Being in a Position to Know
John Hawthorne
Juhani Yli-Vakkuri
1. Introduction
The concept of being in a position to know is an increasingly popular member of the
epistemologist’s toolkit. Some have used it as a basis for an account of propositional
justification. 1 Others, following Timothy Williamson,2 have used it as a vehicle for
articulating interesting luminosity and anti-luminosity theses. It is tempting to think
that, while knowledge itself does not obey any closure principles, being in a position to
know does. For example, if one knows both p and p ® q, but one dies or gets distracted
before being able to perform a modus ponens on these items of knowledge and for that
reason one never gets to know q, one was still plausibly in a position to know q.3 It is
also tempting to suppose that, while one does not know all logical truths, one is
1
See especially Rosenkranz (2016b, 2018).
2
See Williamson (2000: §4.2) and (e.g.) Smithies (2019: Ch. 7). As far as we can tell, the expression
‘in a position to know’ came to be widely used in general epistemology due to the influence of
Williamson’s discussion of luminosity, while earlier it along with variations on ‘can know’ and ‘is able
to know’ were sometimes used as glosses on the epistemic logician’s ‘K’ operator (see note 5).
3
An anonymous referee points out (in effect) that such examples might well be used to argue that
being in a position to know also isn’t closed under any inference rules: It’s natural to describe a sudden
death scenario with a sentence along the lines of: ‘Since he died before he could perform the inference,
he was not in a position to know its conclusion.’ Fair enough, but such examples don’t strike us as very
decisive. The trouble is that there are many senses of ‘in a position to know’, and to do its job the
proposed counterexamples would have to work not just for some sense (content) ‘in a position to know’
expresses in some context—that’s easy—but for all senses ‘in a position to know’ is capable of
expressing that are plausibly the ones that philosophers who think ‘in a position to know’ has a normal
modal logic have in mind. Certainly there is at least a sense of ‘in a position to know’ and a
corresponding sense of ‘could have known’ in which its true that someone who is just about to perform
a modus ponens but is struck dead before completing the inference was in a position to know and could
have known its conclusion, and any philosopher who is dead set on having a notion of ‘in a position to
know’ that obeys a normal modal logic can insist that it is that sense (or one of those senses) that he or
she has in mind. Cf. Lewis (1976: 150): ‘Whenever the context leaves it open which facts are relevant,
it is possible to equivocate about whether I can speak Finnish’—and likewise about whether the
unfortunate subject was in a position to know or could have known the conclusion of the modus
ponens. And if the philosopher proposing the counterexample fixes the context in a way that leaves no
room for equivocation, the philosopher who claims to have a normal modal logic-obeying notion of ‘in
a position to know’ can fix the context another way and say: ‘We’re talking at cross purposes. Yes,
given his untimely death the subject was not in a position to know the conclusion, but given his general
cognitive capacities the subject was in a position to know the conclusion, and when I say ‘in a position
to know’, I mean in a position to know given the subject’s general cognitive capacities’’.
1
nevertheless in a position to know every logical truth. 4 Putting these temptations
together, we get the view that being in a position to know has a normal modal logic. As
evidence of these temptations, we note that the use of normal epistemic logics is
sometimes justified by glossing their epistemic operators using ‘in a position to know
that’ and synonyms or near synonyms such as ‘can know that’ and ‘able to know that’.5
A recent literature has begun to investigate whether it is a good idea to give in to these
temptations.6 That literature assumes very naturally that one is in a position to know
everything one knows and that one is not in a position to know things that one cannot
know. It has succeeded in showing that, given the modest closure condition that
knowledge is closed under conjunction elimination (or ‘distributes over conjunction’),
being a position to know cannot satisfy the so-called K axiom (closure of being in a
position to know under modus ponens) of normal modal logics. In this paper, we
explore the question of the normality of the logic of being in a position to know in a
more far-reaching and systematic way. Assuming that being in a position to know
entails the possibility of knowing and that knowing entails being in a position to know,
we can demonstrate radical failures of normality without assuming any closure
principles at all for knowledge. (However, as we will indicate, we get further problems
if we assume that knowledge is closed under conjunction introduction.) Moreover, the
failure of normality cannot be laid at the door of the K axiom for knowledge, since the
standard principle NEC of necessitation or modal generalization also fails for being in
a position to know. After laying out and explaining our results, we briefly survey the
coherent options that remain and give some reasons for preferring our favorite option,
which is that ‘one is in a position to know’ is approximately synonymous with ‘one can
know’, and thus it’s no surprise that it doesn’t behave like a necessity operator.
2. The logic
Our investigation of the logic of being in a position to know will be conducted using a
language of propositionally quantified modal logic. The language has an infinite stock
4
By a ‘logical truth’ we mean simply any theorem of the logic characterized by the axioms and rules
under consideration; the notion is syntactic, not semantic (model-theoretic). For semantic notions of
‘logical truth’, the idea that one is in a position to know every logical truth has is no more plausible
than the idea that set of ‘logical truths’, in the relevant semantic sense, is axiomatizable (in some cases
it is, in others it isn’t). Part of the appeal of the idea that one is in a position to know all logical truths,
we take it, comes from the idea that one could in principle prove any of them in a finite number of
steps.
