An Editor Recalls Some Hopeless Papers

Author(s): Wilfrid Hodges

Reviewed work(s):
Source: The Bulletin of Symbolic Logic, Vol. 4, No. 1 (Mar., 1998), pp. 1-16
Published by: Association for Symbolic Logic
Stable URL: http://www.jstor.org/stable/421003 .
Accessed: 17/08/2012 08:45

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend access to The
Bulletin of Symbolic Logic.

http://www.jstor.org
THE BULLETINOF SYMBOLICLOGIC
Volume 4, Number 1, March 1998

AN EDITOR RECALLS SOME HOPELESS PAPERS

WILFRID HODGES

?1. Introduction. I dedicate this essay to the two-dozen-odd people whose

refutations of Cantor's diagonal argument (I mean the one proving that the
set of real numbersand the set of naturalnumbershave differentcardinalities)
have come to me either as referee or as editor in the last twenty years or so.
Sadly these submissions were all quite unpublishable;I sent them back with
what I hope were helpful comments. A few years ago it occurred to me to
wonder why so many people devote so much energy to refuting this harmless
little argument-what had it done to make them angry with it? So I started
to keep notes of these papers, in the hope that some pattern would emerge.
These pages report the results. They might be useful for editors faced with
similar problem papers, or even for the authors of the papers themselves.
But the main message to reach me is that there are several points of basic
elementary logic that we usually teach and explain very badly, or not at all.
In 1995 an engineer named William Dilworth, who had published a refu-
tation of Cantor's argument in the Transactions of the Wisconsin Academy
of Sciences, Arts and Letters, sued for libel a mathematician named Un-
derwood Dudley who had called him a crank ([9] pp. 44f, 354). The case
was dismissed. For myself I am more scared of the copyright law than the
law of libel. After taking legal advice I decided not to quote any of the
authors directly. The alternative was to write some letters saying in effect:
'I'm sorry we couldn't publish your paper as a contribution to logic. Can I
please publish parts of it as examples of garbage?' Not much is lost, because
almost all of the papers were written by manifest amateurs who had great
difficulty explaining what they meant, and I freely admit that much of what
follows is my own attempt to discern some thoughts behind the streams of
words. This is in no sense a scientific analysis of experimental data.

?2. Cantor's proof. The authors of these papers-henceforth let me call

them just the authors-seem to have read Cantor's argument in a variety
of places. In my records only one author refers directly to Cantor's own
argument [7]. One quotes Russell's 'Principles of mathematics' [20] later

Received Octobeer 6, 1997; revised October 15, 1997.

? 1998, Association for Symbolic Logic
1079-8986/98/0401-0001 /$2.60

1
2 WILFRIDHODGES

in his discussion;he could havefound the diagonalargumenton page 365

of that book, in wordsthat closely follow Cantor. Anothercites Fraenkel
Abstractset theory'[12] as his source,and anotherrefersto Barrow'The-
ories of everything'[2]. One contentshimselfwith referencesto two earlier
unpublishedpapersof his own. Othersgiveno source.
For definitenesslet me writedown a proof, not in Cantor'swords,which
containsall the points we shallneed to commenton.
(1) Weclaimfirstthat for everymap f fromthe set {1, 2,... } of positive
integersto the open unit interval(0, 1) of the real numbers,thereis some
realnumberwhichis in (0,1) but not in the imageof f.
(2) Assumethat f is a mapfromthe set of positiveintegersto (0, 1).
(3) Write
0 . anl an2 an3 ...

for the decimalexpansionof f(n), whereeach ani is a numeralbetween0

and 9. (Whereit applies,we choose the expansionwhichis eventually0, not
that whichis eventually9.)
(4) For each positiveintegern, let bn be 5 if ann7 5, and 4 otherwise.
(5) Let b be the realnumberwhosedecimalexpansionis
0. bl b2 b3 ...

(6) Thenb is in (0,1).

(7) If n is any positiveinteger,then bn =: a,,, and so b :f (n). Thusb is
not in the imageof f.
(8) This provesthe claimin (1).
(9) We deduce that there is no surjectivemap from the set of positive
integersto the set (0, 1).
(10) Since one can write down a bijectionbetween(0, 1) and the set of
realnumbers(and a bijectionbetweenthe positiveintegersand the natural
numbers,if we want the latter to include 0), it follows that there is no
surjectivemap fromthe set of naturalnumbersto the set of realnumbers.
(11) So thereis no bijectionbetweenthese two sets;in otherwords,they
havedifferentcardinalities.
This is the proofwhichall the authorsattacked.Most authorshad seena
form of the argumentwhichuses a picture:we writeout the decimalsf(1),
(2), ... in a columnwithf(1) at the top, and we traceout the realnumber
b as we walk down the diagonalline
all, a22, * ,

changingthe digits as we go. I shall call this the writtenlist form of the
argument.
None of the authorsshowedany knowledgeof Cantor'stheoremabout
the cardinalitiesof powersets.
 AN EDITOR RECALLS SOME HOPELESS PAPERS 3

