Epistemologists have always recognized the importance of causal processes in accounting for our knowledge of things. In discussions of perception, memory and reasoning, for example, it is commonly assumed that these ways of coming to know are fundamentally causal. We perceive things and thus come to have knowledge about them via complex causal processes; memory is, at least in part, the retention of previously gained knowledge through some sort of causal process; and reasoning is a causal process that takes beliefs as inputs and generates beliefs as outputs.
A causal theory of knowledge is a form of externalism and is based on the fundamental idea that a person knows some proposition, p, only if there is an appropriate causal connection between the state of affairs that makes p true and the person's belief in p. Although this kind of theory has roots that extend to ancient times, contemporary versions attempt to make more precise the nature of the causal connections required for knowledge. The causal theory is closely related to other forms of externalist theories, such as the conclusive reasons theory, information-theoretic views and the various forms of reliabilism.
In the internalist-oriented environment which dominated epistemology from the time of Descartes until the middle of the twentieth century, it was not considered appropriate to refer to the causal history of a belief in providing an analysis, or definition, of the positive epistemic status of that belief (see Internalism and externalism in epistemology). Rather, the epistemologist's job was to provide definitions of concepts such as justification and knowledge independently of any assumed causal connections with the external world, and then to show how, from such analyses, one could argue ('internally') that such causal connections exist. To do otherwise would be to beg the question against scepticism about the external world.
Since the 1960s, there has been a shift away from the internalist position in epistemology. The causal theory of knowing is one of the early versions of externalism, introduced by Alvin Goldman (1967) and conceived primarily as a response to the Gettier problem which had appeared only a few years earlier (see Gettier problems). In Gettier examples, a person, S, has a justified belief in something that is only coincidentally true. This element of coincidence, which is perhaps the most salient feature of Gettier cases, is very difficult to explain without introducing some element of external connection between the individual's belief, the justification for that belief and the state of affairs which is the object of the belief. Goldman's original proposal was to focus on the causal connections that typically obtain between these various epistemically relevant items when a person has knowledge.
Goldman expressed his proposal as a set of truth-conditions for knowledge, in the following schematic form:
[A person] S knows that p if and only if the fact p is causally connected in an 'appropriate' way with S's believing p.
The ways that are 'appropriate' include perception, memory, various other kinds of causal chains and forks, and combinations of these. Goldman adds to this the further condition that the relevant causal connections obtaining between the state of affairs p and one's belief must be 'correctly reconstructed by inferences, each of which is warranted'. This further condition is designed to accommodate the fact that causal chains can sometimes take very unexpected routes. If the route is unusual enough, then even though it results in a true belief that p, it does not provide knowledge. For example, suppose that Sally is perceiving an object only through a complex network of mirrors, or via some holographic imaging device which produces a very realistic image of the object. Then, she might come to believe that the object is in front of her when in fact it is, but she might fail to know that it is there, as she knows nothing about the unusual causal mechanism. If she were in a position to correctly reconstruct the causal chain, however, then she would have knowledge.
One example developed by Goldman that illustrates the intuitive appeal of his proposal involves a person, Smith, who perceives solidified lava lying around a mountain. On the basis of beliefs about this lava, and background beliefs, Smith inferentially comes to believe, correctly, that the mountain erupted many centuries ago. Assuming that Smith's inferences are warranted, does he know? To answer this, we must ask what sort of causal ancestry obtains between the eruption of the mountain and Smith's belief that it has erupted. If the lava that he sees resulted from the eruption, as he imagines, then Smith does have knowledge. However, suppose that unknown to Smith the lava has been placed there by promoters who wish to make it look as though the area was once volcanic. The lava actually produced by the eruption has been completely covered by years of sedimentation, and cannot be seen. Then, there is no appropriate causal connection between eruption and perceived lava, and Smith does not know. The naturally intuitive appeal of such an example is confirmation that Goldman's causal theory captures at least part of what we require for knowledge.
In Gettier examples, the requirement that there be an appropriate causal connection between the fact p, and S's (warranted) belief, is not satisfied, very much along the lines illustrated in the lava case. This can be seen by considering a specific (Gettier-style) example, introduced by Keith Lehrer (1965). Suppose Smith correctly infers that someone in his class owns a Ford, from some true evidence that justifies the false belief that a student, Mr. Nogot, owns a Ford. It so happens that another student in Smith's class, Mr. Havit, does own a Ford, but Smith has no evidence one way or the other for this proposition. The causal theory clearly explains the lack of knowledge in this example, since the state of affairs making it true that someone in Smith's class owns a Ford (namely, Havit's owning a Ford) is not causally related in any of the appropriate ways with Smith's belief. Rather, Smith's belief is caused by states of affairs which make true the evidence on which his belief that Mr. Nogot owns a Ford is based. And, these states of affairs, whatever they are, have nothing to do with Mr. Havit's ownership.
It should also be noticed that if this case were redescribed so that Smith does have evidence that Havit owns a Ford, then it would be clear (all else being equal) that he does have knowledge that someone in his class owns a Ford, for then the required causal connection would obtain. It can be concluded that in Gettier examples, as well as 'ordinary' cases in which we would tend to ascribe knowledge to individuals, the causal theory provides a clear and intuitively appealing account of knowledge.
Despite its merits, there are a number of examples that raise serious difficulties for the causal theory. One class of examples is generated by the phenomenon of causal overdetermination. Suppose Alfred comes to believe that there is a sheep in the field by hearing a recording of the sounds normally produced by sheep while also looking at a distant boulder that looks like a sheep in the field. Even if there is a sheep somewhere in the field not seen by Alfred, he does not know that one is there. Later, should he come to perceive the real sheep, he would come to know that there is a sheep in the field. The problem for the causal theory is that Alfred's later perception of the real sheep would not cause his belief in an appropriate way, because his belief already exists. The causal theory would incorrectly deny that Alfred has knowledge. This kind of case can be accommodated within the general spirit of the causal theory by allowing as 'appropriate' the relation of causal sustaining, a very weak form of causal connection.
