Philosophy Compass 5/1 (2010): 1628.

Russell, His Paradoxes, and Cantors Theorem: Part I

Kevin C. Klement

University of Massachusetts
Abstract
In these articles, I describe Cantors pover-class theorem, as vell as a number ot logical and
philosophical paradoxes that stem trom it, many ot vhich vere discovered or considered (implic-
itly or explicitly, in Bertrand Pussells vork. 1hese include Pussells paradox ot the class ot all
classes not members ot themselves, as vell as others involving properties, propositions, descriptive
senses, class-intensions, and equivalence classes ot coextensional properties. Part I tocuses on Can-
tors theorem, its proot, hov it can be used to manutacture paradoxes, lreges diagnosis ot the core
diculty, and several broad categories ot strategies tor oering solutions to these paradoxes.
1. Introduction
In 1961, W. V. Quine described a philosophical method he dubbed 1he Ways ot Para-
dox. It begins vith a seemingly vell-reasoned argument leading to an apparently ab-
surd conclusion. It continues vith caretul scrutiny ot the reasoning involved. It ve are
ultimately unvilling to accept the conclusion as justied, the process may end vith the
conclusion that, some tacit and trusted pattern ot reasoning must be made explicit and be
hencetorvard avoided or revised (Ways 11,. 1his method pertectly describes Bertrand
Pussells philosophical vork, especially trom 1901 to 1910, vhile composing Principia
Mathematica.
Pussell is most closely associated vith the class-theoretic antinomy bearing his name:
the class ot all those classes that are not members ot themselves vould appear to be a
member ot itselt it and only it it is not. 1his is one ot a large collection ot paradoxes
Pussell discovered or considered that shaped his subsequent philosophy. Many, it not
most, stem trom violations ot Cantors poverclass theorem, the result that every class
must have more subclasses than members. 1ogether, they led Pussell to be increasingly
vary, not only ot implicit reasoning involving class existence, but also ot the very practice
ot taking apparent reterence to mathematical, abstract, or logically complex things at
tace value.
In this, the rst article in a series ot tvo, ve discuss Cantors poverclass theorem, and
hov it can be used to generate paradoxes. We then summarize a number ot paradoxes
thereby generated, either explicitly or implicitly considered by Pussell himselt. We con-
clude vith a briet summary ot the various kinds ot solutions they might be given. In the
sequel article, the impact ot these paradoxes on Pussells ovn philosophy, and his vievs
about their proper solution, are explored in more detail.
2. Cantors Powerclass Theorem, Russells Paradox and Freges Lesson
Cantors poverclass theorem, also knovn as the poverset theorem or just Cantors the-
orem, is the videly-accepted result that every class or collection ot things can be divided
2010 Kevin C. Klement
Russell, His Paradoxes, and Cantors Theorem: Part I 17
into more subgroups or subclasses than it has members. 1he poverclass ot a class is the
class ot all its subclasses, so the theorem asserts that the poverclass ot a class is alvays lar-
ger in size (cardinality, than the class itselt. Georg Cantor established this result in 1891
vith the tolloving argument. Lvery class c has at least as many subclasses as members,
since tor each member a, the class ot a alone is a subclass ot c. 1he core ot Cantors argu-
ment involves shoving that there cannot be equally many subclasses as members. Suppose,
tor reductio ad absurdum, that there vere. In that case, the members and subclasses could
be paired o so there vould be a oneone tunction, f, mapping each subclass s ot c to a
distinct member ot c, vhich ve can call f (s). Some members might be in the subclass
they are mapped trom, others not. It s = {a}, tor a given subclass s and member a, and
a happens to be the object f (s) that s is mapped onto, then a is a member ot its corres-
ponding class, but not so it a = f (s). Consider then the class, w, consisting precisely ot
the members ot c that are not members ot the subclasses that map to them. As w is itselt a
subset ot c, it must be included in the mapping. Hence, theres some member r ot c such
that f (w) = r. Consider nov vhether or not r w. We dened w as the class ot all
members a ot c that are not in the class s such that a = f (s), so in the case ot r, vhich is
f (w), r w just in case r / w, vhich is a contradiction. Cantor concluded that there can
be no such oneone tunction trom subclasses ot c to members, and, theretore, that there
must be more subclasses.
It the class in question is nite or denumerable, Cantors reductio reasoning can be
represented in tabular torm.' Arrange the members, a
0
, a
1
, a
2
, a
3
, . . . ot c horizontally,
and arrange the subclasses, s
0
, s
1
, s
2
, s
3
. . . , vertically utilizing the ordering ot their cor-
responding members, so that f (s
0
) = a
0
, f (s
1
) = a
1
, etc. Place a checkmark vhere
the chart intersects tor the rov ot a subclass and the column tor any member ot it. lor
example:
a
0
a
1
a
2
a
3
a
4
a
5
. . .
s
0
. . .
s
1
. . .
s
2
. . .
s
3
. . .
s
4
. . .
s
5
. . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1he problematic subclass, w, is generated by moving along the chart, diagonally, trom the
upper lett, dovnvards and to the right, and including precisely those members ot c that
do not have checkmarks vhere the diagonal passes through their columns. In the exam-
ple, this vould include a
1
and a
4
, in vhose column no checkmark is tound along the
diagonal, but not the others. Notice, hovever, that as w itselt should be represented by a
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
18 Kevin C. Klement
rov, vhere the diagonal passes through it, there ought to be a checkmark just in case
there is no checkmark, vhich is clearly impossible. Cantors method ot proot here is
theretore called a diagonal argument or diagonalization.