?3. Why this target?Cantor'sargumentis short and lucid. It has been

aroundnow for over a hundredyears. Probablyeveryprofessionalmathe-
maticianalivetodayhas studiedit andfoundno fallacyin it. So thereis every
temptationto imaginethatanybodywho writesa paperattackingit mustbe
of dangerouslyunsoundmind. One shouldresistthis temptation;the facts
don't supportit. On a few occasionsI was able to speakto the authorsof
thesepapers;one or two wereclearlyat sea, but otherswereas sane as you
or me. In the courseof researchingthis paperI came acrossstatementsby
two of the leadinglogiciansof this century,which-read literally-werejust
as crazyas anythingin these attackson Cantor'sargument.Read on and
judge.
There is a point of culturehere. Severalof the authors said that they
had trainedas philosophers,and I suspectthat in fact most of them had.
In English-speakingphilosophy(and muchEuropeanphilosophytoo) you
are taught not to take anythingon trust, particularlyif it seems obvious
and undeniable. You are also taught to criticiseanythingsaid by earlier
philosophers. Mathematicsis not like that; one has to accept some facts
as given and not up for argument. Nobody should be surprisedwhen
philosopherswho move into anotherareatake theirhabitswith them. (In
the days when I taught philosophy,I rememberone studentwho was told
he had failed his coursebadly. He duly produceda reasonedargumentto
provethat he hadn't.)
To anticipatefor a moment, I don't think any of our authors located
anythingdistinctivelybad about Cantor'sargument.The points on which
they trippedup were all things that might have trippedthem in a thou-
sand othermore mundanearguments.Most of the muddleswerenot even
mathematical.Differentauthorsmadedifferentattacks.
It'snothingmorethan a guess,but I do guessthat the problemwith Can-
tor'sargumentis as follows. This argumentis often the firstmathematical
argumentthatpeoplemeetin whichthe conclusionbearsno relationto any-
thing in theirpracticalexperienceor theirvisual imagination.Compareit
with two other simple facts of cardinal arithmetic. First, m x n = n x m.
Wecan see whatthis amountsto by thinkingof a rectanglewith one side of
length m and one side of length n. The picturepoints to the right formal
argumentwhen m and n are finite, and exactlythe same argumentworks
when they are infinite. Or second, 1 + co = co. We don't meet co in our
everydaylife, but we can see how to provethe inequalityby moving each
numberalong by one. (Thepicturelies wellwithinthe rangeof whatwe can
'iibersehen',to quote G6del [14].)
Butthenwecometo Cantor'sresult,andallintuitionfailsus. Until Cantor
first provedhis theorem([6], by a much longer argument,as it happens),
nothinglike its conclusionwas in anybody'smind'seye. And even now we
acceptit becauseit is proved,not for any otherreason.
4 WILFRID HODGES

?4. Not attacking an argument. It was surprisinghow many of our authors

failed to realise that to attack an argument,you must find something wrong
in it. Several authors believed that you can avoid a proof by simply doing
something else.
The commonest manifestation was to claim that Cantor had chosen the
wrongenumerationof thepositive integers. His argument only works because
the positive integers are listed in such a way that each integer has just finitely
many predecessors. If he had re-ordered them so that some of them come
after infinitely many others, then he would have been able to use these late
comers to enumerate some more reals, for example the real number b which
we defined in (5) of the proof.
Other authors, less coherently, suggested that Cantor had used the wrong
positive integers. He should have allowed integers which have infinite decimal
expansions to the left, like the p-adic integers. To these people I usually sent
the comment that they were quite right, the set of real numbers does have
the same cardinality as the set of natural numbers in their sense of natural
numbers; but the phrase 'natural number' already has a meaning, and that
meaning is not theirs.
One or two authors were ready with a counterargument. To say that the
existing concept of natural numbers is incompatible with their numbers is to
say that at least one of their numbers can't be included in the set of natural
numbers. But we can demonstrate that any object whatever can be included
in a natural number series. (Read Benacerraf [4], these authors might have
added-though they didn't.)
This already goes to quite a deep issue about the identity of mathematical
structures. I think there might be some difficulty in putting together an
answer which everybody working in the foundations of mathematics would
accept. But really the question should never have arisen in this context.
There is no way that one can regard Cantor's assumptions about natural
numbers as a mistake in his argument. The existence of a different argu-
ment that fails to reach Cantor's conclusion tells us nothing about Cantor's
argument.
How does anybody get into a state of mind where they persuade themselves
that you can criticise an argument by suggesting a different argument which
doesn't reach the same conclusion?
Well, roughly as follows. Suppose our friend Hugo offers us a proof, by
induction on n, that for every natural number n a man with n hairs on his
head is bald. There are three degrees of response. The most passive is to say
'There must be a mistake somewhere', and leave it to somebody else to find
where the mistake is. (In practice 'passive' is perhaps the wrong word, if we
need to do some work to wall off a safe area of arguments where we never
have to consider Hugo's.) The next is to look at Hugo's argument and try
to find a place where Hugo has made a step which is not cogent. The third
 AN EDITORRECALLSSOMEHOPELESSPAPERS 5

