Analogies and Metaphors to Explain Godel's Theorem

Author(s): Douglas R. Hofstadter

Source: The Two-Year College Mathematics Journal, Vol. 13, No. 2 (Mar., 1982), pp. 98-114
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/3026485 .
Accessed: 27/12/2010 10:17

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=maa. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

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 support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The Two-Year College Mathematics Journal.

http://www.jstor.org
Analogies and Metaphors to Explain
Godel's Theorem
Douglas R. Hofstadter
When I was a graduatestudentat Berkeleyin mathematicsduring1966 and 1967,I
foundout, to my chagrin,thatmathematicswas too abstractforme. I had always
thoughtthatI was a prettyabstractthinker, but what I began to realizeabout that
timein mylifewas that,in fact,all of my thoughtsare veryconcrete.They all are
based on images,analogiesand metaphors.I reallythinkonlyin concreteideas, and
I foundthatI couldn'tattachany concreteideas to some of themathematicsI was
learning.I could learntheformalstatements and theorems,I could provetheorems
formally,but I reallycould not go beyond them.I was just not able to get the
concepts withoutimages, so I turnedaway frommathematicsand went on to
physicsat the Universityof Oregon. Then my career went throughvariegated
phases,and finallyI woundup in computerscienceand artificialintelligence, which
is not exactly an accident because my greatestinterestin artificialintelligence
nowadays is in understandinganalogies. In a way I have come back to study,
throughcomputerscience(and particularly throughthebranchof it called artificial
intelligence),
whattheseanalogiesare thatI thinkwith.So thereis a littleloop there
thatI have closed,and I feelveryhappy doing it.
My book* illustratesparticularlystronglymy own predilectionfor thinkingin
metaphors,analogies, images, and so forth.Many of the images that I have
producedin thebook are connectedwithGddel's theorem.Some of you may know
and some of you may not know exactlywhat Gddel's theoremis. I will tryto give
you a flavorof whatit is in thistalk.tIt is not intendedjust foryou to learnhow to
explainGddel's theoremto otherpeople, but possiblyforyou to learn some of the
conceptsdirectlyrightnow fromme. For thoseof you who alreadydo knowwhat
thetheoremis, I hope you willfindsome new variationsof lookingat it and how to
thinkabout it. Gddel's theoremdepends on the idea of self-reference or level
crossing,whichyou'll see in thispaper. Its major impact upon mathematicsis to
show thatformalsystemshave a certainkind of incompleteness. It reallyinvolves
theconceptof how self-reference, or mixingof different levels,can be broughtinto
mathematics.

* GOdel,Escher,Bach: an EternalGoldenBraid (New York: Basic Books, 1979; VintageBooks, 1980).

tThis paper is a lightlyedited transcript
of the talk presentedat the FourthInternationalCongresson
MathematicalEducationwhichwas held in Augustof 1980 in Berkeley.

98
 WhichBall Is The Most Different?

The obviouslydifferent one is the small whiteone. Two of themare the same color and two
of those two sets is the small
are the same size, and the only one thatis in the intersection
whiteone.

To begin withI will give you a littlepuzzle. It is a puzzle which delightsme

because it was somethingthathappened to me as a real-lifeevent.I was sittingat
the edge of a swimmingpool, and in thepool therewere threeballs, a small black
one, a smallwhiteone, and a largewhiteone. I was lookingat them,and I thought,
"How strange-theobviouslydifferent one is thesmallwhiteone! Two of themare
thesame size and twoare thesame color,and theonlyone thatis in theintersection
of thosetwosetsis thesmallwhiteone. It is themostdifferent because it is themost
same!" Here we have a sort of mixing of levels. At some primitivelevel of
observation,at the perceptuallevel,we have the categoriesof size and color,and
thereis no unique answeryieldedat thatlevel.But we have thismoreabstractview
of the situationthatsays thatsize and color are two equallyprimitivethings,so we
can't choose betweenthem.However,we do have an object that is different. We
have one thatdiffersin size, anotherthatdiffersin color, and so thereis one left
over. So we are somehowchangingin our level of description.We are movingto a
more abstractlevel, and yet somehowwe feel we reallyhaven't cheated. I don't
reallyknow quite how to say this,but somehowthereis a validityto thisanswer.
Althoughit seemsstrange,thereis some senseto thisresponseand it is perfectly fair
to move to this next abstractlevel and give this as our answer.This is the first
example of how level mixingcan appear in a simplecontextand can seem quite
naturaland amusingat the same time.G6del's theoremdeals primarilywiththis
whichis somethingthathas fascinatedpeople fromall times
idea of self-reference,
is somethingthatoccurs in manyjokes: it is general
and all places. Self-reference
and fascinatespeople.

99
 What is the onlyplace in a room thata laser cannot hitifyou shinethe beam? That place is
indicatedby the black dot in the figureon the left.The onlyplace thatlaser cannot hitis its
own rearend. You can shineit to any otherpointexceptthatblack dot.
You can actuallyget it to shineon thatspot,dependingon whatyou mean by "thatspot."
If thatspot is definedin coordinatesrelativeto theroom,then,of course,the laser can shine
in otherwords,in termsof the
throughit. If, though,it is definedin termsof self-reference,
laser itself,then,of course,it can't evershineon thatparticularpoint.

