Craig E Knowledge & The State of Nature An Essay in Conceptual Synthesis 1991
Craig E Knowledge & The State of Nature An Essay in Conceptual Synthesis 1991
Craig E Knowledge & The State of Nature An Essay in Conceptual Synthesis 1991
THE STATE OF N A T U R E
This page intentionally left blank
Knowledge and
the State of Nature
An Essay in Conceptual Synthesis
EDWARD CRAIG
C L A R E N D O N PRESS. O X F O R D
This book has been printed digitally and produced in a standard specification
in order to ensure its continuing availability
OXFORD
UNIVERSITY PRESS
the rationale behind the stretch. With luck it might also reveal the
sources of the indeterminacy or elasticity which dogged the attempts
to answer, or even to ask, the first familiar question.
It can at least be said for this way of creeping up on the concept
of knowledge that we are asking a question that can reasonably
be expected to have an answer. One doesnt have to commit oneself
to a great deal of epistemology or semantic theory, as the standard
approach evidently does, to presume that there is such a thing as
the point of this concept, what it does for us, the role it plays in
our lives. And if this is so, one way to find out must be to form
some hypothesis about it, try to work out how a concept custom-
designed for that role would look, and then see to what extent it
matches our everyday practice with the concept of knowledge as
actually found. We may then have to revise or supplement the
hypothesis from which we began, but that will hardly be surprising,
and certainly no cause for instant despair.
Whilst agreeing that we may expect the concept to serve some
purpose, however, we might doubt whether the consideration of
its purpose will necessarily lead to anything like an analysis, or
to anything that can be measured against the intuitive extension.
We might doubt, in fact, whether it will necessarily lead anywhere
interesting at all. Every language, an objector might reason, has
a word for water. And having that word has an important purpose,
namely, to make it possible to talk about water, something which
every community has an obvious need to be able to talk about.
But no a priori thought about that purpose will bring us any closer
to an analysis of the concept of water, even if the notion of an analysis
be very generously interpreted. Couldn't it just be that knowledge,
like water, is common and important stuff, and that the purpose
of the concept is simply to enable us to think and talk about it?
Though I would be hard put to it to argue the point, I am fairly
confident that this is mistaken. Knowledge is not a given phenome-
non, but something that we delineate by operating with a concept
which we create in answer to certain needs, or in pursuit of certain
ideals. The concept of water, on the other hand, is determined by
the nature of water itself and our experience of it. But probably
a better response here, at any rate a less dogmatic one, is the proof
of the pudding: if some hypothetical but plausible purpose does
issue in conditions of application showing a close fit to the intuitive
extension of know, and does fit well with a variety of facts about
4 SECT. I
1
Plato, 97 ff. (Works are specified by the author's name and, where appropriate,
a number, as listed under References, pp. 168-9.)
8 SECT. I
is not the only, and perhaps not even the chief, thing that we may
want in a true belief.
Even if we confine ourselves to recent literature of the analytic
school the kind of approach I shall recommend is not without prece-
dent. David Pears has suggested something of the kind:
It often happens, as Aristotle saw, that understanding how a thing has
developed from primitive beginnings helps us to understand it in its deve-
loped form. Certainly this is true of knowledge. In this case the primitive
beginning is the ability to make a discriminating response to circumstances.
In the early stages the response will be a piece of overt behaviour. Later,
it may be internalised, and stored for future use. 2
This point, which Pears did not go on to press, may well be useful,
but its primary use is most likely to be that of shedding light on
the concept of belief, and maybe, if we add the idea of the overt
behaviour, when it arrives, being in some sense successful, that of
true belief. It will not in itself bring us very close to the concept
of knowledge as we have it, unless an adequate account of that con-
cept turns out (somewhat against the early odds, most of us will
feel) not to need, beyond true belief, any further apparatus. So we
can't rest content with Pears' remarks, though we can, and should,
take his hint.
I have elsewhere referred to this project as the practical explica-
tion of knowledge. The notion of an explication came into currency,
I believe, through Carnap. To explicate a concept was not exactly
to analyse it; it was to construct a new version of it satisfying certain
standards, with the proviso that to count as a new version of that
concept it had to emerge with many of its principal features intact.
The procedure suggested here is analogous; but it is concerned with
the practical rather than (as in Carnap) the theoretical aspect of the
concept. Hence the title of the earlier paper. But let it not mislead:
Carnaps intentions were normative, the establishment of the con-
cepts fit to form the rational basis of the unified science, whereas
mine are the more purely theoretical ones of shedding light on the
nature and origins of present practice. In that respect at least I shall
follow the traditional approach.
Whilst speaking of precedents, we should notice that my project
can claim membership of another tradition, one which spreads itself
2
D. F. Pears. ' E. J. Craig (1).
SECT. I 9
' C. Radford.
S E C T . II 13
1 2
R. Nozick, (2), pp. 172-8. See also F. Dretske. D. K. Lewis.
20 SECT. I l l
will therefore lead him to hope for an informant who will give him
the truth about/; whichever of all these possibilities is realised. Which
is to say, if you like the jargon, that he wants an informant who
will give him the right answer in a range of possible worlds.
Is it possible that we are not being quite rigorous enough here,
and claiming more than in fact follows from the inquirer's position?
The reply can be made that he is only interested in the actual world
it is just that he doesn't know, out of all those possible worlds,
which one it is. What I need, he might say, is for the informant
to be right in whichever of these possible worlds is the actual one;
the rest can go hangI shall have my true belief.
I doubt whether we can accept that, however. The trouble is
that it leads to no strategy on the part of the inquirer. Imagine
someone about to go out, and wishing to stay dry, but not knowing
whether it will be raining or not, so that he faces one 'possible
world' in which it will, and one in which it won't. Can he say
T only want to keep dry in the actual world; I'm not bothered
about whether I would have kept dry in whichever of those worlds
turns out to be merely possible'? He can say it, and in a sense no
doubt it is true. But if he proposes to do something about it he
will either have to guess which possibility will be realised or take
such action as will work in either case, even though that means
planning for at least one eventuality which will turn out to have
been merely possible. And the same applies to our inquirer: he must
either guess which of the possible worlds he is actually in, or he
must adopt a strategy which works in many merely possible worlds
as well as the actual one.
Some may now have got the impression that I am about to award
a prize to Professor Nozick; for haven't we said, in effect, that the
search for the desired informant will naturally become a search for
someone who is precisely a good 'tracker' of p, someone whose
belief as to whether p is true in all close possible worlds? But I
am not, for two main reasons. For one thing, Nozick does not
select the right range of possible worlds, if by that we mean one
that coincides with the range of interest of the inquirer. For another,
Nozick's tracking condition doesn't reflect the epistemic demands
that the inquirer is bound to make. I shall return to the second
point later, having dealt in some detail with the first.
The first point, then, is that Nozick's range of possible worlds
is not the one which will concern the inquirer; it overlaps with
SECT. I l l 21
it, of course, but it is wider. And just this extra width causes trouble
for his account, seen (in the way he surely meant it) as an attempt
in traditional style to match an intension to the intuitive extension
of 'know'. The practically explicated concept, on the other hand,
picks the right range of possible worlds, and illuminates the reason
for doing so. Here we see it scoring its first points against a presti-
gious attempt at an analysis of the more familiar kind.
Nozick recommends assessing his two counterfactuals by refer-
ence to what is the case in all close possible worlds. Fortunately
it is not necessary for our purposes to specify exactly what 'close'
means here. Roughly speaking, two worlds are said to be close to
each other if they differ only slightly, distant if they differ radically,
and this is accurate enough for the point I now wish to make, which
is that our inquirer will not be interested in all close possible worlds,
but only in those that he cannot rule out as being merely possible,
or non-actual. Suppose he is considering the credentials of a potential
informant whom he can see to be wearing a red shirt. There is a
close possible world (and it surely is close, if the concept of closeness
is to be capable of any work whatever) in which that same person
is wearing a blue shirt; but since the inquirer will be perfectly satisfied
that this world, although both possible and close, is not the actual
world, he will have no interest in it. It is of no concern to him
whether the potential informant would hold a true belief on the
question at issue were he wearing a blue shirt - he already knows
that he isn't.
(Here we should watch out for a point which could cause confu-
sion. If you really thought that I would quite likely be wrong about
p were I wearing a blue shirt, you would be doubtful about employ-
ing me as informant even if you could see that I am wearing a red
one. But that is not be because you are taking the 'blue shirt' world
into account as one in which, for your purposes, I need to be right.
Rather it happens because, your beliefs about the effect that the
colour of one's shirt has on capacity as an informant being what
they are, for almost any value of p the belief that I wouldn't be
good on p were I wearing a blue shirt will make you wonder how
then I can be good onp when wearing a red one.)
If we use the expression 'open possibility' to mean a possibility
which so far as the inquirer knows might be actual, we may put
the point more generally in these terms: the inquirer will want the
informant to be a good tracker of p in worlds that are both close
22 SECT. I l l
and open possibilities for him, the inquirer. This will be a narrower
class of possible worlds than that which Nozick uses for the assess-
ment of his counterfactual conditions. The possibility therefore arises
that Nozick's analysis will rule out some cases of (putative) know-
ledge where an analysis more closely tied to the situation of the
inquirer would allow them. This would happen if tracking held in
all the possible worlds which the inquirer will have an interest in,
but failed for at least one of the wider class of possible worlds which
Nozick would have us take into account.
With this thought in mind, let us look at a case which causes
Nozick a good deal of trouble: the case of the Great Bank Robbery. 3
Jesse James, the reader will recall, is riding away from the
scene of the crime with his scarf tied round his face just below the
eyes in the approved manner. The mask slips, and a bystander,
who has studied the 'wanted' posters, recognises him. The bystander
now knows, surely, that it was James who robbed the bank. But
Nozick has a problem: there is a possible world, and a 'close' one,
in which James' mask didn't slip, or didn't slip until he was already
past the bystander; and in that world the bystander wouldn't believe
that James robbed the bank, although it would still be true that
he did. So Nozick's condition (4)the second of the two counterfac-
tuals - is not satisfied, and he is threatened with having to say that
the bystander doesn't know that it was James, even though the
mask did slip. So his analysis looks like ruling out something which
is as good a case of knowledge as one could wish for.
In the face of this problem Nozick resorts to fudging. He recalls
his previous stipulation that the method by which the knowledge
is acquired be held constant, so that close possible worlds in which
a different method would be used to arrive at the belief that p from
that used in actuality are not to figure in the assessment of the coun-
terfactuals. And he then implies that, had James mask not slipped,
the bystander would have been employing a different method, an
implication which violates the distinction between the method and
the evidence obtained by the use of it (as Graeme Forbes4 has
put it). A natural response to the case of the Great Bank Robbery,
I believe, is that such a manoeuvre ought to be unnecessary; for
what would have happened if the mask had not slipped is wholly
3 4
' R. Nozick, (2), p. 193. G. Forbes, pp. 47-8.
SECT. I l l 23
detailed answer to the query: and what property is that? There could
be almost as many different answers as there are types of thing that
the inquirer might want to know about. If it were not so, choosing
our sources of information would be a very much simpler business
than it is. But even the very general formulation that we can give
is enough to enable us to link the constructed concept to the most
prominent recent attempts at an analysis of the concept of know-
ledge. Let us begin by looking back to the counterfactual analysis
of Nozick and Dretske. It adds as the further condition the two
clauses:
2 3 4
D.Hume, p. 76. A. Goldman, (1), p. 362. Ibid., p. 357.
SECT. IV 29
given that it is so, quite a lot of (small) changes could have been
made to the world as it is without dislodging my belief. So the
truth of the theorem and my belief that it is true satisfy the two
counterfactuals. But to say that they are causally related embroils
us in all manner of ontological difficulties which surely go far beyond
anything we have said in merely agreeing that the counterfactuals
hold: if the truth of a mathematical formula can be the cause of
a belief must it not designate some kind of state of affairs, something
of the sort which could enter into causal relations? And then what
sort of state-of-affairs could it be, and how could its influence on
our minds be a causal one? This is well-trampled ground, but fortu-
nately there is no need for me to trample it still more. My point
is simply that removing all this from consideration, as Goldman
did in his early paper on the Causal Theory, helps to bring causality
and tracking into near-congruence by excluding a principal area in
which they may very plausibly be thought to diverge.