response,the most masterful,is to claim that one step in Hugo'sargument

is wrong simply because theproof won'twork withoutit.
We have all seen responsesof these three kinds to the paradoxes. Bar-
wise and Etchemendyare forcefuladvocatesof the second responsein the
introductionto theirbook on the liarparadox([3]p. 7):
A treatmentof the Liar all too often takes the followingform.
First,variousintuitivelyplausibleprinciplesareset out andmoti-
vatedby a discussionof the commonsensenotionsinvolved.Then
a contradictionis shownto followfromtheseintuitiveprinciples.
At this point the discussionturnsdirectlyto the questionof which
principles can be kept and which must be abandoned ... the Liar
has forcedus to abandonintuitivelyplausiblesemanticprinciples
withoutgivingus a reason,beyondthe paradoxitself, to suspect
theirfalsehood.Wesee thatthey arefalse,withoutunderstanding
why.
Readersof the proof of Theorem6 on theirpage 79 canjudgewhetherthey
might also be an illustrationof the thirdresponse. (Theymotivatevarious
intuitivelyplausibleprinciples,shortof onewhichtheyrejectby an argument
that 'exactlyparallelsthe reasoningusuallytaken to show that the Liar is
paradoxical'.)
I see no differenceof principlebetweenwhat these criticsof Cantorare
doing and what I called the masterfulresponseabove. We dislikethe con-
clusion, so we outlawone of the steps that got us there. Some might feel
that on the moral scale thereis a differencebetweena conclusionwhichis
downrightparadoxicaland one that we happento dislike. But I can't see
how the responseis justifiedin one case if it isn'tin the other.

?5. Attackingan argument.In formallogic we teachpeoplehow to con-

structarguments,and how to checkthe validityof a formalargument.But
we hardlyteachanythingabouthowto assessthecogencyof an unformalised
deductiveargument. Our authorsare makingtheir criticismswithout the
benefitof any trainingin how to do it.
Therearesomegood booksabouthowto assessunformalisedarguments.I
haveby my handAlec Fisher'sexcellentvolume'Thelogic of realarguments'
[11], andI shallquoteit later.Likemost of the genre,Fisherconcentrateson
scientificand moralargumentsratherthan deductiveones. Thus (p. 140):
... it is clearly difficult to apply [traditional formal logic] to real
arguments-to argumentsof the kind one finds for examplein
newspapers,magazinesand learnedjournals.
If Fishermeansto implythat traditionalformallogic, as traditionallypre-
sented,givesthe righttools for analysinginformaldeductivearguments,then
part of my purposeis to sow some doubtsaboutthis.
6 WILFRID HODGES

This may be the moment to mention a passage in Wittgenstein's book

'Remarks on the foundations of mathematics' [22], where he claims (if I
follow him right) that Cantor's argument has no deductive content at all.
The theme of Wittgenstein's book is that mathematical statements get any
meaning they may have from rule-governed activities that involve them. He
singles out Cantor's argument because it would appear to have no relation
to any imaginable activity.
Except for one, namely the activity of writing out lists of complete decimal
expansions of real numbers. This is of course a daft activity, doomed to
failure. Ah, says Wittgenstein, that's what Cantor's theorem must amount
to ([22] p. 129):
Surely-if anyone tried day-in day-out 'to put all irrational num-
bers into a series' we could say: "Leaveit alone; it means nothing;
don't you see, if you established a series, I should come along with
the diagonal series!" This might get him to abandon his undertak-
ing. Well, that would be useful. And it strikes me as if this were
the whole and proper purpose of this method. It makes use of the
vague notion of this man who goes on, as it were idiotically, with
his work, and it brings him to a stop by means of a picture.
None of our authors showed any knowledge of Wittgenstein's critique, or
any sympathy with it. They all regarded Cantor's argument as an attempt
at a deductive proof of a meaningful proposition, and they all assessed it in
these terms.
So how does one assess an unformalised deductive argument? Broadly
speaking, such an argument has three kinds of component:
* There are the stated conclusion, the stated or implied starting assump-
tions, and the intermediate propositions used in getting from the assump-
tions to the conclusion. I shall call these the object sentences.
* There are stated or implied justifications for putting the object sentences
in the places where they appear. For example if the argument says 'A,
therefore B', the arguer is claiming that B follows from A.
* There are instructions to do certain things which are needed for the
proof. Thus 'Suppose C', 'Draw the following picture, and consider the
circles D and E', 'Define F as follows'.
A criticism of an argument might focus on any of these components. For
example it might claim that one of the object sentences is meaningless or
ambiguous; this would be an attack on the object sentences. It might claim
that an object sentence appears somewhere without properjustification; this
would be an attack on the justifications. It might claim that one of the things
we are instructed to do in the proof is impossible; this would be an attack on
the instructions.
In fact none of the authors took issue with the object sentences themselves,
but there were several attacks on the justifications and the instructions.
 AN EDITOR RECALLS SOME HOPELESS PAPERS 7

One authordid findanotherformof attackwhichI mustmention. Sadto

say,it was a flashof unintendedbrilliance,buriedbeneatha dozen pagesin
whichnothinghappened.Thisauthorhadin frontof hima formof Cantor's
argumentwhichused reductioad absurdum.(In some formalsystemsthis
would be neededto pass from (8) to (9) in the proof above.) Let us prove,
he said, that Cantor'sargumentis invalid. We start by assumingthat it is
valid. If it is validwe are entitledto use it; and so we do, downto the point
wherewe get a contradiction. But since we have reacheda contradiction,
our originalassumptionmust have been wrong. That is to say, Cantor's
argumentis invalid.
There is a quick though slightly dishonest refutationof this critique.
Namely, Cantor'sproof also makes an assumption,and when our author
reachesthe contradictionhe only knowsthatat least one of his assumptions
mustbe false;it neednot be the one he madefirst. Thisrefutationis dishon-
est, becauseit fails to point out thatthe assumption'Cantor'sproofis valid'
doesn'tplayanyrolein the argumentwhichfollows.Wearein territoryquite
close to Carroll's'Whatthe tortoisesaidto Achilles'[8],and I leaveit to the
readerto sortout the details.Typically,this refutationof Cantor'sargument
has nothingto do with Cantor'sargumentin particular-if it workedat all,
it wouldworkagainstanyargumentby contradiction,includingthosewhich
the intuitionistsfindvalid.