Now I would like to give you some examplesof self-reference outside Gddel's
theorem.Some of themwillbe verysimple,and some of themmay be veryfamiliar
to you: theyare intendedto give you the flavorof it. My firstexample involvesa
questionabout a laser. What is the only place in a room thata laser cannot hit if
you shine the beam? I have indicatedit by the black dot in the figure.The only
place thatlasercannothitis itsown rearend. You can shineit to any otherpointin
theroomexceptthatblack dot. However,thismatteris just a littlebit trickierthan
that.You can actuallygetit to shineon thatspot,dependingon whatyou mean by
"thatspot." If I defineit in coordinatesrelativeto theroom,then,of course,thelaser
can shinethroughit. If, though,I defineit in termsofself-reference, in otherwords,
withrespectto the laser itself,then,of course,it can't evershineon thatparticular
point.This is reminiscent of an epigrammaticstatementthatwas made about 200
years ago by the German man of lettersGeorg Lichtenberg.He said thatthe one
place in a room thata flycan land withsafetyis on thehandle of a flyswatter.

Georg Lichtenberg'sinvulnerablefly.These days,perhapstheinvulnerableflysitson top of a

helicopter!
malathion-spraying

100
 Here are threeclassic examplesof self-reference
thatoftenappear in littlesigns
thatyou can buy in stores.

THIMK I NEVER PLANAh

MAKE MISTEAKS ead
1. 2. 3.
The firstone is the mostconcise versionof Gddel's theoremthatI have ever seen.
The second one does in factcontaina mistake,since "mistake"is misspelled.In the
thirdone, it's implied that the signmakerdid not plan ahead (but on second
thought,of course, that's false). These are all variantsof self-reference, but an
interesting versionof self-reference,in that the self-reference
is indirect.In each
case, it is leftto theobserverto perceivethatwhatthesignis sayingis relatedto the
signitself.It doesn'tsay anythingsuch as "This signcontainsa mistake."It doesn't
referto itselfdirectly. Yet it does have an indirectself-reference.
Here are some directly self-referential
statements:

Hofstadter's Law

It always takeslongerthanyou expect,even whenyou

take into account Hofstadter'sLaw.

The nextone is verysimilarto Hofstadter'sLaw, and is, unfortunately,

a sad fact:

One of the lessonsof historyis thatno one ever learns

the lessonsof history. l

Finally,we come to whatis oftenregardedas the centralself-referential

statement,
namely,the EpimenidesParadox:

EpimenidesParadox

This sentenceis false.

Probablyall of you have thoughtabout theEpimenidesParadox,and realizethatif

it is true,thenit is false.And if it is false,thenit is true,and thatgetsyou into a
bind. Gddel's theoremis reallybased upon the EpimenidesParadox. I would now
like to tellyou whatGodel actuallydid, and thenfurther illustrateGddel's theorem.

101
 In orderto describewhatGodel did, I have to set thestagea bit and describethe
stateof thefoundationsof mathematicsat theend of thelast century.Mathematics
was beingaxiomatizedin an attemptto make veryclear whatdid,and whatdid not,
constitutea proofof a statement.Various axiomaticsystemshad been developed
forEuclidean geometry, non-Euclideangeometry, projectivegeometry,etc. At that
time,mathematicians such as Peano, Hilbert,and Fregewereinvolvedin axiomatiz-
ing mathematics.Perhaps the most ambitiousattemptswere made by Russell and
Whiteheadnear the beginningof this centuryin theirPrincipiaMathematica,in
whichtheyattemptedto develop all of mathematicsfromthe notionsof logic and
sets.The importantthingabout an axiomatizationof any mathematicalsystemis
thatit takes the set of conceptsthatone uses forthinkingabout thesethingsand
reducesthemto a fixedand finitevocabulary,a finiteset of symbols,and a finite
set of axioms; or ifyou want,a finiteset of axiom schemas(whereschemasare like
a mold for axioms) and a finiteset of rules of inference.Thus, everythingis
collapsed down to a formalsystem,which involvesexpressingthingsin a fixed
vocabularyaccordingto a fixedgrammarand thenevolvingtheoremsfromaxioms
accordingto fixedrules.In otherwords,theresultis a sortof typographical way of
reasoning:reasoningis turnedinto a mechanicalprocedure-verysimilarto what
computersdo nowadays in manipulatingsymbolswithinthemselves.This is what
PrincipiaMathematicawas all about: it triedto evolve all of mathematicsin one
system.
Then, Gddel, in 1930-31, twentyor so years afterthe publicationof Principia
Mathematica(1910-1913), realized that therewas somethinggoing on here that
could lead to some profoundconsequences.His idea was thattheaxiomatizationof
any branch of mathematicscreates a very interestingformalobject or formal
structure, namely,theformalsystemitself.In otherwords,PrincipiaMathematicais
not just a systemin whichmathematicsis being done, but moreoverit itselfcan
become a mathematicalobject,in the sense that one can look upon its axioms,
theorems,rules of inferenceand so forthas mathematicalobjects. The rules of
inferenceare reallythingsthatare manipulatingobjects,theobjectsbeingstringsof
symbols.So thisobservationcould have led Godel to say,"Maybe I shouldinventa
mathematicsthat applies to stringsof symbols."He would have then been the
inventorof somethingresemblinga modern programminglanguage, like Lisp,
Snobol,and otherlanguagesthatare called "stringmanipulationlanguages."But he
didn't do that. He thoughtto himself,"There is no branch of mathematicsthat
studies the propertiesof stringsof symbols.However, thereis a branch called
numbertheorywhichstudiesthepropertiesof integers.I can just replaceall of these
symbolsby integers, and thatway I can turnthestudyof PrincipiaMathematicaas
a mathematicalobject into a branch of numbertheory.I can just replace all the
stringsin PrincipiaMathematicaby numbers,and thenI can describewhathappens
on each page as a sequence of transformations of numbers."So it becomes a little
branch of numbertheory,which is an ironybecause PrincipiaMathematicawas
supposedto be a systemin whichall of mathematics was developed.But hereGodel
is in a way turningaround and saying,"But PrincipiaMathematicaitself,its own
structure,just formspartof numbertheory.If you thinkof thebook and thesystem
of theorems,the system in which the symbols are being manipulated,as a
mathematicalobject,thenit itselfhas been sortof "swallowed"by numbertheory,
whichis one of the subjectsthatit is supposed to be studying."That was thereally
trickyinsightthatGodel had, the idea thatone could turna mathematicalsystem

