Theory of Knowledge

Introductory Post

This will be the web log for Theory of Knowledge, PHI 3300, for the spring 2010 term at UF.  I'll post here announcements, and follow ups on class discussion typically with open comments.  If you would like to initiate a discussion, then send me what you want to post.

Final Post

I returned the second paper by e-mail.  If you didn't get it, and happen to read this, you can send me a note to let me know.  I included the final exam grade and course grade.

Have a good summer.

Bealer's Response to Scientific Essentialism

Bealer's response to Scientific Essentialism is not to reject a posteriori necessities but instead to argue that these involve a special class of terms which he calls semantically unstable terms.  In particular, natural kind terms like 'gold', 'water', 'tiger', 'lead' are he says semantically unstable terms.

A term is semantically unstable if its extension (in a world and across possible worlds) can vary across circumstances which are epistemically the same for the users of the term and in which we keep fixed the conventions attached to the term.  If Putnam is right, 'water' is semantically unstable because the conventions and epistemic situation of people on Earth and on Twin Earth is the same in 1750 but 'water' here means H2O and 'water' on Twin Earth means XYZ.

Bealer basically claims that the central concepts that philosophy is concerned with are not semantically unstable but rather semantically stable terms.  These include terms such as

property, proposition, relation, category, cause, event, object, at least determinable concepts of space and time, justice, right, evil, consciousness, belief, evidence, justification, meaning, concept,

and so on.  He says, for example, it is hard to see how 'property' could be thought to vary its extension across possible worlds fixing epistemic position and conventions for its use.  Similarly for 'consciousness': if we try a Twin Earth thought experiment on this, he says, it goes nowhere, for finding being with silicon neurons, for example, but who behaved just like us, would not cause us to think that they weren't or couldn't be conscious.

If Bealer is right, then necessities involve the concepts  philosophy is centrally concerned with (not natural kind concepts, he claims) would not be a posteriori necessities, and so one could admit that there were such necessities without thinking it undermines the autonomy and authority of philosophy.

Scientific Essentialism

What is scientific essentialism?  This is the thesis that there are a posteriori necessities, that is, necessities that can be discovered only by empirical research.  Something which is a priori can be known a posteriori--you could learn by being told that there is no greatest prime number.  You know then a posteriori, by authority, but not a priori.  But it is still an a priori truth because it can be known a priori.  An a posteriori truth is one that can only be known a posteriori.  So an a posteriori necessity is a necessity that can only be known a posteriori.

The traditional view was that all necessities are truths of reason--Kant being an exception to this in his view that there were synthetic a priori truths.  But even Kant thought that all necessities were a priori.  Thus, the claim that there are a posteriori necessities runs against a strong current.

The view, however, is not a recent one.  It goes back at least to John Locke's Essay Concerning Human Understanding in which he distinguishes between nominal and real essences.  Locke was writing at the height of the Scientific Revolution and his thinking was informed by contemporary developments in science, and especially by Newton and Robert Boyle who was a proponent of the corpuscular theory of matter, that all the dynamics of objects could be explained in terms of the motions, shapes, and sizes of their parts.  Locke's distinction between real and nominal essences was a response to this idea.

A nominal essence is the verbal definition of a thing.  It real essence is the underlying nature of thing that explains the features by which we pick it out.  Sometimes these can coincide.  They do so for geometical kinds like 'triangle'.  The nominal essence is being a plane figure with three interior angles and this fixes also all its essential properties and so this is also it real essence.  But in the case of natural kinds, kinds in the world, like lead or gold, Locke held  that these could come apart.  We might define 'gold' as a yellow, malleable, noncorrosive metal that can be dissolved in acqua regia (nitro-hydrochloric acid).  Its real essence has to be discovered because that is to be sought in the details of how its particles and their natures induce on gold its superficial properties.  Since the real essence is supposed to pick out the real nature of the thing, it is supposed to be an essential properties of the kind gold, though it is not discoverable from the nominal definition. 

This idea fell out of favor, but was resurrected in the early '70s by Saul Kripke and Hilary Putnam.  Putnam developed a style of argument for this that relied on what have come to be called Twin Earth thought experiments.  Let us run through this quickly with the example Putnam uses initially, namely, with the term 'water'.

First, we consider whether 'water' mean the same thing in 1750 before the development of the atomic theory and now, that is, before we discovered the chemical structure of the water in our lakes, rivers, oceans, etc.  It is natural to say 'yes'.  This is a crucial first step in the argument.

Now, we consider the possibility that somewhere in the galaxy there is a planet that to all superficial appearances is a twin of Earth.  Call it Twin Earth.  There are people there to look just like the people here, they speak a language that sounds like English, some of them are students at a university they call 'the University of Florida', and one of the faculty there is named 'Kirk Ludwig' and teaches a philosophy course in which they are discussion Twin Earth thought experiments.  And there is a clear, potable, liquid in the lakes, rivers, oceans, etc., which falls from the sky in rain storms, and so on.

A spaceship from Earth reaches Twin Earth.  The scientists are amazed!  They go to the beach.  They take shelter  from the rain. They get on with their testing of substances on Twin Earth.  Lo and behold, they discover that the stuff in the lakes, rivers, oceans, and rain has a complicated molecular structure which is not H2O but something we will abbreviate as XYZ. 

