Honestly, I don't really 'get' a lot of this paper, but I don't think it's because it's particularly poorly written. I think it's just over my head now. I'm referencing philosophers and propositions that were ready-known to my audience at the time. Look for the real-philosopher-style proposition towards the end. Written for Theory of Knowledge, senior year undergrad.
Bonjour's objection in “A Critique of Foundationalism” is clearly the strongest to Descartes' indubitability. By promoting his own form of epistemic regression, Bonjour shows that the concept of a foundational statement is philosophically inconsistent and the use of such statements actually works counter to the pursuit of truth.
Bonjour seems the first to recognize that the statements made in defense of a position in epistemology must themselves stand up to epistemological scrutiny: does the philosopher know the propositions he uses as evidence? He is right in asserting that an epistemological argument must not only be logically sound but epistemologically consistent, or else one risks deriving (philosophical) knowledge from propositions that are (or may) not be known. The root of Bonjour's perspective can be seen to reside in his theory of “epistemic responsibility”.
“To accept a belief in the absence of such a reason [to think a belief is true] ... is to neglect the pursuit of truth ...” (page 186). Bonjour is (with good reason) attempting to reground epistemological arguments as philosophical, not linguistic, objects. He claims that strong foundationism, by calling a proposition “foundational” and accepting it without justification, abandons not only the epistemological prime directive, justified true belief, but the point of philosophy as a whole.
Bonjour's regression consists of the articulation of a sequence of nested 'why' statements intended to bring logical solidity to the justification process. Descartes investigated the process of confirming knowledge of a proposition in his own mind, and he found that, when traced by intuition, the nested why's terminate with some empirical belief that is (agnostically) unjustifiable, owing to his experiment of skepticism.
Descartes is okay with this, since his God allows him to believe his senses. But Bonjour cries foul from an agnostic perspective, claiming that indubitability at any level of epistemic regression changes the game, allowing a seemingly arbitrary set of propositions to be exempt from the justification requirement.
Essentially, the current problem consists of finding a condition that when added to justified true belief yields knowledge. Lehrer and Paxson suggest that one's justification for belief should not be open to defeat by another true statement. This condition solves many tough special cases of not knowing by applying Goldman's causal knowledge theory at the point of justification. One need only seek a defeating statement to discount the act of justification.
Unfortunately, the defeasibility criterion was engineered to illuminate only certain special cases not included by the current theories; it actually proves to be too strong to be utilized. Harman shows that there will always be a defeating statement for any justification chain. His counterexample is that if one were presented with two equally likely propositions and told that one or the other was true, one would not be justified in believing either one, since the false proposition would be a defeater.
Since this situation exists with every arbitrary pairing of a fact and a reasonable fiction, knowledge cannot be possible under the indefeasibility solution. Obviously the epistemic definition of “defeat” must be weakened.
Goldman offers to revamp the entire theory. If the scope of possible defeating statements is narrowed to only those alternatives that are relevant, enough justified true beliefs are credited as knowledge to mesh with intuition. As a result, though, this notion of “relevance” must be defined with at least as much precision as the notion of “defeat”; Goldman offers the avenues of relevance by way of likelihood and relevance by way of context. He is still holding fast to his previous theory of the causal connection between fact and belief, but he instead applies it to the generation of alternative states.
Goldman uses knowledge to mean chiefly the distinguishing of the truth of believed propositions from alternative propositions that are deemed relevant, so he is not dissuaded by the extra step in his epistemic process of telling relevant from idle.
Goldman himself has declared that his first causal theory cannot account for barn-type examples, and this theory that weights the validity of the causal process that produces a proposition is too closely related to be removed from that possibility.
Consider Hirsch's wording of Goldman's hypothesis. “[S knows that p if] For any q that is a “relevant alternative” to p, if q would have been the case instead of p, then either S would not have believed p, or S's belief that p would have resulted from a causal process different from the one that actually occurred.”
In the barn example, S knows that p is a barn since if it had actually been a barn facade, S's belief would have resulted from a flawed perceptual process (according to Goldman). But Goldman misses that the process of causality is immaterial; S is believing the evidence of her senses, and this evidence has no causal connection with the truth of the barn.
The cases of deception, like the barns and coins examples, prove to be significant since they both begin the epistemic process as a less-than-perfect-information game. Intuition demands that S knows there is a barn because if no attempt had been made to hoodwink S, S's proposition would have been true. If S had been aware that there was a possibility of encountering a facade, S would not have held the belief (in the epistemological sense) “there is a barn”. Notice that an intuitive assessment of this case does not need to resort to causal language; regardless of the object in front of the car, S's beliefs will always been formed (or “caused by”, if one wishes) by evidence (including memory).
A similar situation exists in the case of two people in S's office: A, who secretly owns a Ford, and B, who does not own a Ford but presents a body of evidence M that he does. Let P be the proposition “a person in S's office owns a Ford.” By definition, S does not know the truth and by intuition, S does not know P. But note that only M has any causal power; if A was not there, or if B was not there and A had presented his evidence under an assumed name, or if A did not own a Ford, S's belief would be identical and his knowledge would still total zero.
Goldman's argument makes the bold move of transferring the determining factor of knowledge to an external source, namely, "likelihood" or "the context". Such a shift in paradigm clearly runs counter to intuitions about knowledge. Beyond the initial truth of the proposition, one feels that the epistemic burden should lie internally, whether from simple arrogance or a genuine desire to define knowledge as a uniquely human (or perhaps just biological) capacity. Tracing causality vectors proves too problematic in cases of subterfuge, and so we may take these cases as counterexamples to any non-evidential external-causal theories of knowledge.