/LMAR=0/XLINE=3/FONT#0=BAXL30/FONT#1=BAXM30/FONT#2=BAXB30/FONT#3=SUB/FONT#4=SUP/FONT#5=BASL35/FONT#6=NGR25/FONT#7=MATH30/FONT#8=FIX25/FONT#9=GRK30



␈↓ ↓H␈↓␈↓ ε(draft





















␈↓ ↓H␈↓α␈↓ β[ASCRIBING MENTAL QUALITIES TO MACHINES

␈↓ ↓H␈↓Abstract:␈αAscribing␈αmental␈α
qualities␈αlike␈α␈↓↓beliefs,␈↓␈α
␈↓↓intentions␈↓␈αand␈α␈↓↓wants␈↓␈α
to␈αa␈αmachine␈α
is␈αsometimes
␈↓ ↓H␈↓correct␈αif␈αdone␈αconservatively␈αand␈αis␈αsometimes␈α
necessary␈αto␈αexpress␈αwhat␈αis␈αknown␈αabout␈αits␈α
state.
␈↓ ↓H␈↓We␈α∞propose␈α∞some␈α∞new␈α∞definitional␈α∞tools␈α∞for␈α∞this:␈α∞ definitions␈α∞relative␈α∞to␈α∞an␈α∞approximate␈α
theory
␈↓ ↓H␈↓and second order structural definitions.






␈↓ ↓H␈↓(this draft of MENTAL[F76,JMC]@SU-AI compiled at 11:26 on December 12, 1977)
␈↓ ↓H␈↓␈↓ εH␈↓ 91


␈↓ ↓H␈↓αINTRODUCTION

␈↓ ↓H␈↓␈↓ α_To␈αascribe␈αcertain␈α␈↓↓beliefs␈↓,␈α␈↓↓knowledge␈↓,␈α
␈↓↓free␈αwill␈↓,␈α␈↓↓intentions␈↓,␈α ␈↓↓consciousness␈↓,␈α ␈↓↓abilities␈↓␈α
 or␈α␈↓↓wants␈↓
␈↓ ↓H␈↓to␈α
a␈α
machine␈α
or␈α∞computer␈α
program␈α
is␈α
␈↓αlegitimate␈↓␈α
 when␈α∞such␈α
an␈α
ascription␈α
expresses␈α∞ the␈α
 same
␈↓ ↓H␈↓information␈α∂ about␈α∂ the␈α∂machine␈α∂that␈α∂it␈α∂expresses␈α∂about␈α∂a␈α∂person.␈α∂    It␈α∂ is␈α∂ ␈↓αuseful␈↓␈α∂ when␈α∞ the
␈↓ ↓H␈↓ascription␈α helps␈α us␈αunderstand␈αthe␈αstructure␈αof␈αthe␈αmachine,␈αits␈αpast␈αor␈αfuture␈αbehavior,␈αor␈αhow
␈↓ ↓H␈↓to␈αrepair␈α
or␈αimprove␈α
it.␈α It␈αis␈α
perhaps␈αnever␈α
␈↓αlogically␈αrequired␈↓␈αeven␈α
for␈αhumans,␈α
but␈αexpressing
␈↓ ↓H␈↓reasonably␈α
briefly␈α
what␈α
is␈α
actually␈α
known␈α
about␈α
the␈α
state␈α
of␈α
a␈α
machine␈α
in␈α
a␈α
particular␈α
situation
␈↓ ↓H␈↓may␈α∩require␈α∩ascribing␈α⊃mental␈α∩qualities␈α∩or␈α⊃qualities␈α∩isomorphic␈α∩to␈α⊃them␈↓∧1␈↓.␈α∩ Theories␈α∩of␈α⊃belief,
␈↓ ↓H␈↓knowledge␈α
and␈α
wanting␈α
can␈α
be␈α
constructed␈α
for␈αmachines␈α
in␈α
a␈α
simpler␈α
setting␈α
than␈α
for␈αhumans␈α
and
␈↓ ↓H␈↓later␈αapplied␈αto␈αhumans.␈α Ascription␈αof␈αmental␈αqualities␈αis␈α␈↓αmost␈αstraightforward␈↓␈αfor␈αmachines␈αof
␈↓ ↓H␈↓known␈α∞structure␈α∞such␈α∞as␈α∞thermostats␈α∞and␈α∂computer␈α∞operating␈α∞systems,␈α∞but␈α∞is␈α∞␈↓αmost␈α∂useful␈↓␈α∞when
␈↓ ↓H␈↓applied to entities whose structure is very incompletely known.