Query: is that stuff on Twin Earth water or not?

Putnam, Kripke, and many other philosophers say intuitively the answer is 'no', even though it fits every description someone competent in English in 1750 could have given of water.  It isn't water, they say, because it is not H2O. 

We get the same result, they hold, if we merely consider a counterfactual world in which there is a community with the same conventions for the use of words as in English, in a world superficially like Earth, except that there is XYZ in the oceans, rivers, lakes, rain etc.  Intuitively, they say, that stuff is not water because it is not H2O.

Now, we might have initially thought that water is not necessarily H2O because we seem to be able to imagine that we have discovered that it have had some other chemical composition.  But this they explain away as our conceiving of a situation that is epistemically the same as ours but in which we discover that what we would there call 'water' was something else.  So they have an explanation of the apparently contrary intuition: it is really not about water being something other than what it is but about 'water' in an epistemically indistinguishable circumstance picking  out something other than H2O.  And they say there is no debunking explanation of the intuition that the stuff on Twin Earth (call it 'twater') is not water. 

They conclude then that necessarily water is H2O but that this could be discovered only empirically.

Notes on "Two Dogmas"

Notes on "Two Dogmas" are available under the 'Notes' heading in the right hand column and also here.

Final Examination Posted

The final examination is available in the left hand column under 'Course Info'.

'7+5=12' is analytic

Kant argued that there are synthetic a priori truths in part by example.  He thought the truths of geometry and arithmetic were a priori, but not analytic.  He said a judgment was analytic just in case the concept of the predicate was contained in the concept of the subject.  For example, he held that 'Bodies are extended' is  analytic because the concept of extension is contained in the concept of body, but that 'Bodies are heavy' is not analytic because the concept of heaviness is not contained in the concept of body.  But he held that such truths as that the shortest distance between two points is a straight line and 7 + 5 = 12 are not analytic because the concept of a straight line is not contained in the concept of shortest distance between two points and the concept of 12 is not contained in the concept of the sum of 7 and 5. 

Part of what is going on here is that he had a very narrow notion of analytic judgment.  For example, it does not have room for sentences of the form 'p or not p' as expressing analytic judgments, because they are not of subject predicate form and the test of containment does not apply.  But also he was thinking of a particularly simple way of thinking about containment, modeled by 'Bachelors are unmarried'.  The picture is this.  The concept of a bachelor is a conjunctive concept in the sense that it can be spelled out in terms of a conjunction of features.  x is a bachelor iff x is male and x is unmarried and x is of marriageable age, and x is not a widower, and x is not a priest, etc.  Then 'Bachelors are unmarried' is true because anything that is male and unmarried and ... is of course unmarried.  And it is not easy to see how this model fits the examples above. 

But if we have a more general account of the analytic, namely, now applying the term to sentences rather than judgments, as sentences whose truth can be proven from statements about the meanings of the contained terms, then we can show that the examples that Kant has in mind are analytic.  Let me just do with this '7+5=12'.  Here goes with the definitions, then.

Primitive singular term:

0  (zero; primitive)

One place operator:

S(x)  (read ‘the successor of x’; the result of replacing ‘x’ with a singular term in ‘S(x)’ is a singular term, and a primitive)

Binary operators:

+, *   (addition sign and multiplication sign, respectively; the result of placing a singular term on either side of a binary operator and enclosing the result in parentheses is a singular term; defined contextually below)

Binary predicates:

=, <   (the identity relation and the less than relation, respectively; the result of placing a singular term on either side of a binary predicate is a sentence; ‘=’ is primitive; ‘<’ is defined below)

 Def. 1       For any x, x is a number iff x = 0 or x is the successor of a number.

Def. 2       For any n, m, if n and m are numbers, then n + 0 = n and n + S(m) = S(n + m).

Def. 3       For any n, m, if n and m are numbers, then n * 0 = 0 and n * S(m) = (n * m) + n.

Def. 4       For any n, m, if n and m are numbers, then n < m iff m is a successor of n.

Def. 5       For any n, m, if n and m are numbers, n < m iff there is a number r (=/= 0) such that n + r = m.

Def. 6       1 = S(0), 2 = S(1), 3 = S(2), etc.

Now, let's tackle '7+5=12'.  First, by def. 6, 12 = S(S(S(S(S(S(S(S(S(S(S(S(0))))))))))))).  Now, by def 2,  7+5 = 7+S(4) = S(7+4) = S(7+S(3)) = S(S(7+3)) = S(S(7+S(2))) = S(S(S(7+2)))) = S(S(S(7+S(1))))) = S(S(S(S(7+1))))) = S(S(S(S(S(7+0)))))) = S(S(S(S(S(7)))))).  And since 7 = S(S(S(S(S(S(S(0)))))), substituting in, we get S(S(S(S(S(S(S(S(S(S(S(S(0))))))))))).  So, we have

S(S(S(S(S(S(S(S(S(S(S(S(0))))))))))))) = S(S(S(S(S(S(S(S(S(S(S(S(0)))))))))))!