5
For example, Berto and Hawke (2018: 6) use ‘in a position to know’ to gloss the epistemic operators
when describing what they call ‘[t]he standard approach to (multi-agent) epistemic logic’. See also
Hilpinen’s (1970) use of ‘in a position to know that’ and Williamson’s (1990: 5-10) use of ‘the subject
is able to activate knowledge that’. Admittedly some characterizations of being in a position to know
are less suggestive of a normal modal logic: notably Williamson’s (2000: 95) remark that what one is
in a position to know is ‘open to one’s view, unhidden, even if one does not yet see it’ could easily be
taken to suggest that a subject lacking in logical acumen is not able to know some logical consequences
of what he knows due to poor eyesight. In what follows we hope to offer more decisive objections to
the normality assumption. (Thanks to an anonymous referee for discussion here.)
6
See See Heylen (2016) and Rosenkrantz (2016a, 2016b, 2018: 317–8).
2
of atomic sentences, an infinite stock of propositional variables p, q, r, …, the standard
truth-functional connectives, a universal quantifier "p for each propositional variable
p, the propositional operators KP (‘one is in a position to know that’), K (‘one knows
that’), £ (‘necessarily’), @ (‘actually’), and the usual formation rules and
metalinguistic abbreviations (thus ¯𝜑 is ¬£¬𝜑 and $p𝜑 is ¬"p¬𝜑). We will assume
the logic characterized by the following rules and axioms, where the final two are
standard axioms for the logic of actuality.7
(Taut)
All tautologies.
(MP)
Modus ponens.
(UG)
If ⊢ 𝜑 ® 𝜓, then ⊢ 𝜑 ® "p𝜓, where p is not free in 𝜑.
(UI)
"p𝜑 ® 𝜑[𝜓/p], where p is free for 𝜓 in 𝜑
(TK)
K𝜑 ® 𝜑
(KP/¯)
KP𝜑 ® ¯K𝜑
(K/KP)
K𝜑 ® KP𝜑
(T£)
£𝜑 ® 𝜑
(K£)
£(𝜑 ® 𝜓) ® (£𝜑 ® £𝜓)
(NEC£)
If 𝜑 is provable using only the axioms and rules already specified,
then ⊢ £𝜑.
(T@)
@𝜑 ® 𝜑
(RIG@)
𝜑 ® £@𝜑
We will call this logic ‘L’, and will also write ‘⊢ 𝜑’ for ‘𝜑 Î L’.8
Here are the further candidate axioms and rules that will be discussed below.
(KKP)
KP(𝜑 ® 𝜓) ® (KP𝜑 ® KP𝜓)
(NECKP)
If ⊢ 𝜑, then ⊢ KP𝜑
7
See Crossley and Humberstone (1977). In higher-order logic, these axioms can be derived from the
axiom of choice: see Goodsell and Yli-Vakkuri (2020).
8
Here is a further axiom that may strike some readers as plausible, which says that being in a position to
know is factive:
(TKP) KP𝜑 ® 𝜑
(TKP), however, plays no role in our results. We also lean towards the view that (TKP) has false instances,
and that its apparent factivity is the result of something like a presupposition: see §4.
3
(DISTK)
K(𝜑 Ù 𝜓) ® (K𝜑 Ù K𝜓)
We will call the logic that results from adding any axioms or rules X1, …, Xn to
L ‘L + X1 + … + Xn’, and we will write ‘X1, …, Xn ⊢ 𝜑’ when L + X1 + … + Xn
includes 𝜑. We will say that the logic of a propositional operator P is normal in L + X1
+ … + Xn when L + X1 + … + Xn includes
P(𝜑 ® 𝜓) ® (P𝜑 ® P𝜓)
(KP)
and is closed under the rule
(NECP)
If ⊢ 𝜑, then ⊢ P𝜑.
Thus, for example, the logic of KP in L + (KKP) + (NECKP) is normal, and L + (KKP) +
(NECKP) ⊢ KP(K𝜑 ® 𝜑), since ⊢ K𝜑 ® 𝜑.
3. Our main results
In this section we will present our main results informally. The formal proofs are
included in the Appendix.
The sentence
a:
"p(p « @p)
will play a starring role in our discussion. a has two interesting features. First, a is a
logical truth. Second, the truth a expresses is extremely modally fragile: if things had
been different in any way, it would have been false. After all, a says that everything is
as it actually is, and if things had been different in any way, things would not have been
as they actually are. Owing to this fragility, one only gets one shot, so to speak, at
knowing a.9,10 If one doesn’t know a, then it is impossible for one to know a, since if
one doesn’t know a, then, if one had known a, things would have been different than
they actually are and a would have been false. Thus, if one doesn’t know a, then one
could not have known a. Ditto for the conjunction of a with anything else: if one
doesn’t know a Ù 𝜑, then one cannot know a Ù 𝜑. Formally:
(i)
⊢a
(ii)
⊢ ¬Ka ® ¬¯Ka
9
When we say that one ‘knows 𝜑’, where 𝜑 is a sentence, we mean that one knows the proposition
expressed by 𝜑 —that is, we mean what is formalized by K𝜑. The sentence a, of course, would have
expressed a truth as long as it had its actual character (in the sense of Kaplan 1989), but it would not
have expressed the same truth as it actually does.