102
on itselfso thatit could become itsown object of study-slightlyindirectbecause it
involvesreplacingor codingsymbolsby numbersthatstandforthem.That's called
G6del numbering. That whole systemhad to be worked out in great detail to
convincemathematicians thatwhat he was doing was quite rigorousand not some
made-up sleight-of-hand. Once you realize that PrincipiaMathematica(which
includesnumbertheoryas one of itssubjects)can talkabout PrincipiaMathematica
itselfthroughthiscode called Godel numbering,thenyou can get sentencesthat
have twolevelsof interpretation ratherthan one. Earlier,a givensentencewas only
thoughtof as speakingabout numbers. (It could say somethinglike "641 is prime.")
But now, thereemergesa secondlevel of interpretation, because numbersrepresent
statements. Someonecould say,"Really,thisstatementsays somethingabout strings
in PrincipiaMathematica"!
The final trickis to finda specificstringthat can say somethingabout itself.
What Godel foundwas thatit was possibleto finda sentencethatsaid this: "This
sentenceis not provable." And this is where we come back to the Epimenides
Paradox. The EpimenidesParadox states,"This sentenceis false." In mathematics
up until that time the idea of truthwas exactly equated with provability.In
particular,ifone took thesystemof PrincipiaMathematica(whichwas supposed to
be all-inclusive),the idea of provability withinPrincipiaMathematicawould have
been synonymous withtruth.So, to say "This sentenceis not provable"would have
been synonymouswith saying "This sentenceis not true,"or, "This sentenceis
false." But if that were reallywhat Gddel's sentencesaid, then PrincipiaMathe-
maticawould have a statementin it thatwas neithertruenor false.This would be a
statementabout numbers,neithertruenorfalse,and thatseemsimpossible.It seems
that if it's true,then it's false, and if it's false, then it's true. That just seems
contradictory, and people had to thinkvery hard to figureit out. (But Godel
perhapsdidn'thave to thinktoo hard.)
Even afterGodel had made it veryclear in his paper what the consequencesof
thiswere,a lot of people debated for many years about the differences between
truthand provability. What Godel showedwas thattherewas actuallya distinction
betweenprovabilitywithinany specificsystemand truth.Consider Gddel's sen-
tence,forexample:

Godel' s Sentence

"This sentenceis not provable."

(More precisely,"This sentence is not provable in
formalsystemX.")

Thereare twopossibilities.Eitherit is provable,or it is not provable. Those are the

twopossibilities.If it is provable,thenit is true,so whatit sayshas to be true,and it
says it's not provable.That is reallya contradiction, an absolute contradiction.If
thingswere thatway, thenmathematicswould be inconsistent, and thatwould be
prettymuch theend of mathematics. Thus thatleaves us withonlyone alternative,
namely,thatthesentenceis notprovable(and that'swhatit says about itself).So it
mustbe true. Now, here is the distinction-hereis the trick.It could be trueand
yetnot be provable,whichwould mean the systemwas not quite as strongas one
had thought.That is, PrincipiaMathematicaactually had a defect-it was not

103
 One reallyhas to specifywhatone means by
capable of provingall truestatements.
"provable." You can't just say "provable" in some etherealsense-you have to
specifythesystemyou mean.
In passing,I shouldalso mentionthatLeon Henkininventeda different kindof
sentence:

Henkin's Sentence

"This sentenceis provable."

Henkin'ssentence,by theway,is not thenegationof Gddel's sentence.Do you see

why?
I now want to give you an image to hold in your mind of a formalsystemin
whichone can expressand prove mathematicalstatements.As in my book, I call
the systemTypographicalNumber Theory (TNT). That's appropriatebecause it
sortof "blows up." I'm notgoingto presentit by any meansin fullbecause it would
take too much space, and I just want to give you the flavorof it. Here are five
axioms of arithmetic writtenin its notationjust so you have a sense of the way in
whichone can expressstatementsin a formalsystem.