1his reasoning is validated vithin most torms ot set theory, and is dicult to counter.
Hovever, it is not completely incontrovertible. In particular, the supposition that w
corresponds to a vell-dened subclass ot c might be open to doubt, since it is dened in
terms ot a tunction vhose domain is cs poverclass, and perhaps there is a vicious circle
in this it w is to be included in that very range. More on this belov.
Pussells initial reaction to Cantors theorem vas to regard it as guilty ot error.` Cantor
himselt concluded trom the theorem that there vas no greatest cardinal number, since
tor any number ot things, the number ot their subclasses vould be greater. Pussell and
others have regarded this as paradoxical (and indeed the problem here has sometimes been
called Cantors paradox or the paradox ot the greatest cardinal,. Certain classes such
as the universal class containing everything, or the class ot all classes it vould seem,
cannot have more subclasses than members, because all their subclasses are members. In-
deed, at rst blush, it scarcely seems possible that any collection could be larger in size
(cardinality, than such huge classes as the universal class or class ot all classes. Pussell
attempted to contravene the alleged impossibility ot mapping each subclass ot the class
ot all classes to a member by mapping each subclass to itself, i.e., letting f (s) be s itselt.
Cantors diagonal class w is then the class ot all classes ot classes not included in them-
selves. 1his class too is mapped to itselt, and a contradiction results by asking it it is a
member ot itselt. Drop the assumption that w need only contain those classes of classes
that are not members ot themselves, and this becomes Pussells paradox in its tamous
torm. Pussell vas explicit in many places that Cantors theorem vas his inspiration.`
Pussell soon communicated it to Giuseppe Peano and Gottlob lrege, vhose logical sys-
tems it rendered inconsistent.
Cantors diagonalization method generalizes beyond mappings involving classes or sets.
Given certain assumptions about the nature ot properties (or predicates, attributes, uni-
versals, etc.,, it establishes that the number ot properties applicable (or not, to a certain
logical kind ot thing must alvays exceed the number ot things ot that kind. It ve recon-
ceive a
0
, a
1
, a
2
, a
3
, . . . in the chart as the things ot the kind in question, and s
0
, s
1
, s
2
, s
3
, . . .
as properties applicable to them, and viev the checkmarks as indicating vhich things in-
stantiate vhich properties, ve are prompted to ask vhether or not there is such a property
as not instantiating the corresponding property in the mapping. It so, it should be included in
the mapping, but then the object that corresponds to it instantiates it it and only it it does
not. Here, hovever, the supposition that there must be such a property, merely because
ve seem able to describe its instantiation conditions, is even more open to doubt.
Gottlob lreges reaction to the inconsistency in his logical system, published in an
appendix to volume II ot his Grundgesetze, usetully illustrates the matter. Pather than
properties, lrege spoke ot vhat he called concepts, understood as a kind ot tunction
trom objects to truth-values. 1hinking ot these tunctions extensionally, lrege equated
concepts satised by all and only the same objects. lrege traced the presence ot Pussells
paradox in his system to his Basic Iav V, vhich could be vritten:
(
,
F() =
,
G()) (x)(F(x) G(x))
1his states that the extension ot the concept F ( ), or class ot all F s, is identical to the
extension ot the concept G( ), or the class ot all Gs, it and only it, all and only F s are
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
Russell, His Paradoxes, and Cantors Theorem: Part I 19
Gs. lrege understood the notation
,
(. . . . . .) as representing a second-level tunc-
tion, a tunction that takes a concept as argument and returns an object as value. Ori-
ginally, it vas to stand tor the tunction that takes a concept as argument, and returns
as value its corresponding class, or extension. In the appendix, lrege argues that the
lett-to-right halt ot this biconditional must come out talse, regardless ot vhat second-level
tunction
,
(. . . . . .) is taken to represent. While not explicitly presented as such, the
reasoning is straighttorvardly Cantorian. Pevrite s
0
, s
1
, s
2
, . . . in the chart above
as F
0
( ), F
1
( ), F
2
( ), . . . tor dierent concepts, and revrite a
0
, a
1
, a
2
, . . . as

,
F
0
(),
,
F
1
(),
,
F
2
(), . . ., and the connection becomes clear. It it vere possible
to map concepts to objects to yield distinct objects tor distinct (i.e. non-coextensive,
concepts, then, by vhat amounts to diagonalization, ve could alvays produce a contra-
diction: just consider the concept ot being an object in this mapping that does not fall under
the concept from which its mapped, i.e., the concept ot being an x such that:
(F )(x =
,
F () & F (x))
1he object that results by applying the tunction
,
(. . . . . .) to this concept vould be
such as to tall under that concept it and only it it does not. 1his is equally true vhether

,
(. . . . . .) is interpreted to yield the extension ot the concept to vhich it is applied,
or vhether it yields some other kind ot object that vould be dierent tor dierent or
non-coextensive concepts.
Iet us call the result that it is impossible to generate a mapping trom concepts, prop-
erties or (propositional, tunctions to objects that results in distinct objects tor dierent
or non-coextensive properties, lreges lesson. We might be tempted simply to take this
lesson in stride, except that there are many cases in vhich it seems possible to generate
a distinct object tor distinct properties. In such cases, one must either react by explain-
ing vhy the initial impression that a distinct object can be generated tor each property
vas mistaken, or explain hov this might be possible vithout diagonalization leading to
contradiction.
3. A Plethora of Paradoxes
Pussell telt the impact ot lreges lesson early. In a Sept. 1902 letter to lrege, Pussell
despaired that trom Cantors proposition that any class contains more subclasses than
objects ve can elicit constantly nev contradictions. It is vorth listing several examples.