10
One should read our ‘only gets one shot’ in a modal rather than temporal way: in the temporal sense,
if one doesn’t know a, one always gets another shot at knowing a (later), but it is always the case that
if one doesn’t know a, it is impossible for one to know a.
4
(iii)
⊢ ¬K(a Ù 𝜑) ® ¬¯K(a Ù 𝜑)
Let us say that one conjunctively knows iff one knows the conjunction of 𝜑 and
some proposition ($pK(p Ù 𝜑), abbreviated as KÙ𝜑11). Our first result is:
(1)
(KKP) + (NECKP) ⊢ KP𝜑 « KÙ𝜑
That is: If the logic of being in a position to know is normal, then one is in a position
to know something if and only if one conjunctively knows it.
Informal argument: Suppose that the logic of being in a position to know is
normal, so that one is in a position to know all logical truths (that is, suppose (NECKP)),
and being in a position to know the premises of a modus ponens entails being in a
position to know its conclusion (that is, suppose (KKP)). Let 𝜑 be anything one is in a
position to know. Since a is a logical truth, so is 𝜑 ® (a Ù 𝜑). By (NECKP), then, it
follows that one is in a position to know 𝜑 ® (a Ù 𝜑), and by (KKP), that one is in a
position to know a Ù 𝜑. Since one is in a position to know something only if it is
possible for one to know it (i.e., by (KP/¯)), it is possible for one to know a Ù 𝜑. But,
as we just saw in the previous paragraph (by (iii)), it is possible for one to know a Ù 𝜑
only if one actually knows a Ù 𝜑, and so only if one actually knows the conjunction of
𝜑 with something. It follows that if one is in a position to know 𝜑, then one
conjunctively knows 𝜑. What we have just seen is that, if the logic of being in a position
to know is normal, then if one is in a position to know something, one conjunctively
knows it. We can also use normality to establish that one is in a position to know
anything one conjunctively knows. If one knows a Ù 𝜑, one is in a position to know a
Ù 𝜑. Since (a Ù 𝜑) ® 𝜑 is a logical truth, one is in a position to know it (by the (NECKP)
part of normality). Then one is in a position to know 𝜑 (by the (KKP) part of normality).
Assuming normality for being in a position to know we have now established both
directions of the biconditional: one is in a position to know something if and only if one
conjunctively knows it. This is result (1).
An immediate consequence of (1) is (2):
(2)
The logic of KÙ in L + (KKP) + (NECKP) is normal.
That is: if the logic of being in a position to know is normal, then the logic of
conjunctive knowledge is normal.
What happens if the logic of being in a position to know is normal and
knowledge distributes over conjunction, in the sense that one who knows a conjunction
knows each conjunct? The result is arguably even more disturbing than (1). By (1) it
already follows that one is in a position to know something if and only if one
conjunctively knows it. If knowledge furthermore distributes over conjunction, then
11
In order to smoothen the presentation, we will pretend that KÙ is an operator in the language. KÙ is an
operator in the standard loose sense in which a formula like T(x) in one free variable is a predicate. If
we had l-expressions in the language we could avoid this loose talk by proving things about the
operator lp.$qK(q Ù p), but that would require adding the axioms of the l-calculus to the logic and
applying them in the derivations in the Appendix—a significant increase in complexity with little
payoff.
5
one knows everything one conjunctively knows, and it follows that one is in a position
to know something if and only if one knows it. This is our third main result:
(3)
(KKP) + (NECKP) + (DISTK) ⊢ KP𝜑 « K𝜑
Our fourth main result states an obvious corollary:
(4)
The logic of K in L + (KKP) + (NECKP) + (DISTK) is normal.
That is: if the logic of being in a position to know is normal and knowledge distributes
over conjunction, then the logic of knowledge is normal.
Let us next see what we can show about the individual components of a normal
modal logic for being in a position to know, (NECKP) and (KKP), beginning with the
former.
Suppose, as (NECKP) states, that one is in a position to know every logical truth.
Let l be an arbitrary logical truth. It follows that a Ù l is a logical truth, and so that
one is in a position to know a Ù l, and so that it is possible for one to know a Ù l. But,
once again, it is only possible for one to know a Ù l if one actually knows a Ù l, and
so actually conjunctively knows l. It follows that, if one is in a position to know every
logical truth, then one conjunctively knows every logical truth. This is our fifth main
result:
(5)
L + (NECKP) is closed under (NECKÙ).
Our sixth main result states the obvious corollary:
(6)
L + (NECKP) + (DISTK) is closed under (NECK).
That is: if one is in a position to know every logical truth and knowledge distributes
over conjunction, then one knows every logical truth.
Let us finally turn to the hypothesis that being in a position to know is closed
under modus ponens (that is, (KKP)). That is, if one is in a position to know the premises
of a modus ponens, then one is in a position to know its conclusion. Our seventh main
result is:
(7)
(KKP) ⊢ KP(𝜑 ® 𝜓) ® (KP𝜑 ® ¯K𝜓)
This amounts to the observation that this entails that, if one is in a position to know the
premises of a modus ponens, then—because being in a position to know entails the
possibility of knowing—it is possible for one to know its conclusion.