Which, you can see ;), involves the same designator on each side of the '=' sign.  So we see that '7+5 = 12' is true as a matter of definition alone.

Access and State Internalism

This is a brief note on the relations between access and state internalism (Accessibilism and Mentalism).  Goldman says that the Guidance-deontological conception of justification is what leads to internalism.  He thinks it does so by two stages.  The GD conception holds that to be justified in believing something is to believe responsibly, i.e., in accordance with one's epistemic duties.  He thinks this motivates the view that what justifies belief must be something that the agent can himself drawn on, for otherwise he would not be able to carry out his epistemic duties and so be epistemically responsible in believing.  This gives rise to the first step, as Goldman sees it.  To paraphrase:

KJ: x is justified in believing that p only if the justifiers for x's belief are readily knowable by x.

He thinks we get internalism by adding to this that only conscious mental states or those plus what we can remember of past mental states are readily knowable.  These states are conceived of as non-relational states, and, hence, as internal states.  (A state is a relational state if having it implies the existence of a distinct contingent individual (not a part of oneself).  So being six feet tall is a non-relation state and being married is a relational state.) 

Feldman and Conee distinguish two versions of internalism.  The first is Accessibilism, according to which what justifies one in believing something must in some appropriate sense be accessible to you, something you can come to know about or perhaps be justified in believing obtains.  The second is Mentalism, which holds that the justifiers for belief are all non-relational states of the agent.  Goldman's first stage on the way from GD is in fact a version of Accessibilism.  So he seems to recognize only the second sort.  However, many philosophers who are internalists, for example, Laurence Bonjour, seem to be internalists primarily in the first sense.

What is the relation between the two sorts?  I said they were independent.  Why?  If they are independent, then one can hold either without the other consistently. 

Suppose you are an access internalist.  Then you think that if you are justified in believing that p, then what makes you justified is something that must be in principle accessible to you.  But this doesn't entail that the justifier has to be a fact about a mental state of yours.  The justifier might be some fact, for example, about the external world.  It just has to be something that you can come to know about it.  So Accessibilism by itself does not entail Mentalism.  This is reflected in Goldman's argument.  For he adds to KJ (which is a kind of Accessibilism) the thesis that the things we can readily know are facts about our mental states, to arrive at Mentalism.  So Accessibilism by itself does not entail this.

What about the other direction, from Mentalism to Accessibilism?  Is there an entailment here?  It doesn't seem so.  Mentalism says the justifiers are internal.  And it may be that the most plausible candidates are also things we can know about directly, so that they satisfy the standard required by Accessibilism, but this isn't required by Mentalism as such.  You might think that, for example, we can be justified in believing only things about our own minds and that what justifies us are various reliable belief producing processes that are internal to the agent, without requiring that we be able to determine or tell or know that the processes that produce our beliefs are reliable.  In this case, one would be state internalist without endorsing Accessibilism.  You might protest that these  don't sound like mental states, but maybe the processes in question are mental processes, and so involve a series of mental states; or we could just identify the justifiers with non-relational mental states and declare that you don't have to know you have them for them to justify other things.  Perhaps some of them, for example, are deeply unconscious mental states, and one might even say, in principle inaccessible.  The point is not that this is plausible, but that it is allowed by the advocate of mental state internalism, and so shows that this  view  does not entail Accessibilism.

Goldman's Historical Reliabilism

I want to summarize in this post some of the critical remarks I made about Goldman's historical reliabilism.

Goldman wants  to give a reductive analysis of the concept of justification.  His initial thought is to identify as a sufficient condition what seems in many cases to be a necessary condition on being justified, namely, that a belief be produce by way of a reliable process.  This is why his position is called historical reliabilism.

To simplify it a bit, the basic idea is that

S's belief  that p is justified iff it was produced immediately by a belief independent unconditionally reliable process or it was produce by a belief dependent conditionally reliable process whose inputs are justified.

Goldman himself notes some problems with this, of a sort lots of people have noticed.  The first has to do with whether a process is required, because that implies the passage of time.  If you prick my finger with a pin, I seem to know I am in pain as soon as I am.  There is a reliable connection: I reliably get it right (or so it seems).  But is there a process?  Goldman says yes, even if it doesn't take much time.  But  even if there is such a process, the question is whether it is conceptually required: if there weren't really a process, but simultaneity of belief and pain, would we deny the belief that you are in pain was justified?

The second has to do with the possibility that some processes that are unreliable might have been reliable, but not produced justified beliefs.  For example, wishful thinking.  Wishing thinking is coming to believe something because you wish it to be so.  (It is quite common in various rather innocent ways, as in the so-called Lake Woebegone effect: people tend to overestimate their own knowledge and competence relative to others--and there is evidence that it is inversely proportional to their actual knowledge and competence (the Dunning-Kruger effect).  In a survey of faculty at the University of Nebraska (in 1997), 68% estimated that they were in the top 25% of teachers at the university.  No doubt that leads to greater confidence which enhances performance.)

