In%20Defence%20of%20a%20Dogmatist

1
In Defence of a Dogmatist

• Brian Weatherson
• May 2006

2
What is Dogmatism?

• Well be primarily concerned here with dogmatism
  about knowledge, though well also discuss
  dogmatism about justification
• For some propositions p and evidence E, you can
  know p on the basis of E without having
  independent or antecedent justification for E ? p

3
What is Dogmatism?

• Im using evidence here in a very broad sense, so
  it includes anything that justifies beliefs,
  whether or not we are aware of it, or able to
  appeal to it in conscious reasoning
• Well often be interested in more restricted,
  hence stronger, dogmatist claims that say this is
  true for some p, E in a particular area
  (perception, induction)

4
What is Dogmatism?

• Im not going to say much about independent or
  antecedent justification
• But I will insist that when E is all of our
  empirical evidence, a belief in q is justified
  independently or antecdently of E iff q is
  justified, and knowable, a priori
• So the dogmatist has an expansionary view of
  actual justification, and a restrictive view of a
  priori justification

5
The Sceptical Argument

1. If I know I have hands, then Im in a position to
   know that Im not a handless brain in a perfectly
   functioning vat (BIPV)
2. Im not in a position to know that Im not a
   handless BIPV
3. So I dont know I have hands

6
The Sceptical Argument

• Not the strongest sceptical argument
• Premise Two seems just wrong to me
• I can deduce that it is wrong from the fact that
  I know lots of stuff that is incompatible with my
  being a BIPV
• A stronger sceptical challenge is this knowledge
  a priori or a posteriori?

7
The Sceptical Argument

1. If I know I have hands, then Im in a position to
   know that Im not a BIPV
2. Im not in a position to know a priori that Im
   not a handless BIPV
3. Im not in a position to know a posteriori that
   Im not a handless BIPV
4. So I dont know I have hands

8
The Sceptical Argument

• Im assuming here that any knowledge which is not
  a priori is a posteriori
• Both these technical terms are vague, but I
  insist this is a penumbral connection between
  them
• So this argument is valid, and it is far from
  obvious which premise is wrong
• The sceptic has an argument for 2 and 3

9
The Sceptical Argument

• To introduce it, we need one more technical term
• Say a HBIPVE is a handless brain in a perfectly
  functioning vat with the evidence I actually have
• This seems to be the character that is meant to
  worry me in normal sceptical arguments how do I
  know Im not him?

10

1. If I know I have hands, then I can know that Im
   not a HBIPVE, and if I can know (a priori) that
   I have E ? Im not a HBIPVE, I can know (a
   priori) that Im not a HBIPVE
2. It could have turned out that Im a HBIPV
3. If it could have turned out that Im a HBIPV,
   then it could have turned out that Im a HBIPVE
4. If it could have turned out that Im a HBIPVE,
   then I cant know a priori that Im not a HBIPVE
5. If I can know a posteriori on the basis of my
   evidence E that Im not a HBIPVE, then I can know
   a priori that I have E ? Im not a HBIPVE
6. So I cant know that I have hands

11
The Sceptical Argument

• I think this is a fairly interesting argument
  because epistemologists divide up so widely on
  what is wrong with it
• As well see, versions of this argument to
  alternative sceptical conclusions are even more
  dramatic in this respect
• We could divide the premises even further, but we
  wont today

12
Premise One Closure

• Our argument needs a conjunctive closure premise,
  but anyone who accepts the original closure
  premise should like this
• I dont have anything to add to the debates over
  closure, so Ill just note this argument

13
Premise Two Possibility

• This is denied by the so-called semantic
  response to scepticism
• It has some plausibility in the case of external
  world scepticism
• As well see, however, it doesnt seem to
  generalise to other sceptical arguments

14
Premise Three Evidence

• This is denied by externalists about perceptions
• I can see that I have hands, so I have different
  evidence to a HBIPVE
• Williamson denies this premise in all sceptical
  arguments
• This denial seems too strong