The next step is to see how close we now are to another popular
formula, that which declares knowledge to be true belief attained
by a reliable method. For some writers, indeed, there seems to be
no distance at all between this formula and one or other of those
we have looked at already. Ramsey5 once defined knowledge in
terms of the 'reliable method' formula, but then quickly said that
what he meant by a reliable method was, or at any rate included,
causal connection with the fact believed in. More recently, David
Armstrong6 made similar opening moves, and then characterised
reliability in terms of satisfaction of the Nozick-Dretske tracking
conditionals. But even without adopting so direct or stipulative a
stance we can easily establish contact between the account in terms
of reliability of method and those in terms of tracking and causal
relationships.
One's first thought might be that if a certain belief satisfies the
tracking conditionals, and if in addition there is such a thing as
the method by which it was acquired, then that method must be
a reliable method, in the sense of one which will produce the belief
that p if p is the case, and will not produce it otherwise. But the
first thought needs a supplement: it must be that the tracking con-
ditionals are satisfied because the belief was acquired in that way.
1 6
F. P. Ramsey. D. M. Armstrong, ch. 12, esp. p. 169.
30 SECT. IV
Were that not so, then although the belief might be called reliable,
there would be nothing particularly reliable about the method.
Now for the vast majority of human beliefs, if not indeed for
all, there is such a thing as the method of acquisition. We do not
just have beliefs, we come by them in specific ways. Perhaps in
creatures whose mental life is much more primitive and much more
instinctual than ours there are beliefs which they have not in any
real sense acquired; perhaps even some very few human beliefs are
like that too. But be that as it may, the vast majority of human
beliefs are acquired and something, at least, can be said in description
of the process by which their subject acquired them.
It does not strictly follow from this that when we trust a belief
we trust it because it was acquired by such and such a methodone
with a long and successful career behind it. But although that does
not strictly follow it is nevertheless a fact that what convinces us,
in an enormous number of cases, to place confidence in a certain
belief, is a further belief about the way in which its ownermaybe
another, maybe ourselfcame to hold it. And even in those cases
in which what did the trick was better describable as something
about what the belief-holder was than as how he came to believe
ita teacher, perhaps, or a policeman, or a doctor-very often some
view about how such people come to hold their beliefs on the appro-
priate subject-matter underlies and underpins our confidence in their
opinions. For all such cases, therefore, the idea that what we are
looking for is true belief reliably acquired will fit the bill; and since
they form so large a proportion of all the cases we are likely to
consider it should not be felt surprising if that formula has seemed
to offer an acceptable definition of knowledge in general. For in
virtually every case in which we take the counterfactual conditions
to be met, we also take it that the reason why they are met lies
in the nature of the method used.
It seems then that where there is tracking in consequence of the
method used we have a reliable method. But the relationship between
reliabilism and the tracking and causal analyses would fall apart if
the converse failed, in other words if a method could be reliable
without producing beliefs that satisfied the counterfactuals. That
would be so if a method which generated true beliefs just in all
actual cases could thereby qualify as reliable. And so long as we
do not take into account the point of the concept of knowledge
we might think that reliability could be so defined. But a reliable
SECT. IV 31
ofbut the point that follows doesn't need it to mean any more
than that.)
In the case of a (true) belief acquired perceptually, for example,
S will normally be aware of having had such and such visual (or
other) experiences. It will be true that he has indeed had them,
he will believe that he has had them, that he has had them will
raise the likelihood that his belief about the state of his environment
is a true one, and he will believe that it raises the likelihood. So
the above conditions (l)-(4) will be met. They will also be met
if we consider the standard cases of coming by beliefs about math-
ematics. There will be some proposition q to the effect that 5 has
been through a process of proof or calculation that terminated in
p, S will believe it, its truth will raise the likelihood that p is true,
and S will believe that as well. The position is the same if we consider
inferentially acquired beliefs of an a posteriori kind. Suppose I
believe, truly and with some confidence, that there has been a dog
in the garden. I will normally have come to this belief via the belief
that there are paw-prints in the flower-bed or something similar.
This something, whatever it is, will nearly always be true, since
the occasions on which I will come to a true belief as the result
of a false belief will be very much in the minority. The first two
conditions are, therefore, satisfied. So are the others, for the truth
of q (in this example: that there is a paw-print) significantly raises
the likelihood of p and I will virtually always believe it to do so.
Again, suppose that I believe that there is a dog in the garden,
this time because I see it. Whether we describe my reason (the propo-
sition q of conditions (l)-(4) above) as 'my having seen the dog'
or rather as 'my having had a visual experience as of a dog' is a
question we can postponeindefinitely, if we so wish. It remains
true that (1) some such thing has happened, (2) in very nearly all
cases I will be aware that it has happened, (3) its having happened
increases the likelihood that there is a dog in the garden, and finally
(4) I will be aware that it increases the likelihood that there is a
dog in the garden.
If, then, 'having a reason' is captured by (l}-(4), normal human
consciousness (and self-consciousness) of the processes leading to
their beliefs will ensure that they have reasons for nearly all those
beliefs with respect to which they are reliable informantsor which
they 'know' by the lights of the constructed concept; a fortiori i the
same holds if 'having a reason' is sufficiently covered by (l)-(3).
SECT. IV 33
Like true belief that tracks the facts, or true belief causally connected
to the facts, or true belief attained by a reliable method, true belief
with a reason is not at all far off the mark. Once again it becomes
understandable that philosophers, especially philosophers interested
in emphasising the role of rationality in human life, should have
presented the concept of knowledge in this dress.
It may be worth pausing at this stage to draw a certain parallel.
There is a similarity between some of these thoughts and a proposal
once made by H. P. Grice7 in connection with another question.
Grice was roasting one of the old chestnuts of philosophical logic,
the relationship between 'If. . . then. . .' as it occurs in everyday
speech, and the material implication (p*q) of formal truth-
functional logic. Can one equate them? There are, of course, excel-
lent and over-rehearsed reasons for thinking that one cannot: the
material implication comes out true, for instance, if any pair of pro-
positions, however remote and unconnected, are substituted for p
and q, provided just that the first of them is false, or both are false,
or both true; whereas 'If I am writing about Grice, then the Moon
is about 240,000 miles from the Earth' is surely not trueat any
rate it is extremely weird and would never be said. But if they meant
the same then the second would have to be true if the first wasso
they do not mean the same. Grice undertook to show that this
well-worn argument was indecisive. He formulated certain general
principles allegedly governing all conversation, and so having
nothing in particular to do with the meaning of 'If.. . then.. .'. He
then proposed that an everyday conditional, 'If ... then .. .', means
no more than the corresponding material conditional, and that those
features of its use which suggest otherwise do not result from its
meaning but from the operation of these general principles.
My procedure with respect to the concept of knowledge is to
some extent analogous. I allow the situation and needs of the inquirer
to generate a concept, or, as it might be better to say, a description
of a prototypical instance, roughly: 'true belief plus some property
indicative of true belief, and then suggest, in effect, that we take
this to be the core of the concept of knowledge, as Grice suggested
that we regard the material implication as the core meaning of the
expression 'If ... then . . .'. We are then confronted with a number
7
H. P. Grice. (Note that this reference offers mainly material about the general
principles governing conversation, not about the specific issue of conditionals.)
34 SECT. IV
of analyses, none of them, evidently, very far from the mark; and
I make no attempt to deny that there is some justification for the
additions they each make to the minimal concept, much as Grice
accepted that the actual use of the everyday conditional does differ
from that which formal logic lays down for material implication,
and that an attempt to reflect these differences in a different account
of its meaning consequently has some foundation. But these very
genuine differences, Grice thought, were not best described as a
difference in the meaning of the expressions; and in much the same
way I suggest that the differences between the 'minimal' analysis
and the competing 'standard' analyses are not best seen as the out-
come of the concept of knowledge. What makes them seem plausible
is not the concept of knowledge, but certain very general beliefs
which we all hold. These are, in particular, beliefs about the extent
to which the world is a system of causally inter-related states, more
specifically beliefs about the extent to which belief-states are them-
selves the end-product of a causal process; the belief that for nearly
all human beliefs, there is such a thing as the method by which
they were acquired; and the fact that human beings are usually con-
scious of certain stages of the processes by which they arrive at
beliefs. The effect is that when the conditions laid down in the mini-
mal concept are satisfied, it will almost always be believed that vari-
ous further conditions are satisfied too: in particular, those of Nozick
and Dretske, those of Goldman's causal theory, those of the reliabi-
list in so far as they genuinely differ from these; and finally (to
reverse the chronology) those of the traditional analysis of know-
ledge as true belief with a reason.
V
We now need to pick up a hint from the last section, and take
a closer look at a distinction which I have so far smudged over,
though it is an important one. There are informants, and there are
sources of information. Or, to arrange the terminology differently,
among the various sources of information there are on the one hand
informants who give information; and on the other there are states
of affairs, some of which involve states of human beings and their
behaviour, which have evidential value: information can be gleaned
from them. Roughly, the distinction is that between a person's telling
me something and my being able to tell something from observation
of him. Of course, in this evidential sense it is far from being only
persons that are sources of information. A tree is a source of infor-
mation on its age, since one can tell its age by counting the growth-
rings ; in fact, anything is a source of information on a great variety
of mattersgiven a suitably equipped observer who knows which
inferences to draw.
In general terms it can be said that the concept of knowledge,
as we operate it in everyday practice, is tied to informants rather
than to sources of information in the sense just (approximately) char-
acterised. We don't speak, even metaphorically, of a tree as knowing
how old it is; and if Fred enters dripping wet, although he may
well know that it is raining, we don't say that he knows it just
because we can tell it by looking at him.
Just to draw this distinction isn't enough; we need some idea
as to why it should have been felt worth drawing, otherwise we
shall find that our 'practical explication' leaves us still a very long
way from the concept of knowledge as we find it in use. Why bother
to build into the vocabulary a distinction, amongst sources of infor-
mation, between 'informants' who 'know' on the one hand and other
sources which do not know on the other? The answer cannot lie
in the distinction between agents and non-agents, for the second
example of the previous paragraph proves that an agent can be a
36 SECT. V
2
Above, Section II, last para.
SECT. V 39
knew that this music was by Bartok even though he got the answer
right, and at the same time enabled the inquirer to conclude confi-
dently that it was indeed Bartok he had been listening to. The ascrip-
tion of knowledge, in other words, seems to stop just where specific
knowledge on the part of the inquirer is needed if he is to make
use of the 'informant'; and this coincides with the distinction we
intuitively make between using someone as an informant and using
their behaviour (including their utterances) as a source of infor-
mation, or more precisely as evidence. Notice that in case B(ii) our
inquirer would have got the information he wanted (that the music
was by Mozart) had he been told that the music was by Haydnso
long as he believed that the 'informant' would not at any rate take
Bartok's music for Haydn's. Once we have the crucial distinction
in mind we see that the informant is not clearly functioning as an
informant at all.
Ascriptions of knowledge may, as we shall later see, be in a certain
sense relativised to an inquirer's concerns. But they are not relati-
vised to any special knowledge of the circumstances surrounding
his inquiry.
Alvin Goldman3 has drawn attention to facts about the phenome-
nology of 'know' which are surely germane here. Henry, driving
in the countryside, sees a barn, and we may imagine circumstances
to be such that we unhesitatingly agree that he knows that that
is what it is. (Circumstances are, in other words, pretty much nor-
mal; and nobody has recently been putting any of the sceptic's
favourite possibilities into our heads.) Goldman now asks us to sup-
pose that we hear that, unknown to Henry, this stretch of country
is full of papier mache barn-facades, very plausible ones, which
Henry would not be able to distinguish from a real barn given the
amount of attention he can spare from the driving seat. Now do
we think he knows that this barn (this one really is) is a barn?