Wishful thinking is not in fact reliable (as we think) but if it were, though we didn't know it, would that mean that wishful thinking led to justified beliefs?  Most people think not.  But then that is a counterexample because an analysis is supposed to give  you the necessary and sufficient conditions for the application of the concept, and so is committed to anything which in principle fits the conditions being a justified belief.  

Goldman says we could reject the intuition (but isn't that just ad hoc?), or we could insist it is only processes that are actually reliable that count in any possible world (but isn't that just ad hoc, and won't it rule out beliefs from being justified by processes that could have been developed but have not been, or different forms of perceptual experience that we have?), or we could just give up the "standard format of conceptual analysis" in favor of trying to explain why we call what we do justified. 

With respect to the last of these, what does that mean?  Giving up on the analysis?  Then isn't that to admit defeat?  In any case, if our use of "justified" doesn't actually track just being reliably produced, how could such an explanation expect to be successful?

Goldman also looks at another counterexample to the analysis as sketched above.  Jones remembers things that happened as a child but his psychiatrist and parents tell him that he is wrong and those are 'false memories'.  He stubbornly persist and ignores the counterevidence.  His belief is reliably produced but how can he be justified when he ignores the counter evidence?  Any reasonable person would retract his belief.

Goldman says that what we should do is say that someone is justified in believing something if it is produced by a reliable process and there was no other reliable process available to him to use that would have had a different result if employed instead or with the one in question. 

But what makes a process available?  We don't want just anything to count or processes which won't be developed for years may be relevant and that seems bizarre, or ones that the person couldn't in any reasonable sense access.  And we don't want  to count getting new evidence because sometimes we are justified but wouldn't be if we gathered some more perhaps misleading evidence.

Goldman says the sorts of things we should have in mind are things like reviewing the evidence we have or going over it more carefully.

But if that is how we understand the availability of a process, then it is understood in terms of an epistemic notion, that of evidence, and Goldman ends  up violating one of his own conditions on the adequacy of an analysis.

Second paper topics available

The second paper topics are available in the left hand column under 'Course Info'.  Due 4/23 by email by midnight.

Revised Schedule

A revised schedule for the last  four weeks is available here and in the left hand column under 'Course Info'.

Counterexamples to Nozick's Analysis of Knowledge

I wanted to record the counterexample to Nozick's analysis I discussed in class but did not have on the handout.  I'll add some additional ones. 

Nozick's initial analysis goes as follows: 

For any x, x knows that p if and only if

(1)   p is true

(2)   x believes that p

(3)   if p weren’t true, x wouldn’t believe that p

(4)   if p were true, x would believe that p.

Nozick says that this handles the original Gettier cases because they fail condition 3.  However, it seem relatively easy to construct a case in which we have a Gettier counterexample to knowledge but Nozick's analysis gets the call wrong.

Suppose Smith, Jones and Brown all work at GM headquarters.  Brown has just bought a Ford.  Alone of among his coworkers he has always wanted one.  But he naturally doesn't want anyone to think that he drives a Ford.  In this, he is similar to everyone else at his workplace with one exception, namely, his friend Jones.  Smith, who lives close to Brown, spots Brown driving his Ford with his friend Jones.  Brown convinces his friend Jones to pretend that the Ford is his.  Jones is going to retire in a few years and has no ambition for advancement, so he doesn't care if his coworkers think he owns a Ford, though he would not in fact consider buying a Ford, because he doesn't think they have good quality control.  Smith comes to believe that Jones own a Ford on the basis of Jones's behavior and what he says.  He infers one of his colleagues owns a Ford.  He has a justified true belief in this last, but he doesn't know that one of his colleagues owns a Ford.

However, Nozick's conditions for knowledge are satisfied.  Smith's belief that one of his colleagues owns a Ford is true, and if it weren't true he wouldn't believe it, because none of his coworkers besides Brown would own a Ford, and the only reason Jones pretends to own a Ford is because Brown bought one and got spotted driving it.  On the other hand, if his belief were true, he would believe it, because it would only be because Brown bought a Ford, and in that case, it was pretty likely that Smith would have spotted Brown driving it.

This is a counterexample also to the revised version of Nozick's analysis that relativizes to the method by which the belief is acquired.

For any x, x knows that p if and only if

(1)   p is true

(2)   S believes, via method or way of coming to believe M, that p.

(3)   If p weren’t true and S were to use M to arrive at a belief whether (or not) p, then S wouldn’t believe, via M, that p.

(4)   If p were true and S were to use M to arrive at a belief whether (or not) p, then S would believe, via M, that p.

In addition, we can modify the case involving the grandmother who comes to believe that her grandson is well by seeing him.  Suppose that again the family would not want her to hear if her grandson were ill because of concern for her health.  So they do some contingency planning and hire an impostor who will step in and pretend to be her grandson if he becomes ill.  Suppose her eyesight is not what is used to be, and someone who looks a lot like her grandson and can imitate his voice flawlessly, so that the deception will not be too hard to pull off.  Surely she knows her grandson is healthy, even though if he were not, she would still believe it using the same method.  She thus knows though she doesn't satisfy Nozick's conditions for knowledge.