15
Premise Four Empiricism

• In earlier work, I said that a rationalist
  response to scepticism is one that says we can
  know a priori something that could have turned
  out to be false
• This is not the same as saying that we can know a
  priori some contingent things
• Some necessary truths could have turned out to be
  false, e.g. water is molecular

16
Premise Four Empiricism

• More importantly, some contingent truths could
  not have turned out to be false
• For example, that water is watery
• Things that could not have turned out false are
  deeply necessary to use Evanss term
• I borrow the phrasing could have turned out
  from Stephen Yablo
• By inclination, Im a rationalist

17
On to Dogmatism

• What is interesting about the dogmatist position,
  as I see it, is that the dogmatist says that
  premise five is the false one
• The dogmatist thinks that the sceptical
  possibilities are possible
• And she thinks that sometimes knowledge outruns
  evidence
• And, most importantly, shes empiricist

18
On to Dogmatism

• The primary argument for dogmatism, as I see it,
  is that for some sceptical arguments, the other
  premises are true
• But the sceptical conclusion is crazy
• Hence dogmatism!
• This requires rejecting a priori knowledge (and
  justification) of things that could have turned
  out false

19
On to Dogmatism

• There is a long tradition, especially in
  Anglophone philosophy, of being very sceptical of
  the a priori
• That would include being sceptical of claims to
  know were not HBIPVE a priori
• I think any good argument against dogmatism
  should respect this attitude
• Personally, I reject dogmatism because I simply
  lack the attitude

20
On to Dogmatism

• If we are to argue for dogmatism this way, we
  need a sceptical argument whose other premises
  are plausible
• External world scepticism doesnt help
• The premise about evidence is rather implausible
  I think
• The inductive sceptic argument is better
• Roger White is, I think, first to explicitly
  discuss dogmatism about induction

21
Inductive Scepticism

• p is some proposition that I can know
  inductively, e.g. that it will be hotter in
  Austin than Anchorage next year
• E is my evidence, and E is the proposition that
  my evidence is E
• hT is the history of the world to date
• So p hT E is a sceptical possibility, where
  the history of the world and my evidence is the
  same, but p is false

22

1. If I know p, then I can know (p hT E), and
   if I can know a priori that E ? (p hT E),
   then I can know a priori that (p hT E).
2. It could have turned out that p hT
3. If it could have turned out that p hT, then it
   could have turned out that p hT E
4. If it could have turned out that p hT E,
   then I cant know a priori that (p hT E)
5. If I can know a posteriori on the basis of my
   evidence E that (p hT E), then I can know
   a priori that E ? (p hT E)
6. So I cant know that p

23
Inductive Scepticism

• The first two premises seem very plausible to me,
  but I dont have anything new to say about them
• What makes the inductive sceptic interesting is
  that premise 3 is plausible
• Williamson denies it because of EK
• But it follows from the (very plausible)
  principle that evidence supervenes on causal
  history, so Ill accept it

24
Inductive Scepticism

• So three premises look good
• But the conclusion is untenable
• So its down to rationalism or dogmatism
• Thats an interesting conclusion already!
• If I had to pick, Id pick rationalism
• But I dont think the existing arguments against
  dogmatism are that good

25
Dogmatism

• Im going to discuss three arguments
• Some of these are primarily arguments against
  dogmatism about justification
• But they also seem to tell against dogmatism
  about knowledge
• One, the bootstrapping argument, is primarily an
  argument against perceptual dogmatism

26
Dominance

• Roger White discusses the following kind of
  objection (slightly modified)
• The dogmatist says I cant know, or even
  justifiably believe, a priori that(p hT
  E)
• But a priori I can run a dominance argument that
  has as its conclusion (p hT E)

27
Dominance

• Either I will get evidence E or I wont
• If I do, then Ill be justified in believing p
• Ill then be able to infer (p hT E)
• If I dont, then I wont even have E so Ill be
  able to infer (p hT E)
• So I can know a priori that either way Ill be
  justified in believing (p hT E)
• So Im justified a priori in believing this