Most of us will think not, in spite of his being, in fact, right in
this case. Had he been 2 miles further down the road, let us say,
he would have looked out of the window, unhesitatingly identified
a barn, and been wrong.
Now vary the conditions. Suppose that there is only one such
fake barn, but that it is very close to the barn that Henry has (cor-
rectly) identifiedperhaps the very next one down the road is the
3
A. Goldman, (2), esp. Section I.
42 SECT. V
one that will catch him out. Decide how you feel, then try another
variation: there is only one such fake barn, and it is nowhere near
Henry's route. Or there was once such a 'barn', but it was destroyed
5 years ago. Or there never was such a barn, but someone once
thought of making one and then decided not to bother. Or it is
just that it is a physical possibility to make one.
My own reaction here (and I take it that it will be widely shared,
or that I will quickly be notified if it isn't) is that at the beginning
of the sequence I am reluctant to say that Henry knows that he
is seeing a barn, but that thereafter my reluctance falls off sharply.
How does this compare with the reactions of an inquirer weighing
Henry up as a possible informant?
We need to be very careful here as to exactly what the inquirer
knows at the outset. If he just knows that the fake barn is very
close to Henry, and here is Henry now announcing a barn, then
obviously he will be unwilling to use Henry as an informant, much
as we are unwilling to allow that he knows. But then comes an
anomaly. If the inquirer knows all that and a little bit more, namely
that the fake barn is not, as a matter of fact, the thing that Henry
is now looking at, won't he then be happy to rely on Henry to
tell him whether or not the thing he is looking at is a barn? Surely
he will, so long as he can eliminate the possibility that it is the
fake barneverything else around here Henry can perfectly well
distinguish from a barn. Yet the knowledge that it isn't the fake
barn that Henry is looking at doesn't increase our willingness to
say that he knows that the thing he is looking at really is a barn.
There seems to be a mismatch here between our intuitive use of
the concept of knowledge and the inquirer's view of the adequacy
of an informantso the discrepancy has to be explained.
We can explain it by calling once again on the distinction between
the informer and the source of information. Henry ceased to be
a desirable informant when we heard that there was, close to him,
a fake barn-facade which he would not manage to distinguish from
a real one. Receiving the further news that whatever Henry was
looking at when he announced a barn was at any rate not the said
barn-facade, and being satisfied that there was nothing else that he
would mistake for a barn, we could make use of him after all. But
notice that 'make use of seems to be about the right way of putting
it: it is only because of some special piece of knowledge which our
imaginary inquirer happens to have (and Henry doesn't) that he
SECT. V 43
can take Henry's affirmation as a basis for the belief that there is
a barn there. Because what he comes to believe is identical with
what Henry told him there is some temptation to think that Henry
is here informing him of the barn. But in view of the way in which
the inquirer uses his special ancilliary information, and wouldn't
form a belief without it, the case is just as easily assimilated to that
in which Henry announces a barn and the inquirer, aware that at
any rate there is no barn there, infers that it was the barn-facade.
Tentatively, we may make a more general point. Clear is at any
rate this: the fact that whether the object he sees were a barn or
a barn-facade Henry would say that he saw a barn, is something
that will not in practice not bother our inquirer. More accurately:
it will not bother him unless he assigns a significantly high prob-
ability to the alternative that it is a facade. This, I conjecture, is
what is at work on us in the series of situations just envisaged.
The mere logical possibility of a indistinguishable object gives no
boost to the thought that it is the fake that Henry is looking at,
the physical possibility (by itself) doesn't do so either; even the
knowledge that someone once conceived the intention of making
one has very little effect. If we are told that they actually did it,
then something stirs; and if we are told that they did it near here
the matter begins to get serious. If, finally, we hear that there are
several of the things along this very stretch of road, then we cease
to look to Henry to tell us where the (true) barns are. And our
judgement as to whether Henry knows that it is a barn (given that
in fact it is) tends to keep step with this progression: at the beginning
he does, at the end he doesn't, in between we waver in varying
degrees.
Now, in one of the cases in which the likelihood that this is the
barn-facade seems to have grown large enough to matter, imagine
our inquirer to be in possession of additional information which
lowers that likelihood again. Our trouble was at first that he may
still use Henry as an informant, but we then saw that this is not
really what he is doing: he is using Henry in conjunction with his
own additional information to arrive at a belief, and so Henry is
serving him as a source of information. When we asked earlier what
gave us a practical interest in the distinction between informant and
source of information, a central component of our answer was that
some routes to reliable belief were open only to particular inquirers,
those specially equipped with background information and
44 SECT. V
:
B. A. O. Williams, (1), p. 7.
48 SECT. VI
Another possibility:
3. There is something about the present case which makes the
continuation of the correlation accidental in this instance.
Notice here, for clarity, that we must carefully distinguish between
the circumstance of the (near) universality of the correlation having
been accidental in the pastin which case no inference to the present
case would be warrantedand its being accidental in the present case
in spite of not having been accidental in the past. (This might be so
if some of the essential conditions which made it happen in the past
have been removed, but it happens nevertheless, and indeed this
would be just the structure of the standard type of Gettier-example. 3 )
We should ask, what is wrong with the correlation's being acciden-
tal in this instance, if de facto it holds? After all, the inquirer will
take it to hold, in virtue of its having held up to now, and since
it does hold in the present case (no matter how that comes about)
he is led to pick an informant who will in fact tell him the truth.
So he gets what he wanted; is there any reason to be dissatisfied?
One reply is this: that although in one sense he gets what he
wanted, that is the truth as to whether p, there is a good deal that
he would have liked to get with itand doesn't. He would have
liked, for instance, the assurance that he could use that informant
again on similar questions; but quite the contrary, he is actually
warned against it, and must now begin to look elsewhere. This reply,
however, is for two reasons not sufficient. First, it leaves one won-
dering why the fact that our inquirer did not get something else
should so cloud the fact that he did get sound information about
p. Of course, if it were clear that inquirers were never looking for
anything so specific as information, on that occasion, as to whether
p, but always for informants who were right in general on that kind
of question, then it would be clear what they felt to be missing.
But far from being clear, it doesn't even sound likely. It may be
(we shall shortly consider the question more carefully) that being
right on one issue nearly always goes along with being trustworthy
on a range of others, but why that should stop someone counting
as a good informant on the single issue in circumstances in which
that property doesn't draw the more general reliability along with
it still awaits explanation.
I mean that to be taken at face value, not as a rhetorical denial
3
E. Gettier.
SECT. VI 49
That point does not of course touch the question how we assign
these comparative probabilities in the first place, but the example
which follows in Blackburn's paper gives some indication. There
Mirv and Pirv have the same evidence, but crucial to what they
are going to make of the evidence is a certain question about the
likely behaviour of a professorand on this point Mirv has the
true belief and Pirv the false one. On the basis of this description
of the situation we give Mirv the better chance, for we quite reason-
ably take it that someone who, in reasoning, is operating with true
beliefs has a better chance of success than someone who is at some
point relying on a falsehood. What is at work here, in other words,
is the very same principle as governs the response to Gettier's exam-
ples. As with the various competing analyses, here too our approach
unifies prima facie distinct features of the literature.
VII
who knows his own name but is no good at finding out anyone
else'swhich suggest that the 'single proposition' analysis may have
something to be said for it. Can the technique of practical explication
be used to throw any light on the matter?
We can arrive at much the same question from another direction,
one highly pertinent to the approach I am recommending. McGinn
makes a suggestion2 about the pragmatic value of the concept of
knowledge. Knowing that p, according to his view, involves being
good at getting true beliefs on a range of associated questions, so
'if we know that S knows that p we can infer . . . a number of truths
about the world given information about S's other (relevant) beliefs'.
Now I attempt to construct the concept by considering the situation
of someone who is looking for the truth on the sole question whether
pmy inquirer as presented is not looking for the truth on a range
of issues similar to whether/;.
This raises the question: ought I to present my inquirer as someone
who is seeking a good source of information, not just for whether
p, but for this wider class? I would have to answer the question
why he should (almost) always be doing thatthe stronger the
description of him the greater the problem of arguing that virtually
any member of any community must satisfy it. But there may be
a better route. For there seems to be at least a possibility that the
concept constructed from my original basis will include the idea
of the wider competence, or will be such as to imply this competence
in nearly all cases. Then we will be able to say that the concept
fashioned to meet the needs of my inquirer will put him in a position
to declare a bonus: not only will he have found a good informant
on the question whether phe will usually have found an informant
who is trustworthy over a range of />-like issues. Perhaps we shall
end up saying that these two turn out the same, and that looking
for ap-informant has to be equivalent to looking for a p-like inform-
ant. But that is not clear a priori, and my first guess is that these
two are not the same but only concomitant, and then only nearly
always rather than invariably.
Let us investigate the guess. What the inquirer wants, we have
seen, is someone possessing a (readily detectable) property which
correlates well with being right as to whether p. More than that,
the inquirer must believe that it correlates well, otherwise he can
2
Ibid., p. 540.
56 SECT. VII
3
Ibid., p. 536.
S E C T . VII 57
ing') holds for p in particular but not for other propositions 'very
like' p. Nozick's conditions are then satisfied, but we are unwilling
to attribute knowledge of p. The example must be so constructed
that the putative knower's capacities, and those features of his situa-
tion which might be thought to be making him reliable as to p,
are such that one would be led to expect reliability with respect
to these others as well. When it turns out that the subject is not
reliable with respect to them, this makes us feel that it cannot really
be those capacities etc. which are operative in respect of p, in spite
of the fact that his belief 'tracks' its truth-value. We needn't therefore
say that the reason why this is not knowledge is because knowledge
of p calls for knowledge of the other ('jp-like') propositions as well;
we can say that even with respect to p the subject is not really
exercising those capacities which make him reliable, so that we have
to regard it as a fluke that he meets Nozick's conditions. And the
trouble with flukes, from the point of view of the inquirer, is that
they aren't predictable; which makes them of no use to him, since
he needs to be able to tell in advance that there is a high chance
of learning the truth.
This is paralleled by the line of thought that arises if we go back
and look at McGinn's case from the perspective of our inquirer:
if we now consider someone trying to use that person as an informant
about whether p (whether a certain stick is bent or straight), we
can see that they will judge him to be a good informant because
they will see him looking at the objects in question, and since the
properties he is judging them to have are (we suppose) simply percep-
tually detectable they take it that he possesses a property correlating
well with being right as to whether p. What they learn, however,
when they learn of his incapacity to give the right answer to the
other questions (whether the other sticks are bent or straight) is
that there is special reason to mistrust the correlation in this particu-
lar case. The question whether theywho don't at this stage, let
us remember, know whether/" or notshould regard him as a good
informant is back in the melting pot. Unless they can find some
further property which he possesses and which (i) correlates well
with true belief as to p and (ii) doesn't correlate equally well with
true belief as to all these other propositions about whichas we
have now found outhe doesn't hold true beliefs, then they must
suppose that his chances of telling them the truth about p are, for
all they can tell, no better than random. And then, since their own
60 SECT. VII
chances are presumably at least that good, they must cease to regard
him as a potentially useful informant.
VIII
but adds that there must be knowledge of the two preceding con-
ditions as well.
That an externalist will return an immediate' 'No' to Iteration
is clear enough; just a little more thought is needed to appreciate
the force of the temptation for the internalist to answer 'Yes'. If
to know that/? is to satisfy the conditions:
(!)/
(2) S believes that/;
(3) S has X
then knowing that one knows is to know that the three conditions
hold, that is to say:
(!') S knows that/;
(2') S knows that S believes that/?
