Internalist Echoes in an Externalist Theory — Epoche B2
Not an Early Reliabilism: Goldman's Causal Theory and the Generality Problem A misnamed ancestor Alvin Goldman's 'A Causal Theory of Knowing' (1967 [1] ) is routinely introduced as early reliabilism, the first rough version of a view he later polished. This is not what happened. The 1967 paper proposes that knowledge is true belief appropriately caused by the fact believed, and says nothing about reliability. Reliabilism is a different theory, arriving in 'Discrimination and Perceptual Knowledge' (1976) and stated properly in 'What Is Justified Belief [2] ?' (1979). Goldman did not refine the causal theory into it [3] . He abandoned the causal theory [4] , for reasons that had nothing to do with insufficient externalism. Getting the sequence right matters because it relocates the interesting question. This note argues that reliabilism does contain an element that cannot be specified in third-person terms, but that the element is not lurking in the word "appropriate". It is in the individuation of belief-forming process types, and the claim defended here is that no purely objective specification fixes the type, so reliability itself cannot be assessed without appeal to what the subject was going on. Anyone who disagrees has a definite task: supply a principle of type individuation that makes no reference to the subject's sensitivities. The causal theory, and why Goldman dropped it Edmund Gettier's 1963 paper showed that a belief can be true and justified while its truth is a matter of luck [5] . Goldman's diagnosis was that the luck consists in a missing connection: in Gettier's cases the fact that makes the belief true plays no role in producing it. His proposal was that a subject knows a proposition when the fact that makes it true is causally connected, in an appropriate way, with the subject's believing it. Perception, memory and inference from things one perceives all supply such chains; Gettier cases do not. The theory failed in both directions, and Goldman came to say so. It is not necessary. I know that all humans are mortal, and that the sun will rise tomorrow. General facts and future facts do not cause my beliefs about them, since a fact about all cases is not a particular event standing at the head of a chain, and a future event has not happened yet. Goldman tried to accommodate general knowledge and never made it comfortable. It is not sufficient, and this is the decisive case. In the 1976 paper Goldman presents a scenario he credits to Carl Ginet. Henry drives through a district in which almost every barn-like structure is a façade erected for a film, and stops in front of the one genuine barn. He looks at it, sees it, and believes there is a barn there. The causal chain is exactly the chain that would satisfy the causal theory in any ordinary case: the barn causes the light, the light causes the visual experience, the experience causes the belief. Henry does not know. What is wrong is not the chain but the neighbourhood — the alternatives Henry could not have discriminated. Reliabilism is Goldman's response to that, and it is a theory about types of process rather than about token chains. That is a change of subject matter, not a change of emphasis. Reliabilism, and the problem inside it The 1979 statement is simple. A belief is justified when it is produced by a reliable belief-forming process, a process whose outputs are true more often than not. Goldman adds a further condition, easy to overlook: there must be no reliable process available to the subject which, had it been used, would have led her not to believe as she does. The simplicity is deceptive, because reliability is a property of types and every actual belief-forming event belongs to indefinitely many types. Suppose a student in Seoul attends a cram school and comes to believe a claim about the subject because the instructor asserts it. Which process produced the belief? Accepting testimony. Accepting testimony from a paid instructor. Accepting testimony from an instructor who has taught this material for a decade and whose students' examination results are published. Accepting testimony about a factual claim from someone with an interest in seeming authoritative. Every one of these is a type the event instantiates, and their reliability rates differ enormously. The last is unreliable and the third may be more reliable than the student's own reading. The theory returns a different verdict for each, and it has not yet said which one to use. This is the generality problem [6] , pressed most forcefully by Earl Conee and Richard Feldman in 'The Generality Problem for Reliabilism' (1998). Their conclusion is not that reliabilism is refuted but that it is incomplete in a way its defenders have not repaired: without a principled rule for selecting the type, the theory does not determine an answer in any case at all. What fixes the type Notice what pulls us towards one description rather than another. If the student is responding to the instructor's manner and would have believed just as readily had the instructor been bluffing, the relevant type is something like accepting confident assertion. If she is responding to the fact that this material has been checked against the examinations, and would have withheld belief had the instructor strayed into an unfamiliar area, the relevant type is narrower and more reliable. The difference between the two students is a difference in what they are sensitive to — in which variations would have changed the belief. Those are facts about the subject. They are not facts about her reflective access to her reasons, and nothing here requires that she be able to articulate anything; a subject may be finely sensitive to cues she could not name. But they are facts about her rather than about the world, and they are what make one type-description correct and the others idle. If that is right, reliabilism cannot be stated in wholly externalist terms: its central notion is not defined until