28
Dominance

• The problem with this argument is that it slides
  between facts about evidence I have, and facts
  about what Im justified in believing about my
  evidence
• It is possible that Ill not have evidence E, but
  not be justified in believing that I dont have
  evidence E, let alone knowing this
• So the second half of the disjunctive argument
  seems to fail

29
Bootstrapping

• This is primarily an argument against a certain
  kind of perceptual dogmatism
• In Pryors version, the dogmatist endorses the
  following principle
• If it appears to the agent that p, then in the
  absence of defeaters, she is justified in
  believing that p, even if she isnt justified in
  advance in believing that appearances are reliable

30
Bootstrapping

• This seems to run into a version of Stewart
  Cohens easy knowledge argument
• Cohen is primarily concerned with attacking
  reliabilism
• But the objection, if sound, seems to tell
  against the dogmatist as well

31

• Imagine my 7 year old son asking me if my
  color-vision is reliable. I say, Lets check it
  out. I set up a slide show in which the screen
  will change colors every few seconds. I observe,
  That screen is red and I believe that it is red.
  Got it right that time. Now its blue and, look
  at that, I believe its blue. Two for two I
  trust that no one thinks that whereas I
  previously did not have any evidence for the
  reliability of my color vision, I am now actually
  acquiring evidence for the reliability of my
  color vision. But if Reliabilism were true,
  thats exactly what my situation would be.

32
Bootstrapping

• The argument is directly targeted at the
  dogmatist about perception
• But we can imagine a similar argument against the
  inductive dogmatist
• I sit down and make a number of inductive
  inferences
• I then conclude, from the truth of those
  conclusions, that Im a reliable inductor

33
Bootstrapping

• This reasoning is bad
• Some say dogmatists must endorse it
• But in fact there are a number of reasons to
  reject it even given dogmatism
• Worries about projection
• Worries about randomness of sample
• Worries about radiance
• Radiance and randomness

34
Projection

• The bad inference has the form, all Fs so far
  have been Gs, so generally Fs are Gs
• But perhaps being an F isnt a natural kind
• I shouldnt, for instance, conclude from the
  reliability of my visual perception that I have
  an accurate sense of smell
• So the argument only works if Im concluding that
  a single method is reliable

35
Projection

• Arguably, I need to know that Im evaluating a
  single method
• If for all I know, I use many perceptual, even
  visual perceptual, and inductive methods, this is
  a bad inference
• Since I cant tell from the armchair what is a
  single method, I cant learn about my own
  reliability from the armchair

36
Randomness

• The dogmatist only says that we can conclude that
  things are as they appear in cases where there
  arent defeaters
• That suggests that the cases where we can infer
  from appearances to reality wont be a random
  sample of all cases
• But we need a random sample to do the enumerative
  inductive inference being considered

37
Radiance

• Those two responses are far from telling
• I think the big worry with the argument is that
  it makes a radiance assumption
• The idea of radiance is based on Williamsons
  discussion of luminosity
• A property is luminous iff whenever it is
  instantiated, we are in a position to know that
  it is instantiated

38
Radiance

• A property is radiant iff whenever it is
  instantiated, we can justifiably believe that it
  is instantiated
• Williamson argues against luminosity
• His arguments dont carry over to arguments
  against radiance
• Thats because the arguments centrally appeal to
  the factivity of knowledge

39
Radiance

• Still, there is good reason to doubt in many
  cases that properties are radiant
• In particular, it is possible that we could have
  evidence, or appearances, or beliefs, without so
  much as being justified in believing we have them
• Perhaps ideally our evidence, beliefs or
  appearances would be radiant, but we cant assume
  things are ideal

40
Radiance

• If appearances arent radiant, then Cohens
  version of the argument fails
• When I get a red appearance, Im justified in
  believing Im looking at something red
• But it doesnt follow that Im justified in
  believing things are as they appear
• Because I might not be justified in believing
  that the thing appears red

