Sober on Epistemology

Curtis Brown

This outline more or less follows the main line of thought in Elliott Sober, Core Questions in Philosophy (Fourth Edition), Lectures 12 - 17.

I. Knowledge

A. A proposed definition of knowledge (the "JTB" account, for "justified true belief"):

S knows that p if and only if:
(1) S believes that p
(2) p is true
(3) S is justified in believing that P

B. Problems for this definition:

1. Gettier examples (the person who will get the job has 10 coins in his pocket . . .)
2. the stopped clock
3. the lottery

all three cases appear to be cases in which the three conditions are met, i.e. they appear to be instances of justified, true belief, but it also seems that we don't know p in these cases. So the examples provide evidence that the conditions are not jointly sufficient.

C. One proposal  to revise the definition (the Certain True Belief account):

S knows that p if and only if:
(1) S believes that p
(2) p is true
(3') S is absolutely certain that p

(the only change is to the third condition. We could regard this either as a replacement condition or as an explication of what it is to be justified. We could call this the CTB account, for Certain True Belief.)

D. A problem with the revised definition is that it seems to lead to skepticism about knowledge, the view that we can know little or nothing.

Descartes's increasingly radical skeptical arguments in Meditation I seem designed to show that it is possible that I am mistaken about almost everything I believe. If I recognize that it is possible for me to be mistaken, then I cannot be certain that I am correct. So the arguments of Meditation I seem to show that I do not know anything except perhaps that I exist and what my current experiences are.

Of course, Descartes tries to overcome these skeptical doubts later in the Meditations, via his proof of God's existence in Meditation III. But most readers find the skeptical arguments more convincing than the proof of God's existence, so we seem to be left with the worry that if knowledge requires certainty, then we can know hardly anything.

E. A Third approach to defining knowledge (the Reliable True Belief account):

S knows that p if and only if
(1) S believes that p
(2) p is true
(3'') S's belief that p is a reliable indicator that p: in the circumstances S is actually in, S's belief that p can be caused only by p.

Notice that the new condition 3 is an objective, not subjective condition: that is, it says something about the circumstances S is actually in, not (just) about S's subjective mental state.

Notice also that it follows from this definition that one could know that p without knowing that one knows that p. That is, I could believe p, p could be true, and I could be in circumstances in which that belief could only be caused only by the fact that p, even though I did not in fact know that I was in such circumstances.

If there is no evil demon, and no other systematically deceptive feature of my circumstances relevant to my belief that p, then arguably if p is an observational belief (I see a hand in front of me, the car is green, etc.), then it is such that I could only come to believe it on the basis of the fact it represents.

F. Problem cases revisited.

1. Gettier examples. Although the belief that the person who will get the job has ten coins in his pocket is justified and true, it does not seem to be a reliable indicator (in the sense of our third version of condition (3)). The process by which the belief was arrived at (namely deduction from a false premise) could just as easily have led to a false belief as to a true one.

2. The stopped clock. Again, although in this case my belief based on the clock reading turns out to be true, it is not a reliable indicator of the time. I could just as easily have passed the clock a little earlier or later, in which case my belief would have been false. (This does raise the issue of how broad the circumstances we need to consider are, though. Sober points out that to some extent this will be relative to the context in which we ask the question. But it seems that there will be very few contexts in which the relevant circumstances include the exact time at which one forms a belief.)

3. The lottery. My belief that I won't win is justified and true. However, again, it is not a reliable indicator of the truth in the relevant sense. A belief is a reliable indicator of a fact only if, given my circumstances, I could only acquire the belief if it were true. But in the lottery case it is entirely possible (though not likely) that I could acquire the belief even though it were false. After all, the person who actually wins the lottery had equally good reason to believe that he or she would not win.

So it looks like the RTB gives the answers that seem right in the cases that the JTB account had trouble with. What about the kind of case that gave the CTB trouble?

4. The evil genius. If Descartes' arguments that God exists and isn't a deceiver are not successful, then it appears that, according to the CTB, I cannot know (for example) that there is a computer monitor in front of me. This is because it's not certain (indubitable) that there's a monitor in front of me, since an evil genius could feed me monitor-like experiences even if there wasn't a monitor there.

What about the RTB? We need to distinguish two cases. (1) There is an evil genius. In this case, my belief that there's a monitor in front of me is not a reliable indicator (even if there is in fact a monitor in front of me), because it could be false just as easily as true. In that case, I don't know that there is a monitor in front of me. (2) There is no evil genius (or other weird sceptical possibility). In that case, in the circumstances I'm actually in, my belief that there's a monitor in front of me could only be caused by . . . a monitor in front of me. So it's a reliable indicator, so I know that there's a monitor in front of me.

I can't be certain, of course, that I'm in situation (2) rather than situation (1). But if, as I believe, I am in fact in situation 2, then I do know that there's a monitor in front of me. The mere possibility of an evil genius isn't enough to prevent me from knowing: only an actual evil genius could do that!

II. Justification of Belief

In addition to skepticism about knowledge, Sober discusses skepticism about whether our beliefs are ever justified. 