TYPOGRAPHICAL NUMBER THEORY (TNT):

A formalsystemin whichone can express
and provemathematicalstatements

Five Axioms

1. Va: -Sa = O
2. Va:(a+O)=a
3. Va: Va': (a + Sa') = S(a + a')
4. Va: (a O) = O
5. Va: Va': (a . Sa') = ((a . a') + a)

and 18 rulesof manipulation.

The upside-downA, "V" is a quantifierand means "forall," the colon is just a

punctuationmark,and thelittleswirl,"-," called a tilde,means "not," and finally
thecapital"S" standsfortheidea of "the successorof,"or "one morethan."So the
firstaxiom says: "For all a, it is not thecase thatthe successorof a equals 0," and
in more understandableEnglish,it says: "No number'ssuccessoris 0." Well, you
could say "Minus one has a successor,and it is 0." But what thisaxiom is doing is
tellingyou what the domain is, and it is saying,"Minus one is not in the domain.
There is no numberwhose successoris 0." So it is basicallysaying"0 is thelowest
number."
The secondone is easier.It says: "For all a, a plus 0 equals a." The nextone says
somethingabout theway additionworks.It says: "If you have any two numbersa

104
and a', thena plus the successorof a' is equal to the successorof a plus a'." The
next one says: "Multiplyingby 0 gives you 0 always." Then the next one is a
primitive formof thedistributivelaw. It says: "If you multiplya withthesuccessor
of a', thenyou geta timesa' plus a." These axioms are to be consideredas strings,
as inviolateobjects that are simplycapable of being manipulated.You can start
manipulatingaccordingto the 18 rules of manipulation(whichI am not going to
exhibit).They simplyinvolvemovingsymbolsto the rightor to the leftor taking
symbolsout and puttingothersin, and so forth. They are veryformal,somewhat
like a computerprogramactingon objects. Presumablythose 18 rulesof manipula-
tion are well thoughtout, so that theynevertake you to a false statementif you
startedwithtrueones. That's the idea of TypographicalNumberTheory.

THE GODEL ISOMORPHISM

0 ... ... 666

S ... ... 123
=... ... 111
+... ... 112
... ... 236
( .... 362
)... ... 323
... ...212
> ... ...213
. . . 312
. . . 313
a . 262
... 163
A . 161
V. 616
D. 633
. . 223
3. 333
V. 626
.. 636

The idea thatGodel had was thatbecause thisis such a formalprocess,one can
see thisas a mathematicaloperation-the idea of manipulatingsymbols.Now he
did not inventthe idea of a stringmanipulationlanguage. He simplysaid "Let's
replace all thesesymbolsby numbers,and thenwe'll have a numbermanipulation
system."The way he chose to do it is not theway I am goingto show you,but the
way I am goingto showyou is just as good. It worksjust as well.One has to replace
each symbolby some number.In the box, I show each symbolof TNT (and there
are 20 of them)corresponding to a three-digitnumberso thatone can "translate"a
stringof TNT symbolsintoa long numeralthatconsistsof a sequenceof three-digit
numerals.(I'll soon show you how.) Once you graspthisidea of replacingsymbols
by numbers,long stringsof symbolsby biggernumbersand so forth,thenyou have
theidea thatthe systemcan speak about itselfin code. Once you understandthat,
you can go on to hypothesizethatmaybesome stringcould be inventedthatspeaks
about itself,and says about itselfthatit is notprovable.I am goingto come back to

105
thatquestionof how a statementcan be constructedthatspeaks about itselfat the
end of the talk.Let's take thatforgrantedforthemoment.If you will assume that
one can constructsuch a statementin the languageof PrincipiaMathematicaor in
thelanguageof TNT, or in any formalsystemof thissort,thenyou mightask this
question,whichis a veryimportantquestion:Is Gddel's sentencereallya statement
about numbers, or a statementabout itself?Rememberit is a stringin some formal
system,and ostensiblythatformalsystemcontainsstatements about integersas did
TNT.
For example,"a times0 is 0 foreverya." That's a statementabout integers.Or
you could say somethinglike "17 is prime,"in thatlanguage.Basically,thatsounds
like a statementabout 17. So if you have a verylong and complexstatementthat
says somethingverycomplicatedabout numbers,how can it also say something
about itself?Does it reallysay somethingabout itself?It is a questionof levelof
descriptionand that'sa veryimportantidea.
Let me now take a minute to talk about computers.Everybodyspeaks of
computers,at bottom,as manipulating"bits." "What you have in computersare l's
and 0's,"-that's what people say. "You move these l's and 0's around in fancy
ways, and that'sreallyall a computercan do." That's sort of a strangeway of
speakingabout it because we are also veryhappy to say thata computermanipu-
lates letters.We talkabout "text-handling programs"all thetime.Is it wrongto say
thatcomputersmanipulatetexts?Do theyreallyjust manipulatel's and 0's? Are
theyreallyjust doing binaryarithmetic? Of course,if you say theyare just doing
binaryarithmetic, whyshouldyou stop there?You can say, "There are no I's and
0's in a computer.All that'sthereis currentgoingoffand on. And so reallyall that
a computercan do is manipulateelectrons.It can't manipulatel's and 0's. It can't
do any arithmetic at all! It can just manipulateelectrons!"Thereis a pointat which
you wantto stopgoingdownin thehierarchy of description. You mightwantto say
thata computercan manipulate1's and 0's. Or if it is convenient,you mightwant
to stopabovethatlevel.You mightwantto say,"A computercan reallymanipulate
letters."
You mightalso want to stop fora minuteand changegears to consideranother
image.You mightask: "What is a novel?Is a novel reallya set of letters?Or is it a
set of words?Or is it a set of sentences?Or is it a set of ideas?Or is it a set of events?
What is a novel?" It would certainlybe wrongto say a novel is a set of lettersor
words,because we say the novel could be translatedinto another language.
Ultimately,whenwe remembera novel,we don't rememberany of the letters,we
don't rememberany of thewords(or onlya fewof them).We remembertheevents.
But then-doesn't stylecount?It is veryhard to determinethe rightanswerto the
question"What is a novel?" In some sense,all of the different answersare right.I
am tryingto make certain you understandmy point of view. I'm not saying
anythingis wrongwithsayingG6del's sentenceis a statementabout numbers, but
by thesame tokenone shouldn'tfeelbad about sayingit is a statementabout itself.
It's bothof thosethings!It is bothof thesethingssimultaneously. Neitherviewof it
is morerightor morewrongthan the other.
To giveyou a humorousversionof Gddel's theorem,considerthe"Wild Dance"
versionof G6del's theorem.This is a real-lifeexamplewhichhappened at a party.
At the party,certainrecordscould not be played. Actually,mostof themcouldn't
be played,because people would dance to them.As soon as theydanced, the floor
would startto vibrate,therecordplayerwould shake,the needle would bounce all