The example of a serves as a useful reminder of how problematic (7) is.
Replacing 𝜓 with a Ù 𝜑 in (7), we get:
(!??)
KP(𝜑 ® (a Ù 𝜑)) ® (KP𝜑 ® ¯K(a Ù 𝜑)).
There are two main ways to generate counterexamples to (!??). First, insofar as we are
willing to countenance any unknown logical truths at all, we should accept that there
6
are cases in which one is in a position to know 𝜑 ® (a Ù 𝜑) as well as 𝜑 but one does
not know either 𝜑 ® (a Ù 𝜑) or a Ù 𝜑. The details can be filled in in a variety of
plausible ways. Perhaps 𝜑 is some humdrum truth (such as ‘John and Juhani had lunch
at Scott’s Seafood Restaurant in Mayfair on December 7th, 2019’) that one knows, and
therefore is in a position to know, but, although one is in a position to know 𝜑 ® (a Ù
𝜑), one has never considered the issue, and for that reason one knows neither 𝜑 ® (a
Ù 𝜑) nor a Ù 𝜑. Second, and perhaps even more decisively, consider someone who
knows, and so is in a position to know, the logical truth 𝜑 ® (a Ù 𝜑) but is merely in a
position to know 𝜑, knowing neither 𝜑 nor a Ù 𝜑. Both kinds of case are
counterexamples to (!??): by (!??), if one is in a position to know both 𝜑 and 𝜑 ® (a Ù
𝜑), it is possible for one to know a Ù 𝜑, which in turn entails that one does know a Ù
𝜑.12
4. Alternatives
We have seen that the principle that knowledge entails being in a position to know and
that being in a position to know entails possibly knowing, which are axioms of our
logic:
(K/KP)
K𝜑 ® KP𝜑
(KP/¯)
KP𝜑 ® ¯K𝜑
produce disastrous results when combined with either component of a normal logic for
being in a position to know along with a minimal logic of necessity and actuality. We
see two main lines of retreat, one of which we find clearly preferable to the other.
But first we will briefly address a general methodological concern that a number
of people have expressed to us in conversation. The concern is that we do not offer a
‘semantics’, specifically a ‘possible-worlds semantics’, for our logic. Some
contemporary philosophers are so enamored of the set-theoretic apparatus introduced
by Kripke for investigating modal logics13 that they will only accept our proofs—and
in particular our proof of a14—after we have given them a possible-worlds semantics
on which they are valid. The demand is misguided. The practice of using possibleworlds semantics for justifying axioms and rules of inference is itself badly in need of
justification, and it is justified by axioms and rules of inference like those of L if it is
justified by anything: see Dorr, Hawthorne, and Yli-Vakkuri (2021: Ch. 0) for a sketch
of how such a justification might go.
The first line of retreat involves rejecting axiom (KP/¯), according to which
being a position to know entails the possibility of knowing, and accepting that being in
12
The formal result that underwrites these remarks is this:
(KKP) ⊢ KP(𝜑 ® (a Ù 𝜑)) ® (KP𝜑 ® K(a Ù 𝜑)).
13
See Kripke (1959) and (1963).
14
Lines 13–14 in the proof of (3) in the Appendix. On our view the axioms of actuality derive their
ultimate justification from the higher-order axiom of choice. See Goodsell and Yli-Vakkuri 2020 for
relevant results.
7
a position know has a normal modal logic. 15 Here is one natural way to develop this
thought: The logic of being in a position to know is normal, and one is in a position to
know a proposition if and only if one knows it under an idealization to perfect
rationality. It need not be metaphysically possible for that idealization to hold, and,
under a metaphysically impossible idealization, one may know some propositions
without knowing all propositions.16 This way of thinking may have some precedent in
15
There are precedents for giving up (KP/¯) in the work of David Chalmers and Declan Smithies.
We have in mind Chalmers’ proposal in Constructing the World regarding ‘ideal a priori
warrant’: in some cases ‘there exists an (ideal a priori) warrant for believing p even though the warrant
cannot be used to know p.’ Chalmers’ example, too, involves the actuality operator. However, he has in
mind a sentence that expresses a proposition necessarily equivalent to the proposition that snow is
white in any world (holding its character fixed) in which snow is white:
When p is the proposition expressed by the semantically fragile sentence S discussed at the
end of the third excursus (‘Snow is white iff actually snow is white’), one can argue that there
exists a proof of p even though it is impossible to use it to prove p. In particular, there exists
an abstract proof of S using the logic of ‘actually’. S expresses p in the actual world, so this
abstract proof of S is also an abstract proof of p. But if one were to use the proof to prove S, S
would express p' rather than p, so one would not prove p (Chalmers 2012: 93).
Smithies’ idea is that ‘being in an epistemic position to know’ is can be a ‘finkish’ disposition
to know under certain circumstances—a disposition that would disappear if those circumstances
obtained:
As a rough heuristic, you’re in an epistemic position to know that p just in case you would
know that p if your doxastic response to your epistemic position were sufficiently rational.