A. Hume's argument that induction cannot be rationally justified.

Two versions of this argument:

1. The first version: Induction presupposes the Principle of the Uniformity of Nature, the idea that the unobserved parts of the universe (those parts in the future, or a long time ago, or a long way away, for instance) are more or less similar to the observed parts. But there is no way to justify this principle except by induction itself.

2. The second version abandoned the Principle of the Uniformity of Nature, and was expressed in terms of inference rules instead. There are two problems with inductive inference rules.

a. It is difficult (impossible?) to formulate a precise rule of inductive inference. Deductive inference rules, by contrast, are easy. For instance: if you have the premise that if p then q, and the premise that p, then you may infer that q. 

What would a corresponding inductive inference rule look like? Perhaps something like this: If you have the premise that all observed A's have been B's, then you may infer that all A's are B's. More generally, from the premise that n% of observed A's have been B's, infer that n% of all A's are B's.

But clearly that rule is not satisfactory. Some instances of the rule seem OK: for instance, from the premise that all observed emeralds have been green, it seems reasonable to infer that all emeralds are green. However, other instance of the formula do not fare so well. For instance (Nelson Goodman's "new riddle of induction"), from the premise that all observed emeralds have been grue, it does not seem acceptable to infer that all emeralds are grue.

b. Suppose we somehow managed to construct an inductive inference rule that seemed intuitively to be acceptable. How could we show that it is reliable? In the case of deductive inference rules, we can show (by using truth tables, for instance) that the rules can never lead us from true premises to a false conclusion, and hence are legitimate deductive inference rules. In the case of an inductive inference rule, we would like to show that the rule will usually lead from true premises to a true conclusion. But Hume's circularity worry seems to be a serious problem here. It seems that to infer that induction is usually reliable, we would need to make an inference something like this:

most observed uses of induction have been reliable
most uses of induction are reliable

. . . which is an instance of the inductive pattern, and so amounts to using induction to justify our use of induction.

2. One response to these problems with the justification of induction is to argue that they presuppose a foundationalist conception of justification. On the foundationalist view, beliefs come in distinct layers. On the bottom, there are indubitable beliefs, for example about one's own existence and mental states. On the next level up, we have observational beliefs about the world. A level higher than that, we have beliefs about unobserved entities. The foundationalist view is that every belief on any given level must be justified entirely on the basis of beliefs at the next lower level (so that ultimately everything is justified on the basis of indubitable beliefs).

Sober suggests that foundationalism is mistaken. Perhaps we need a more coherentist conception of justification. Instead of saying that we are justified in believing p only if we can derive p from lower-level beliefs, perhaps we should say something like this: I am justified in believing that p unless I have some concrete reason to doubt it. (A concrete reason for doubt could be observational evidence that conflicts with p, or it could be a conflict between p and other things I believe.) Notice that the evil genius does not count as a concrete reason to doubt my beliefs unless and until I have some reason to believe that there actually is an evil genius.

3. Another response to the worries about induction may be to suggest that induction as it is usually described is not in fact an inference rule we actually use. We should notice three things about our inferences from observed to unobserved phenomena:

a. sometimes many positive instances are not enough to lead us to believe a general proposition. (The turkey's induction, the water balloon induction, the iron-is-always-a-solid induction, etc.)

b. sometimes a single positive instance is enough. (I wonder whether the red area on the stovetop feels nice. I touch it and burn myself. I do not need to do several more trials to convince myself that it will always burn me if I touch it.)

c. even when we do believe a generalization on the basis of observation of positive instances, it does not seem to be the mere repetition of positive instances that explains our confidence in the generalization.

These observations may lead us to wonder whether many inferences that could be thought of as inductive are not better described as abductive instead. For instance, I notice that all the emeralds I have observed have been green. What is the best explanation of this fact? I may consider various possible explanations: all emeralds are green, all emeralds are grue, I've observed a highly biased sample and in fact a small percentage of emeralds is green, and so on. It seems that, rather than just mechanically extend my observations to a broader sample, I in fact try to determine which of these various possibilities is the best explanation of my observations.

An interesting and related twist on the issue is the view of Karl Popper. Popper also held that induction is not an inference procedure we actually use. However, he would also have rejected the idea that we use abduction. Popper held that the only form of reasoning we use in deciding what to believe is deductive reasoning. We can refute a theory by deductive reasoning if we encounter an observation that is incompatible with what the theory predicts: T -> E, but not E, therefore not T is a valid deductive argument form. Popper thought that we never have reason to hold that a particular theory is true; the best we can say is that we have not yet refuted it. In his view, science is not about establishing the truth of the right theory, but about trying as hard as we can to refute the ones that aren't true. All we can say about the ones that survive the testing is that we don't yet have any reason to think them false. Popper defends this view in "Conjectural Knowledge: My Solution of the Problem of Induction," chapter 1 of his book Objective Knowledge.)

Last update: October 19, 2007.
Curtis Brown  |  Introduction to Philosophy  |  Philosophy Department  |  Trinity University