41
Radiance Randomness

• It might be objected that usually appearances and
  beliefs are radiant
• But unless they are always radiant, we wont know
  the cases we know about form a random sample
• Further, unless the property of not being
  appeared to redly is radiant, for all we know
  there may be many cases where we have an
  inaccurate appearance of red

42
Radiance Randomness

• We also get randomness problems if we arent
  always aware of our appearances
• If sometimes I have a red appearance, but dont
  believe I have one, then it might be that the
  cases where I believe I have a red appearance are
  a non-random part of the sample
• So the argument needs a self-awareness assumption

43
Radiance Bootstrapping

• But the big problem isnt that the few actual
  failures of radiance threaten the randomness of
  the sample
• Rather, it is that the assumptions needed to get
  the anti-dogmatist argument going are
  inconsistent
• These are that appearances are radiant, that we
  are aware of appearances and that bootstrapping
  arguments are bad

44
Radiance Bootstrapping

• Assume that appearances are radiant
• So if Im appeared to redly, I believe that Im
  appeared to redly, and this belief is justified
• So I can reason as followsI believe thats a
  red appearanceAnd it is a red appearanceIntrospe
  ction works again!

45
Radiance Bootstrapping

• Given radiance and self-awareness, I can do this
  over and over again
• Eventually, Ill have a large sample of cases
  where my introspective beliefs are accurate
• So I can conclude on the basis of this little
  reflection that my introspective beliefs are
  generally accurate

46
Radiance Bootstrapping

• But this is crazy
• It takes serious psychology to know that my
  introspective beliefs are accurate
• A little armchair reasoning like this wont cut
  it
• This reasoning is as bad as the reasoning that
  Cohen parodies
• Indeed, it is just that reasoning, with a
  different target

47
Radiance Bootstrapping

• If we really want to block all bootstrapping
  arguments, we have to reject not just dogmatism,
  but also the radiance and self-awareness
  assumptions
• Otherwise an introspective bootstrapping argument
  will be licensed
• But dogmatism without radiance and self-awareness
  doesnt allow bootstrapping, so dogmatism cant
  be faulted here

48
Bayesian Objection

• Our final argument comes from principles of
  Bayesian epistemology
• The dogmatist thinks that when we get E, our
  credence in E ? p can go up
• Or at least, she thinks that we can go from not
  being justified in believing E ? p to being
  justified in believing it
• But this cant happen in Bayesian models

49
Bayesian Objection

• Well work through a simple version of this
  argument
• Let A be the proposition it appears to me that
  there is a hand
• H is the proposition there is a hand
• F is the proposition it falsely appears to me
  that there is a hand, i.e. A ? ?H.
• And Pr is our prior probability

50

1. Pr(A) lt 1 premise
2. Pr(F) gt 0 premise
3. Pr(F A) Pr(A) Pr(F ? A) probability defn
4. Pr(F ? A) Pr(F) from definition of F
5. Pr(F A) Pr(A) Pr(F) from 3, 4
6. Pr(F A) gt Pr(F) from 1, 2, 5
7. Pr(F A) 1 - Pr(F A) probability theorem
8. Pr(F) 1 - Pr(F) probability theorem
9. Pr(F A) lt Pr(F) from 6, 7, 8

51
Bayesian Objection

• All that is mathematically sound
• But it usually gets a philosophical gloss
• The last line is read as saying that the
  probability of F goes down when we get evidence
  A
• This relies on an undefended claim, namely that
  we should update credences by conditionalisation

52
Bayesian Objection

• If we grant conditionalisation, things look bad
  for the dogmatist
• The dogmatist says that we can know, and
  justifiably believe F when our evidence is A,
  but not before
• But it seems plausible that evidence that
  justifies a belief shouldnt make its credence go
  down

53
Bayesian Objection

• Can the use of conditionalisation here be
  defended?
• I think it cant
• Conditionalisation usually is the right approach
  to updating
• But it isnt always the right approach in cases
  where uncertainty is relevant