?6. Attackson the justifications.In a well-respectedtextbookrecentlyI

noticedthis sentence:
In all cases (0, 0) is a point of order 2 since any point of order
2 has the form (x, 0), where x is a root of the cubic equation
0 = x3 + ax.
Thislooks verylikean attemptto argue'P(a), becauseforall x, ifP(x) then
Q(x)'. It'sextravagantto supposethat the authormade a mistakeof logic.
Morelikelyhe meantto say 'sincethe points of order2 are exactlythose of
the form ...', and he wasn't too careful about the exact wording because he
expectedthe readersto thinkit throughin theirown termsanyway.This is
a kind of conversationallooseness. I doubt if AlfredTarskiwas everguilty
of it, but probablymost of the restof us havebeen on occasion.
It'squite a differentmatterwherea writerdirectlyaddressesthe question
whetherQ follows logicallyfrom P, and gets it wrong. Thereweretwo of
ourauthorswho saidtheydisagreedwithCantoraboutwhatfollowsdirectly
fromwhat.
The firstof these authorsdeniedthe step from (8) to (9) in the proof. In
fact he agreedthat Cantorhad provedthat
The imageof any map fromthe set of positiveintegersto the set
(0, 1) is a propersubsetof (0, 1).
But he denied,severaltimesover,that it followsthat
8 WILFRID HODGES

There is no injective map from the set of positive integers to the

set (0, 1), whose image includes all of (0, 1).
(I repeat the caution that I'm not using the author's exact words. But he
was reasonably proficient in set theory, and he should accept these quoted
statements as equivalent to his formulations.) On the face of it, this author
is denying that the inference
(*) Vx 3y --?(x, y) - 3x Vy q(x, y)
is valid.
The second author maintained that Cantor had proved something so
strong that the result was paradoxical, though Cantor had failed to recognise
this. He claimed that (8) directly implies
The number b is not in the image of any map from the positive
integers to (0, 1).
He had no trouble in showing that this is absurd. I suppose he was using the
fallacious inference
(**) Vx 3y +(x, y) F 3y Vx 0(x, y).
This fallacy is familiar from examples of the form 'Everything has a cause;
therefore there is something that causes everything'.
There don't seem to be any recognised systems of logic in which (*) is
invalid or (**) is valid. So I suppose these are just mistakes, not evidence
for variant logics. I looked to see whether the psychological literature on
mistakes of logic could throw any light. First let me quote Lance Rips'
([19] p. 392) list of factors which cause errors in experimental tests of logical
reasoning:
If Q follows from P according to some logical theory T but sub-
jects fail to affirmthat Q follows from P, that could be because (a)
T isn't the appropriatenormative standard; (b) subjects interpret
the natural-language sentences that are supposed to translate P
and Q in some other way; (c) performance factors (e.g., memory
or time limits) interfere with subjects' drawing the correct conclu-
sion; (d) the instructions fail to convey to subjects that they should
make their responses on the basis of the entailment or deducibility
relation rather than on some other basis (e.g., the plausibility or
assertibility of the conclusion); (e) response bias [i.e., subjects'
guesses about how the experimenter set up the test] overwhelms
the correct answer;or (f) the inference is suppressed by pragmatic
factors (e.g., conversational implicatures). If Q does not follow
from P according to T but subjects affirm that Q follows from
P, that could be because (a)-(e) hold as above; (g) subjects are
interpreting the task as one in which they should affirm the argu-
ment, provided only that P suggests Q, or P makes Q more likely,
 AN EDITORRECALLSSOMEHOPELESSPAPERS 9