106
 Gidel's Theorem(Wild Dance Version)
"This recordcannotbe playedat thisparty"

/ (((X1
K I

- Loudspeaker-* Air-* Eardrums

Grooves- Needle - Electricity >
Phonograph<- Table <- Floor <- Muscles <- Nerves<- Brain GD

Playinga recordat a partycan createa "strangeloop"-a sortof generalized,level-crossing

feedback-which unexpectedlypreventsthe record frombeing played. This is a real-life
analogue to Godel's constructionof a self-referential
(and self-underminingsentence)in a
formalmathematicalsystem.Both involvethe fact thatsymbolicactivity(musical vibrations
or provingof theorems)has simultaneously and theselevelsof
severallevelsof interpretation,
far frombeing just "differentways of looking at one thing,"can actually
interpretation,
witheach otherand cause near-paradox(G6del) or wreakhavoc (at theparty).
interfere

over the place, so it wouldn't be reproducingthe music any more. It was a

self-destroyingdance! This givesus a questionto ask. We have a set of "levels of
translation"of thegrooves.The questionis: "Is theplayingof a recordreally: a set
of vibrationsin a loudspeaker,or waves in the air, or shakingsof the floor,or
motionsof people's feet,or quiveringsof thetable thattherecordplayeris on?" It's
all of thesethingssimultaneously. Wheneverone of thesethingshappens,the next
one is setin motion.If you have dancersthere,people willstartdancing.When they
startdancing,the floorvibrates.When the floorvibrates,the table vibrates.When
the table vibrates,thephonographwillvibrate.When thephonographvibrates,the
recordwillvibrate.So theloop willclose. All of theseare valid descriptionsof what
a record,when played, is doing. This particularexample of the recordis a very
interestingone because one can turn it into a kind of paraphrase of G6del's
theoremby saying:"For everyrecordplayer,thereare recordsthatit can't play."
One doesn't even requirethe dancers.Afterall, the sounds all by themselveswill
vibratethe recordplayeritself.So one can say thereis automaticallysome sortof
self-destroyingrecord.For everyrecordplayer,thereis a recordthatwill make it
vibrate and fall apart. After all, every record player has a special resonant
frequencyat which it will startvibratingmore and more,and eventuallyit will
break.You mightsay, "Oh, no, it doesn't.It mightjust be a verywell made one."

107
But certainlyifyou make a loudenoughnoise,anything willbreak.If you reproduce
the sound of an atomic bomb in perfectfidelity,it will destroythe recordplayer.
Givenany recordplayer,thereis some bombthatwilldestroyit You simplyrequire
that the recordhave the noise of that bomb in its groovesand then the record
playerwillnot be able to play thatrecordin fullfidelity.This is anotherimage of
how to thinkof G6del's proof.

(a) A trin in a shuntingyard.Or-is it a kbg neral on a of paper? Or-is it the

numberof cigarettessmokedin all the galaxiesin the universesincethe ig mBag?

I have one moreimage of G6del's proofthatI would like to giveyou. The idea
here involveslikeningsymbolmanipulationto the movingof railroadcars in a
shuntingyard. We have an engineerin a locomotive,and thisengineerhas been
in theshuntingyard.He is blindlyfollowingthese
toldto followcertaininstructions
The instructions
instructions. involvethe numberson the sides of the boxcars.He
doesn'tknowwhatis insidetheboxcars.All he knowsis thathe mustfollowcertain
rules,and he can move cars around in one way or anotherin theyard,and he tries
to createcertaintrainsaccordingto the rules.For example,thereis one rule that
saysyou can detachany threecars thathave thepattern,6 I, i, 9[ i
wherethemiddleone can be anything(see figure).That's one rulethathe can obey.
He can thendetachthemand theygo away. He is leftwiththeset of cars as shown
in thefigure.