1he rst ve have already discussed:
Intuition 1. Obviously tor any tvo properties that are not coextensive it is possible to
generate distinct objects: their extensions or corresponding classes'
Diagonalization result: Russells class paradox. Consider the property an extension has
just in case it does not have its dening property (or, equivalently, is not a member ot
itselt,. 1his property has its ovn distinct extension, but that extension has that property
just in case it does not.
Many philosophers believe that properties or concepts can be considered objects or
logical subjects in their ovn right. 1hose that do must ponder the tolloving:
Intuition 2. Obviously tor any tvo properties (vhether coextensive or not, it is possible
to generate distinct objects, viz., the properties themselves.
Diagonalization result: Russells predication paradox. Consider the property a property has
just in case it does not instantiate itselt. Does it, as an object, instantiate itselt It does
just in case it does not.
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
20 Kevin C. Klement
Pussells early ontology included propositions understood as mind-independent com-
plex entities, the bearers ot truth or talsity. Many other philosophers believe in similar
intensional entities, though vith videly varying details and vocabularies (e.g., thoughts,
states-ot-aairs, possible tacts, beliet-contents, etc., Consider:
Intuition 3. 1here are as many propositions as there are properties thereot. lor each
property ot propositions, one can generate a distinct proposition, such as the proposition
that every proposition has that property, or the proposition that all propositions vith that
property are true.
Diagonalization result: the propositions paradox. lasten on any one ot these mappings: take
the latter. Consider the property, , a proposition in this mapping has vhen it does not
have the property ot propositions ot vhich it asserts all instances are true. lor example,
the proposition all atomic propositions are true is not itselt an atomic proposition, so it has
, vhereas the proposition all true propositions are true is itselt true, so it does not have .
Consider then the proposition all propositions with are true: does it have It does just
in case it does not.'
lrege, tamously, but also many other philosophers, including Pussell prior to On
Denoting, believe in special abstract semantic objects: senses, meanings, individual
concepts, denoting complexes, and so on. At least some (and perhaps all, ot these entities
are understood as picking out their reterents or denotations in virtue ot their unique
possession ot some property or other.
Intuition 4. 1here are as many descriptive senses as there are properties. lor each
property, ve can generate a descriptive sense that picks out as denotation vhatever object
(it any, uniquely holds that property. While the object picked out may be the same tor
descriptive senses generated trom distinct properties, the descriptive sense itselt vould be
dierent tor dierent properties.
Diagonalization result: the Russellian descriptive Sense Paradox. Consider the property, H,
vhich a descriptive sense has vhen it lacks the property in virtue ot vhich it presents a
denotation, it any. 1he sense the author of Waverly did not vrite Waverly, hence, it has H.
On the other hand, the sense the self-identical thing is a selt-identical thing, so it lacks H.
Nov consider the sense the H, does it have H It does just in case it does not.''
By a slight variation, ve could consider an old tashioned intension understood as a
semantic entity that represents the entire class ot things having a certain property, rather
than just the thing having it, as above. Larly Pussell called these concepts ot a class,'`
but I shall call them class-intensions instead.
Intuition 5. 1here are as many class-intensions as there are properties. lor each prop-
erty, there is a class-intension that represents the class ot things having that property.
While the corresponding extension or class may be the same tor dierent class-intensions,
the class-intensions themselves are dierent tor dierent properties.
Diagonalization result: The Russellian class-intension paradox. Consider the property, K,
a class-intension has vhen it lacks the property it uses to collect together its class, it any.
1he class-intension (all) teaspoons is not a teaspoon, hence, it has K. 1he other hand, the
class-intension (all) class-intensions is a class-intension, so it lacks K. Nov consider the
class intension (all) class-intensions having K, does it have K It does just in case it does not.
(It class-intensions simply are properties, then this paradox collapses into the predication
paradox above.,
We neednt necessarily generate a distinct intension tor each property dening a class,
its enough to generate one corresponding to the equivalence class ot coextensive proper-
ties, since it still vould hold that ve vould get dierent ones tor non-coextensive proper-
ties. Indeed, a version ot the paradox could be tormulated dealing vith that equivalence
class itselt:
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
Russell, His Paradoxes, and Cantors Theorem: Part I 21
Intuition 6. We can map properties to equivalence classes ot properties vhere the asso-
ciated equivalence relation is coextensionality. lor any tvo non-coextensive properties,
the equivalence class to vhich they vould be mapped vould be dierent. 1he property
ot having a heart vould be mapped to the same equivalence class as the property ot having
a kidney, but not to the same equivalence class as the property ot being a featherless biped,
though the latter vould be mapped to the same equivalence class as being human.
Diagonalization result: the Russellian property equivalence class paradox. Consider the prop-
erty that an equivalence class ot coextensive properties has just in case it doesnt have
any (or, it you preter, all, since theyre coextensive, ot the properties it contains. Nov,
consider the equivalence class ot all properties coextensive vith : does it have It it
does have , then it doesnt have , since its one ot the properties in the equivalence
class. It it does not have , then it must have at least one property coextensive vith ,
in vhich case, it must have as vell so ve get a contradiction either vay.
Other examples could be given, but the above suce to establish the general pattern
ot hov Cantors theorem, or, more specically, lreges lesson, generates Pussell-style
paradoxes almost ad nauseam.'` One need only mention a category ot entity most
likely, an abstract entity that can be correlated or related to properties (or classes, in
a systematic vay' and vhere the identity conditions are ne-grained enough that the
entities correlated vith non-coextensive properties can be distinguished.