More precisely, you would know that p if you were to properly base a doxastically justified
belief that p on your propositional justification to believe that p. This is only a heuristic
because there are finkish cases in which you cannot respond rationally to your epistemic
position without thereby changing it (Smithies 2019: 349-350).
The paradigm cases of finkish dispositions to A in circumstances C discussed in the literature, however,
are all ones in which it is metaphysically possible for the bearer of the disposition to A in C: see Lewis
(1997) and Yli-Vakkuri (2010).
Note that Smithies’ picture seems to be one on which being in a position to know requires
being in fact propositionally justified. This fits poorly we the kind of externalism we favor, one
according to which one might be disposed to form a safe belief in 𝜑 and thus be in a position to know
but currently have no evidence for 𝜑 and, thus, arguably, no propositional justification.
16
Another way to develop the thought is to go for something along the following lines: Being in a
position to know that 𝜑 does not entail possibly knowing 𝜑 but rather possibly knowing some
proposition suitably similar to 𝜑. But this strategy seems even less promising. It is very much out of
step with how the concept of being in a position to know is used it the literature, and we have no good
idea of how to develop this thought in a systematic and satisfying way.
Yet another strategy borrows an idea from the literature on Fitch’s paradox: being in a
position to know 𝜑 entails possibly knowing @𝜑 (see Edgington 1985 and Schlesinger 1985: 103-6).
Our concerns about this idea very much mirror Williamson’s concerns about the Edgington/Schlesinger
proposal (see Williamson 2000: 292-5). One concern is that, on a fine-grained conception of
propositions it is extremely difficult to know @𝜑 in a counterfactual situation, because, while @ is a
convenient guise for singling out the actual world in the actual world, there is no convenient guise for
singling out the actual world in counterfactual situations. Of course, on a coarse-grained conception of
propositions (according to which necessarily equivalent propositions are identical) this problem doesn’t
arise, but then @𝜑, if true, will be identical to every necessary truth—there being only one—and the
8
natural and social science. An idealization to frictionless surfaces may have an
explanatory point even if frictionless surfaces are metaphysically impossible. And
perhaps an idealization to market economies that are free of certain ‘noise’ has a point
even if such economies are metaphysically impossible. Similarly, for example, one who
does not know a logical truth containing a may know a under an idealization to perfect
rationality even when it is metaphysically impossible for one to know a. Drop the
assumption that being in a position to know entails the possibility of knowing, and
weaken our logic by dropping the corresponding axiom (KP/¯), and you will no longer
face the logical problems we have presented for far.
The reader should not underestimate the difficulties here, however. The problem
with this proposal is that it doesn’t allow one to know that one doesn’t know a particular
logical truth. As long as the logic of being in a position to know is normal—indeed, as
long as (NECKP) holds, with or without (KKP) or (KP/¯)—we have the following result.
(12)
If ⊢ 𝜑, then ⊢ K¬K𝜑 ® (KP𝜑 Ù KP¬K𝜑).17
By (12), whenever 𝜑 is a logical truth one knows oneself to not know, one is in a
position to know 𝜑 and one is in a position to know that one doesn’t know 𝜑. This is
inconsistent with the conception of being in a position to know we are entertaining,
since, according to that conception, the truth of
KP𝜑 Ù KP¬K𝜑
amounts to the truth of the inconsistent
K𝜑 Ù K¬K𝜑
under an idealization to perfect rationality. That idealization may be metaphysically
impossible, but it is not (or so its advocates should hope) inconsistent.
Note that this is not only a problem for the ‘idealization to perfect rationality’
conception of being in a position to know. As long as the logic of being in a position to
know is normal, we can prove, even without (KP/¯):
(13)
If ⊢ 𝜑, then ⊢ K¬K𝜑 ® KP(𝜑 Ù ¬K𝜑).
new principle would be a terrible surrogate for the old one, since it will say that being in a position to
know a fact entails knowing the necessary truth. There is a lot more to say about how coarse-grained
conceptions of propositions interact with the logic of @, but we won’t say it here: see Hawthorne and
Yli-Vakkuri (in preparation) for more.
A third way to develop the thought is to say that being in a position to know 𝜑 entails the
possibility of knowing not a proposition similar to the proposition actually expressed by @𝜑 but a
proposition under some guise one associates with the sentence @𝜑. On one version of this proposal the
guise could is just that sentence itself, on another it is the Kaplanian character of @𝜑, on a third it is
the primary intension of 𝜑, on a fourth it is the language of thought sentence one correlates with 𝜑.
17
Suppose K¬K𝜑 and ⊢ 𝜑. By (K/KP), KP¬K𝜑. Since ⊢ 𝜑, by (NECKPK), KPK𝜑.
9
By (13), being in a position to know that one does not know a logical truth 𝜑 implies
being in a position to know the conjunction: 𝜑 and one does not know 𝜑.
The idealization picture suggests a further principle:
(NECKPK)
If ⊢ 𝜑, then ⊢ KPK𝜑.