[3} [i],[p],so he does

(b) The engineerhas a ruk by whichhe can detachthreecars whosesidesexhibitthepattern
i leavinga fiver ain. This is describable ihas a
the
railroad shuntingoperation.or as .do g ."ka manipulationof some numwerals
(Itnup of digits).or Ially as thec-ksdao- of one number(magnitude)gi anote
numaberacordin to an wHinetggI ndre. The suble differee liet the la two i
vay sagnifiantL

Anotherrule that he can obey is this one: "Replace 262 by 666." Then he can
replacethatone as shown.The engineeris an intelligentguy,and as he is shunting
thesecars around,he realizesthat,in fact,he can thinkofjust the numberson the
sides. He doesn'thave to actuallydo thephysicalmanipulationof thecars.

108
 [ [ [
[
(c) Another suchas [i
rulesaysthata pattern ... can be substituted

for . So in the five-cartrain,the engineerperforms

the simplest
of all such

substitutions: just [1
namely. for [ Thuswe havemathematically
converted
one
verylargenumber,in twosteps,intoa somewhatsmaller,but stil huge,number.Or is it
thatwe rearrangedsomefreight,in a railroadshuntingyard,intoa different
order?And
whatkindof freightmightit be?

He can just thinkof thesenumberson theside of thetrainand do it all in his head!

Ratherthan talk about "assemblabletrains"(because some trainsare assemblable
according to these rules, and others are not), he can just as well talk about
" produciblenumbers."What I mean by a "produciblenumber"is thelong number
thatstands forany whole assemblabletrain.In thiscase (see figure),the number
would be 223,123,666,111,666.It would be a produciblenumberbecause he started
withthetrainin one positionthatwas assemblableand did certainthingsaccording
to therules,and he came up withanothertrain(i.e., number).Thus he could start
thinkingabout thenumbersthathe was producing-theselong,giganticnumbers-
and ask if theyhad some number-theoretical propertythatdistinguishesthemfrom
thenon-producible numbers.Of course,theanswerwould have to be "yes",because
ifyou thinkabout it you willsee all thathe is doing are formaloperations.Butany
formaloperationon these numbersis really a mathematicaloperation.It is an
operationthatbelongswithinthedisciplineor the framework of numbertheory.It
is a number-theoreticaloperation.The set of numbersattainable throughthese
manipulationsconstitutesa set of numbersdefinablein numbertheory.Thus, he
can become a numbertheoristinsteadof an engineer,and decide whichtrainsare
assemblableand whichones are not.This is somethinghe findsquitefascinating. So
he starts investigatingthe propertiesof these huge producible numbers,and,
flippingthecoin, of the non-produciblenumbers.

(d) View froma helicopterof thetrainin (a). Here is our answeras to whatkindof freight is
involved.Each car is carryinga symbolof theformalsystemTNT! Shuntingcars in the
railroadyard is simultaneouslyan act of TNT symbol-manipulation (seen fromabove),
and numericalcomputation(seen fromtheside) Theorem-proving in a formalsystem,as
seen by G6del (whose numbersadorn the sides of the cars) is equivalentto number-
theoreticalcalculationaccordingto certaincomputationalrules.

109
 (e) Helicopterviewof the trainin (c). This trainexpressesa theoremof TNT thatis a direct
consequenceof theaxiom seen in (d). The theoremsays: -One is notzero" The engineer
created this theoremnot by thinkingabout how to deduce logical consequences of
axioms,but by followingrulesgoverningnumberson the sides of thecars. Yet froman
outsider'spointof view,theyare equivalent.No one could tellwhichway the engineer
was thinking about his activity!This equivalencebetweendoing reasoningabout mathe-
matics and doing mere calculation withinmathematicsyields a startlinglevel-rossing
loop. allowinga mathematicalsystemto talk about itself.It's the gistof G6del's insight
paradoxicalstatementscan be translatedinto formalmathematical
thatself-referential.
systems.by meansof codes wherenumbersstand forsymbols.