Consider the brain in a vat case in which the BIV thinks it is a BIV.  Imagine that the particular experiment that is being done involves seeing whether or people become paranoid if they come to think that they are BIVs.  It is cheaper to run these experiments using BIVs, so that's how it is done.  No other BIV experiments are being run at this time.  It seems that condition 3 is met, and also that condition 4 is met, for if anyone were a BIV, he would be induced to believe that he was. This is not the result Nozick wants, however, because he wants  the BIV to lack knowledge that he is a BIV.

The difficulty for the analysis  lies in how easy it is to manipulate the background story so as to get the subjunctive conditions to come out the way you want them to.  It is all a matter of describing background causal connections that ensure the right result but in a way that doesn't affect the ground we have for saying someone knows or doesn't know something.

Why is externalism not a response to skepticism?

I have some reason to think that it was not clear in everyone's mind why externalism isn't a response to the traditional problem of skepticism about the external world.  Here is the idea in a  nutshell.  The skeptical challenge is to show that we have more reason than not to believe what we do about the external world.  This requires that we present reasons to ourselves which we can see to involve facts that show that our beliefs are more likely than not to be true.  The externalist holds that one is justified if one's belief have a certain feature such as being produced by a reliable belief producing mechanism, which if a belief has it will mean that it is more likely than not to be true.  But the externalist does not require that one be able to ascertain that those conditions  obtain.  And so the externalist is not in fact offering a response to the challenge but attempting to sidestep it.  Nothing the externalist offers amounts to showing us that we have more reason than not to think that our beliefs are true.  The conditions he suggests are in fact on the reality side of the traditional appearance/reality gap, and as inaccessible to us as any other fact about the external world if the skeptic's assumptions are right.  So the externalist does not really respond to the challenge at all.

The Harman Chronicles, Part 2

I've been meaning to come back to Harman's defense of the no false lemmas response to the Gettier problem, since it is a particularly, shall we say, convoluted one--though that is  not to say it not interesting.

Harman's initial thought was that in inductive reasoning we infer typically from our evidence E not just what might be thought to be the explanation but rather a statement that the thing that is the explanation explains our evidence. 

It goes like this: I have this evidence re relations between Jones and Fords (he said he owns one, he drives one, etc.): therefore the best explanation of my evidence E is that Jones owns a Ford.  Therefore, Jones owns a Ford.  In Gettier cases, the idea is, the best explanation of my evidence turns out not to be the thing I think it is, and so my reasoning involves a false lemma.

Harman goes on to discussion a number of examples involving what he calls misleading evidence one does not possess. 

1. The library dick (LD) sees Tom take a book from the shelf and put it in his backpack and walk out of the library without checking it out.  He knows Tom by sight.  He reports this to the honor court and leaves.  Then Tom's mother testifies that Tom has a twin Buck who is a bad seed and Tom is a thousand miles away.  Though no one on the court is aware of it, Tom's mother is lying.

The LD has justified true belief but does he know?  Many people say he doesn't, but if that is right, what is the false lemma?  There doesn't see to be one, even if he reasons as Harman says he does: my evidence re Tom and the book is best explained by Tom stealing the book.  Therefore, Tom stole the book.  All of that is true.

2. Don goes to Italy and tells you he be there the whole summer teaching.  While there, he decides he wants you to think he is in San Francisco.  He contacts a confederate in SF to whom he mails letters he writes in Italy which are to be sent to you by his friend in SF which say he has returned, is now in SF, etc., to be mailed at weekly intervals.  You go on vacation, and when you return they are put in the in box on your desk but you have not read them yet.  You have a justified true belief that Don is staying in Italy that summer: the best explanation of the evidence you have regarding his location is that he is in Italy.  You infer from that that he is.  There are it seems no false lemmas in your reasoning.  But do you know?  Many people would say you don't.  If so, then that is a problem for Harman's account.

3. A country's leader is assassinated.  A reporter at the scene files a report which is published by his paper.  You read it and go on vacation immediately afterwards, camping far from civilization, and hear nothing further about it.  But the leader's confederates to keep people from panicking, so they have the newspaper print a retraction and have the television stations broadcast a report that it was not the leader but one of his security detail who was killed by the assassin.  You reason that the best explanation of the newspaper report is that the leader was assassinated, and therefore that he was.  You have a justified true belief, and it seems to rely on no false lemmas.  But many people judge that in this circumstance you do not know.  If that's right, it again seems to be a problem for Harman's defense of the no false lemmas response to the Gettier counterexamples.

So the general form of the problem is that sometimes it seems that evidence that is misleading which is not in your possession can undermine your knowledge.  And if so, there are counterexamples to the JTB analysis of knowledge that the no false lemmas account cannot handle--or so it seems.

Harman says someone might try to argue that we just need to require that to know there be no evidence which if you knew about it would make it unreasonable for you to believe what you do.  But this is too strong a requirement.  There will always be something you could learn that would make it unreasonable to continue to believe what you do.  For there are always very unlikely things that are happening.  Let 'p' express one of these (e.g., that a particular ticket wins the lottery).  Suppose that  what you know is that q.  Then if you learned that q only if p, you'd not be reasonable in holding q, because p is more unlikely than that q explains your evidence.