␈↓ ↓H␈↓␈↓ α_These␈α views␈αare␈αmotivated␈αby␈αwork␈αin␈α artificial␈αintelligence␈↓∧2␈↓␈α(abbreviated␈α AI).␈α They␈αcan
␈↓ ↓H␈↓be␈α∂taken␈α⊂as␈α∂asserting␈α∂that␈α⊂many␈α∂of␈α∂the␈α⊂philosophical␈α∂problems␈α∂ of␈α⊂mind␈α∂take␈α∂a␈α⊂concrete␈α∂form
␈↓ ↓H␈↓when␈αone␈α takes␈α seriously␈α the␈αidea␈α of␈α making␈αmachines␈α behave␈αintelligently.␈α In␈αparticular,␈αAI
␈↓ ↓H␈↓raises␈α⊃for␈α⊃machines␈α⊃two␈α⊃issues␈α⊃that␈α⊃have␈α⊃heretofore␈α⊃been␈α⊃considered␈α⊃only␈α⊃in␈α⊃connection␈α⊃with
␈↓ ↓H␈↓people.

␈↓ ↓H␈↓␈↓ α_First,␈αin␈αdesigning␈αintelligent␈αprograms␈α and␈αlooking␈αat␈αthem␈αfrom␈α the␈αoutside␈α we␈αneed␈α to
␈↓ ↓H␈↓determine␈α the␈αconditions␈α
 under␈αwhich␈αspecific␈αmental␈α
and␈αvolitional␈αterms␈α are␈α
applicable.␈α We
␈↓ ↓H␈↓can␈α exemplify␈αthese␈α problems␈αby␈α asking␈α when␈α
might␈αit␈αbe␈αlegitimate␈α to␈α say␈αabout␈α a␈α
machine,
␈↓ ↓H␈↓␈↓↓" It knows I want a reservation to Boston, and it can give it to me, but it won't"␈↓.

␈↓ ↓H␈↓␈↓ α_Second,␈αwhen␈αwe␈α want␈αa␈α␈↓αgenerally␈αintelligent␈↓␈↓∧3␈↓␈αcomputer␈αprogram,␈αwe␈αmust␈αbuild␈αinto␈αit␈αa
␈↓ ↓H␈↓␈↓αgeneral␈α∃view␈↓␈α∃of␈α∃what␈α⊗the␈α∃world␈α∃is␈α∃like␈α∃with␈α⊗ especial␈α∃attention␈α∃to␈α∃facts␈α∃ about␈α⊗how␈α∃the
␈↓ ↓H␈↓information␈α
required␈α
to␈αsolve␈α
problems␈α
is␈α
to␈αbe␈α
 obtained␈α
and␈α
used.␈α Thus␈α
we␈α
must␈α
provide␈α  it
␈↓ ↓H␈↓with␈α
  some␈α
  kind␈α
  of␈α
  ␈↓↓metaphysics␈↓␈α
(general␈α
world-view)␈α
 and␈α
␈↓↓epistemology␈↓␈α
(theory␈αof␈α
knowledge)
␈↓ ↓H␈↓however naive.

␈↓ ↓H␈↓␈↓ α_As␈α⊂ much␈α⊂ as␈α⊂ possible,␈α⊂ we␈α⊃ will␈α⊂ ascribe␈α⊂ mental␈α⊂ qualities␈α⊂separately␈α⊂from␈α⊃ each␈α⊂other
␈↓ ↓H␈↓instead␈α
of␈α
bundling␈α
 them␈α
 in␈α
a␈α
concept␈α
of␈α
mind.␈α
  This␈α
is␈α
necessary,␈α
 because␈α∞present␈α
machines
␈↓ ↓H␈↓have␈αrather␈α  varied␈α little␈α  minds;␈α the␈α  mental␈α qualities␈α  that␈α can␈αlegitimately␈α be␈αascribed␈αto
␈↓ ↓H␈↓them␈αare␈αfew␈α and␈αdiffer␈α
from␈αmachine␈αto␈αmachine.␈α  We␈α
 will␈α not␈α even␈αtry␈α to␈α meet␈α
 objections
␈↓ ↓H␈↓like, ␈↓↓"Unless it also  does X, it is illegitimate to  speak of its having mental qualities."␈↓