(3') S knows that S has X
Now if an internalist so understands (3) that he does not really
distinguish it from (3'), then little else is needed to effect the passage
from knowing, { 1 & 2 & 3 } , to knowing that one knows: {!' &
2' & 3'}. Since (3) is true, our internalist will take it that (3') is
true. Since ex hypothesii the conjunction {1 & 2 & 3} holds, and
he takes that conjunction to define knowing, he will take it that
(!') holds. Now only (2') remains to be accounted for. Our interna-
list might take that step as a result of believing something like the
traditional thesis of the transparency of our mental states to con-
sciousnessthat we always know what we believe. But without
accepting that dubious thesis, he might get to (2') as a common-sense
consequence of believing (3'). For if his chosen A'is 'has good reason
to believe that />', and S knows that he has itwhich is what (3')
saysthen S is hardly likely to miss the fact(2)that he believes
that/;. So (2') will hold, and with it the full set of conditions for
knowing that one knows. If the chosen X,, as in the causal theory
or the tracking analysis, actually mentions the belief that/7, the pas-
sage to (2') is even more obvious. So it should not be in the slightest
bit surprising if the internalist takes the view that 'S knows that
/>' and 'S knows that S knows that p' are equivalent (though he
might, because of the status of (2'), resist the conclusion that the
equivalence is strictly speaking a two-way entailment).
Now at least with regard to the 'If Kp then KKp' principle we
can certainly find examples of just the confusion under discussion.
SECT. V I I I 67
Here are two from Hintikka's Knowledge and Belief, one quoted
(approvingly) by him, the other actually used. 3 Schopenhauer, in
On the Fourfold Root.. ., writes:
. .. just try. . . to know without knowing that you know .. , 4
a challenge the force of which must surely rely on the hope that
your attempts to know will include attempts to assure yourself that
you know, and the further hope that this will convince you that
knowing itself includes knowing that you know. The joker here,
of course, is the second word. If we try to know that p we are
going to concentrate our minds on the question whether we know
that p and what we have to do to bring that state of affairs about.
Then, if our efforts go well, we very likely shall finish up knowing
that we know that p. At least it will be the case that we know
that p, we shall believe that we know that p, and we shall be in
a state such that people in that state who believe that they know
that/) virtually always believe truly.
But it is not the knowing that has produced the iteration; it is
the trying. Think of someone who, instead of trying to know that
p, tries to find out simply whether p. If he does the job well, he
will end up knowing that p (or not-/? as the case may be). At any
rate, he will arrive at a true belief and a state which correlates well
with truth in beliefs of that type. But whether he will know that
he knows is quite another matter, depending on how self-conscious
his investigation of/> was; he may, but quite likely he won't.
Hintikka himself uses a similar route:
. . . all those circumstances which would justify one in saying 'I know'
will also justify one in saying 'I know that I know'. 5
The concept of the good informant, I have argued, has the property
that once we try to state specific necessary and sufficient conditions
for it, once we add some specific condition Y to the requirement
that the informant believe the truth on the matter at issue, it proves
possible to 'outflank' the attempted analysis: we can imagine further
facts, compatible with the informant's possessing F, such that when
they are included in the picture his rating drops to an unacceptable
level; so the suggested conditionstrue belief plus Ywere not,
after all, sufficient.
We noted that the recent literature on the analysis of the concept
of knowledge suggests something suspiciously similar: when a speci-
fic analysis is proposed, sooner or later someone devises a counter-
example to demonstrate the insufficiency of the proposed conditions.
A very general and approximate description of these counter-exam-
ples might be: they all present possible circumstances in which the
further condition X is satisfied, but satisfied in such a way that
it becomes accidental that the subject holds a true belief as to whether
p. What, with more precision, does this really amount to?
Gettier addressed himself to the 'justified true belief (JTB) analy-
sis. A condition of the construction of his type of counter-example,
he pointed out, was that it should be logically possible to have
a justified belief in a proposition that is in fact falseS is justified
in believing that p is not to entail thatp is true. This being granted
(it is not easy to resist it without having to admit that justification
is an unnecessarily strong requirement) Gettier proceeds to invent
situations in which the subject is indeed justified, but in which,
given the manner in which the justification arises, the likelihood
that he is right as to whether p is not thereby increased.
This makes the JTB analysisgranted Gettier's principle about
justificationstructurally very similar to the conditions on a good
informant: true belief plus possession of some property which lends
high probability to truth of belief, but does not (because such a
70 S E C T . IX
(1) p is true
(2) S believes that/)
74 SECT. IX
First, A had to choose the spoken word for conveying his infor-
mation; secondly, when he shaped up to tell his lie he had to hit
on precisely that country the English name of which would sound
to B like 'Guinea' when pronounced with a Gumean accent. Nigeria
wouldn't have done, nor would Ethiopia or Zimbabwe or (probably)
anything else, butit just happened to be 'Ghana' that he picked
on. Well, well! Relative to the evidence in our possessionand that
is what judgements of probability are relative toB was extremely
unlikely to acquire a true belief. But the unlikely can happen; and
this time it did.
Make the comparison: let there be a chair here, and suppose that
I look at it attentively. Now, what has to happen for me to fail
to acquire the belief that there is a chair here? Answer: something
pretty unusual. Let there be a Guinean here intent on deceiving
me about his origins, and suppose that I listen to what he tells
me and believe what I understand him to say. What has to happen
for me to acquire the right belief (without benefit of knowing that
he is lying, without benefit of experience of the Guinean accent)?
Answer: something pretty extraordinary. Wasn't he unlucky? Of
course he wasthe odds were piled house-high in his favour.
We find much the same if we look again at one of Williams' other
examples, that of the Chairman of the Board. People who are
depressed characteristically have a certain type of facial expression,
a certain heaviness of bodily movement, a certain style of speech,
an absence of energy and initiative, a certain 'slant' on life and its
questions. Someone who detects these features in the normal way
and infers, if he finds them, that their possessor is depressed, has
a reasonably good chance of forming a true belief. But our chairman
didn't do this; he came by his (true) belief only because the account-
ant, of all the ways in which he might have manifested his depression,
happened to come up with one (of the very few, we may presume),
which had the effect of making him (the chairman) depressed. And
he had to come up with it for the reason that he (the accountant)
was depressed, rather than for the reason which must be at least
equally common: that (in his everyday professional state of mind)
he had seen that the firm's state was actually parlous. The chairman's
chances of ending up with a true belief would have to be assessed
as very slender by anyone who did not already know the whole
story, including the facts about the accountant's state of mind.
What seemed to cause a problem here was the wrong approach
SECT. IX 77
2
See G. Harman, pp. 479 andpassim, and K. Lehrer. It is interesting to compare
F. P. Ramsey.
78 SECT. IX
4
See Section V.
80 SECT. IX
(or even believed it) had he had the sheer bad luck to have the
totally misleading matter of D brought to his attentionand brought
to his attention by itself, so that he was not in a position to judge
that D was, just this once, nothing but a red herring.
In summary: the No false lemma principle may be taken in a
form in which it is indeed necessary, but then adding it does not
yield sufficient conditions. Or it may be taken in a form in which
it combines with JTB (and, I suggest, with pretty well anything
else ever seriously advanced) to give a set of sufficient conditions;
but in that form it is not necessary. The conjecture derived from
our 'practical explication', that one will achieve sufficiency only
by including conditions too strong to be necessary, survives this
test.
X
There are various ways in which a candidate might fail these con-
ditions. As regards the first, he might just not be here when I want
him. Notice that in this context 'here' does not quite mean 'in the
same place as I am'; it stretches to include all places which our
systems of communication link with the place where I am. But at
this stage we must also build in 'or soon can be'for another subjec-
tively varying factor will be the urgency of my situation; the less
urgent, the laxer we can afford to be over the spatio-temporal
requirements.
Equally, there are various ways of failing condition (2). One of
them, of course, is by simply not being likely to be right about
p. Another, if it is possible (I shall not try to resist those who feel
that it isn't), is this: whilst being likely to be right about/), possessing
no property whatever from which anyone, however knowledgeable
and acute, could infer that one was likely to be right about it. Neither
of these, clearly, have anything in particular to do with the state
of our particular inquirerif they hold, they hold for every
inquirerbut there are many that do. For supposing that the candi-
date does possess a property which correlates well with being right
about p, inquirers may (and mostly will) differ in point of their
capacity to detect the property, and in point of knowledge of the
correlation. One needs both of these to be able to detect the potential
of the potential informant, so some inquirers will be better at it,
some worse. Such differences may depend on sensory acuity, intel-
lectual facility, theoretical background; they may also depend on
altogether more transient features of inquirers, like their spatial pos-
ition or bodily orientation. But all introduce a subjective element
86 SECT. X
mimity, and are in some degree at least helpful to others and respon-
sive to their needs. And even if I, as one of the community, am
not so inclined, I shall still need an appreciation of their point of
view if I am to be any good at getting them to help me. From
such facts arises a pressure towards the formation of 'objectivised'
concepts, concepts which separate as it were the common core from
the multitude of accretions due to particular circumstances and parti-
cular persons and so varying with them.
To see how this may come about, consider that I shall hope that
others will on occasion recommend informants to me. (This includes
the case in which someone recommends himself.) I am not, of course,
asking them to recommend people who, without their recommenda-
tion, would in any case have struck me as good informants. (This
is one way in which my appreciation of their point of view comes
in: if I am to be any judge of whom to ask for a recommendation,
I shall need some idea of what sort of person might look like a
good informant to them.) Quite the contrary: I want them to recom-
mend as informants persons whom but for their help I could not
have recognised. There are many detailed reasons why others may
be in a position to do me such a service, but in general they come
down to this: that they can detect properties of the informant which
I cannot detect, or have more knowledge than I have of which
properties correlate well with being right on the topic at issue. In
practice this will mostly be a matter of their being 'better placed'
than I am. As I seek to discover the close-of-play score there may
be nothing about Fred to suggest to me that he would be a good
person to ask. But Fred's friends are aware that he was at the ground
when play finished; so their advice will help me. Amongst the best-
placed observers of Fred is of course Fred himselfwhich is why
people are so often in a good position, indeed often the best position,
to advertise themselves as good informants.
As well as properties of Fred which I adventitiously cannot detect,
there may be some which I am in general poor at detecting, and
in the event of any of them correlating closely with Fred's capacity
to form true beliefs on whatever it is I am curious about, I will
hope that others, better endowed, will be prepared to point me
towards Fred. In either case I am constructing the idea of someone
who is a good informant, but whom for one reason or another I
cannot detect as such. And I can go further, and conceive of someone
who has a property correlating excellently with possession of the
SECT. X 89
sinks to zero, as one might think it had when we form the conception
of a state of affairs which we are totally incapable of detecting, the
force of a 'state of nature' explanation will sink to zero too, unless
it takes the liberty of helping itself to additional explanatory princi-
ples. This is not to say, with the verificationist, that there is no
such conception, only that a certain style of explanation is impotent
to account for its existence, if it does exist.
But since for the moment we need only concern ourselves with
covert and not with strictly undetectable properties we don't need
to tread on any verificationist toes, or (what would worry me more)
to go beyond the effective range of our method in order to use
these thoughts about objectivisation for clarifying some of the prob-
lem cases. They indicate sufficiently why we will come to distinguish
between the properties of the object which make it suitable for a
certain use, and the properties of the user which make him capable
of using it. To put it briefly: the two facts that human life is social,
and that the members of a human society have differing capacities
and perspectives, make it obligatory to form the separate conceptions
of the state of the object, which is invariant with respect to different
individuals, and the states of the 'consumers' of the object, which
are not. We are in the same area here as are those ethical theorists
who point out the social advantages of adopting moral principles
which prescind from facts specific to individuals.
What happens to the concept at the centre of our investigation,
that of the good informant, as objectivisation proceeds? The require-
ment of a true belief remains, and so does that of a property correlat-
ing well with truth of belief on the issue in hand, but that of the
detectability of the property will be diluted. When we began the
individual inquirer was looking for something that he could recog-
nise there and then, preferably with a minimum of effort. That,
of course, is what he will still hope for, but it is not what will
be embodied in the public concept that now develops. For one thing,
what is effortlessly available to him then and there will not be a
matter of public interest, nor will it be the only thing that is of
interest to him; for another, it will be in his interest to bear in
mind the possibility of others having different (which will include
more powerful) powers of detection, since that will at times be of
use to him, let alone to them. The more we get used to the existence
of such powers, the weaker will the detectability requirement, as
reflected in the public concept, become. The concept of knowing,
SECT. X 91
our hypothesis must now run, lies at the objectivised end of the
process; we can explain why there is such an end, and why it should
be found worth marking in language.