4. Kinds of Solutions
In a 1903 paper entitled On Some Diculties in the 1heory ot 1ransnite Numbers
and Order 1ypes, largely dedicated to Pussells paradox, Pussell identied three broad
approaches tor nding a solution. It is tair to say that most contemporary approaches
can still be seen as talling under one ot these categories, though ve shall discuss some
exceptions belov. Pesponses to the other Cantorian paradoxes can be sorted under
roughly the same headings. 1he three categories, as Pussell dubbed them, are (i, the
theory ot limitation ot size, (ii, the zigzag theory, and (iii, the no classes theory. We
discuss these in turn.
.+. +nrorirs or iixi+:+iox or sizr
Consider, again, those classes that led Pussell to suspect an error in Cantors proot: the
universal class, and the class ot all classes. Cantor himselt called such things inconsistent
multiplicities (in a letter to Dedekind, meaning that their size is too large tor them to
be considered one thing. Axiomatic set theories nov prevalent among mathematicians,
such as Zermelolrnkel (Zl, set theory, also disavov the existence ot a universal set,
or set ot all sets. 1his is keeping vith an iterative conception ot a set, vhereupon sets
are thought to be built up out ot successive applications ot poverset and union opera-
tions.' More complicated theories, such as von NeumannBernaysGodel (NBG, set
theory, distinguish sets trom proper classes, vhere the latter are considered too large to
be members ot any set or class. Here, although there may be a class ot all sets, it is a
proper class, and hence not a member ot itselt, nor are those subclasses vhich are also
proper classes members. Insotar as it has a poverclass at all, it vould only contain sets
that are subclasses ot it, not all subclasses vhatever. Lven here, then, there is no class ot
all classes, both improper and proper.
1hese theories escape contradiction by denying that a distinct object can be generated
tor every property or characteristic ot sets (or at least classes, in the case ot NBG,. 1hose
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
22 Kevin C. Klement
properties that are true ot too many things have no class (or no distinctive class, as-
sociated vith them. 1here are only as many classes as there are properties that are not
too videspread. 1his general line ot avoiding inconsistency is perhaps clearer in the case
ot the Iimitation ot Size-based set theory developed more recently by George Boolos
(Saving lrege and elsevhere,, tormulated in a second-order logic, vhere lreges Basic
Iav V is replaced vith (Nev V,:
(
,
F() =
,
G()) ((Big(F) & Big(G)) (x)(F(x) G(x)))
1his commits us to as many classes (Boolos calls them subtensions, as there are non-big
properties. Boolos denes a big property as one that is instantiated by as many things
as there are things though not necessarily by all things. Whether one adopts Booloss
proposal or a related one, the condition ot not being a member ot oneselt, x x, is
thought to represent a property that holds ot too many things to have a distinctive class
associated vith it, and hence, according to this approach, there is no such class as that
vhich vould be involved in Pussells class paradox. It also denies the existence ot a uni-
versal class or class ot all classes, thereby escaping the vorries Pussell initially entertained
about Cantors theorem.
1he limitation-ot-size approach has not been pursued much, or as directly, vith regard
to the other paradoxes listed in the previous section. Indeed, it is not entirely clear hov
to extend this approach to them in a plausible vay. We shall return to this issue in the
sequel article.
.:. +nr ziz: +nror.
1he previous approach requires that, contrary to our intuitions, it is untrue, atter all, that
ve can generate a nev object tor every property (or every subclass, ot our original group
ot things, tor those that apply to too many things, there is no distinct associated object ot
the type suggested by the intuition. 1he zigzag approach vorks dierently. It grants the
intuition that tor every property or at least, tor every unexceptional property ve can
generate a distinct object. Hovever, it denies that tor every grammatically vell-tormed
condition, ve have the kind ot unexceptional property to vhich the intuition correctly
applies. In particular, the conditions used to specity the vould-be diagonal subclasses
or properties are thought to be exceptional or problematic in some sense, and that this
undermines the diagonal reasoning behind Cantors theorem.
Pecall that Cantors argument begins by assuming that a oneone mapping exists
betveen subclasses ot c and members ot c, and then uses that very mapping to dene
another subset w ot c vhich, it is alleged, cannot be included in the mapping. 1his
is because nothing in the mapping could be it, given hov w is described. 1he argu-
ment concludes the mapping does not exhaust the subsets ot c. One might counter by
questioning vhether or not just any description ot a subclass ot c necessarily corresponds
to a genuine subclass ot c. In eect, one could exploit the impossibility ot ws occurring
in the map in question in the other direction, arguing that there can be no such subclass
as w. 1he so-called subclass that vould be generated trom reversing the arrangement ot
checkmarks along the diagonal is no actual subclass at all, but merely an empty descrip-
tion to vhich nothing need ansver. 1o provide a complete solution, one vould need
to specity conditions under vhich a description ot a subclass (i.e., specication ot condi-
tions tor inclusion in that subclass, ot a given class can or cannot be guaranteed to dene
a subclass.