The idea is that one not only knows but knows that one knows each logical truth under
an idealization to perfect rationality. But if we have (NECKPK), we certainly cannot have
a normal logic for being in a position to know. If we had both, we could prove, again
without (KP/¯):
(14)
If ⊢ 𝜑, then ⊢ K¬K𝜑 ® KP^.
That is, we would get the result that one is in a position to know that one does not know
a logical truth only if one is in a position to know a contradiction.
We see no happy way out of these problems.
The second line of retreat, which we endorse, involves giving up both
components of normality for being in a position to know: (NECKP) and (KKP). This
approach is easy to motivate once one notices that ‘in a position to know’ is a
composition of two operators, ‘in a position to’ and ‘know’.
We begin with the former. We propose that ‘x in a position to F’ is
approximately synonymous with ‘x can F’ or ‘x is able to F’. The standard of
synonymy we have in mind here is roughly sameness of Kaplanian character. It’s not
that these expressions have the same semantic content in a context-independent way,
but rather that they are capable of expressing approximately the same range of
semantic contents in different contexts—plausibly, including varieties of deontic
possibility. It’s not hard to imagine contexts in which
You are not in a position to say what she just said
and
You can’t say what she just said
both seem to express the same true proposition—for example, the proposition that, in
view of a’s legal obligations, a can’t say what b just said, where a is the person being
addressed and b is the person referred to by ‘she’.18 Thus, ‘one in a position to know
that’ has approximately the same context-independent meaning (character) as
lp.¯Kp,
with ¯ interpreted as expressing the modality determined by the context of speech or
writing. When the modality is deontic, axiom (K/KP) (K𝜑 ® KP𝜑) will not hold, but L
is a good enough logic of ‘in a position to know’, since the senses of ‘in a position’ that
18
See Lewis (1976) and Kratzer (1977) on the flexibility of ‘can’ and its relation to ‘in view of … can’
and related constructions like ‘given … can’.
10
are at issue in epistemology when ‘in a position’ combines with ‘know’ are never
deontic, as far as we can tell.
Note what follows: even in non-deontic senses of ‘in a position to’, being in a
position to know is not factive. And indeed it is easy to find examples of its nonfactivity. For example, each of us is in a position to thump this table. And each of us
is also in a position to thump this table while knowing that he is thumping it. But then,
since neither of us in fact thumps this table, each of is in a position to know something
(namely, that he is thumping this table) that is actually false.
Why is the view that ‘in a position to know’ is factive so tempting? Note first
that ‘can know’ displays the same appearance of factivity. One does not normally
say—at least without further elaboration—either that one ‘can know’, ‘could know’
(etc.) or that one ‘cannot know’, ‘could not have known’ (etc.) 𝜑 unless one takes 𝜑
to be true. Consider:
John could not have known that we were out of cocktail olives.
One does not even normally assert a subjunctive conditional with ‘might/could have
known 𝜑’ either in the consequent or the antecedent unless one takes 𝜑 to be true:
If you had read the emails, you might have known that the meeting was
rescheduled.
If I could have known that the meeting was rescheduled, my lateness would not
have been excusable.
We are not looking at a special feature of ‘in a position to know’ but at a special
feature of factive verbs, such as ‘realize’, ‘learn’, and ‘acknowledge’. The result of
combining any of these with either a possibility modal or ‘in a position to’ is an
operator that appears to be factive and moreover ‘projects’ this appearance of factivity
out of a variety of constructions when embedded. It’s natural enough to describe this
kind of ‘projected’ content as a ‘presupposition’. In fact, the term ‘factive’ was
originally introduced for the kind of presupposition triggered by ‘know’ and other
factive verbs, and that this class of verbs is the paradigm of the presupposition trigger
in the literature.19
19
In a survey article, David Beaver and Bart Geurts begin their (non-alphabetic) list of ‘lexical classes
widely agreed to be presupposition triggers’ with:
•
factives (Kiparsky and Kiparsky, 1970)
Berlusconi knows that he is signing the end of Berlusconism.
→ Berlusconi is signing the end of Berlusconism (Beaver and Geurts 2011).
Here they are citing the article that initiated the study presupposition in linguistics, Paul Kiparsky and
Carol Kiparsky’s ‘Fact’, which in turn begins:
The object of this paper is to explore the interrelationship of the English complement system.
Our thesis is that the choice of complement type is in large measure predictable from a
number of basic semantic factors. Among these we single out for special attention
presupposition by the speaker that the complement of the sentence expresses a true
proposition (Kiparsky and Kiparsky 1971: 345, emphasis in the original).
11
Now we see the solution to the logical mystery: Of course being in a position to
know does not have a normal modal logic. Its logic is exactly that of lp.¯Kp (‘one can
know’), with ¯ restricted by some condition. For a variety of restricting conditions and
a variety of propositions p, one can know p and can know ¬p, but one can never know
p Ù ¬p, as would be required by a normal modal logic.
