in Barcelona.
Clark could make his account equivalent to Goldman's by requiring that S's evidence include beliefs about the causal relations in the chain. Then those beliefs would have to be true, in which case there is a correct reconstruction of the causal chain. But Clark placed no such restriction on what must count as evidence.
A counter-example to Clark's analysis, due to Saunders and Champawat, is much like the examples Goldman has been giving at each point in his analysis. Smith's belief that q, Jones owns a Ford, is true, but the reason it is true has nothing to do with why Smith believes it. Smith has sold his Ford to buy another car, but he has just won a new one in a raffle.
It is easily explained on Goldman's analysis why there is no knowledge. The causal connection to the belief was broken when Smith owned the Ford, and no new causal connection to Smith's belief arose from the the fact of Smith's ownership of a new Ford. Goldman notes that if Clark had required that among Smith's true beliefs ones to the effect that his inferences mirror the causal connection. As Goldman put it, Smith would have to have a belief such as that "Jones's owning a Ford yesterday would result in Jones's owning a Ford now."
A second kind of counter-example was given by Saunders and Chapawat, as well as by Keith Lehrer. The first kind was to the effect that Clark's condition is too weak for knowledge. Lehrer's example shows that it is also too strong: that it rules out legitimate cases of knowledge.
Consider a slightly different p: Someone in Smith's office owns a Ford. Smith might have had redundant information about someone in his office owning a Ford. Some of that information (That Brown owns a Ford) is false, so some of his grounds for believing that someone in the office owns a Ford are false. But this should not block him from having knowledge.
The causal analysis blocks this result and allows knowledge because it does not require that every causal connection leading to the belief that p contains nothing but true beliefs as links. All that is needed is at least one. Another way in which Goldman's account is superior is that it allows that some background assumptions might be false, as long as the belief that p is strongly-enough supported.
Lehrer's specific example shows the need for a further refinement in Goldman's account of knowledge. Suppose it is a fact that someone in Smith's office owns a Ford. This is what might be called an indefinite or indeterminate fact, as opposed to the definite or determinate fact that Jones owns a Ford. The latter kind of fact seems to be the sort that is involved in real causal connections, while the former is involved only in the sense that it is logically implied by a determinate fact.
So Goldman proposes to broaden once again his account of a "causal chain" to include what is "logically" related to an actual cause. He illustrates this further by showing that a belief in a conjunction is not caused by the conjunction, but by the conjuncts that make it up. So, he will allow that the conjunction is part of a causal chain, since it is logically related to facts that are causes.
A final form of facts that can be accommodated by the device of causal chains is universal facts. That all men are mortal could be known because it is logically related to the facts of the perceived deaths of many people, and the solid inference to the mortality of all men that is based on this.
With all the expansions on the notion of "causal chain" in hand, we can give a final analysis of knowledge.
S knows that p if and only if:The "appropriate" was of connection are at least these:
The fact p is causally connected in an "appropriate" way with S's believing that p.
Although this analysis is obviously stronger than the traditional analysis, it is also weaker, in the sense that its requirements may fall short of what is demanded in "justification." Specifically, many accounts of justification require that S be able to state what justifies him in his belief. But if one has forgotten the background information that warranted some of his inferences used in reconstructing the causal chain, one can still know, because there still is a causal chain. "It seems clear that many things we know were originally learned in a way that we no longer remember. The range of our knowledge would be drastically reduced if these items were denied the status of knowledge."
Another way in which one might have knowledge without being able to justify it is if processes like extra-sensory perception turn out to be "appropriate." A person who had extra-sensory perception, was caused to have a belief on that basis, and correctly reconstructed the causal chain leading to the belief by warranted inferences would have knowledge. So Goldman's account is extensible, due to the underdetermined status of what is appropriate.
Goldman hesitates to get involved in the difficult question of knowledge of ourselves, but he says it cannot be avoided. There is no apparent problem in cases where a mental state is distinct from a belief about being in that state, since there could be said to be a causal chain leading to the belief.
But in some cases, it seems that the mental state and the belief about it are identical or barely distinguishable. The case of pain is an example. "My being in pain and my believing that I am in pain are hardly distinct states of affairs." A technical way of handling such cases is to count them as "degenerate" or "limiting" cases. Goldman thinks this is actually a good result, since there is a lot of dispute over beliefs about these states. Some take them to be a paradigm case of knowledge, while others think they are not knowledge at all.
Goldman recognizes that his proposed analysis flies in the face of tradition. In the first place, it is entirely silent regarding skepticism. The skeptical question is just irrelevant to whether one knows, given the causal account. More importantly, it does not confine the analysis merely to "questions of logic or justification." Instead, it focuses on "causal or genetic questions." Although people did not use the expression at the time, we can say now that he is proposing an "externalist" account of knowledge.
In his later work, Goldman abandoned the letter of the causal analysis, but not its spirit. In the 1979 paper, "What is Justified Belief" (reprinted in our text), Goldman allowed that "principles of justified belief must make reference to causes of belief." But rather than focus on actual "causal chains," he turned to the reliability of the processes that cause our beliefs. Thus he was one of the early advocates of "reliabilism" in epistemology, though the idea can be traced back at least to 1929, when F. P. Ramsey gave an analysis of knowledge that made reliability a condition in a very brief paper entitled "Knowledge."
[ Course Home Page | Menu of Lectures ]