Earlier in this section we looked also at the third factor and the
determinants of its most subjective versions, so to speak its 'me-here-
now' shapes. How important is it to the particular inquirer, in his
particular circumstances, to get the truth? What is his attitude to-
wards risk? These are equally ripe for objectivisation. Just as I may
have an interest in being able to sit down where and when I want
to, without knowing exactly where or when that will be, so I may
have an interest in collecting information while it is available, without
knowing when, or why, or under what pressures, it may be needed.
In addition, I shall be concernedwithout knowing in detail what
their circumstances and purposes arethat others should make
judgements as to who is a good informant. That may be because
I have enough altruism to wish that their enterprises in general,
hence also their enterprises of belief-acquisition in particular, will
succeed. But even without it I shall still want them to make assess-
ments of informants, because that may turn out to be useful to me.
I shall not suppose, however, that in making such assessments they
have my particular circumstances in mind: if they tell me whom
to ask about the times of trains to London, I shall not expect them
to take into account how important it is to me to get the right
answer, what I shall lose if I don't arrive on time. Perhaps I don't
yet know that myself; perhaps it isn't even on my own behalf that
I am trying to find out, but for someone else whose exact concern
with the information is unknown to me.
All this is going to edge us towards the idea of someone who
is a good informant as to whether p whatever the particular circum-
stances of the inquirer, whatever rewards and penalties hang over
him and whatever his attitude to them. That means someone with
a very high degree of reliability, someone who is very likely to
be rightfor he must be acceptable even to a very demanding
inquirer. So of the worlds that we cannot quite definitely exclude,
we shall want to include in our assessment of him even those that
we regard as very improbable. Moreover, we shall be motivated
to take a pretty careful look at those which we 'can quite definitely
exclude'is that really as many as we think? These thoughts take
us further down the road of objectivisation. Knowledge, so the
hypothesis goes, lies at the end of it.
92 SECT. X
The first step in writing about scepticism must be to say what one
is writing about. A negative attitude towards almost any generally
received truth can be called scepticism, and in the course of the
history of thought a wide variety of positions have been so called.
Certain ancient sceptics were known to recommend abrogation of
all belief in furtherance of inner peace. Hume, for many a paradigm
sceptic, regarded such a recommendation as potentially disastrous,
but happily impossible to follow. His scepticism consisted primarily
in a drastic re-evaluation of the powers of reason to underpin belief.
In the twentieth century the sceptic has mostly been thought of
as someone who denies that we know anything much; how close
to Hume he may be depends on Hume's concept of reason and
its relation to this sceptic's concept of knowledge.
In offering to show how the existence of a debate about scepticism
is a product of the concept of knowledge, I do not of course have
anything like this range of material in mind. That all the positions
commonly called scepticism form an intellectual natural kind, about
which one can aspire to generalise without falling into triviality,
is unlikely almost beyond consideration. What I shall try to do
is to account for the role played in this debate by a certain type
of argument, familiar to every student of philosophy from the first
of Descartes's Meditations, where it is exemplified by the thought
of the deceitful demon. The strategy is to invent a hypothesis with
two properties: first, it would make no noticeable difference for
us if it were true; secondly, if it were true virtually all our present
beliefs would be false. Then it is urged that, because of the second
property, we must either show that the hypothesis is not true or
abandon virtually all claims to knowledge; and that because of the
first property we cannot show that it is not true. Nowadays the
demon has a colleague in the scientist -whose computer feeds perfectly
'normal' stimulation to an excised brain floating in nutrient fluid.
Technology has advanced but the problem remains the same: how
SECT. XII 105
1 2
See G. E. Moore. See E. J. Craig, (3).
3
See E. J. Craig, (2), csp. chs. 1 and 2.
106 S E C T . XII
of knowledge certifies.
As the reader will have recognised, I am now in danger of making
things much too easy for myself, for what these superlatives ('worst'
and 'highest') mean is still unclear, and they may mislead. The danger
is of slipping into saying that for knowledge, given that the concept
is governed by objectivisation in the way just envisaged, 'absolute'
objectivisation is called for, meaning by that the demand that the
informant be certain to have the right answer in any conceivable
world, including those 'sceptical worlds' of demons and brains in
vats; that the 'highest' standard, in other words, means a standard
capable of withstanding any conceivable mishap. That would explain
how and why the concept of knowledge brings scepticism in its
train; but of course it is not within a mile of being warranted by
anything we have established so far. It is rather as if we had said,
with Peter Unger, 6 that knowing (given its genesis) will come out
as 'being right as to whether p, and not by accident', and then
assumed permission to say that someone who is right, but would
be wrong if he were a brain in a vat, or a victim of the demon,
is only right to some degree by accidentleaving ourselves just
as open to the charge of misusing the notion of an accident as the
sceptic is to misusing that of knowledge. Perhaps he doesn't misuse
it, and perhaps this would not be a misuse of the concept of the
accidentalbut on that issue this cheap manoeuvre does not advance
understanding one inch. The hypothesis was that knowledge 'lies
at the end of the road of objectivisation', but where does that road
come to an end, or when have we gone far enough? The whole
issue turns on finding a justified answer to that question.
It should be said straight away that Unger's spectacular and notori-
ous argument for scepticism7 did not involve the cheap and ques-
tion-begging manoeuvre, but rather made use of a far subtler thought.
He began by pointing out a category of what he called 'absolute
terms'. Being flat, he said, means being perfectly flatnot at all
bumpya standard which, for all we know, quite likely nothing
whatever achieves, but which in all practical contexts is relaxed in
accordance with the degree of flatness needed for whatever purposes
are under considerationa rugby pitch, a lawn, a billiard table.
A necessary condition of knowing, he then argued, is being certain.
7
' See P. Unger, (1). See P. Unger, (2) and (3).
110 S E C T . XII
paralyse our powers of decision and action. That isn't yet to say,
as some opponents of scepticism have said, that they belong to the
pathological side of our mental life; but it is to say that they have
a role in it only under very special circumstances, which certainly
don't include the ordinary social practice of gathering and passing
on information, or certifying informants. In deciding whether Fred
can be advanced as an acceptable informant as to whether/) nobody
considers whether he would answer correctly were he (or they) sub-
ject to some systematic Cartesian illusion and then, having decided
that he would not, tries to calculate how much damage the fact
does to his credentials by asking how likely it is that that 'possibility'
will be realised. Just as importantly, nobody, however vital to them
it is that they get at the truth about/;, expects anyone else to consider
such matters before recommending an informant; nor does anybody
consider them before relying on one of their own beliefs. Unger
himself makes a remark which only sharpens the difficulty:
. . . with respect to the matter of whether there are elephants, for practical
purposes there is no important difference between whether you know that
there are elephants or whether you are in that position with respect to
the matter that you actually are in. 8
Why do we have the absolute concept of knowledge, must we not
ask, rather than the concept of this other state (the one we 'actually
are in'), or some vaguer concept taking in both, if 'for practical
reasons there is no important difference' between the circumstances
in which they apply? Or how do we arrive at it if the thoughts
it embraces (such as those of the classical sceptical possibilities) have
no role whatever in everyday mental life? The reply might be made
that even if we cannot answer those questions we can still point
to various 'absolute' concepts which uncontroversially have been
formedso there is no insuperable difficulty about their formation,
even if we do not see how it can happen. But this takes us only
half way, for in the case of knowledge there is a problem which
is not nearly so acute when we are thinking of emptiness or flatness.
I do not want to insist that 'flat' and 'empty' definitely are absolute
terms in Unger's sense. But if they are, as is on first impression
quite plausible, at least one can think of (reasonably) everyday pro-
cesses of thought which might have made them that way. Notice
that whatever processes our explanation calls upon do have to be
8
Ibid., (2), p. 201.
112 S E C T . XII
going to drop, no matter what the test told him about their status.
If that is so, the position is different. It isn't exactly that there is
no practical motivation for the test, but rather that the motivation
is balanced by the inability to take advantage of it. There may be
excellent reason for me to go to the shopsI need to buy something.
But then there is the countervailing reason that I am too mean to
buy anything when I get there. So I may just as well not go; and
likewise if there are certain beliefs which I am not going to change,
I may just as well not bother to investigate them.
The to-and-fro of the debate doesn't finish there, however. Other
possibilities remain uninspected. For one, perhaps he couldn't have
seen, without applying the test, that certain beliefs were going to
stay with him no matter what its verdict. For another, perhaps in
order to test those beliefs which he was capable of changing, he
had to operate test procedures radical enough to induce doubt, or
in some cases pseudo-doubt, in respect of all his beliefs, immovable
ones included. For a third, even in the case of such 'immovable'
beliefs, there may be practical benefits arising out of our attitude
towards them and other beliefs consequential on them if they are
subjected to a very rigorous test and are seen to pass it; in that
event there could be a practical point in applying the test to them,
even at the risk that they would fail it.
We are ill-equipped to deal with these thoughts, partly because
we have too ill-defined a notion of what constitutes a practical ben-
efit, and partly because (as in the second and third possibilities just
mentioned) we are not clear enough about the boundary-conditions
of the whole inquiry. But appeal to our ignorance, of which there
is plenty in this area, will only serve to stave off an attempt to
show that no practical motives could lead to the formation of a
concept of knowledge capable of supporting, or in some way permit-
ting, radical scepticism based on the argument from the thought
of the demon. It cannot help with the positive task of showing how
practical pressures could give rise to such a concept. That task is
evidently beset by all manner of difficulty; there is no obstacle here
to understanding the feeling that the argument from the possibility
of the Cartesian demon invites us to abandon the very purpose of
speaking of knowledge.
XIII
one knows that p; and this is very well founded, and completely
harmless, and really quite boring. Its content is no more than this:
that successful investigations result in knowledge.
There may however be a line from the first-person stance to scepti-
cism which does not pass through the Iteration Principle. Perhaps
it is not quite the form of scepticism we have had in mind in the
paragraphs immediately preceding, but it is still quite sceptical
enough to deserve the name. If we have adopted the first-person
restriction and are interested in whether we know that p then it
will not satisfy us just to know that p so long as the fact that we
know that p remains opaque to us. Suppose that someone is in
a position to assure mequite likely it will be on the basis of an
externalist analysis of knowledgethat provided I do have hands
(provided neither demon nor computer-aided scientist is at work)
I know that I have hands. Nowhe will typically continueyou
believe, indeed you have no shadow of doubt, that you do have
hands, so without casting any greater shadow of doubt you may
proceed to believe that you know it. I am offered a conditional
with the observation that I am incapable of doubting the antecedent;
and if I accept the conditional I had surely better abandon hope
of ever showing that I don't know that I have hands, since for that
I would have to show that I don't actually have any. But something
seems to have gone missing, for I knew when I began that I had
all these beliefs and could raise no real doubt about them; but I
was worried by the thought that I could be in that state and the
beliefs be false, for instance if I were subject to the demon, or were
the consciousness of a brain in a vat. I wanted to get behind the
fact that I have these beliefs and cannot doubt them, to see whether
I could find a guarantee of their solidity which indubitability alone
cannot provide, not, Moore-like, to re-affirm the beliefs and then,
via an externalist analysis of knowledge, to conclude that they were
known to be true as well as true simpliciter. And in that I failed:
faced with the hypothesis of the demon and its like, I found myself
just emptily repeating my own convictions without being able to
point to anything on which they were based which would not equally
be present were they false and any of the traditional sceptical hypoth-
eses true. Isn't this scepticism? And doesn't it emerge from the first-
person stance?
I shall not spend time on the first of those questions. To the
second I would say 'No'. No, because the first-person stance by
124 SECT. X I I I
itself does nothing to explain why the fact that I can give no dis-
tinguishing marks by which to rule out that sort of possibility should
have any significance for me at all. And once we allow that sort
of possibility a place in our thoughts we will get the same effect
whether we operate from the first-person stance or not. It can easily
come to look otherwise, for if we ask 'Does Fred know that/;?',
why should we not use our knowledge that p as a baseline and
start off by observing that p is true? None at all (that we have seen
so far), unless we are going to allow the sceptical hypotheses their
traditional role. Once we do that we shall find ourselves stuck on
the thought that our decision that Fred does or does not know
that/) rests on our decision about p, which in turn rests on a belief
which we hold in spite of the fact that, in the sceptic's story, it
would be false although to us the world was indistinguishable from
the world we are actually inas we at any rate immovably believe.