Philosophers are of course free to use ‘in a position to know’ as a term of art,
and we have nothing to say about the myriad meanings they might assign to that term,
but we do have word of warning about this approach. Philosophers who take it face two
main dangers if they are not very explicit about what they mean by the term of art and
what its logic is. The first danger is that the claims they make using the term of art may
be too vague to be interesting. The second danger is that, unless there are very specific
guidelines about the use of the term, its users will tend to fall back on their
understanding of the ordinary English ‘in a position to know’.
Kiparsky and Kiparsky introduce the phenomenon using ‘Two syntactic paradigms’, two lists of
‘predicates which take sentences as their subjects’, labelled ‘Factive’ and ‘Non-factive’ (ibid.). Lauri
Karttunen’s (1971) classic paper, in which he argues, contra Kiparsky and Kiparsky, that the choice of
sentential complement-taking verb does not alone determine whether the speaker presupposes the truth
of the complement, begins:
There is a class of verbs that are commonly called ‘factive’ verbs. …
There is a general agreement that factive verbs involve presuppositions, though it seems that
nobody quite understands what we mean by the term ‘presupposition’ (Karttunen 1971: 55).
And ‘the hallmark of presuppositions, as well as the most thoroughly studied presuppositional
phenomenon, is projection’ (Beaver and Geurts 2011)—projection of the kind we saw examples of in
the main text.
12
Appendix: Proofs of the main results
We begin with
(3) (KKP) + (NECKP) + (DISTK) ⊢ KP𝜑 « K𝜑.
Proof. Below is an abbreviated derivation of KP𝜑 « K𝜑 in L + (KKP) + (NECKP) +
(DISTK), where ‘K’ (boldface) indicates provability using classical logic and (NEC£)
from (K£), and
a
:=
"p(p « @p)
a*
:=
¬K(a Ù 𝜑)
1. a* ® £@a*
(RIG@)
2. a* ® £("p(p « @p) ® (a* « @a*)) (UI), (NEC£)
3. a* ® £((a* « @a*) « a*)
1, K
2, 3, K
4. a* ® £("p(p « @p) ® a*)
5. a* ® £(a ® a*)
4 abbreviated
6. ¬K(a Ù 𝜑) ® £(a ® ¬K(a Ù 𝜑)), 5, with a* unabbreviated
7. ¬K(a Ù 𝜑) ® £((a Ù 𝜑) ® ¬K(a Ù 𝜑)), 6, K
(TK), (NEC£)
8. £(K(a Ù 𝜑) ® (a Ù 𝜑))
9. ¬K(a Ù 𝜑) ® £¬K(a Ù 𝜑)
7, 8, K
9, K
10. ¬K(a Ù 𝜑) ® ¬¯K(a Ù 𝜑)
10
11. ¯K(a Ù 𝜑) ® K(a Ù 𝜑)
12. ¯K(a Ù 𝜑) ® K𝜑
11, (DISTK)
13. p « @p
(T£), (T@) (RIG@)
14. "p(p « @p)
13, (UG)
P
14, (NECKP)
15. K (𝜑 ® (a Ù 𝜑))
P
P
16. K 𝜑 ® K (a Ù 𝜑)
15, (KKP)
(KP/¯)
17. KP(a Ù 𝜑) ® ¯K(a Ù 𝜑)
18. KP𝜑 ® ¯K(a Ù 𝜑)
16, 17
P
12, 18
19. K 𝜑 ® K𝜑
20. K𝜑 ® KP𝜑
(K/KP)
19, 20
21. KP𝜑 « K𝜑
13
(1) (KKP) + (NECKP) ⊢ KP𝜑 « KÙ𝜑
Proof. To show this, replace line 12 in the above with K(a Ù 𝜑) ® $pK(p Ù 𝜑), where
p is not free in 𝜑 (which is equivalent by contraposition and quantifier duality to an
instance of (UI)), and proceed:
11. ¯K(a Ù 𝜑) ® K(a Ù 𝜑)
12. K(a Ù 𝜑) ® $pK(p Ù 𝜑)
UI
13. p « @p
(T£), (T@) (RIG@)
14. "p(p « @p)
13, (UG)
14 abbreviated
15. a
16. 𝜑 ® (a Ù 𝜑))
15
16, (NECKP)
17. KP(𝜑 ® (a Ù 𝜑)))
(KKP)
18. KP𝜑 ® KP(𝜑 Ù a)
19. KP(𝜑 Ù a) ® ¯K(𝜑 Ù a)
(KP/¯)
P
20. K 𝜑 ® $pK(p Ù 𝜑)
11, 12, 18, 19
P
(K/KP)
21. K(p Ù 𝜑) ® K (p Ù 𝜑)
(NECKP)
22. KP((p Ù 𝜑) ® 𝜑)
P
P
P
23. K ((p Ù 𝜑) ® 𝜑) ® (K (p Ù 𝜑) ® K 𝜑) (KKP)
22, 23
24. KP(p Ù 𝜑) ® KP𝜑
P
25. K(p Ù 𝜑) ® K 𝜑
21, 24
P
25
26. ¬K 𝜑 ® ¬K(p Ù 𝜑)
27. ¬KP𝜑 ® "p¬K(p Ù 𝜑)
26, (UG)
27
28. ¬"p¬K(p Ù 𝜑) ® KP𝜑
P
28, definition of $p
29. $pK(p Ù 𝜑) ® K 𝜑
30. $pK(p Ù 𝜑) « KP𝜑
29, 20
(2) The logic of KÙ in L + (KKP) + (NECKP) is normal.