So he says what we need is to focus only on unpossessed undermining evidence.  But he doesn't say how to draw the line.  He just says there must be such a line, and gives a label to the evidence that is misleading.  

But now what is he to do about defending the no false lemmas account?  What he does is refine his account of broadly inductive reasoning.  He says we infer not that our evidence E is best explained by such and such, but that we infer that E and there is no undermining evidence to this conclusion because of such and such.  Then one infers such and such.  And when there is undermining evidence, this intermediate conclusion or lemma is false, and so the no false lemma's account is seen to handle these further cases. 

Of course, this response is only as good as the correctness of the claim Harman makes about how we actually reason, and his main reason for urging that we reason this way is that he thinks the no false lemmas account is correct.  This makes his defense look ad hoc and question begging.

What lies behind the dream argument?

One of the questions on the midterm is about what picture of the nature of dream experience lies behind the dream argument.  We discussed this in class and my own paper in a way aims to bring it out, but I have reason to think that it went by so fast when the course was so new so long ago that it may not be so clearly fixed in your mind as to be easy to recall.  So here is a refresher on all of that (I’ll indicate when it gets especially relevant to the question about the picture lying behind the dream argument):

Let’s begin with a look at Stroud’s account of the basic structure of the skeptical argument, specifically with respect to skepticism about the external world.  Stroud, I think, provides the materials for two distinct arguments, which I label in my reconstructions argument #1 and argument #2.   Argument #1 (throughout we restrict substitution for ‘p’ to sentences about the external world) goes as follows.

K. For any x, t, x can know that p at t only if, for any q, if at t x knows that p is incompatible with x knowing that q, x can know that q is not the case.

1. For any x, if x dreams that p, then x does not (thereby) know that p.

2. For any x, t, if x is conscious at t, then x can know something about the external world at t only if x can know that x is not (merely) dreaming that p at t (from 1&K).

3. If 2, then for any x, t, x cannot know that x is not dreaming at t.

4. For any x, t, x cannot know anything about the external world at t (from 2&3).

And Argument #2  goes as follows:

1. For any x, if x dreams that p, then x does not (thereby) know that p.

2. For any x, t, experience e, if e is x’s at t, it possible for x to have an experience e′ at t which is a dream experience and which is indistinguishable (for x) from e at t.

3. If 1 & 2, then for any x, t, x cannot know that x is not dreaming at t.

4. For any x, t, if x is conscious at t, then x can know something about the external world at t only if x can know that x is not dreaming at t. [1,3, K]

5. For any x, t, if there is no time at which x is conscious at which x can know something about the external world, then x cannot know anything about the external world.

6. For any x, t, x cannot know anything about the external world at t.  [1-5]

We’ll consider mostly the first, as Stroud, though he supplies the materials for the second, shifts to the first when he concentrates on drawing a conclusion.

But first a remark on the question whether we can know the premises independently of having knowledge of the external world.  Any argument to show that we don’t know anything about a domain of facts F had better not include any premise that is about that domain of facts, or its soundness would show that we had no reason to believe it because we did not know the premises.  Such a skeptical argument would be self-defeating.  You could know the conclusion of it by way of the argument only if it were not sound.  It follows that you cannot know the conclusion by way of the argument.  Why suspect the arguments above of this?  Because of the first premises of each.  For is it clear that the concept of dreaming rules out its being a way of knowing?  Given what we know about dreaming, we take it that someone cannot know something by merely dreaming it.  But is this really implied by the concept of dreaming as opposed to an empirical fact about it?  And if it is part of the concept of dreaming, how and why?  If I fell asleep and dreamt of what was going on around me while my eyes were closed, and indeed I dreamt truly and we could show that there was some sort of causal account of this, would dreaming be inferior to ordinary perception as a method of coming to know about the world?  Suppose that it had always been like that for everyone?   

Turning to the main point, what establishes the crucial assumption (3) in argument #1?  Stroud’s idea is that if one can know that p only if one knows (can know) that not q, and that not q is among the things that ‘p’ stands in for, then (2) entails that to know that one is not dreaming one would have to already have established that one was not dreaming.  But that is impossible.  So (2) by itself entails that one cannot know that one is not dreaming. But wait just a minute!  Where did this ‘already’ come from?  And why couldn’t one already know?  Stroud is reading the ‘only if x knows that not q’ in (2) as ‘only if one can establish first and independently that not q’.  In general, a conditional of the form ‘x can know that p only if x can know that not q’ does not mean ‘x can know that p only if x can know first and independently that not q’. 

For example, I can know that I am conscious only if I can know that I am not dead can be true without it being true that I can know that I am conscious only if I first and independently establish that I am not dead.  My best reason for thinking I am not dead may well be that I am conscious. Is there something special about not dreaming that makes it different from not being dead?  What is driving Stroud is the idea that knowing one is not dreaming would be something that would be determined by a test the results of which one could know and from which one could infer that one was not dreaming.  But if one could dream one performed the test, it might seem as if one would first have to establish that one was not dreaming to use the test.  But so far we do not have enough on the table to force this result.  We could, as Moore does, insist that we do know that we performed the test, and so do know that we are not dreaming, by virtue of the fact that the result of the test is incompatible with our dreaming.  There is no inconsistency in this.  This is shown by reflection on the case of anoxia, a condition in which oxygen deprivation leads someone into a state in which his judgment is severely degraded, so much so that he is not in a position to tell that it is.  Here, while one is suffering from anoxia, there is no way to tell that one is, for even if one performed a test, one would not know whether one had performed it correctly or correctly assessed its results.  But it doesn’t follow that if one is not suffering from anoxia that one cannot tell that one is not, for one can tell by the clarity of one’s thinking.  