54
Bayesian Objection

• Can the use of conditionalisation here be
  defended?
• I think it cant
• Conditionalisation usually is the right approach
  to updating
• But it isnt always the right approach in cases
  where uncertainty is relevant

55
Bayesian Objection

• Game plan for the last few slides
• Distinguish risk from uncertainty
• Describe a model for representing uncertainty
• Describe a new approach to updating in that
  model, one that blocks this argument
• Give a dogmatist defence of that approach

56

• By uncertain knowledge, let me explain, I do
  not mean merely to distinguish what is known for
  certain from what is only probable. The game of
  roulette is not subject, in this sense, to
  uncertainty nor is the prospect of a Victory
  bond being drawn. Or, again, the expectation of
  life is only slightly uncertain. Even the weather
  is only moderately uncertain. The sense in which
  I am using the term is that in which the prospect
  of a European war is uncertain, or the price of
  copper and the rate of interest twenty years
  hence, or the obsolescence of a new invention, or
  the position of private wealth owners in the
  social system in 1970. About these matters there
  is no scientific basis on which to form any
  calculable probability whatever. We simply do not
  know. Nevertheless, the necessity for action and
  decision compels us as practical men to do our
  best to overlook this awkward fact and to behave
  exactly as we should if we had behind us a good
  Benthamite calculation of a series of prospective
  advantages and disadvantages, each multiplied by
  its appropriate probability, waiting to be
  summed. (Keynes 1937 114-5)

57
Representing Uncertainty

• Traditional Bayesian models didnt have an easy
  way to represent uncertainty
• There is a common way this is done
• Instead of representing credal states by a single
  probability function, we represent them by sets
  of probability functions
• Say S is the agents representor iff it is the
  set of functions representing her

58
Representing Uncertainty

• The idea is that more uncertain p is, the larger
  this set is x For some Pr in S, Pr(p) x
• Now we can represent uncertainty in a distinct
  way from risk

59
Updating Uncertainty

• This kind of model is widely used in the
  literature
• It is usually assumed, without much argument,
  that we should update S on evidence E by
  conditionalisation
• That is, the new representor is derived from the
  old one by conditionalising each element

60
Updating Uncertainty

• That is what I want to reject
• I think the agent might sometimes use her new
  evidence to realise that she need not have been
  as uncertain as she was
• So well get from the old representor to the new
  one by first cutting some functions from the set,
  and second conditionalising the remaining members

61
Updating Uncertainty

• The benefit of this approach is that it lets us
  say something quite attractive about the a priori
  rational credal state
• It is somewhat intuitively that a priori we
  should treat all possibilities symmetrically
• It is notoriously hard to implement this
  intuition consistently in a traditional Bayesian
  theory
• But we can if we allow for uncertainty

62
Updating Uncertainty

• The a priori rational credal state is represented
  by the set of all analytically coherent
  probability functions
• A probability function is analytically coherent
  iff it assigns probability 0 to anything that
  could not turn out to be true and conforms to the
  Principal Principle

63
Updating Uncertainty

• If we accept the usual conditionalisation
  approach to updating, we cannot accept that this
  is a rational a priori state
• Thats because updating this state by
  conditionalisation does not allow for ampliative
  learning
• But if updating involves culling the set, then we
  can learn things that go beyond our evidence

64
Updating Uncertainty

• And dogmatists should be very pleased with this
  idea for representing the a priori
• Our discussion of dogmatism started with the idea
  that we shouldnt allow for a priori
  justification of substantive beliefs
• Any initial credal state other than this one
  seems to adopt some substantive claims, at least
  about relative probability

65
Updating Uncertainty

• If when the agent sees that she appears to have
  hands, she culls as well as conditionalises her
  representor, it wont follow that her credence in
  F will lower
• So that evidence might ground her belief that F
• In short the Bayesian objection to dogmatism
  seems to show up a problem for the Bayesian, not
  the dogmatist