Quite coincidentally,he also happens to have studied TNT (Typographical

NumberTheory),and he is veryinterested in that,too, but thatis just anotheridea
to him.One day he happensto be takinga helicopterride,and he getsa top viewof
whathis trainslook like. He getsto see what'sinsidethe boxcars(see figure).This
one in particularhas one of the fiveaxioms of TNT in it: "For all a, it is not the
case thatthesuccessorof a equals 0." Now rememberthatone of thethingshe was
allowed to do was removethesethreecars so thatit lefthimwitha shorterstring,in
whichhe thenreplacedthe262 by a 666. Well, the666 cars all have 0's inside(see
figure).In fact,thecorrespondence betweena car's side numberand thesymbolit is
carryingis exactlytheone thatI exhibitedearlier-the one thathad all thesymbols
of TNT withtheirthree-digit codes, so thatall the 666 cars carry0's inside.So, in
fact,what he had unknowingly startedout withwas an axiom-carrying train;and
by manipulating a littlebit,he came up withthisnew theorem-carrying train,which
seen fromabove says, "It is not the case that I equals 0"-a true sentencethat
followsfromthe otherone because the otherone says: "No number'ssuccessoris
0." This one says: "It is not thecase thatthesuccessorof 0 is 0," and I just call the
successorof 0 "one". This new trainsays "It is not thecase that 1 equals 0." This is
a revelationbecause now he sees thatalthoughhe had thoughtthatall he was doing
was constructingproducible numbers,it turns out that from another (literally
higher-level)viewpoint,he was actuallymanipulating stringsof TNT. In fact,he was
producingstatementsof numbertheorywhen seen fromabove! Fromtheside, he
was just producingnumbers; but fromabove he was producingstatementsof
numbertheory(and theywere all true).Thus, he could conceive of the idea that
some train,some verylong train,mightbe talkingabout its own number,its own
code number,and it mightbe saying somethingabout the producibilityof that
number.Rememberthatproducibility (like,say, primality)is a number-theoretical
notion,and if our formalsystemis at all powerful,then it can certainlyreason
about this notion of produciblenumbers.Thus, some very long train mightbe
talkingabout whetheror not it itselfis producibleand mighteven say,"This trainis
not producible;thistrainis not assemblable."Now thatis a verystrangeidea. I
would like to suggestto you how thatcould happen.

110
 This is veryhard to figureout; it is the last trickthatGodel used. "How do you
geta stringof some formalsystemnotjust to speak about otherstringsin theformal
system,but to speak exactlyabout itself?If you trynaivelyto make somethingthat
speaks about itself,you mighttryto quote a sentencewithinitself.(I am speaking
now of English.)If I tryto make a sentencethat talks about itself,thereare two
ways I mightuse. One is to say somethinglike,"This sentenceis false." Now how
does thatwork?It is based on the conventionthatthephrase"thissentence"refers
to thesentenceit is containedin. Or I could say somethinglike,"I am lying."That
is using the conventionthat"I" stands forthe speaker.We are using in a certain
sense a sleightof hand. I don't know what you want to call it, but it is a kind of
convention.There is anotherway of achievingself-reference, and thatis more like
the one that said "THIMK," wherethe self-reference is more indirect,wherethe
thingthatis expressingsomethinghas thesameformas thethingit is talkingabout.
If you wantto have a sentencethattalksabout its own formwithoutreferring to
you have to have some sortof way of quotingit withinitself,and you
itselfdirectly,
come to somethinglike thisexample(see figure).We have herean attemptto create
a sentencethatspeaks about itself,and it works.It talksabout a quoted sentence
whichis infinitely long. So the sentenceitselfhas to be infinitely long,and lo and
behold! The quoted sentenceis the sentenceitself! It is a sequence of nested
sentences,each of whichis quoted withinthe outerone, and since it is an infinite
structure we have theoutermostone identicalto theone thatis one levelin, and it's
identicalto the sentenceone level further in, and so forth.But thisis not goingto
satisfyus if we are tryingto producea finitesentence.
So how do you do it in a finitesentence?It mightseem impossible.Godel found
a way to do it, and basicallyit involvesthe idea of diagonalizationas in Cantor's
argumentthatshowsthereal numbersare uncountable.

The sentence
"The sentence
"The sentence
"The sentence

0.~~~~~~~~~~~~~~~9

is infinitelylong."
is infinitely
long."
is infinitely
long."
is infinitely
long.

I am going to show you a way of 0.~ doing it~in~ English.