or P is inductivegroundsfor Q; (h) subjectstreatthe arguments

as an enthymemethat can be filledout by relevantworldknowl-
edge; (i) subjectsascribetheir inabilityto drawthe inferenceto
performancefactorsand incorrectlyguessthat P entailsQ; or (j)
subjectsare misledby a superficialsimilarityto some valid infer-
ence from P' to Q' into supposingthat thereis a valid inference
fromP to Q.
Is this list helpful?Wehavealreadyruledout the possibilitythatthe authors
werecallingon somevariantlogic, or thattheywereassessingthe argument
as anythingbuta strictlydeductiveone. Thisdisposesof (a), (d), (f), (g) and
(h). The authorsweren'tup againstlimits of time, and they didn'tregard
Cantor'sargumentas an experimentaltest of theirreasoningpowers;so out
go (c), (e) and (i). This leaves(b) and (j).
Case (b) wouldapplyif the authorsmisinterpreted some sentencein Can-
tor's argument. Looking at what they say, I am sure this is not what has
happened. In fact their mistakesare about the logical relationsbetween
sentenceswhichthey themselveshavewritten.
Case(j) is obscurelystated.Does Ripsmeanthatthe subjectshavemisread
the inferenceas beingof some otherformwhichhappensto be similar?Or
does he mean that they havecorrectlyidentifiedthe form, but incorrectly
guessedthat the form must be valid becausea similarone is? Eitherway
round,I don'tseewhyRips givesthis only as a causeof mistakenlyinferring,
not as a causeof mistakenlyfailingto infer.
What is missingfrom Rips' first list (a)-(f) is the case wherea person
correctlyunderstandstwo sentencesbutfailsto noticethe logicalconnection
betweenthem.
Somewritershavearguedthat if B followslogicallyfromA and a person
reallyunderstandsboth A and B, then that personmustsee that B follows
from A. (For example,one could make a case that this is a criterionof
whetherthe person'reallyunderstands'the two sentences.)This was never
plausiblefor the cases whereit takes a lengthyargumentto get from A to
B. If we can fail to notice distantlogicalrelationships,then it must at least
be possiblefor us to fail to notice close ones. (And of course it happens.
A few years ago an algebraist,now dead, publisheda long paper which
seemedto be a majorcontributionto an importantproblem.His argument
dependedon findinga familyof numberswhichsatisfya certainvery large
set of equationsand inequalities. Sadly it came to light that a particular
smallsubsetwasunsatisfiable; the factwas obviousonceit hadbeenpointed
out, but it was easily missed.) Though this is speculation,it seems to me
the most naturalexplanationof our author'sfailureto acceptthe entailment
(*).
The correspondingexplanationof (**) would be that the author failed
to notice the differencebetween two of his formulations;he thought he
10 WILFRIDHODGES

was paraphrasingwhen in fact he was reversingtwo quantifiers. I have

an impressionthat the authormay have been thrownby the fact that the
diagonalnumberis alwayscalledb (in our proof), as if it was independent
of f. One symbol,one object.
Rips' book containsa wealthof informationaboutthe errorsthat people
do make in logical reasoning. I had hoped that he would give some data
on (*) and (**). Unfortunatelyhe chooses to 'represent'sentencesby first
findingequivalentsentencesin prenexformandthenusingSkolemfunctions
for existentialquantifiers([19] pp. 90ff, 185ff). Thus (*) and (**) become
respectively
for (*) Vx --o(x, ax) H-Vx -~(x, a,)
and
for (**) Vx )(x, ax) t-Vx +p(x,a).
The steps involvedin these reductionsare at least as elaborateas either(*)
or (**) on their own. The reductionof (*) removeseverythingof interest.
Rips doesn'tmention any experimentaltests of the reducedform of (**);
probablyit's too blandto be tested. So Rips' book left me disappointed.
I turnednextto therivalworkof Johnson-Laird andByrne[16]. Thisbook
I readwith caution. I knowI am not alone in findingits accountsof logical
theoryalmostincomprehensible(see my briefreviewin [15]). Nevertheless
the book does reportsome veryinterestingexperiments.
Johnson-Lairdclaims in [16] and elsewherethat our normal mode of
deductivereasoningis proofby cases;thatwe representthe casesby whathe
calls 'models'(theyarenot whatmodeltheoristscall 'models');and thatwe
haveno systematicprocedureforfindingthe neededcases. A majorcauseof
mistakesin deductionis failureto find the rightcases. The more cases are
needed,the moremistakespeoplemake.
Johnson-Lairdand Byrne[16]havea chapter(Chapter7) on 'Manyquan-
tifiers:reasoningwith multiplequantification'.This shouldbe the placeto
findsometreatmentof (*) and (**);in particularwe look thereto findwhat
Johnson-Lairdand Byrnethink the relevant'models'are. But it transpires
that all the examplesin thatchapterhavethe form
Qix Q2y xRy
where Q1, Q2 are relativisedquantifiers(like 'all of the musicians','some
of the authors')and R is knownto be an equivalencerelation. We get the
'models'by sketchingsomeequivalenceclassesandputtingmarkersto show
(a) what types of person or object occur in each and (b) which of these
types are universallyquantified.The resultis a kind of Venndiagramwith
quantifiers.I didn'tsee how to extendthis formatto our situation.
A later chapterin the book (Chapter9) claims to describea procedure
for constructing'quantifiedmodels'. Much is obscure;the authorslimit
 AN EDITORRECALLSSOMEHOPELESSPAPERS 11
themselvesto a languagein whichthe only relationsymbolis equality.But
one can imaginethat a refinementof theirprocedurewould throwup the
standardfour-elementcounterexampleto (**) as a modelof the sentenceon
the left. I supposethatthe Johnson-Lairdpositionwouldbe thatthe author
who thought(**) wasvalidhadgeneratedonly'models'of thepremisewhich
were also 'models'of its conclusion,and failed to realisethat other cases
are possible. The problem(and I think it is Johnson-Laird's problem,not
is this to
ours) to bring claim a formwhichis (i) testableand (ii) significantly
differentfromthe baldstatementthatthe authoror the experimentalsubject
has failedto realisethat the conclusiondoesn'tfollowfromthe premise.
Turningto the valid inference(*), we run into a new problem.Johnson-
Lairdcan explainhow peoplemakefalseinferencesby failingto considerall
the cases. But it was not at all clearto me how his theoryexplainspeople's
failureto makecorrectinferences,or how he reachesany predictionsabout
how hardpeople will find it to performone or anothercorrectdeduction.
Johnson-Lairdand Byrnedo discussin detailan exampleof 'suppressionof
validdeductions'([16]p. 81ff). But this turnsout to be an examplewherea
pragmaticallymisleadingsecondpremisecausesthe subjectsto misinterpret
the firstpremise,a phenomenonwhichseemsto havenothingto do with the
analysisin termsof 'models'.In sum,Johnson-Lairdand Byrnealso left me
disappointed.

