Reliabilism and Virtue Epistemology

A thermometer is good when its readings track the temperature. Nothing else is required of it, and no question arises about whether the thermometer has reasons. The externalist proposal is that believers are evaluated the same way, and that the long search for the right relation among beliefs was a search in the wrong place.

This article gives that proposal its two main developments. The first treats the believer as a mechanism and asks whether the mechanism is accurate. The second treats the believer as an agent and asks whether the accuracy is to their credit.

Justification as reliability

The theory's virtues are substantial and easy to underrate because they are structural rather than clever.

It dissolves the regress. The trilemma assumed that justification is conferred by reasons, so that the question of what justifies the reasons always recurs. On process reliabilism justification is conferred by a process, and processes do not require justification, so there is nothing for the regress to climb. Foundationalism, coherentism, and infinitism are all answers to a question that no longer arises in that form.

It extends knowledge to creatures who deserve it. The child, the dog, and the woman who can identify Porto all have reliable processes, and the theory says all three have justified beliefs without requiring any of them to produce a defence.

It explains why justification matters. The connection to truth is not something the theory has to argue for afterwards; it is what justification is defined as.

Goldman's own formulation adds a refinement worth noting. Belief-forming processes divide into those that take only non-belief states as input, such as perception and memory, and those that take beliefs as input, such as inference. The first kind is justification-conferring when reliable. The second is justification-conferring when it is conditionally reliable and its input beliefs are themselves justified. The recursion terminates in the first kind, which is why the theory has a foundationalist shape without a foundationalist epistemology.

The problem of specifying the process

This is the objection the theory has never disposed of, and it is worth stating precisely because it is often stated loosely.

You see a bird at the feeder and believe it is a goldfinch. What process produced that belief? Perception, which is highly reliable. Visual perception, which is somewhat less so. Visual perception of birds at a distance of five metres in good light, which is reliable if you know birds. Visual identification of small yellow birds by someone who has read one field guide, which may be poor. Every one of these descriptions is true of the very same episode, and they have different reliability scores, so the theory issues no verdict until one of them is selected.

The problem is not that the descriptions are hard to compare. It is that reliabilism needs a principled rule for picking one, and the rule must be stated in terms the theory has available. Picking the description that yields the answer we already believed is circular, and picking the narrowest description is fatal, since the narrowest description of any token process is satisfied only once and so is either perfectly reliable or perfectly unreliable depending on whether that one belief is true.

Richard Feldman pressed this in 1985 and the responses have been technical rather than decisive. Some appeal to the descriptions that figure in cognitive science, on the grounds that there is a fact about which process types the mind actually implements. Others accept that the relevant type is fixed by context, which brings a contextual element into an otherwise austere theory. Neither response has produced a rule that settles the goldfinch case.

Adding the agent

The shift from process to agent is small in appearance and large in effect. A process is something that happens; a virtue is something a person has. Once the bearer of evaluation is the person, two things become sayable that reliabilism could not say.

The first concerns credit. In a fake barn district Henry's perception is functioning normally, and on a straightforward process reliabilism his belief comes out justified. Ernest Sosa's framework distinguishes a belief that is accurate, one that is adroit in the sense of manifesting competence, and one that is apt, meaning accurate because adroit. Henry's belief is accurate and adroit but not apt, since the explanation of its accuracy runs through his luck in which building he faced rather than through his competence. An archer's shot that reaches the target because a gust corrected a bad release is the analogy, and it is exact.

The second concerns the responsibilist strand, developed by Linda Zagzebski and by Jason Baehr, on which the virtues are character traits rather than faculties. This connects epistemology to the ethics of belief and to questions about intellectual education that the faculty version leaves untouched. Its difficulty is coverage: ordinary perceptual knowledge does not manifest open-mindedness or courage, and most of what anyone knows is ordinary perceptual knowledge. Most responsibilists therefore take themselves to be explaining a distinctive high-grade achievement rather than analysing knowledge in general.

What the two approaches share

Both are externalist, both make truth-conduciveness constitutive rather than instrumental, and both decline the structural dispute. Where they differ is in what carries the evaluation, and the difference shows up mainly in the cases.

Reliabilism handles the child and the animal most cleanly, since no agency is required. Virtue epistemology handles the fake barns and the clairvoyant most cleanly, since the notion of a competence explaining a success has the resources to say why Norman's conviction is defective while the woman's identifications of Porto are not: Norman's belief does not issue from a competence he has integrated with the rest of his cognition, and Sosa's reflective knowledge is precisely that integration.

The bill comes due at the same place for both. Reliabilism must say which process, and virtue epistemology must say which competence, and the generality problem is the same problem in both currencies.

Next

The modal conditions. If what excludes Gettier luck is that the belief could not easily have been false, that thought can be stated as a condition on knowledge in two ways, and the two come apart.