~ ~Instead
~ ~of showing
~~~~1 you via
Cantor, the mathematician,I'll show you throughQuine, the philosopher.Quine
invented the followingmethod. The idea is of self-reference via form not via
conventionsuch as "I" or "thissentence.""Quining"is whatI called it in mybook.
(He certainlydidn'tcall it that!)Quiningis an operationthatI defineon any string
of English.Take thestring,put it in quotes,thentake anothercopy and put it right
afterthe quoted one.
So you have twocopies of a string,and thefirstone is in quotes-that's all. Here is
an exampleof a quined phrase:

"is a sentencewithno subject"

is a sentencewithno subject.

It is a perfectlytrue sentence.Neitherhalf is a sentence,but the full thingis a

sentence.(And it is a sentencewitha subject,incidentally!)The nextone, though,is
and is one whereyou get real self-reference.
a littletrickier

"yieldsfalsehoodwhenprecededbyitsquotation"
yieldsfalsehoodwhenprecededbyitsquotation.

What is thisreallysaying?The second part of the sentenceis active: it's the thing
that is speaking.It is referringto some phrase in quotes. What is it saying?It is
sayingthat something,when you precede it by its quotation,yields a false state-
ment.Well, what is thatthing?That thingis the verythingthatis speaking.And
when you precede it by its quotation, you more or less "accidentally" have
reconstructed the sentenceitself.So, in fact,thissentenceis talkingabout itselfby
means of form,not by convention.
Godel figuredout a way of doing thatin mathematicalsystems.The way is very
simple,because quoting somethingis very much like takingits Godel number.
Convertinga stringinto a quoted stringis verymuch like converting a stringto its
Godel number.Godel realized that one could parallel this constructionprecisely
withina formalsystemand create self-reference thisway. Actually,of course,it
happened the otherway around. Godel didn't see this sentenceby Quine. Quine
inventedit afterGddel's work.It was just meant to explicatewhat is going on in
Gddel's proof.
My last pictureshowsa parallelphenomenonthathappensinsidelifeforms.The
way self-reference happens in mathematicsor formalmathematicalsystemsis very
similarto the way self-replication happens in livingorganisms.How does a fish
reproduceitselfalong withits DNA inside?To give you a glimpseof what I am
going to say, the fishis sortof like the unquoted string,and its DNA is like the
quoted string.So firsttake the DNA out by itself(see figure).Now let the DNA
develop according to certain chemicalprocesses. (This is the analogue to the
typographicalprocessof quining.)The DNA develops,and we see it startto develop
here,developfurther, further,and you see whatresults?I willjust say thisis a sort
of an answerto an age-old question.It is a verysexistquestion. It asumes that
femaleova play no role in reproduction of thespeciesat all, thattheyarejust there
forfood,and thatreallythe spermscontainall the hereditary material.Justmake
thissimplifying assumption:thata man containssperms,and since his spermsare
goingto turnintopeople, theytoo mustcontainsubsperms,and theymustcontain
subsubsperms,etc., etc., ad infinitum.The medieval puzzler is: "Does a man
containall of his futureprogenyforall generationsin some sortof infiniteregress?"
The answeris, no, notreally,not evenifreproduction wentthatway-no morethan
a fish'sDNA is actuallya fish.

112
 DNA
DWA

of thefish,but also programsits own replication.The

DNA not only codes forthe structure
methodby whichthisis accomplishedis analogous to the way in whichself-referencecan be
achievedin a formalmathematicalsystem.

was a GuggenheimFellow at Stanford

of Indiana University
Douglas R. Hofstadter
Universityduring 1980-1981. In 1980 he won the Pulitzer prize in the category of
general nonfictionforhis book "Godel, Escher, Bach: an Eternal Golden Braid."
He was born in New York Cityin 1945, but spent most of his childhood years at
Stanford,California.He was always interestedin mathematicsand physics as a child.
At age 12, he began learning French and became fascinated by languages. In college
he studied several other languages, as well as computer science. As an undergradu-
ate, he used computers to experiment withseveral ideas about language structure,
including foraysinto artificialintelligence research. He also used the computer to do
what he calls "experimental math," particularlynumber theory "experiments." The
result was a general fascination withrecursion and logic.
He began his graduate studies at the Universityof California,Berkeley, in mathe-
matics, but later switched to physics at the Universityof Oregon.
Music has been equally dominant in his life and includes piano playing and
composition.

113
 His many interests-logic, computers, languages, music, and the mind-
recombinedin 1972 and set offwhathe describesas a "mentalfire"thatburnedfor
severalyears. The outcomewas "Godel,Escher,Bach," and, afterfinishing his Ph.D.
in physics,a returnto his old interest:
artificial
intelligence.
He recentlytalkedabout his work."Mycurrentresearchin artificial intelligence
involvesattemptsto get a machine to see patternsin simple structuresand to
generalize those patternsin "natural"or "artistic"ways. You could call this an
attemptto program"artificial intuition."Mypassion is to gain insightintotheintuitive
processes of thought."

TelegraphicReviews
Edited by
Peter A. Lindstrom

Telegraphic Reviews are designed to give the TYCMJ readers short summaries of the contents of
recently published mathematics texts, supplementary materials, mathematics laboratory materials, and
professional readings. Such information can be helpful in deciding whether or not to order an
examination copy and whether or not the text,supplementary material, or professional reading material
might be purchased for a mathematics laboratory, college library,or personal library. Publishers are
denoted by standard abbreviations; complete addresses may be found in Books in Print.

Codes Used in Telegraphic Reviews

T is Textbook * is Special Positive Emphasis

S is Supplementary Material ? is Special Negative Emphasis
L is LibraryPurchase 1, 2, 3, 4 is Number of Semesters
ML is Mathematics Lab Material for Use of Textbook
P is Professional Reading

APPLICATIONS(S**,1): Toolsfor Teaching, Compiledby the UMAP CentralStaff.

Birkhauser Boston,Inc.,1981,ix+ 727pp.,$35.00.
This book containstwenty-four UMAP (Undergraduate Mathematics and Its Applica-
tionsProject)modules.Each moduleis a self-contained, instructional
lesson-length, unit
fromwhichundergraduate students
can learnprofessional of mathematics
applications and
to suchfieldsas biomedical
statistics sciences, American
economics, earthscience,
politics,
Thisis onebookthatwillNOT collectduston yourbookshelf
and thesocialsciences. as you
willuse itoften.PAL

ALGEBRA(COLLEGE) (T*,1): CollegeAlgebra,by Jimmie Gilbert,JamesSpencerand

LindaGilbert. P-H, 1981,xv+ 432pp.,$16.95.
Includesrealnumbers, polynomials,exponentsand radicals,linearand quadraticequa-
tionsand functions,conicsections,
systems of equationsand inequalities,
determinantsand
matrices,exponentsand logs. Informalapproachwitheffective use of discussionand
illustration.
Numerousand well-explained examples.Practicetestsfolloweach chapter.
Definitions
and keypointsareemphasized stated.LMC
and carefully
[Continuedon p. 163]

114