?7. Attacksontheinstructions.Thisbringsus to the thirdpoint of attack,

the instructions.
One authorcomplainedthat Cantor'sproof requiresus to write out an
infinitediagram.But that'sa thingwe can'tdo; the authorconscientiously
provesthis as follows. As we makethe list, it becomesinfiniteeithergradu-
ally,or suddenly,or not at all. The idea that it becomesinfinitegraduallyis
incoherent;at any stageit is eitherdefinitelyfiniteor definitelyinfinite.If it
suddenlybecomesinfinite,thereis a stageat whichit becomesinfinite.But
thisis false;at everystagein the constructionof the list, it is finite.Therefore
it neverbecomesinfinite.
Of coursenobodywouldsuggestthatin orderto carryout Cantor'sproof
you actuallyhaveto writeoutthe infinitediagram,wouldthey?Wouldthey?
Now supposethat
X, X1, X2, X3, ...

is an infinitelist or enumerationof some but not necessarilyall

of the real numbersbelongingto the interval. Writedown one
belowanothertheirrespectivenon-terminatingdecimalfractions
... [and here follows a diagram with some dot-dot-dots].
This is fromKleene's'Introductionto metamathematics' ([18]p. 6). Taken
literally,what Kleene is
says quite mad. Of course one can avoidtakingit
12 WILFRID HODGES

literallyby sayingsomethinglike (3) in my versionabove. But it was clear

thatmanyof the authorshad difficultiespassingfromthe writtenlist version
of the argumentto somethingmore abstract. With hindsightit may have
beenunkindof Kleeneto dumpthis on the unsuspectingbeginner,six pages
fromthe startof his book.
Wemove on quickly.A commonfaultin argumentsis to takefor granted
somethingwhich should have been proved. One of our authors accused
Cantorof doingthis;he complainedthat Cantorassumesthatthereis a map
from thepositive integers to (0, 1). See (2) in our version above.
It shouldbe easyto dealwith this. Cantoris assumingP in orderto prove
somethingof the form 'If P then Q'. This is a standardmovein arguments.
One assumesP 'for the sake of argument'.Nobody interpretsthis kind of
assumptionas a claimto knowP, or evento believeP, do they? Do they?
EvertBethis one of the few logicianswho haveseenproblemsin this form
of argumentand takenthem seriously.He reportshis conclusionson pages
36f of [5]. On page 17 of the same essayhe had given a naturaldeduction
argumentwherea premisenumbered(2), viz.,
(Ey)[S(y)&M(y)],
is assumedand then dischargedlater. Referringback to that argumenthe
comments
... the (possibly false) assumption, which at a certain moment
has been introduced, is eliminated later on ... However, if we
wish exactlyto know whatis going on, then we ought to consult
the semantictableau. In the formalderivationof Section4, we
know by premiss(2), that some individualfulfils the condition
S(y)&M(y), and we agreeto givethis individualthe name'a'.
Thislastsentenceis completelymad. Bethimpliesthatwe knowthatpremise
(2) is true. But in the first place, nowherein the articledoes he give any
evidencewhateverthat (2) is true; in fact he has describedit as 'possibly
false'just a few lines earlier.In the secondplace,it is a stringof symbolsin
an uninterpretedlanguage(as far as we know-Beth has explainedon p. 11
how to interpreta language,but he has said nothingto suggestthat he has
in mind any particularinterpretationfor this one); so no questionof truth
or falsehoodarises. The last clauseof the sentenceis mad too: if we know
that 'Someindividuallives in Neasden',it makesno senseto "agreeto give
this individualthe name 'a'" until we havepickedout one such individual.
But Beth has done nothingto pick out an individual.
Beth'smistakeshere seem to me of the same orderas any made by our
criticsof Cantor. He gets awaywith it becausehe is a brilliantlogician,he
writesa convincingstyleand we believehis conclusions.ThoughI havehim
in my sights here, probablymost of us have said or writtenequallycrazy
thingsat one time or another. And in this particularcase, Beth'saccount
 AN EDITOR RECALLS SOME HOPELESS PAPERS 13
has the meritof highlightingtwo of the main problemsabout assumptions
'for the sake of argument'.First, when we assumeP, we proceedas if we
knew P. Second,whenwe assumethereis x such that Px, we proceedas if
we haveidentifiedsuch an x. Whatis going on here?
This is not the place to answerthat questionat length. But let me put
down some pointersin this strangelyunchartedterritory.
Assumptionsin argumentsappearin at least the followingfourguises:
(a) The writersays 'I assumeP becausewe alreadyknow P'. Here the
assumptionservesas a lemma.
(b) The writersays 'In the followingdiagram,assumeA is the such-and-
such,B is the such-and-such'etc., and then uses the diagram.
(c) The writersays 'AssumeP', deducesQ, and concludes'If P then Q'.
(d) Thewritersays'AssumeP', deducessomethingknownto be false,and
concludes'Not P'. (Or the 'nots' could be the other way round.) This is
reductioad absurdum.
This list is not complete. For exampleI am ignoringthe ancient and
renaissancerule of false position ('regulafalsi'), wherewe solve an equa-
tion by makingtwo possiblyincorrectguessesabout the solutionand then
calculatingthe errors;see Smith[21]p. 437ff.
Form (a) is unproblematicand I say no moreaboutit.
Form (b) occurs most often in geometricarguments,but one meets it
elsewhere.On the Johnson-Lairdtheorywe use a versionof it all the time.
Gelernter[13]makesit thebasisfora computerimplementationof geometric
reasoning.Peoplehavefoundit problematicfromearliesttimes,becausethe
objectsin the diagrammight havedifferentpropertiesfrom the thingsthat
they represent. (Maybewe are provingpropertiesof equilateraltriangles,
but our hand slips and the trianglewe drawis scalene.) Aristotleraisesthe
matterbrieflyin his Metaphysics[1] (book XIV 1089a24ff);his view seems
to be that thereis no harmdone as long as the false assumptionis not 'in'
the proof. No doubtone can elaboratethis into a reasonabletheory,though
theremay be moreto say in particularcases.
Naturaldeductionconflatesthe two forms (c) and (d). Leavingasidethe
caseswhichworrythe intuitionists,it's agreedthat both formsof argument
are valid-these ruleswon'tlet us down. The problemis to explain(c) and
(d), not as formalrulesbut as meaningfulpiecesof discourse.Until we can
do this, I'm not surethat we havegivena just and fair answerto the author
who criticisedstep (2) of Cantor'sargument.
There is a chapter entitled 'Suppose for the sake of argument that ...'
in Fisher'sbook alreadymentioned[11] (Chapter6). Fisher gives many
sensibleexamples,and maybethey would be enoughto soften the heartof
Cantor'scritic. Thushe comments:
A mathematicianwho presentsthe standardEuclideanproofthat
thereareinfinitelymanyprimenumbersbeginsby supposingthat
14 WILFRID HODGES