Another point arises here. Once we have decided to admit the
requirement that we be able to distinguish the circumstances involv-
ing the demon from those which, as we believe, actually obtain,
we do not then need the first-person approach as an extra: we get
it thrown in. For if we see the challenge in those terms there will
be no point in responding by asking about someone else's know-
ledge, let alone trying to take someone else's opinion as ground-level
evidence: even that there is anyone else will be a belief we can only
be entitled to make use of once the challenge has been met. That
Descartes first introduced himself in isolation and then thought of
the demon is from this point of view an expository accident; he
could just as well have thought of the demon and then have realised
that this was one he was going to have to cope with on his own.
How may we summarise? It isn't (obviously) that first-persona-
lism in epistemology doesn't exist, nor that it has nothing to do
with scepticism. The point is that both first-personahsm and scepti-
cism result from thinking about certain possibilities (fully compre-
hensive dreams, demons, brains in vats, in general any hypothesis
on which our cognitive faculties would give a systematically distorted
view of large tracts of reality) which are normally ignored. If we
want to understand scepticism we need to understand how and why
such possibilities come to be taken seriously.
The other suggestion I want to discuss in this section is the idea
that scepticism might be the outcome of the attempt to achieve an
'absolutist' view of truth. Bernard Williams writes of the 'absolute
SECT. X I I I 125
implies in the quoted paragraph, leads us to the first, and the first
(this seems to be said in the final sentence) gives us grounds for
scepticism. We can't reach Nowhere, so the possibility of being
wrong is ineradicable.
We need to be clear just how the train of thought is supposed
to run. Why, for instance, should we be so keen on that last sentence
(of my last paragraph)? Our beliefs are formed by usthat seems
to be the literal content of the metaphor about Nowhere. Now
if that thought, unaided by hidden premises, has sceptical conse-
quences, then any belief-forming being must be subject to scepticism.
The idea that God stands above scepticism, which afflicts only crea-
tures, will have to be dropped, and the doctrine of his infallibility
declared incoherent. One can avoid this conclusion only by identify-
ing the facts with God's beliefsso no wonder that something of
the kind appears to have tempted some theologians. The hidden
premise, that the belief that p is distinct from the fact that p, is
the further, minimally realistic, premise which we are to add to
'Our beliefs are formed by us'. Nagel, it appears, is happy to add
the slightly stronger premise that we form these beliefs in interaction
with the facts, but that strengthening can hardly be the decisive
factor in the production of scepticismwhat if we formed them
without interaction with the facts?
That premise is also part of the argument from the objectifying
theoretical advance. We are to envisage a sequence of events, thus:
(1) The World (W) appears to me to be Fand I accordingly
believe it to be F.
(2) I form a theory (T) about Wincluding mewhich explains
(amongst many other things) how W comes to appear F to
me.
(3) According to T, W is not F.
(4) I come to believe Ton account of its great explanatory
powerand give up my belief that W is F. (Whether or not
it continues to appear to me to be F.)
Now it is at this stage that Nagel arrives (see the passage quoted
above) at the thought: 'But this idea [I take it, the corpus of beliefs
that constitute believing T], since it is we who develop it . . .' He
goes on: Tf the initial appearances cannot be relied upon because
they depend upon our constitution in ways that we do not fully
understand, this more complex idea should be open to the same
SECT. X I I I 127
No more than Nagel's (to all intents identical) argument does this
lead to radical scepticism. What it describes is a particularly sophisti-
cated version of the familiar process of correcting our own mistakes.
It might sharpen our consciousness of the claims of fallibilism, and
make us more alive to just how many of our present beliefs might
have to go by the board before cognitive stability is reached. But
that it can be reached, and that our investigative practices are leading
us in the right direction even if the road may be longer and more
disturbing than we thought, this need not be brought into question.
Just why any view should be open to the reflection that it is 'only
one particular representation' still stands to be explained. It is not
explained by the undoubted fact that some views invite that reflec-
tion.
The idea that is here broachedthat of the attempt to get a view
of the world which will enable one to take account of the perspectives
of othersdoes however have something to do with the debate about
scepticism, and does have something to do with an important source
of encouragement for the sceptic. In this Nagel and Williams are,
I believe, right. But what does the trick is the way in which this
procedure of objectivisation, as I have called it, acts to form the
concept of knowledge; whereas they see it as it were pushing our
conception of the world out of reach, always one step beyond the
latest advance of our inquiry. What does not emerge (and cannot
emerge from that style of argument) is why they think it impossible
in principle for us ever to catch it up.
5
Williams (2), p. 65.
XIV
tually effective strategy; but one can perfectly well hope for an effect
without being in the slightest bit concerned whether or not it would
have occurred had circumstances been in any way different. Indeed,
hoping that it would occur under circumstances differing from the
actual sounds like anotherand rather strangeenterprise, perhaps:
hoping that its occurrence is probable. What the point of doing
that might be, as distinct from hoping that it would occur, isn't
at all clear: what might impel us to entertain a hope which would
be satisfied if the event proved to have been likely, even if it didn't
actually happen? I don't say that there could never, in any particular
instance, be an answer to that questionsometimes one might be
less concerned to be right than to avoid looking foolish. But that
is not an answer so natural and comprehensive as to make us expect
the counterfactual component of the concept to be as it is even if
no active truth-seeking strategies were involved.
Notice that this argument does not call upon the fact that, gener-
ally speaking, we want our beliefs for certain purposes, to enable
us to act successfully in certain ways. No doubt that is so, no doubt
it has some significance for the theory of knowledge, and a great
deal for the sociology of knowledge. But all that the present argument
makes use of is the fact that we are active, mentally and otherwise,
in seeking true beliefs; and it would hold even for a being whose
interest in true beliefs, once acquired, was purely theoretical.
That brings us back to asking: why should we be interested in
believing truths? This wider question has a clear place here, and
a fundamental one. Counterfactuality may be a central feature of
the concept of knowledge, but we arrived at it only as an inevitable
feature of any well conducted search for truth. So why are we search-
ing?
Imagine someone who watches certain events without the slightest
need or wish to intervene in themthey do not affect him in any
way that would cause him to prefer that the series should be just
as it is rather than otherwise, or vice-versa. His natural equipment,
let us suppose, causes him to form beliefs about what is happening.
Since, ex bypothesi, it is a matter of complete indifference to him
which events occur, and we have so far provided him with no other
evaluatory principles, there is as yet no reason in sight why he should
prefer these beliefs to be true rather than false, or indeed why he
should be pleased (or sorry) that he has beliefs at all. What do we
have to add to this neutral state for it to begin to make some difference
132 S E C T . XIV
to him whether his beliefs about what happens are true or not?
For a start, let us simply suppose that he prefers some events
to others. If he is right in believing that A is about to occur, and
he likes A, then something nice is coming up, if he is wrong then
it isn'tand that makes a difference to him. But so long as we take
it that he cannot actually affect what is going to happen, and don't
feebly abandon our inquiry by assuming that he values being right
per se, this still doesn't throw any light on why he should want
his beliefs to be true. If A happens, then something nice happens,
but whether he antecedently predicted A does not, in this very simple
model, make A any nicer, or the non-occurrence of A any nastier.
As soon as we get pleasures of anticipation and pains of disappoint-
ment into the model, the situation will of course change in this
respect, so this one might call the first grade of cognitive involve-
ment.
Given this grade of involvement, it does become the case that
if he can exercise any sort of control over what he believes, he can
affect something which it is in his interest to affect. Now he has
a motive for selecting, if he can, between different ways of acquiring
beliefs, but it is not yet by any means clear that he will choose
to select beliefs for their likelihood of truth. Having a true belief
that something nasty is inevitably on the way may well make life
a good deal nastier than the nasty thing will in due course make
it anyway; for that matter, having a false belief that something nasty
is coming will be no better, up to the moment when it is falsified.
He may decide that the pains of anticipation of the nasty are worse
than those of disappointment at the non-eventuation of the nice,
and opt so far as he can to believe in a thoroughly rosy future irrespec-
tive of its likelihood; or he may find it so enjoyable when things
turn out better than expected that he aims at a deeply pessimistic
set of beliefs in the hope of thus maximising the number of pleasant
surprises; or he might prefer to avoid believing anything as far as
possible. To explain the preference for truth from this standpoint
we would therefore need a supplementary, and I would have thought
far from compelling, hypothesis about the balance between these
various (and presumably variable) psychological forces.
A second grade of involvement comes if we suppose that he can
in some cases affect what happens. Then at last it does appear that
he must come to value true beliefs. For he needs true beliefs about
what will happen if he undertakes such and such a course of action,
SECT. XIV 133
as well as about what will happen if he does nothing, but just lets
things slide. Once this stage is reached he must develop an interest
in training himself to form his beliefs in ways which reliably lead
to truths. And if we also, as for any hint of realism we must, suppose
him to be partially ignorant of the conditions under which he is
operating, reliability must include the counterfactual property that
the methods would still have produced true beliefs had the world
been somewhat different.
Truth, and the counterfactual effectiveness of the processes used
to reach it, are two of the principal features of both the concept
of knowledge and that of a good informant. Active involvement
enters twice, on my account: we want true beliefs because we are
agents; and we actively seek the truth, which is why we must try
to 'track' it, not merely hope to hit it. I cannot decisively rule out
the possibility that these two features may be convincingly explained
without calling on the fact that the concept of knowledge is a concept
formed and operated by active beings who need to direct their
activity. And I grantat least I can find no argument to the con-
trarythat a pure spectator could in theory operate it. But why
should he? To ignore the question is to accept that there is a practice,
common to the vast majority of human beings, which mysteriously
exists without any basis in the human situation. Answering it, on
the other hand, won't be easyfor anyone who thinks that episte-
mology deals with a purely spectatorial side of our nature.
XV
using as our informant on this question and not Mabel at all. Earlier,
in Section III, we considered the case of someone who holds a true
belief that p but is useless as an informant because we can only
tell that his belief is true if we already know that/) ourselves. What
we have here is the same thing, but just as it were one link further
down the chain of communication. What is needed is something
or other about Mabel which enables us to judge what it was that
Fred told her from what she tells us, something which we can detect
in her without first knowing what Fred said; that is to say: at the
very least some kind of reaction to Fred's assertion. Perhaps we
have here the beginnings of the intuition that if Fred's knowledge
is to be transmitted to Mabel, so that she too now knows that p,
some kind of 'uptake' on her part is required.
What kind? We have implied that it will have to be at least enough
for her to be disposed, as a result of being told that p by Fred,
to tell us that p rather than that not-p, or nothing at all. Shouldn't
we consider the freakish case in which we know that Mabel systema-
tically distorts this kind of information, and so can gather from
what she says that/) is true although she actually tells us something
else? But we have considered this sort of thing before, and we should
make the same response now that we made in Section V. Such exam-
ples may at best succeed in showing that a certain condition is not
strictly necessary, a point which, though worth noticing, shouldn't
affect our conception of the prototypical case. But does this one
even do that? If we have to work our way back to the proposition
that p from Mabel's assertion that q it is at best doubtful whether
she really is functioning as an informant as opposed to a source
of information, and hence whether there is any real informant in
this story apart from Fred. We can therefore confirm the implication:
Mabel must react to Fred by acquiring the disposition to tell others
what he told her.