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
Russell, His Paradoxes, and Cantors Theorem: Part I 23
Apart trom Pussells ovn experiments vith this approach, vhich ve leave tor the
sequel, and reconstructions thereot, the most thorough examination ot an approach along
these lines is perhaps Quines system Nl, vhich takes a torm similar to nave set theory,
except that the class abstraction schema:
(y) (x)(x y . . . x . . .)
rather than holding tor all open sentences . . . x . . . not containing y tree, is only alloved
tor instances in vhich the open sentence . . . x . . . has certain syntactic properties. In
particular, it needs to be stratied, i.e., a tunction must exist assigning natural numbers to
terms anking the membership sign vherever it occurs in . . . x . . . so that the number
assigned to the term lett ot is one lover than that assigned to the term on the right. In
Nl, Cantors theorem is unprovable (and indeed, demonstrably talse tor many instances,,
as the diagonally generated class w in Cantors proot vould be dened by an illegitimate
tormula. Instead, Nl embraces classes that have all their subclasses as members, including
a universal class and a class ot all classes. Pussells class paradox is also blocked, as it too
vould be dened by an illegitimate tormula, vhich is not surprising given that it can be
thought ot as generated by diagonalization.
Simply taken as a solution to Pussells class paradox, the overall strategy is neutral
betveen an interpretation according to vhich the problematic diagonal condition does
correspond to a property, albeit an exceptional one vith no corresponding subclass,
and an interpretation according to vhich the condition, although it can be stated in a
syntactically vell-tormed vay, does not comprehend a genuine property at all. lor the
general approach, hovever, to solve some ot the other paradoxes mentioned in Sec. 3,
particularly the predication paradox, something more like the latter interpretation seems
more promising. (Note that tor the other paradoxes, admitting the property but denying
a vell-dened subclass vont be enough, since some other entity, or even the property
itselt, is involved instead., 1his interpretation could then allov that tor every property,
there is a distinct corresponding object (itselt, or some proposition about it, or some
descriptive sense involving it, etc.,, but deny that there are such diagonal properties as
non-selt-instantiation, H trom the descriptive sense paradox, or trom the propositions
paradox. Again, to be tully plausible, the theory vould need to explain under vhat
conditions the specication ot the exemplication conditions tor a vould-be property
does or does not suce to guarantee the existence ot a property so delineated.
On the other interpretation, a property is admitted, but is regarded as exceptional in
some sense, and theretore does not have a unique corresponding object. It is dicult to
assess vhich ot these interpretations is right tor Quine, vhose nominalistic tendencies
steer him avay trom speaking in terms ot properties rather than linguistic tormulas.
Quine states the limitation on vhat conditions dene classes in the metalanguage, and
in terms ot syntactic teatures ot the open sentence used to describe a class. lor those vho,
unlike Quine, embrace second-order logic, the requirement could instead take the torm
ot object language qualiers tor sorting out those properties that determine a corresponding
object trom those that do not. lor a theory involving vhich properties dene classes, this
tack is compatible vith certain neo-logicist set theories that adopt a genericized version
ot Booloss (Nev V,, in vhich talk ot properties too big to generate classes is replaced
by more neutral talk about properties that are bad or non-distinct-class-generating (ct.
Shapiro 63,:
(
,
F() =
,
G()) ((Bad(F) & Bad(G)) (x)(F(x) G(x)))
1o count as a zigzag theory, badness vould need to be spelled out in terms ot the
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
24 Kevin C. Klement
internal properties ot a property rather than, e.g., the range ot its applicability. Again,
there has been very little by vay ot exploration ot approaches along these lines as applied
to other paradoxes.'
.. +nr xo ci:ssrs +nror.
1he third, and most radical, kind ot solution to these paradoxes involves escheving the
kind ot vould-be entity that appears to violate Cantors theorem altogether. 1hinking ot
the classes paradox, Pussell called this approach the no classes theory. Here, one vould
deny that there are any such things as classes, and suggest that discourse apparently about
classes, to the extent that it is not meaningless or contused, is not to be taken at tace value.
Such discourse vould be meaningtul precisely to the extent that it is possible to revord it
in a torm in vhich no explicit mention ot a class is made. lor example, the claim that the
class of sedans is a subclass of the class of cars can be revorded simply by saying that all sedans
are cars. 1he solution to Pussells class paradox comes in insisting that certain kinds ot talk
about classes cannot be so revorded. In particular, the claim that a class is a member ot
itselt is to be regarded as meaningless, along vith derivative expressions, such as that a
class is not a member ot itselt. Hence the description used to dene Pussells paradoxical
class is not meaningtul, and theretore does not determine a condition or property that
denes a class.
lor the other paradoxes, it vould be more appropriate to speak ot the no properties
theory or the no propositions theory, and so on. Ot course, philosophers are likely
tamiliar already vith arguments shoving that there is no such thing as Pedness, or
(talse, propositions such as Jupiter is in my pocket. 1o be tully plausible, hovever, these
approaches must make sense ot the apparent discourse about these entities that seems
unquestionably true, such as the claim that, Luclid proved the proposition that there are
innitely many primes. It also must ensure that the paraphrase given ot such discourse is
not by itselt enough to generate the paradoxes. Again, the suggestion is likely that vhile
some discourse about these apparent entities can be revorded in a torm in vhich they
are not mentioned, the discourse giving rise to the paradoxes cannot. Notice that it is
not enough simply not to take the entities as sui generis. Peplacing abstract propositions
in tavor ot, e.g., classes ot synonymous sentences, does not help solve the paradoxes it
enough such classes are posited to violate Cantors theorem.
Nevertheless, it is approaches ot this stripe that, by and large, Pussell himselt grav-
itated tovards, especially trom late 1903 and attervards (atter discovering his theory ot
descriptions,. We shall take up his vievs in the sequel article.