there are only finitely many. He is not asserting (telling us) that
there are only finitelymany primes (becausehe knows full well
that this is false) but he is askingus to considerthe proposition
with a view to drawingout its implications.
NeverthelessI think there is somethingmissing. Fisher has told us the
purposeof forms (c) and (d). But there are other ways of achievingthis
purpose. We can draw out the implicationsof a propositionP without
assumingP. For examplewe can use Frege'spreferredstyleand stickto the
format'If P then ... '. WhatFisherhas not told us is, first,exactlywhatwe
are being told to do when the argument says 'Assume ...', and second, why
this is a good way of achievingthe statedpurpose.
Writersof a psychologicalcast sometimesspeakof assuming'forthe sake
of argument'as a kind of mentalactivity.Thus Rips ([19]p. 7f) conflatesit
with 'imagininga situation'. Some form of this view must be correct. For
examplein a debateone speakermay say to the other:
(+) Whenyou say 'Q and R', areyou assumingthat P?
Normallythe secondspeakerunderstandsthe questionand knowswhether
the correctansweris Yes or No. 'Assuming'is somethingthat we do with
our minds,and normallywe can tell whetherwe aredoingit.
But this activityof assuminghas some odd properties.Firstof all, we can
assumethingsthat we could neverconceivablyimagine.Thus to provethat
thereis no greatestinteger,we start by assumingthat thereis one. If any
readerknowshow to imaginethatthereis a greatestinteger,I'dbe interested
to hearhow they do it and whatit feels like. But in any case, this approach
to assumingmustbe barkingup the wrongtree.Thevalidityof an argument
can neverdependon you or me doingsomeparticularthingin the privacyof
ourimaginations.(Ourimaginationsmighthelpus tofind a validargument,
but this is a differentmatter.)
Second, in the debatejust mentioned,the second speakercould quite
meaningfullyanswerthe question(+) by saying
I am assumingit for Q but not for R.
So assumingis not a thing that we do at a particulartime;it's a thing that
we do at a particularstagein an argument,and in respectof certainthings
in the argument.This pulls 'assuming'out of the worldof brutefacts, and
givesit an intentionalorjuridicalfeel.
Third,one can assumethingswhicharenot evenmeaningfulpropositions,
since they contain symbols for which no referencehas been given. The
extremecase of this is whereone makesassumptionsin naturaldeduction
arguments,using an uninterpretedfirst-orderlanguage,as in the example
from Beth above. But there is alreadyan example at (2) in our proof
of Cantor'stheorem,wherewe assumethat f has some propertywithout
takingany stepsto specifywhatf is.
 AN EDITOR RECALLS SOME HOPELESS PAPERS 15