To get a little nearer completeness, we may also spend a moment
on the case in which the disposition acquired is simply to repeat
Fred's words on request, maybe uncomprehendmgly, as one might
parrot a message in an unknown tongue (a reaction even more mini-
mal than what I just now called the absolute minimum). Here any
hesitation one may have felt over the previous case is surely absent:
the only informant in the piece is Fred, and Mabel is no more func-
tioning as an informant than is a page of a book, or a strip of magnetic
tape on which Fred's speech is recorded. She must, it seems, be
SECT. XV 137
1 2
M. Welbourne, p. 93. Section V, above, and C. Radford.
138 S E C T . XV
Many will surely feel that she wouldn't. Mabel is suffering from
the notorious false lemma, for if she were brought to realise that
Fred isn't double bluffing, what would she then believe? Would
she believe that not-p, because she would then think that he was
merely (single) bluffing? Or that/?, because she would then think
that he wasn't bluffing at all? Or maybe nothing, because she then
wouldn't know what to think he was up to? So far as our story
goes, it could be any of these. Lucky old Mabel then, to have got
the answer right after all.
Consider now the parallel question as to whether Mabel is a good
informant. We, who want a good informant and are hoping that
Mabel will do the job, believe that she is telling us what Fred told
her; further that Fred was potentially a good informant, and if sincere
in what he said actually was one. Probably at this stage -we rate
her chances of being right very high, but then we hear that she
thinks that Fred was double-bluffing, and that in fact he wasn't.
How are we now to assess the likelihood that what she tells us
is the truth? We have to assess the relative likelihoods of two possibi-
lities: that he was only bluffingin which case she will be wrong
and that he wasn't bluffing at allin which case she will be right.
And all we have to go on is that something or other made her think
he was double bluffing. We are all at sea.
Suppose, however, that we are in a position to make some such
estimate. Suppose we become virtually certain, and rightly, that
Fred was being perfectly sincere. Now -we know, since we can work
out the effect of her view that he was double-bluffing, that Mabel
is just as likely to be right as Fred was, which means very likely
indeed. But now we are in the position of knowing much more
about the situation than Mabel does, and it is very doubtful whether
we are using her as an informant as to whether p. What we are
doing, rather, is using her as an informant (or maybe as a source
of information) to find out what it was that Fred told her. On that
quite different question she is a good informantthough she didn't
get the title by inheritance from Fred, but by using her own eyes
and ears. And once we have that piece of information, we can do
the rest, reliably. She could not; from what we know about her,
her route is altogether too hazardous.
That, then, is why it is not enough for Mabel to believe what
Fred tells her, not even if she believes it because it is what he told
her. What we want, it now seems, is that Fred should have been
S E C T . XV 139
sincere, and that Mabel should believe him to have been sincere
and believe what he told her (i.e. thatp) for that reason. This appears
to be what Welbourne calls 'believing Fred', and holds to be necess-
ary for the transmission of knowledge. And as far as Mabel's contri-
bution is concerned, he takes it to be sufficient: 'All that is required
of a listener who understands a knowledgeable teller if the knowledge
is to be successfully transmitted to him is that he believe the teller'.3
Necessary it may be, but I doubt whether it is sufficient (even sup-
posing that Fred satisfies all required conditions on the teller), for
we can repeat the recipe used at the previous stage and insert another
'deviant' link: let Mabel's (true) belief that Fred was being sincere
(and was 'speaking from knowledge' etc.) be reached by some weird
route involving a whole lot of false beliefs on her part; so structure
the story, if you will, that only an amazing series of coincidences
prevented her from discovering their falsity and thereupon taking
Fred to be lying, or falling into complete confusion as to whether
he was lying or not. Then whether the exchange leaves her knowing
what Fred knows becomes as doubtful as ever.
Perhaps, however, it isn't Welbourne's intention to define 'believ-
ing the speaker'; perhaps he is offering it as an undefined intuitive
notion. But if that is to succeed, then on my diagnosis it would
have to be just an unrecognised attempt to exclude all deviance from
the route by which Mabel arrives at her belief, to ensure by covert
stipulation that she is not prey to any false lemma. It would be
preferable, because ultimately more illuminating, not to accept it
as primitive but rather to ask what was involved. Then we would
come to see what work it was really doing.
3
M. Welbourne, p. 5.
XVI
(hogy) = know (that) and ismerni + direct object. Again, the word
for the theory of knowledge derives from the second: ismeretelmelet.
Nor is any restriction to the direct object form implied in the French
equivalent theorie de la connaissance, although connaitre is used
with the direct object construction. The theory of banking, which
is understood to some degree by both economists and aircraft
designers, is quite a different businesstwo quite different busi-
nesses, in fact.
It should be added that when we go outside Europe the outlook
begins to change: English no longer looks like the odd man out.
In the handful of African and Asian languages on which I could
find a good informant, all without exception turned out; to share
the English pattern: they commonly translate 'know (that)' and
'know (somebody)' by the same word. In view of this there is no
question of shelving the issue, saying that it is altogether too paro-
chial a linguistic fact to expect our style of pragmatic synthesis to
account for it, rather as if one had asked State of Nature Theory
in political philosophy to explain some idiosyncracy found in only
half a dozen of the world's constitutions. On the contrary, it is
something very widespread. The obvious exceptions ('obvious', that
is, when judged from the western edge of Europe), as I have just
said, are only dubiously such; to use them as a smoke-screen would
be to duck a pressing issue.
There are, however, a couple of methodological points on which
we ought to dwell a little, since they will be important for the subse-
quent section as well as this one. My type of explanation is best
suited to explaining why a certain concept or linguistic practice
exists, why languages need expressions to fit certain pragmatically
created slots, so to speak. Now if we regularly find the same word
filling two slots it is fair to conclude that the slots must be related,
and to try to specify the relationship. But we must be clear: there
is no commitment to the view that, if two slots are related, one
will find the same expression in each. Be the difference only gramma-
tical (here a 'that' clause, there a direct object for instance), no practi-
cal explication, however impressive an example of its kind, can yield
the result that all languages must use the same word in both places.
If native speakers can make the distinction, in any way at all, then
they could use two words to mark it. It may be said that if the
roles are conceptually very similar the verbal distinction will in due
course evolve away. Perhaps, but this has the unsteady feel of a
142 SECT. XVI
mation in these areas is hardly ever found except in those who have
frequently been sensorily acquainted with them. That would bridge
the gap between 'knowing Fred' and having information (of this
special category) about him, and it would bridge it in a way which
allows us to assimilate another fact of usage: that knowing Fred
comes in degrees. For we may know Fred very well, quite well,
or hardly at all, so to account for the locution we need some quantity
of which someone can have a lot, or a bit, or very little, and if
knowing Fred does after all have to do with having information
about him, then how much information one has will fit this bill
very nicely. The most that our earlier arguments have shown, then,
is that this range of information is not correctly specified if we just
call it, with complete generality, 'information about Fred'.
Admittedly, there is an alternative way of accounting for these
facts which is less favourable to our project. It says that for certain
values of X 'knowing X' means being from time to time sensorily
related to X, perceiving X, or perhaps more generally 'being in
the company of A". Thus for non-perceptual objects, or more gener-
ally non-locatable objects, such as the law, knowing them is indeed
to be understood in terms of possessing information about them.
But for some objects, especially persons, knowing them is primarily
a matter of being together with them, and related to having infor-
mation about them only because and in so far as that is the normal
outcome of being in their company. And being in someone's com-
pany has the required property of being (roughly) quantifiable: one
may spend a lot, not much, or hardly any time with Fred.
It is gratifying, however, to see that this account fails decisively.
Knowing Fred, whatever it may be, is not something that rises and
falls with the length of time the subject is together with, or experienc-
ing, Fred. I may have spent hours with that fellow-commuter, yet
scarcely know him at all. On the other hand there are people in
whose proximity I haven't spent anything like so long and yet know
(clearly because of the manner of the contact with them) quite well.
The refutation of the 'acquaintance' theory doesn't prove that the
informational approach must be correct, but when one thinks of
the sort of example that refutes the former one cannot but incline
to the view that the latter is a very strong candidate: surely the
key lies in how much, and the kind of, information about Fred
that the subject's contact with him has yielded?
The impression is reinforced when one remembers that lack of
SECT. XVI 145
very few exceptions, only people who are literally 'acquainted' with
him possess. In seeking a good informant on such topics we therefore
take ourselves to be looking for someone who has spent a good
deal of time withmeaning in much the same place asFred.
Whether that condition is strictly necessary or not is a question
that might be best avoided, but if we must ask it the right answer
is probably 'not'which shouldn't lead us to think that an adequate
account of the concept would not mention it at all. The position
is much like the one we saw earlier, 2 when asking whether it is
a necessary condition of knowing that the knower believe what he
knows, and what we are in danger of getting blocked by is too
much respect for the old analytic format of logical necessity and
sufficiency, with all else consigned to silence.
Apart from terms designating persons, the most common values
of X for which 'knows X' raises the issue of sensory acquaintance
are expressions referring to places'knows London', and the like.
They pose no fresh problems, and in one respect they are less compli-
cated. It is rather easier to state, in general, what sort of information
about London we shall normally be looking for: how to get from
A to B, where certain sorts of facility are to be found. The point
of principle is unaffected, namely that this sort of knowledge is vir-
tually never found in anyone who has not spent a good deal of
time in London. Once we have seen this we need no longer be
too worried about the prospect of establishing close contact between
knowing that p and knowing X, close enough to shed a good deal
of light on the linguistic data.
But still there is cause to go a little further. A small observation
may show that the idea of acquaintance, or perhaps it should be
interaction with, a person, has found its way deeply enough into
the concept of knowledge to have an effect on the outward grammar
of the direct-object form. When we are thinking of knowing things
about Fred, the demise of Fred does not affect the tense in which
we cast the verb: there are people around now who know a lot
about Schubert (died 1828). But were that Viennese bicentenarian
still alive today he would not know Schubert, though it might be
true that he knew him, and quite likely that he would know a lot
about him. The same holds if we substitute the name of a town
for that of a person. True, an example of a town that no longer
2
Section II, above.
SECT. XVI 147
exists may be somewhat harder to light on; but one that has changed
a lot will do just as well.
Our deliberations so far come to this: whilst there is clearly a
heavy informational component to 'knows X', the past-tense pheno-
menon that we have just seen indicates another element, one which-
unlike possession of information about Xceases with the cessation
of X. One thing you can no longer do to X when X has ceased
to be is, obviously enough, perceive it. But there are many other
barriers to perceptual contact. One of the most common is spatial
separation, and we do not change 'I know' to 'I knew' just because
an acquaintance is now far awaywe do that only when we feel
we have 'lost touch'.
As to what this might in essence amount to, I offer a conjecture
which at least has some of the right properties. What 'knows Fred'
suggests (in addition to information) is a capacity which might be
broadly described as the capacity to interact with Fred, and to do
so more smoothly and successfully than is generally the case when
two more or less randomly selected persons come into contact with
each other. (Note, now we are talking of capacities, that the capacity
to recognise Fred isn't of much use here. I don't know Queen Eliza-
beth II, but I would be pretty good at recognising her; probably
some people who do know her would be no better.) Only in the
rarest cases will such a capacity exist in the absence of quite a lot
of information about Fredso the two ideas intertwine.
Compare now two other examples of 'knows X': 'knows Lon-
don', and 'knows German'. The former surely implies an ability
to get about in London, to find places one wants, and so on. It
also implies possession of a good deal of information about London,
but this is not really a separate thing, since an adult human who
can find his way about virtually always knows what he is up to,
and so can present his capacity to himself and others in informational
form, even if only rather crudely. To ask whether capacity or infor-
mation is primary would probably be idle, and in so far as there
are pointers one way or the other they are variable and context-
dependent. In 'knows German' the emphasis is slightly different:
it seems far more closely connected to knowing how to do something
than to knowing whether something is the case, much more like
an ability to act than a capacity to inform, though this may well
be just because people are on the whole less skilled at articulating
their linguistic than their geographical abilities. It is worth noting
148 SECT. XVI
that there are languages which commonly use the word for 'can'
rather than 'know' in this connection: 'er kann Deutsch' is thoro-
ughly colloquial for 'he knows German', and Welsh has a very similar
construction.
The form 'knows X' can therefore sometimes imply (or maybe
even just straightforwardly affirm) capacities to act in certain ways.