More difficult to accommodate are overdetermination cases in which an individual is acquainted with some state of affairs which is causally sufficient for the state of affairs p, but which is not in fact any part of the cause of p. Abigail might, for example, be aware that Jones has taken a fatal dose of poison, with no antidote, and thereby come to know that Jones is dead even though in fact (but unknown to her) Jones died of other causes. The only way to accommodate such an example is to further extend the scope of 'appropriate causal connections' to include cases in which one infers p from something which is causally sufficient for p even though not the cause or any part of the cause of p. It is not obvious, however, that this kind of modification is in the spirit of the causal theory, for it abandons the idea that knowledge requires a causal connection between belief and known fact.
Equally troubling for the causal theory are situations involving logical and/or mathematical facts. One kind of situation is very similar to the example just given. Suppose Mark observes that an owl is on top of a flagpole, which he already knows to be 15 feet high. He also observes a mouse to be 12 feet from the bottom of the flagpole. Mark correctly infers that the mouse is a little over 19 feet from the owl. But, the mouse's being a little over 19 feet from the owl is no part of the cause of Mark's belief in that fact, nor are the facts from which he makes the inference themselves causes of the mouse's distance from the owl. Rather, this is a case in which the fact concluded follows logically from the observed premises. To take another example, suppose that the object of belief is itself a logical or mathematical truth, such as '2 + 2 = 4'. Surely a theory of knowledge should allow for knowledge of such truths, but the causal theory faces serious obstacles in attempting to provide an account. Whatever the nature of the facts which make true logical and mathematical truths, they do not seem to be the sorts of facts that are parts of causal chains. If not, then there can be no appropriate causal connection of any kind between such facts and one's beliefs. It should be noted, however, that there have been some serious efforts, particularly by Mark Steiner (1973) and Philip Kitcher (1984), to accommodate mathematical and logical knowledge within the framework of a causal theory.
In his own effort to accommodate examples involving inferences based on logical connections, such as the one concerning the owl and mouse, as well as knowledge of mathematical and logical truths, Goldman (1967) proposed that the scope of 'appropriate causal connection' be extended even further to include logical and mathematical connections. Even those philosophers who are inclined to be sympathetic with the previously mentioned extensions of the notion of a causal connection have baulked at this idea. Peter Klein (1976) argues that there is no adequate way of formulating a version of the causal theory which allows logical and mathematical connections to be counted as causal.
The problematic examples discussed thus far have raised questions about the necessity of the 'appropriate causal connection' requirement. There are also difficult examples that raise doubts about the sufficiency of the 'appropriate causal connection' requirement, even in its most refined forms. Perhaps the most famous of these examples is one introduced by Goldman himself (1976), leading to his own ultimate abandonment of the early version of his causal theory.
In this example, we are to suppose that Henry is driving through the countryside looking at the scenery. One of the things he sees is a barn, and on the basis of this perfectly ordinary perceptual evidence, along with his background beliefs about barns, Henry comes to believe that there is a barn there. Unknown to him, however, the region is populated by papier-mâché barn facsimiles, of the sort found on movie-studio lots. These facsimiles, which are propped up by sticks from behind, would be mistaken by Henry, or anyone else, for real barns when sighted casually from the road. It is only by good fortune, or coincidence, that Henry has perceived a real barn rather than a facsimile. Given this, Henry cannot be said to have knowledge that the object he sees is a barn, for he could just as well have been perceiving a facsimile. The problem is that Goldman's causal theory of knowledge is satisfied: Henry's belief that there is a barn there is caused in a perfectly straightforward way by the presence of the barn, and we may suppose that Henry has correctly reconstructed through warranted inferences the causal chain leading to this belief. So, if Henry does not have knowledge, the conditions of the causal theory do not appear to be sufficient for knowledge.
In light of these and other difficult counterexamples to both the necessity and the sufficiency of the causal theory, it has largely been abandoned, at least in the form originally suggested by Goldman in which a causal connection between state of affairs and belief is required.
The causal theory of knowing still survives in many of the externalist theories which have arisen since Goldman's early proposal. Among the more prominent theories in this category are the 'conclusive reasons' approach, exemplified in the works of Dretske and Armstrong, and various versions of Reliabilism, exemplified in the later works of Goldman, and in works by Swain, Alston and Plantinga.
Goldman (1976) proposed a modification of the early causal theory in which it is required not only that the individual's belief that p be caused in an appropriate way by the state of affairs p, but also that the individual be a reliable discriminator with respect to p and other, alternative states of affairs that might causally substitute for p. When this discriminatory capacity is lacking, or is not functioning properly, then the individual fails to know even if the causal connection requirement is met. In this way, examples such as the one involving Henry can be accounted for.
But even with this modification, the causal theory cannot handle examples in which there is no causal connection of the required sort, such as the mathematical and logical cases illustrated above. Ultimately, the requirement that there be an actual causal chain linking the state of affairs that makes p true with the belief that p must be abandoned. Taking its place is the notion of an 'appropriate causal ancestry' of a belief. A belief is an instance of knowledge, according to this kind of causal theory, provided it is produced in an appropriate manner, where the manner of production is appropriate just in case it is reliable. A theory which holds this is called a 'reliability theory', and is the main kind of successor to the early casual theories (see Reliabilism).
See also: Causality and necessity in Islamic thought; Information theory and epistemologyMARSHALL SWAIN