1
Reliability belongs to a process
A process type, such as vision in good light, is reliable when it produces mostly true beliefs over many uses. It is a property of the kind of process, not of the one belief it produced this time.
Lesson · Knowledge and Evidence · Analysing knowledge
Reliabilism and the Clairvoyant
Reliabilism replaces the believer’s evidence with the reliability of the process that produced the belief. Norman, who has a reliable and undetected power of clairvoyance, makes the definition too broad, and the envatted twin makes it too narrow as a theory of justification. The lab names the direction of each failure.
reliable process: right in most uses
Norman: reliable, no evidence, believes
vat twin: evidence, unreliable, believes
failures run in opposite directions
Two theories, two kinds of counterexample
Core ideas
What this lesson establishes
Reliabilism and evidentialism are two answers to what makes a belief justified. Each is a definition, and each is tested the same way.
2
Evidence is what the believer can access
The rival view says a belief is justified by what the believer has to go on: experiences, memories, reasons they could cite. Two people with the same evidence are alike in justification.
3
The two views fail on different cases
A case with reliability and no accessible evidence tests one. A case with accessible evidence and no reliability tests the other.
The material
Reliable process against accessible evidence
Two definitions, four cases, and the direction each failure runs.
The reliabilist keeps the shape of the earlier definitions and changes the justification condition. A belief is justified when it was produced by a reliable process. A person knows when the belief is also true and held, so in the lab’s symbols the definition is R ∧ T ∧ B, with R for a reliable process. The evidentialist keeps accessible evidence in its place: E ∧ T ∧ B.
Reliability is easily misread. It does not mean the process gave the right answer this time. A coin toss can deliver a true belief once, and the process of guessing by coin is not reliable for that. A good barometer can give a false reading on a rare day and remain reliable. The measure is the proportion of true beliefs the process delivers across its uses.
Norman
Norman has a faculty of clairvoyance that is, as it happens, reliable. One day he finds himself believing that the President is in New York. He has no evidence for it and no evidence that he has the faculty. The President is in New York. The belief is true, held, and produced by a reliable process.
case R E T B knows?
-----------------------------------------
perception 1 1 1 1 yes
Norman 1 0 1 1 no
envatted twin 0 1 0 1 no
stranger's guess 0 0 1 1 no
Reliabilism says Norman knows, because R, T and B all hold. The verdict, as recorded, is that he does not. The definition is too broad on that row. Evidentialism says Norman does not know, because E = 0, and agrees with the verdict. When the lab searches for formulas that fit every row, it finds two, R ∧ E and E ∧ T, and the first it lists is R ∧ E: reliable and with evidence.
The brain in a vat
The envatted twin is a person whose experiences are exactly like yours but who is a brain in a vat, fed those experiences by a computer. Nearly all of the twin’s beliefs about the external world are false, so as a case of knowing the twin is simply a no, and both definitions agree. The case matters for the justification condition, which is where the two theories differ. If the twin’s beliefs are as well justified as yours, then justification cannot be the same as reliability, because the twin’s process is not reliable.
case R E justified?
-------------------------------------
perception 1 1 yes
Norman 1 0 no
envatted twin 0 1 yes
stranger's guess 0 0 no
Run on this second table, “justified iff reliable” fails twice. It is too broad on Norman, who is reliable and not justified by this verdict, and too narrow on the twin, who is justified and not reliable. “Justified iff has evidence” fits all four rows. That does not make evidentialism correct. The reliabilist replies with the chicken sexer, who sorts chicks reliably, cannot say how, and seems to know; his row would be R = 1, E = 0, the same values as Norman, with the opposite verdict, and no formula can fit two rows like that.
Interactive
The method of cases
A definition tested against every case in the table
Method
Testing a theory of justification
The same procedure as before, with the added care of asking which question the table is about.
1
Fix the target
Decide whether the table is about knowing or about justification. A case such as the vat twin can be no on one and yes on the other.
2
Write each case as values of the conditions
Give the reliability of the process and the accessibility of the evidence separately. The cases only separate the theories when those two come apart.
3
Record the verdicts before applying either theory
Norman and the twin are the cases whose verdicts are contested. Write yours down, and note which readers might disagree.
4
Apply each theory and read the direction
Where a theory says yes and the verdict says no it is too broad. Where it says no and the verdict says yes it is too narrow.
5
Check what the search finds
The lab lists the formulas that fit every row. A theory that fits is not thereby right, but one that has no fitting formula at all needs a new condition.
Reliability and evidence
Score 0 / 0Choose the answer you can justify.
Norman and the twin, in the lab’s terms
Use the justification table with the definition R.
perception: R=1 verdict yes agrees
Norman: R=1 verdict no too broad
vat twin: R=0 verdict yes too narrow
guess: R=0 verdict no agrees
R alone: too broad and too narrow
A single definition can fail in both directions at once, on different rows. The lab reports the first row that disagrees and the verdict for the whole definition, and the second is the one that records both.
Common mistakes
1
Reading “reliable” as “right this time”Reliability is a rate for a kind of process. The stranger’s guess in the table gives a true belief, so it was right this time, and the process is not reliable: R = 0. A reliable faculty can also give a false belief on a rare occasion without ceasing to be reliable. Norman’s case is unsettling because the faculty is reliable and nothing in his experience says so.
2
Concluding that a theory which fits the table is trueThe evidentialist definition fits all four rows, and the lab’s search finds several fitting formulas. The rows are a few chosen cases. The chicken sexer, left out of the table, is a case no formula can fit alongside Norman, so the choice of rows has already decided what fits.
3
Confusing the failure of a definition with its refutationA definition that is too broad on one row has a counterexample, and a counterexample is a reason for revising it, not necessarily for dropping the idea. Reliabilists answer Norman by adding that the believer must have no reason to doubt the faculty, which is a new condition tested the same way.
Check your understanding
✓
Finish when you can say which direction each case pushes.
Given a theory of justification and three cases you should be able to compute which rows it fits, name the cases on which it is too broad and the cases on which it is too narrow, and say what a case that no formula fits shows about the table.
Note: The twin is only a counterexample to reliabilism if you record its justification as yes. Some reliabilists record no and explain the intuition away. The lab accepts either verdict and reports what follows from it.