What this shows is that the real basis of the problem in reflection on dreams has not yet been brought out (Argument #2 really is motivated these further considerations, and this is all now relevant to question 4 on the exam.)  Later in the chapter, Stroud present a picture of our epistemic position according to the skeptic which makes sense of the claim that we could never employ a test to show that we were not dreaming, namely, the idea that we are trapped behind the veil of perception, that “our sensory experience gives us no basis for believing one thing about the world around us rather than its opposite, but our sensory experience is all we have got to go on” (p. 20).  It is the relation of dreaming to this idea, the Lockean veil of perception (or ideas, in Lockean terminology), that needs to be examined.  This is what is expressed by premises (2) and (3) of argument #2. 

The skeptic invites us conceive of dreams in a certain way to bring out facts about the role of sensory experience as a subjective phenomenon in our knowledge of the world around us.  The idea is that in dreaming we are in the position of having experience which at least can be just like waking experience, and that we can at least be in the position of rationally reflecting on it.  But as dream experience, the idea is, it is not connected in the right way (perhaps because only accidentally with what it is about) to yield knowledge or justification.  We may infer from it various things, but without some additional knowledge about how the experience is connected with what we infer from it, our conclusions can’t be justified.  We are like the man in the room Stroud describes who has access to televisions which purport to show things going on somewhere else, perhaps outside the room, but who has no assurance that they actually reflect how things are outside the room or anywhere else.  Experience as such then comes to look like our primary source of evidence about how the world is but without any guarantee that it is hooked up with the world in the way that we suppose on the basis of that very experience.  And if it is our only source of evidence, then we cannot validate it without relying on it, and that sort of validation is no better than any argument that runs in a circle.  If that were good reasoning, then I could prove anything—so the skeptic charges.

That something like this is in the background, and that dreaming is not the central issue, is shown by the fact that skeptics typically invoke a variety of different kinds of scenarios to show that we don’t know what we think we do, the Evil Demon hypothesis, the brain in a vat hypothesis, the virtual reality suit hypothesis (maybe we didn’t cover that one but you’ll get the idea), and so on.  What they all have in common is that they suggest or presuppose a certain conception of our epistemic position which, if correct, would make it impossible for us to rule out these scenarios which we recognize to be incompatible with our knowing what we take ourselves to know, namely, the idea that all we have to go on is the contents of present consciousness and it is compatible with this that the world (as the counter-possibilities are supposed to bring out) is very different from how we think it to be.  This then is embodied in the two assumptions we finally isolated as underlying the problem of our knowledge of the external world, which we called the assumption of the logical independence of the mind and external world and the epistemic priority of thought to the world (assumptions (L) and (E) in my paper on skepticism). 

Sosa's Criticism of the Doxastic Ascent Argument

This is a note about Sosa's response to the Doxastic Ascent Argument ('doxastic' just means having to do with belief--he might as well have said 'Belief Ascent Argument'), and in particular what role his thesis that the epistemic supervenes on the non-epistemic plays in his response to it.  From feedback, I have concluded I did not do a good enough job of explaining it when we were discussing Sosa.  So here is a fuller discussion which I hope will help with making it clearer what it comes to.

Let me begin by rehearsing the Doxastic Ascent Argument.  I'll spell this out in more detail than Sosa does by reference to Bonjour's discussion of it.

Bonjour says (this is a paraphrase):

1) Justification aims at truth.
2) If justification aims at truth, then to be justified in believing that p it must be the case both that the belief is likely to be true and that one has to have reason to think that it is true or likely to be true.

Given this, he argues as follows:

3) If foundationalism is true, then there are basic beliefs that terminate chains of inferential justification which are themselves justified but not inferentially justified (by definition).
4) If a basic belief is justified for an agent, then it must have some property F that makes it likely to be true (by 1).
5) But a belief is justified for an agent only if the agent who has it has a reason to think that it is likely to be true (by 1&2).
6) Therefore, it is not enough for a basic belief B to be justified for an agent x that B has F; x must be justified in believing that B has F and that a belief's having F makes it likely to be true, so that x has a reason to think that B is likely to be true (5).
7) But if a belief is justified for an agent only if the agent has reasons for thinking it likely to be true, it is not a basic belief (by definition).
8) Therefore, the conditions for something being a basic belief cannot be met, compatibly with 1 and 2 (1,2,4-7)
9) Therefore, there are therefore no basic beliefs (8).
10) Therefore, foundationalism is false (9&3).

Sosa renders this in a more compact form:

"A belief B is foundationally justified for S in virtue of having property F only if S is justified in believing (1) that most at least of his beliefs with property F are true, and (2) that B has property F.  But this means that belief B is not foundational after all, and indeed that the very notion of (empirical) foundational belief is incoherent" (p. 155).