.. ioic:i +.rrs
Another broad kind ot approach, not listed by Pussell in Some Diculties, though,
ironically, otten attributed to him, posits logical types ot things.' Strategies ot this sort
can be seen as attempting to maintain modied or more sophisticated forms ot the intuitions
listed tor the paradoxes, vhich, in the end, are tound not to be inconsistent vith Cantors
theorem. Maintaining that entities and the properties applicable to them tall into distinct
logical types, and in keeping vith the intuition behind each paradox, one might suggest
that tor each property applicable to entities in a given category, it is possible to generate a
distinct nev entity, but insist that this nev entity is in a separate logical category trom the
entities to vhich the original group ot properties vere applicable. Hence, any property
applicable (or not, to these nev entities is not among the original group.
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
Russell, His Paradoxes, and Cantors Theorem: Part I 25
lor the classes paradox, tor example, it amounts to dividing classes into type 1, or
classes ot individuals, type 2, or classes ot classes ot individuals, type 3, or classes ot classes
ot classes ot individuals, and so on, vhere it is not simply talse but meaningless to ask
vhether a b unless b is ot a type one higher than a. 1hen, ve are tree to postulate
a distinct class tor every property applicable to individuals, but this class is not one ot
the entities to vhich that property may or may not apply. Diagonalization never gets o
the ground, since the properties involved in the mapping are not such as to apply, or not
apply, to the entities to vhich theyre mapped, and so no system ot checkmarks (to recall
our visualization earlier, is appropriate.
1he approach has in common vith the no classes theory the suggestion that expres-
sions ot the torm x x or x / y are not meaningtul. In this, they contrast vith
the limitation ot size and zig-zag theories insotar as the latter regard such constructions
as at least grammatically vell-tormed, even it they do not dene classes. Despite this
similarity, their explanations tor their meaninglessness dier. In the no classes theory,
no sentence ot the torm a b is to be taken as about some entity ot any type denoted
by b, instead, the entire sentence as a vhole must be revorded into a torm in vhich no
class is mentioned, vhich is deemed impossible in this instance. In the kind ot theory
mentioned here, b is an independently meaningtul expression, it simply diers in vhat
kind ot thing it means trom a, and it engenders nonsense to attempt to say the same
things ot a one vould say ot b.
lreges theory ot levels ot concepts, according to vhich there are objects or saturated
entities, rst-level concepts (under vhich objects may or may not tall,, second-level
concepts (vithin vhich rst-level concepts may or may not tall,, etc. could be used to
provide a response ot this stripe to Pussells predication paradox.' lor each rst-level
concept (i.e., property,, or concept applicable to objects, there is indeed an entity, that
concept itselt, but that concept is not itselt an object, and the question as to vhether it
applies to itselt is meaningless, one can only ask vhether or not second-level concepts
are applicable to it.
Addressing the other paradoxes vith this kind ot strategy vould involve, tor example,
arguing that although a distinct proposition can be derived tor each property, the resulting
proposition is not ot the right sort either to have or not to have that property. But this
is precisely vhat the denition ot trom the propositions paradox assumes, and hence,
it is poorly dened. Similarly, vhile distinct descriptive senses might be generated trom
all properties, they vould not be the kind ot thing to vhich such properties may or may
not apply.
.. o+nrr :rrro:cnrs
Iastly, there are kinds ot solutions that tall into none ot the above categories. 1hese
include radical approaches as might be taken by a dialethist vho simply embraces the
contradictions as true, vhile trying to insulate their harmtul eects by means ot a non-
explosive paraconsistent logic. Such approaches raise other philosophical issues ve cannot
tully explore here.
Another less radical approach, hovever, might stem trom noting that Cantors theorem
or lreges lesson is not automatically violated by just any tunction that maps properties ot
things to things. It the same object may result as value tor non-coextensive properties as
argument, then the tunction doesnt postulate as many objects as subclasses. One might
then hope to maintain the spirit ot the intuitions lying behind the paradoxes, but vithout
the supposition that the entity generated is alvays distinct.
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
26 Kevin C. Klement
Notice, hovever, that it is not enough to allov that sometimes dierent properties may
generate the same entity in the mapping, one must allov that sometimes non-coextensive
properties may generate the same entity. With regard to classes or extensions, pushing this
line ot response is more or less tantamount to arguing that non-coextensive properties may
have the same extension, a proposal vhich sounds absurd on its tace. Nevertheless, lrege
himselt endorsed such a proposal in his appendix on Pussells paradox, and Pussell himselt
vas tor a time attracted to it.' Hovever, vithout philosophical support provided by an
independent theory,` the supposition that non-coextensive properties may determine the
same class, or, in the case ot the Property Lquivalence Class paradox, be included in the
same equivalence class ot coextensive relations seems bevildering. 1he situation is even
vorse vith the paradoxes involving intensional entities. Intensions are supposed to be
ner-grained in their identity conditions than extensions, yet to solve the descriptive sense
and class-intension paradoxes ved have to allov that the F and the G be identical senses in
some cases even vhen F and G arent even coextensive, or that tvo distinct classes may
be generated by the same class-intension.`' 1here does not seem to be much to be said
in tavor ot such approaches.``
Acknowledgements
1hanks to an anonymous reteree tor helptul comments on a previous dratt.
Short Biography
Kevin C. Klement is an Associate Protessor ot Philosophy at the Lniversity ot Massachu-
setts, Amherst. He is the author ot Frege and the Logic of Sense and Reference, and has
published articles on lrege, Pussell, the history ot analytic philosophy, the history ot
logic and intormal logic. He is currently engaged in a research project regarding the
development ot Bertrand Pussells philosophical logic.
Notes
* Correspondence: Department ot Philosophy, Lniversity ot Massachusetts, 332 Bartlett Hall, 130 Hicks Way,
Amherst, MA 01003, LSA. Lmail: klementphilos.umass.edu
1
Lven vhen the class isnt denumerable, it is otten vorthvhile to imagine impertectly the resulting chart ab-
stractly anyvay, just as a heuristic. 1he tact that such a chart isnt technically possible doesnt invalidate the core
argumentative strategy.