One possible responseis that the letter 'f' is a variablebound by an

implieduniversalquantifier'Forall f'. The problemwith this approachis
thatthescopeof thequantifierwouldhaveto reachall thewaydownto clause
(7) of the proof, throughseveralsentences,includingboth statementsand
definitions.To makesenseof this, we wouldneed a semanticsfor discourse
ratherthan for sentencesone at a time. The semanticsof discourseis still in
its infancy.(See Kampand Reyle[17]for a pioneeringattempt.)
Besidesthequestionswhat(a)-(d) areseparately,wecan also askhowthey
are related. Beth (loc. cit.) takesfor grantedthat (b)-(d) are all examples
of the samephenomenon.A revealingpassagein MichaelDummett'sbook
on Frege'sphilosophyof languagesuggestslinks between(b)-(d) and the
activity of assigning a reference:
If, for example,I takesomecolourcounters,and say,'Letthis one
standforthe Government,thisoneforthe Opposition,thisonefor
theChurch,thisonefortheUniversities,thisone fortheArmy,this
one for the TradeUnions, ... ', and so on, I shall be understood
on the presumptionthat I am about to make some arrangement
of the countersby means of which I intend to representsome
relationsbetweenthese institutions,and assertthat they obtain.
If I do not go on to makeany sucharrangement,but simplystart
talking about somethingelse, my earlierdeclarationslose their
original intelligibility ... it is like my saying, 'Suppose there is life
on Mars', and then failing to drawany consequencesfrom this
hypothesis,and, when challenged,saying, 'Oh, I simplywanted
you to suppose that';...
([10]p. 193).
I thinkI know broadlywhatare the rightanswersto thesequestions,but
I don'tproposeto arguethe matterhere. The conclusionI want to leaveon
the tableis that the notion of assumptionsin argumentsis surroundedwith
seriousphilosophicalpuzzles. It can tripup a professionalalmostas easily
as a beginner.

?8. Conclusion.First, contraryto what severalcriticsof Cantor'sargu-

ment suggestedin theirpapers,at least one mathematicianwas preparedto
look at theirrefutationswith some careand sympathy.
Second,a smallnumberof the criticismsarefaircommenton misleading
expositions. A much largernumberof the criticismsare fair commenton
some seriousand fundamentalgaps in the logic that we teach. Even at a
very elementary level-I'm tempted to say especially at a very elementary
level-there are still many points of controversyand many things that we
regularlyget wrong.
Third,thereis nothingwrongwith Cantor'sargument.
16 WILFRIDHODGES

REFERENCES

[1] ARISTOTLE, Metaphysics X-XIV, Loeb Classical Library, Harvard University Press,
Cambridge, Mass., 1935.
[2] JOHND. BARROW, Theories of everything: the questfor ultimate explanation, Oxford
University Press, Oxford, 1990.
The liar, Oxford University Press, New York,
[3] JON BARWISEand JOHN ETCHEMENDY,
1987.
[4] PAULBENACERRAF, What numberscould not be, Philosophy of mathematics, Selected
readings (Paul Benacerraf and Hilary Putnam, editors), Cambridge University Press, Cam-
bridge, 2nd ed., 1983, pp. 272-294.
[5] E. W. BETH,Semantic entailmentandformal derivability,Thephilosophyof mathematics
(Jaakko Hintikka, editor), Oxford University Press, 1969, pp. 9-41.
[6] GEORG CANTOR, Ubereine Eigenschaftdes Inbegriffesaller reellenalgebraischenZahlen,
Journalfur die reine und angewandteMathematik, vol. 77 (1874), pp. 258-262.
[7] , Uber eine elementare Frage der Mannigfaltigkeitslehre, Jahresbericht der
Deutschen Mathematiker-Vereinigung,vol. I (1891), pp. 75-78.
[8] LEWISCARROLL, What the tortoise said to Achilles, Mind, vol. 4 (1895), pp. 278-280.
[9] UNDERWOOD DUDLEY,Mathematical cranks, Mathematical Association of America,
Washington, DC, 1992.
[10] MICHAEL DUMMETT, Frege,philosophy of language, Duckworth, London, 1973.
[11] ALECFISHER,The logic of real arguments, Cambridge University Press, Cambridge,
1988.
[12] ABRAHAM A. FRAENKEL, Abstract set theory, 4th ed., North-Holland, Amsterdam,
1976, revised by Azriel Levy.
[13] H. GELERNTER, Realization of a geometry theorem-provingmachine, Proceedings of
internationalconferenceon informationprocessing, UNESCO House, Paris, 1959, pp. 273-282.
[14] KURTGODEL,Ubereine bishernoch nicht beniitzte Erweiterungdesfiniten Standpunk-
tes, Collected works (S. Feferman et al., editors), vol. II, Oxford University Press, Oxford,
1990, pp. 240-251.
[15] WILFRID HODGES, Critical commentaryon P Johnson-LairdandR.Byrne, 'Deduction',
Behavioraland Brain Sciences, vol. 16 (1993), p. 353f.
[16] P. N. JOHNSON-LAIRD
and RUTH M. J. BYRNE,Deduction, Lawrence Erlbaum, Hove,
1991.
[17] HANSKAMPand UWEREYLE, From discourse to logic, Kluwer, Dordrecht, 1993.
[18] STEPHEN COLEKLEENE,Introduction to metamathematics, North-Holland, Amster-
dam, 1952.
[19] LANCEJ. RIPS,Thepsychology ofproof, MIT Press, Cambridge, Mass., 1994.
[20] BERTRANDRUSSELL,The principles of mathematics, George Allen and Unwin, Lon-
don, 1903.
History of mathematics, vol. II, Dover, New York, 1958.
[21] D. E. SMITH,
Remarks on the foundations of mathematics, Blackwell, Ox-
[22] LUDWIGWITTGENSTEIN,
ford, 1956.

SCHOOLOF MATHEMATICAL SCIENCES

QUEENMARYAND WESTFIELDCOLLEGE
MILEEND ROAD
LONDONEl 4NS, ENGLAND,U.K.
E-mail: [email protected]