Sosa argues that this does not look convincing from the standpoint of formal foundationalism.  Formal foundationalism, Sosa says, holds that there are general conditions that can be specified which are sufficient for a belief to be justified, and in particular, he says, non-epistemic conditions (substantial foundationalism adds something about what the content of foundational beliefs are).  Formal foundationalism, Sosa says, follows from the thesis that epistemic properties supervene on non-epistemic properties.  What does this come to?

We say that one family of properties S (moral, epistemic, aesthetic, mental, etc.) supervenes on another G (non-moral, non-epistemic, non-aesthetic, non-mental, etc.) if and only if whenever something has a property of the first kind, it has a property of the second kind, and it is necessary that something with a property of the second kind has a property of the first kind.  Another way to put it is this: if anything has, say, the property of being good, then anything that is exactly like it in all non-evaluative respects (respects not having to do with the sort of properties in question, here properties like being good or being bad) will be like it in all evaluative respects.  So if the car you own is a good car, then any duplicate of it will be a good car as well. 

When applied to epistemic properties, properties having to do with knowing, which include knowing, being justified in believing, having reasons to believe, being warranted in believing, having evidence for and so on, the idea is that any duplicate of a person who has a justified belief with respect to her non-epistemic properties will likewise have a justified belief.  So if you are justified in believing that the weather will continue to be cold through the end of spring break, anyone who is exactly like you with respect to perceptual experience, history, patterns of belief, etc., will likewise be justified in believing that.

If this is so, then there is some property of you which can be specified in non-epistemic terms, call it NE, which is sufficient for your having a justified belief, that is, necessarily, if you, or anyone else, has NE, then you have a justified belief. 

This is what Sosa then wants to cite as showing that, if you accept that epistemic properties supervene on non-epistemic properties, foundationalism is true.  For foundationalism is true if you have beliefs which are justified but whose justification does not have to be explained by citing some independently (or antecedently) justified beliefs from which you have (or could have) inferred it.  And if having NE is sufficient to have a justified belief, then since having NE is having a non-epistemic property, your having it makes no reference to your having independently justified beliefs from which you infer the belief in question.  For to cite independently justified beliefs is to cite your having some epistemic properties, and NE is a non-epistemic property.  So if NE is sufficient for a belief of yours to be justified, then that belief is justified but not (solely) by its being inferred from independently justified beliefs.  Hence, it is a foundational belief.

So it seems to follow from accepting that epistemic properties supervene on non-epistemic properties that formal foundationalism is true.  Now, what should we say about the Doxastic Ascent Argument in the light of this?

If  the Doxastic Ascent Argument were sound, then foundationalism would be false, and then so would the thesis that the epistemic supervenes on the non-epistemic.  It is not enough to say simply: supervenience is very plausible, so there must be something wrong with the argument.  One must say what is wrong with it.  Sosa does.  He says the trouble is that it can be true that you can be justified in believing p only if you are justified in believing q, though it is not true that what explains why you are justified in believing p is that you are justified in believing q.  He gives the following example: "Take my belief that I am in pain in a context where it is caused by my being in pain.  The property that my belief then has, of being a self-attribution of pain caused by one's own pain is, let us suppose, a nonepistemic property that yields the justification of any belief that has it.  So my belief that I am in pain is in that context foundationally justified.  Along with my belief that I am in pain, however, there come other beliefs that are equally well justified, such as my belief that someone is in pain.  Thus I am foundationally  justified in believing that I am in pain only if I am justified in believing that someone is in pain" (p. 155).  So here I have a foundationally justified belief, that I am in pain, but I cannot have that justified belief without also being justified in believing that someone is in pain.  But, and this is the crucial point, the fact that to be justified in believing that I am in pain requires that or is linked to my being justified in believing that someone is in pain does not mean that my justification for believing that I am in pain derives from my justification for believing that someone is in pain.  Indeed, in this case, it goes the other way around. 

So the idea is that we can grant to the proponent of the argument that 2 is true and that for that reason (and 1) 5 is true, and so 6, but deny that 7 is true.  For we can allow that a belief can be foundationally justified in the sense of there being a fact about it that makes it justified independently of its being derived from antecedently justified beliefs, while allow that to have that justified belief one must also have reason for it.  For we don't have to say that all of the justification for the belief derives from the having of reasons for it.  That's the key idea: though to be justified in believing that p, you may need reasons for your belief, it can still be that what makes you justified is not (or not just) the fact that you have reasons for it, but some non-epistemic property of the belief, which can be specified without reference to your having reasons for the belief.

(Sosa adds that the Doxastic Ascent Argument would apply also against Coherentism, for the coherentist holds that a belief is justified if it coheres with a coherent comprehensive set of beliefs, but this makes no reference to derivation from antecedently justified beliefs.  His point might be put this way: the Doxastic Ascent Argument is not compatible with supervenience.  Coherentism accepts supervenience.  It holds that the non-epistemic property that justification supervenes on is the property of cohering with most of one's beliefs.  So the Doxastic Ascent Argument is not compatible with Coherentism.)