2
See Pussell, Papers v3, xxxii, and Mathematics and the Metaphysicians, tor discussion ot the historical details,
see Coa, Humble Origins, and Grin, Prehistory.
3
See e.g. PoM 101, IMP 136, MPD 138, Auto 130.
4
lrege vould invoke his sensereterence distinction to explain avay apparent problems vith equating coextensive
concepts, see his Comments on Sense and Peterence.
3
Here ve allov ourselves to Pussellize lreges notation somevhat, and restrict our tocus to concepts as opposed
to other tunctions.
6
Notice that it x is a class, and
,
F() is interpreted to mean the class ot Fs then this precisely gives the
condition tor xs not being a member ot itselt.
7
Indeed, he vas avare ot the main gist prior to reading lreges appendix see PoM 103, and his letter to lrege
ot 24 uly 1902, in lrege, Philosophical and Mathematical Correspondence 139.
8
See lrege, Correspondence 147. Lntortunately, lrege does not seem to have tully appreciated the importance
ot the paradox Pussell vent on to describe, and that it threatened his philosophy as much as it did Pussells, see
Klement, Pussells Paradox in Appendix B, and Frege and the Logic of Sense and Reference, Chap. 6.
9
Depending on vhat ve interpret properties here to mean, Pussell discusses dierent interpretations ot this para-
dox in dierent places. Interpreted to mean vhat early Pussell called predicates, by vhich he meant something
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
Russell, His Paradoxes, and Cantors Theorem: Part I 27
like Platonic universals, it occurs in PoM (80, 102,. Interpreted to mean vhat he called propositional tunctions
it occurred only later. See Klement, Origins tor discussion ot the dierence.
10
A version ot this paradox, dealing simply vith classes ot propositions rather than properties ot propositions
vas tormulated by Pussell in PoM (32728,. Pussell also tormulated it in terms ot propositional tunctions
instead ot classes ot propositions in a letter to lrege (see lrege, Correspondence 13960,. Antinomies ot this torm
vere independently rediscovered by ohn Myhill in the context ot evaluating certain later torms ot intensional
logic, see, e.g., Myhill, Problems, Anderson, Semantic Antinomies, and Klement, 1he Number ot Senses,
and Does lrege Have 1oo Many 1houghts It is sometimes called the Pussell-Myhill Antinomy.
11
A paradox ot this torm is discussed at greater length in Klement, Cantorian Argument.
12
In PoM (67,, Pussell distinguishes the concept ot a class, all humans, trom the class-concept human. 1he
dierence is subtle, and ve could generate a Cantorian paradox trom either one, though I think that the class-
concept is really just the property itselt, and so the resulting paradox is just the predication paradox.
13
1here are, to be sure, other important paradoxes in the same tamily that dont t quite as vell into the
rubric provided by our previous discussion, such as the paradox ot relations Pussell discusses in Mathematical
Iogic (22223,, the paradox Kaplan discusses in A Problem in Possible World Semantics, or vhat is called
the classsense paradox in Klement, 1he Number ot Senses. 1he puzzle called Cantors paradox (see Sec. 2,
concerning vhether or not there is a greatest cardinal number is ot course another paradox related to Cantors
theorem that doesnt neatly t this rubric.
14
Above, ve otten say generated trom but this metaphor should not be taken too seriously.
13
1his vay ot describing things derives trom Boolos, 1he Iterative Conception.
16
1hough see Cocchiarella, Pussells Paradox ot the 1otality ot Propositions, and Cantini, On a Pussellian
Paradox, tor exceptions.
17
As I argue in the sequel article, this attribution is contentious at best.
18
Although, actually, I think this description ot the situation is somevhat misleading, given that, tor lrege, the
extension ot a concept has its being in the concept itselt, and various other suggestions in his vork to the eect
that the extension ot a concept simply is the concept treated as a logical subject. lrom this perspective, the class
paradox and predication paradox are indistinguishable tor lrege, vhich is vhy, I think, lrege describes Pussells
description ot the predication paradox in his letter to him only as imprecise rather than erroneous.
19
1his is evinced by the last minute tootnote to PoM (p. 496,, in vhich Pussell calls it very likely the correct
solution, as vell as in manuscripts ot the period (Papers v4, 1737,. Indeed, Pussell even seems to have hoped that
it might vork in other cases too, mentioning it in connection vith the propositions paradox in particular to lrege
in a letter, see lreges Correspondence 13960. lreges ovn proposed solution vas later tound to lead to more
complicated contradictions. See Quine, lreges Way Out and Iandini, Ins and Outs tor turther discussion.
20
Notice that Booloss (Nev V, system is technically a theory alloving non-coextensve big properties to have
the same class, although the explanation ot this makes use ot a dierent sort ot theory, and talk ot extensions
replaced by talk ot subtensions.
21
lor turther discussion ot related issues, see Klement, A Cantorian Argument 73.
22
Merely bearing the possibility ot such solutions in mind, hovever, torces us to be rather more caretul about
the precise tormulation ot the paradoxes. Nicholas Denyer deected vhat amounts to a lregean version ot the
propositions paradox tormulated by Adam Pieger by pointing out that Piegers mapping didnt necessarily gen-
erate a distinct proposition trom each property, hovever, a tairly insignicant modication to Piegers proposal
is immune to like treatment, see Klement, 1oo Many.