Proof. Immediate from (1).
(4) The logic of K in L + (KKP) + (NECKP) + (DISTK) is normal.
Proof. Immediate from (3).
14
(5) L + (NECKP) is closed under (NECKÙ).
Proof. To show this, let (NECKP) ⊢ 𝜑 and continue from line 11 in the proof of (3) as
follows.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
¯K(a Ù 𝜑) ® K(a Ù 𝜑)
p « @p
"p(p « @p)
𝜑
𝜑Ùa
KP(𝜑 Ù a)
KP(𝜑 Ù a) ® ¯K(𝜑 Ù a)
K(a Ù 𝜑)
K(a Ù 𝜑) ® $pK(p Ù 𝜑)
$pK(p Ù 𝜑)
(T£), (T@) (RIG@)
12, (UG)
By hypothesis
13, 14
15, (NECKP)
(KP/¯)
11, 16, 17
UI and quantifier duality
18, 19
(6) L + (NECKP) + (DISTK) is closed under (NECK).
Proof. Immediate from the proof of (5).
(7) (KKP) ⊢ KP(𝜑 ® 𝜓) ® (KP𝜑 ® ¯K𝜓)
Proof. Immediate from axiom (KP/¯).
References
Beaver D. and B. Geurtz (2011). ‘Presupposition’, Stanford Encyclopedia of Philosophy,
https://plato.stanford.edu/entries/presupposition/
Chalmers, D. (2012). Constructing the World. Oxford: OUP.
Crossley, J. N., and Humberstone, L. (1977). ‘The logic of ‘actually’’, Reports on Mathematical Logic
8: 11–29.
Dorr, C., J. Hawthorne, and J. Yli-Vakkuri (2021). The Bounds of Possibility: Puzzles of Modal
Variation. Oxford: Oxford University Press.
Goodsell, Z. and J. Yli-Vakkuri (2020). ‘Higher-Order Logic as Metaphysics’. Unpublished
Manuscript.
Edgington, D. (1985). ‘The paradox of knowability’, Mind, 94: 55–68.
Francesco Berto and Peter Hawke (2018). ‘Knowability Relative to Information’, Mind, published
online-first: https://doi.org/10.1093/mind/fzy045
Hawthorne, J. and J. Yli-Vakkuri (in preparation). ‘Intensionalism and Propositional Attitudes’.
Heylen, J. (2016). ‘Being in a position to know and closure’, Thought, 5: 63–67.
Hilpinen, R. (1970). ‘Knowing That One Knows and the Classical Definition of Knowledge’ Synthese, 21,
109-132.
Karttunen, L. (1971). ‘Some observations on factivity’, Papers in Linguistics, 4: 55–69.
Kiparsky, P. and C. Kiparsky (1971). ‘Fact’, in M. Bierwisch and K. E. Heidolph, eds, Progress in
Linguistics. The Hague: Mouton, 1971. Reprinted in L. Jakobovits and D. Steinberg, eds.,
Semantics: An Interdisciplinary Reader. Cambridge: Cambridge University Press, 1976. Page
references to the latter.
Kripke, S. (1959). ‘A completeness theorem in modal logic’, Journal of Symbolic Logic, 24: 1–14.
Kripke, S. (1963). ‘Semantical analysis of modal logic I: Normal modal propositional palculi’,
15
Zeitschrift für Mathematische Logik und Grundlagen der Mathematik, 9: 68–96.
Kratzer, A. (1977). ‘What ‘must’ and ‘can’ must and can mean’, Linguistics and Philosophy, 1, 337–
355.
Kaplan, D. (1989). ‘Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and
Epistemology of Demonstratives and Other Indexicals’, in J. Almog, J. Perry and H. Wettstein,
eds., Themes from Kaplan. Oxford: OUP.
Lewis, D. (1976). ‘The paradoxes of time travel’, American Philosophical Quarterly, 13: 145–52.
Lewis, D. (1997). ‘Finkish dispositions’, Philosophical Quarterly, 47: 143–158.
Rosenkranz, S. (2018). ‘The structure of justification’, Mind, 127: 309–338
Rosenkranz, S. (2016a). ‘Being in a Position to Know and Closure: Reply to Heylen’, Thought, 5: 68–
72.
Rosenkranz, S. (2016b). ‘Seemings and Inexact Knowledge’, manuscript.
Schlesinger, G. N. (1985). The Range of Epistemic Logic. Aberdeen: Aberdeen University Press.
Smithies, D. (2019). The Epistemic Role of Consciousness. Oxford: OUP.
Williamson, T. (1990). Identity and Discrimination. Oxford: Blackwell.
Williamson, T. (2000). Knowledge and Its Limits. Oxford: OUP.
Williamson, T. (2003). ‘Everything’, Philosophical Perspectives, 17, 415–465.
Yli-Vakkuri, J. (2010). ‘Habitual and conditional analyses of disposition ascriptions’, Philosophical
Quarterly, 60: 664–670.
16