Given that, it cannot be wild to conjecture that this factor may
be in play when X is the name of a person. Persons enter into our
lives in rather a lot of ways, so that we could hardly be expected
to say just what capacities are meant, as one quite easily can where
X names a town, and trivially can where it names a language. But
we already have some idea from remarks I made3 when speaking
of the sort of information which, in our practical dealings with a
person, we might find especially valuable. If we want to enlist Fred's
help a useful person to have on our side will be someone who knows
Fred: he will get a chance to broach the matter, and know how
to go about it, where we wouldn't. If we want to cheer Fred up
(I don't wish to make it sound as if manipulation were all that inter-
ested us), whom would we prefer to send in: a friend or a stranger?
Knowing someone, being acquainted with them and familiar to
them, oils the wheels and has a million uses.
The line pursued in the earlier paragraphs of this section related
'knows Fred' to 'knows that p'. This one relates it to 'know how
to'. We therefore need a view of the relationship between this loc-
ution, which at first sight appears to ascribe a capacity, and the
evidently informational 'knows that p'. Otherwise this attempt to
relate 'knows (direct object)' to both might be thought to split it
incomprehensibly in two. Besides, quite apart from knowing Fred,
enough languages have both 'knows that' and 'knows how to' for
this to be a question in its own right. This section has contained
hints, but now they must be made more explicit.
3
Above, p. 145.
XVII
appears to have a sense in which it means that the subject can perform
a certain kind of action. One way round the obstacle would be
to disperse it by arguing that this capacity sense, although found,
does not have the empirical status to be a problem: whilst it occurs
in some languages, it is not so widely distributed as to be a proper
subject of the present investigation. A second would be to argue
that, first intuitions notwithstanding, the apparent capacity sense
is really nothing of the kind, but informational after all. The third
is to accept both existence and status of the capacity sense and then,
by exhibiting a natural connection between the ideas of information
and agency, explain how knowing thatp and being able to A come
to attract the same word.
How widespread is 'knows how to'? That is the crucial question
for Strategy One. We are to focus on the obvious way, or ways,
of rendering the English 'knows whether p' in a given language
'obvious ways', to indicate that we are not to be sidetracked into
subtle distinctions of nuance, references to the ways the words used
in the respective language feature in specialised contexts, and so
on. (We are not interested, for example, in whether the expression
that corresponds to 'knows' in the Arabic for 'He knows whether
it rained yesterday' is ever used in what is sometimes called the
'biblical' sense.) Having thus located the analogue(s) of 'know' in
'know whether', we consider the resources standardly available to
a speaker of the given language for describing someone who, as
we would say, can swim, and we look to see whether they (all
or some of them) include the 'knows' component from the first
group of expressions (or something closely related to it). If at least
one of them does, then the given language has the capacity sense
of 'know'.
Notice that the exact grammatical structure, knows how to +
infinitive, is not our real topic. What we are really talking about
are capacity-attributions which use 'know' or its obvious correspon-
dents in other languages. It is that, and not any grammatical isomor-
phism, that we are to concentrate on.'
1
The nearest German equivalent, for instance, is 'Er weift wie man schwimmt'.
But what this suggests most strongly is that the subject can tell us the theory of
swimming, not that he can swim. Finnish (I am told) also has a construction verbally
close to the English 'knows how to'; but again the import is theoretical. I understand
that Hungarian, on the other hand, uses the word for 'know' with the infinitive
as its standard colloquial way of saying 'can'.
152 SECT. X V I I
So, does the capacity sense of 'know' have status or not? Here
we need the empirical facts. We won't of course be able to get all
the empirical facts; we shall have to be content with (or exhausted
by) the facts about a few particularly prestigious (i.e. well-known)
dead languages and a representative selection of living ones. Living
languages are pretty numerousa few thousand, on any sensible
criterion for distinguishing them from each otherso we must hope
to be representative without having to run our test on more than
a small fraction of them. Our best bet will be to rely on the classifica-
tions of languages agreed among comparative linguists, and hope
to find the resources (which is to say an obliging, educated, expert
speaker) for testing a few members of each of the mam groups.
These tests, so far as I have been able to carry them out, support
the contention that the capacity-ascribing sense, if we can rely on
our intuitive impressions to detect it, is at the least very widespread.
Since the honest policy, and the only one that can lead to a genuinely
robust theory, is to confront directly anything recalcitrant which
looks as if it probably holds, it seems that we should accept that
there is a prima facie capacity-related component of the concept,
and try to explain the fact.
If anyone is determined not to accept it, they will be able to
create a little room for manoeuvre by playing on a methodological
difficulty in the application of this test. I have already hinted at
it: we cannot work solely with formal features of grammar, but
must rely to a considerable extent on the semantic intuitions of native
speakers. There may be languages2 which have expressions that
are close verbal parallels to the English 'know how to' without its
being clear that they have the same meaning. Assume (for the
moment) that the English 'knows how to' is a synonym of 'can';
it might be that in some languages there is a verbal equivalent of
'knows how to' which is semantically informational in content, that
is to say: it is applied only to subjects who can tell us the way
to perform whatever action is in question, or (a very different thing),
the theory of its performance. In that case the mere existence of
the verbal equivalent would not qualify that language for inclusion.
Worse, there may be problems in determining the facts here; it may
not be clear what sense, capacity or informational, a given expression
has. We can experience the sort of difficulties that arise by consider-
2
And so there are: n. 1 above.
SECT. XVII 153
ing the English expression itself, and we can kill two birds with
one stone by doing it in the context of the second of the three possible
strategies.
The second strategy is to deny that 'knows how to' has a capacity-
sense after all. First appearances notwithstanding, we might argue,
it indicates the possession of information about the performance
of the action specified, perhaps even (we shall soon see that this
is a little more) the capacity to give information about it. There
are two broad types of position that such an argument might aim
at. One would be the view that behind any exercise of ability lurks
the deployment of information: if you can swim, that is always
because you know that such and such is the way to swim, and
can apply that knowledge. When, attributing to you the ability to
swim, we say that you 'know how to swim', the word 'know' is
being used with reference to the essential informational background,
the 'knowing that'; and the change from 'that' to 'how' marks the
fact that we are dealing with information of the type which character-
istically enables a certain kind of action. This position is not far
from the doctrine which Gilbert Ryle3 called, and attacked as,
'intellectualism'we shall shortly look to see whether his attack
leaves any of it standing that might be of interest to us.
That short statement is enough to expose the difficulty I referred
to in the previous paragraph. We wanted to be able to decide whether
or not a given language uses its literal equivalent of 'knows how
to' in a capacity-attributing sense. But, supposing for the moment
the truth of the 'intellectualist' doctrine just outlined, how would
we determine whether the word 'know' really referred to the state
of information, or whether it was primarily a way of attributing
the ability? If a capacity is always based on information, how shall
we decide which of the two the expression 'knows how to' really
describes? It looks as if we may have to accept a commitment to
extensive quantities of semantic theory, something which the project
was designed precisely to avoid.
The proper way to deal with this difficulty, I am convinced, is
simply to bypass the argument and confront the least favourable
alternative head on. Where two descriptions of the situation are
possible, and a decision between them, if not arbitrary, would have
to be grounded deep in the theory of meaning, we should treat
3
See G. Ryle, ch. II, esp. pp. 30-2.
154 SECT. XVII
it either under both, or under the one likely to cause our hypothesis
the most trouble. Thus, going back to our first strategy, we can
see that it will cause the more difficulty the more languages are
deemed to operate a capacity sense of their word for 'know'; so
our principle recommends that we decide in favour of the capacity
sense wherever there are not clear reasons to the contrary. In other
words, we should acknowledge our obligations and commit our-
selves to the third strategy. Otherwise we invite the charge of either
ignoring the linguistic facts or trying to bury them under a pile
of controversial semantic theory.
In any case, the train of thought we were considering is set in
motion by an 'intellectualist' account of knowing how, and in view
of that it may well be asked whether it is worth considering at all.
Hasn't intellectualism been refuted? Ryle, one must admit, gave
strong reason for thinking that, when taken in strict generality as
applying to every capacity, it must be false: it leads to infinite regress.
I accept the argument and its conclusiona little casually perhaps,
but that will do no harm in this context since it doesn't settle the
present issue. For what Ryle showed was that some capacities must
just be things which we can simply do when occasion arises. So
intellectualism (i) can be true of at most nearly all our capacities,
and (ii) cannot be supported by any general argument to the effect
that capacity entails information. Goodbut suppose that it were
true of nearly all our capacities, and that that belief rested on common
observation rather than any grasp of logical connection. That would
be compatible with all that the argument from regress proves, but
it would still be quite enough to explain how a concept which origi-
nated in the need for good informants should grow to encompass
abilities as well.
But is it true, even in this weakened form? More exactly, does
being able to tell us how to A go with being able to A, in our
experience, in a sufficient proportion of cases for there to be any
plausibility in the claim that the concept of knowledge spreads from
the former to the latter precisely because of such concomitance?
That must be very doubtful. Remember that when we speak of infor-
mation we are speaking of being a good informant, not of information
in the sense in which cognitive psychology might use the word.
It is clear that the child who can unfailingly find the way home
must in that sense have a great deal of stored information, and the
same holds of anyone who can swim, ride a bicycle, or do anything
SECT. X V I I 155
3 4
Ibid., p. 91.10-12. Ibid., pp. 30-3.
166 APPENDIX
case of any particular utterance there are various things which are
agreed between invariantist and contextualist. They are agreed that
there is a certain thought on which the speaker brings the hearers
to focus; what they disagree on are the roles played, in the process
by which he does this, by the semantics of his words on the one
hand and the pragmatics of the situation on the other. But if they
are agreed on that, there can hardly be any difficulty of principle
for them in agreeing that there may be some expressions which are
characteristically used to fix attention on pretty much the same
thought (here in the sense: the same stringency of standard) in all
contexts of utterance. Whether a particular expression is like this
or not may be contentious; whether 'know' behaves in this way
may be a point of disagreement between myself and Ungerthough
more of that in a moment. But contextualist and invariantist as such
are not committed either to the view that there are, or that there
are not, such expressions.
What can be said is that if some particular expression is agreed
to be of this type, a contextualist account of it will look a somewhat
round about way of reaching the place which an invariantist reaches
in one step. But even if there were simply no hope for the contextua-
list view in such a case there would be no conflict with Semantic
Relativism. For that doctrine, if I understand it correctly, tells us
that where a range of standards is in use the phenomena can be
equally well explained on invariantist and contextualist models. Of
expressions which are not associated with a range but with only
with one, invariant standard, it tells us nothing, leaving us free to
conclude that there the balance tips towards invariantism. But in
any case it is not true that my account of 'know' leaves nothing
that contextualism can usefully do, and that for two reasons.
The first involves an indirect use of the basic contextualist idea.
It is exactly the use that I have already made of it, though not under
that name. What we were considering in Section X (on 'Objectivisa-
tion') was the way in which contexts, constituted by varieties of
need, prospective outcomes, recognitional capacities, levels of ignor-
ance of prevailing conditions, and so on, guide our assessment of
sources of information. And it was a feature of these contexts
principally, the fact that one often does not know what the infor-
mation is needed for, or will later be used for, or what will turn
on its usethat created an obvious role for a concept tied to some
high standard of reliability. So the contextualist has certainly had
APPENDIX 167
his share of the action, even if that action results in a situation for
which, some might think, only the invariantist need apply.
The second point is more straightforward. To say that knowledge
always implies a very high level of reliability is not to say that the
level is always the same. Many everyday claims to knowledge are
allowed to get by although made with thinner support than many
others, which, occurring in different circumstances, are questioned
and even rejected. Looking at the cases in which we are apparently
laxer, the invariantist can say that these knowledge-claims are false
but that we judge according to context what degree of departure
from strict truth is appropriate and acceptable; the contextualist can
say that they may well be true, since context affects the details of
their semantics. On the question which position, if either, is right,
I am happy to follow Unger's powerfully argued recommendation.
My 'practical explication' or 'state of nature' method leads to an
account of the linguistic practice surrounding the word 'know' and
its near relatives; it does not determine how we are to apportion
the underlying mechanics of the practice between invariant semantics
and contextually motivated pragmatics.
REFERENCES