Works cited
Anderson, C.A. Semantic Antinomies in the Iogic ot Sense and Denotation. Notre Dame Journal of Formal Logic
28 (1987,: 99114.
Boolos, George. 1he Iterative Conception ot a Set. Logic, Logic and Logic. Cambridge, Mass.: Harvard Lniversity
Press, 1998a. 1329.
. Saving lrege trom Contradiction. Logic, Logic and Logic. Cambridge, Mass.: Harvard Lniversity Press,
1998b. 17182.
Cantini, Andrea. On a Pussellian Paradox about Propositions and 1ruth. One Hundred Years of Russells Paradox.
Ld. Godehard Iink. Berlin: Walter de Gruyter, 2004. 23984.
Cantor, Georg. Lber eine elementare lrage der Mannigtaltigkeitslehre. Jahresbericht der Deutschen Mathematiker-
Vereinigung 1 (1891,: 738.
. Ietter to Dedekind, dated 28 uly 1899. From Frege to Gdel: A Source Book in Mathematical Logic. Ld.
ean van Heijenoort. Cambridge, Mass.: Harvard Lniversity Press, 1967. 11317.
Cocchiarella, Nino. Pussells Paradox ot the 1otality ot Propositions. Nordic Journal of Philosophical Logic 3
(2000,: 2338.
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.
28 Kevin C. Klement
Coa, . Alberto. 1he Humble Origins ot Pussells Paradox. Russell 33 (1979,: 317.
Denyer, Nicholas. Piegers Problem vith lreges Ontology. Analysis 63 (2003,: 16670.
lrege, Gottlob. Basic Laws of Arithmetic. Berkeley: Lniversity ot Calitornia Press, 1964.
. Comments on Sense and Meaning. Posthumous Writings. Lds. P. Iong and P. White. Chicago: Lniversity
ot Chicago Press, 1979. 11823.
. Philosophical and Mathematical Correspondence. Chicago: Lniversity ot Chicago Press, 1980.
. Collected Papers on Mathematics, Logic and Philosophy. Nev York: Basil Blackvell, 1984.
Grin, Nicholas. 1he Prehistory ot Pussells Paradox. One Hundred Years of Russells Paradox. Ld. Godehard
Iink. Berlin: de Gruyter, 2004. 349371.
Kaplan, David. A Problem in Possible World Semantics. Modality, Morality and Belief: Essays in Honor of Ruth
Barcan Marcus. Ld. Walter Sinnott-Armstrong et. al. Cambridge: Cambridge Lniversity Press, 1993. 4132.
Klement, Kevin. Pussells Paradox in Appendix B ot 1he Principles ot Mathematics: Was lreges Peply Ad-
equate History and Philosophy of Logic 22 (2001,: 1328.
. Frege and the Logic of Sense and Reference. Nev York: Poutledge, 2002.
. 1he Number ot Senses. Erkenntnis 38 (2003,: 30223.
. Does lrege Have 1oo Many 1houghts A Cantorian Problem Pevisited. Analysis 63 (2003a,: 449.
. 1he Origins ot the Propositional lunctions Version ot Pussells Paradox. Russell 24 (2003b,: 10132.
. A Cantorian Argument Against lreges and Larly Pussells 1heories ot Descriptions. Russell vs. Meinong:
The Legacy of On Denoting. Lds. Dale acquette and Nicholas Grin. Nev York: Poutledge, 2008. 6377.
Iandini, Gregory. 1he Ins and Outs ot lreges Way Out. Philosophia Mathematica 14 (2006,: 123.
Myhill, ohn. Problems Arising in the lormalization ot Intensional Iogic. Logique et Analyse 1 (1938,: 7882.
Quine, W. V. Nev loundations tor Mathematical Iogic. American Mathematical Monthly 44 (1937,: 7080.
. On lreges Way Out. Mind 64 (1933,: 14339.
. 1he Ways ot Paradox. The Ways of Paradox and Other Essays. 2nd ed. Cambridge, Mass.: Harvard
Lniversity Press, 1973.
Pieger, Adam. Paradox vithout Basic Iav V: A Problem With lreges Ontology. Analysis 62 (2002,: 32730.
Pussell, Bertrand. The Principles of Mathematics (PoM). Cambridge: Cambridge Lniversity Press, 1903 (second
edition 1931,.
. Mathematical Iogic as Based on the 1heory ot 1ypes. American Journal of Mathematics 30 (1908,: 22262.
. Mathematics and the Metaphysicians. Mysticism and Logic and Other Essays. Iondon: Allen 8 Lnvin,
1917 (rst published 1901,. 3974.
. Introduction to Mathematical Philosophy (IMP). Iondon: Allen 8 Lnvin, 1919.
. My Philosophical Development (MPD). Iondon: Lnvin, 1938.
. On Some Diculties in the 1heory ot 1ransnite Numbers and Order 1ypes. Essays in Analysis.
Ld. Douglas Iackey. Nev York: George Braziller, 1973 (rst published 1903,. 13364.
. The Collected Papers of Bertrand Russell v3. Ld. G. H. Moore. Iondon: Poutledge, 1993.
. The Collected Papers of Bertrand Russell v4. Ld. Alasdair Lrquhart. Iondon: Poutledge, 1994.
. Autobiography (Auto). Iondon: Poutledge, 1998.
Shapiro, Stevart. Prolegomenon to Any luture Neo-Iogicist Set 1heory: Abstraction and Indenite Lxtens-
ibility. British Journal of the Philosophy of Science 34 (2003,: 3991.
2010 Kevin C. Klement Philosophy Compass 5/1 (2010): 1628.