There are different roles that causal elements and notions like reliability can play in an epistemological theory.
Suppose it looks to you as if there's a clock on the mantel. But then you acquire evidence that your visual experience was produced in an unreliable way, that is, in such a way that it's not likely to be a reliable guide to whether there really is a clock on the mantel. For example, you may acquire evidence that your visual experiences are being produced by an evil neuroscientist. One way of incorporating notions like reliability into your epistemological theory is to say that in cases like this one, the evidence that your experiences are unreliable will defeat your justification for believing that there is a clock on the mantel.
This is the least controversial use of the notion of reliability. Pretty much everyone accepts something like this.
Slightly more controversial is the use of reliability as an ingredient in classical solutions to the Gettier Problem. By "classical solutions"I mean solutions that retain that idea that knowledge requires justified true belief, and try to find some extra fourth condition to block Gettier-type cases from counting as knowledge. On these views, facts about reliability are what you have to add to justified true beliefs, in order to get knowledge.
There are also more radical solutions to the Gettier Problem. These are views that ditch the idea that knowledge requires evidence or justification, and instead just try to explain knowledge as true belief which satisfies some causal or reliability condition. This is the sort of view we just saw Goldman offering. According to these views, reliability is what you add to true beliefs, instead of evidence or justification, in order to get knowledge.
The most controversial use of reliability in epistemology is the view we're going to look at now. This is the view that, instead of replacing the notion of justification in an account of knowledge, reliability is what makes a belief justified. Goldman came to hold such a view several years after he published "Discrimination and Perceptual Knowledge." He defends this view in "What is Justified Belief?"
This fourth view is an externalist account of justification. It says that the facts about your beliefs which determine whether or not they're justified are not "internally available" to you. You can't tell by introspection and reflection alone whether your beliefs have those features. This is what makes the view so controversial. Traditionally, philosophers always thought that what it's reasonable or justified for you to believe was a wholly internal matter. If you were justified in believing P, then all your internal epistemic duplicates would be justified in believing P, too. (Even many proponents of externalist theories of knowledge, like Goldman when he wrote "Discrimination and Perceptual Knowledge," still thought that justification was a wholly internal affair.)
The externalist about justification rejects all that. The externalist about justification says that you and your internal epistemic duplicates can differ with respect to what you're justified in believing.
There are several different kinds of externalist accounts of justification. We're going to look at just one of these, process reliabilism. According to the process reliabilist, whether or not your belief is justified is a function of how reliable the processes which produced the belief are. You don't have to know how your belief was produced, for the belief to be justified. Nor do you have to have any evidence that the belief was produced by a reliable process. According to the process reliabilist, your belief is justified just so long as it was in fact produced by a reliable process.
For brevity, I'll sometimes refer to process reliabilism as just "reliabilism."
In section 1 of "What is Justified Belief?" Goldman discusses several accounts of justification that he finds unsatisfactory. These accounts postulate beliefs which have a variety of special epistemic properties. Some are:
As Goldman discusses, each of these epistemic properties can be interpreted in a number of different ways. We'll be coming back to these epistemic properties again, when we discuss foundationalism.
Goldman offers an account of justification which is very different from the accounts he examines in section 1.
Goldman emphasizes that in his account, he's not going to make either of the following assumptions:
We'll also be coming back to these assumptions again, when we discuss foundationalism.
A belief-forming mechanism is reliable to the extent that it tends to produce true beliefs. Some belief-forming mechanisms yield beliefs as output only when they're given other beliefs as input. Belief-forming mechanisms of that sort count as conditionally reliable when they tend to produce true beliefs as output when the beliefs they're given as input are themselves true.
Goldman's initial account of justification goes as follows:
1. All beliefs produced by reliable processes (of the sort that require no beliefs as input) are justified.
2. All beliefs produced by conditionally reliable processes that received beliefs as input which were themselves justified are justified.
3. No other beliefs are justified.
Notice that in the definition of "conditionally reliable"we're interested in how likely it will be that the output beliefs are true, given that the input beliefs are true. But when we're defining "justification," we say that if a conditionally reliable process takes a justified belief as input, that's enough for the output belief to be justified, too.
There is a problem for this account of justification. Suppose Jones has very good evidence that his beliefs about his childhood were formed unreliably. If so, then it seems epistemically irresponsible for Jones to continue holding those beliefs--even if, as it happens, the beliefs were formed in a reliable way. Regardless of whether the beliefs were formed reliably, if Jones has good reason to believe they weren't, then it ought to be unreasonable and unjustified for him to continue holding them.
This prompts Goldman's revised account of justification:
Say that a belief B of S's is "undermined" iff S believes (or "ought" to be using a process that would generate a belief) that belief B was formed in an unreliable way. Then:
1. All beliefs produced by reliable processes (of the sort that require no beliefs as input) are justified, unless they're undermined.
2. All beliefs produced by conditionally reliable processes that received beliefs as input which were themselves justified are justified, unless they're undermined.
3. No other beliefs are justified.