␈↓ ↓H␈↓␈↓ α_Machines␈α
 as␈α
 simple␈α∞ as␈α
thermostats␈α
 can␈α∞ be␈α
 said␈α
 to␈α∞ have␈α
beliefs,␈α
 and␈α∞having␈α
 beliefs
␈↓ ↓H␈↓seems␈α⊃ to␈α∩be␈α⊃ a␈α∩characteristic␈α⊃ of␈α⊃most␈α∩machines␈α⊃ capable␈α∩of␈α⊃ problem␈α∩ solving␈α⊃ performance.
␈↓ ↓H␈↓However,␈α⊂ the␈α⊂machines␈α⊂mankind␈α⊃has␈α⊂so␈α⊂far␈α⊂found␈α⊂ it␈α⊃useful␈α⊂to␈α⊂construct␈α⊂rarely␈α⊃have␈α⊂beliefs
␈↓ ↓H␈↓about␈α∂beliefs,␈α⊂although␈α∂such␈α∂ beliefs␈α⊂will␈α∂be␈α⊂ needed␈α∂by␈α∂computer␈α⊂programs␈α∂that␈α⊂reason␈α∂about
␈↓ ↓H␈↓what␈α∪ knowledge␈α∪they␈α∪lack␈α∪and␈α∪where␈α∀to␈α∪ get␈α∪it.␈α∪ Mental␈α∪qualities␈α∪ peculiar␈α∀to␈α∪human-like
␈↓ ↓H␈↓motivational␈αstructures␈↓∧4␈↓,␈αsuch␈α as␈αlove␈αand␈αhate,␈αwill␈αnot␈αbe␈αrequired␈αfor␈αintelligent␈αbehavior,␈αbut
␈↓ ↓H␈↓we␈α∞could␈α∞probably␈α∞program␈α∞ computers␈α∞to␈α∞exhibit␈α∞them␈α∞if␈α∞we␈α∞wanted␈α∞to,␈α∞ because␈α∂our␈α∞common
␈↓ ↓H␈↓sense␈α⊂notions␈α⊃about␈α⊂them␈α⊂ translate␈α⊃ readily␈α⊂into␈α⊂ certain␈α⊃ program␈α⊂and␈α⊂ data␈α⊃structures.␈α⊂ Still
␈↓ ↓H␈↓other␈α∂mental␈α∂qualities,␈α∂e.g.␈α∂ humor␈α∂and␈α∂appreciation␈α∂of␈α∂beauty,␈α∂seem␈α∂ much␈α∂harder␈α∂ to␈α∂ model.
␈↓ ↓H␈↓While␈α⊂ we␈α∂will␈α⊂ be␈α∂ quite␈α⊂liberal␈α∂ in␈α⊂ascribing␈α∂␈↓↓some␈↓␈α⊂mental␈α∂qualities␈α⊂ even␈α∂to␈α⊂rather␈α∂prihtive
␈↓ ↓H␈↓machines, we will  try to  be conservative  in our  criteria for ascribing  any ␈↓↓particular␈↓ quality.
␈↓ ↓H␈↓␈↓ εH␈↓ 92


␈↓ ↓H␈↓␈↓ α_The␈α
successive␈α
sections␈α
of␈α
this␈α
paper␈α∞will␈α
give␈α
philosophical␈α
and␈α
AI␈α
reasons␈α∞for␈α
ascribing
␈↓ ↓H␈↓beliefs␈αto␈αmachines,␈αtwo␈αnew␈αforms␈αof␈αdefinition␈αthat␈αseem␈αnecessary␈αfor␈αdefining␈αmental␈αqualities
␈↓ ↓H␈↓and␈αexamples␈αof␈αtheir␈αuse,␈αexamples␈αof␈αsystems␈αto␈αwhich␈αmental␈αqualities␈αare␈αascribed,␈αsome␈αfirst
␈↓ ↓H␈↓attempts␈α⊃at␈α⊃defining␈α⊃a␈α⊂variety␈α⊃of␈α⊃mental␈α⊃qualities,␈α⊃some␈α⊂comments␈α⊃on␈α⊃other␈α⊃views␈α⊃on␈α⊂mental
␈↓ ↓H␈↓qualities, notes, and references.

␈↓ ↓H␈↓␈↓ α_This␈α∩paper␈α∩is␈α∩exploratory␈α∩and␈α∩its␈α⊃presentation␈α∩is␈α∩non-technical.␈α∩ Any␈α∩axioms␈α∩that␈α⊃are
␈↓ ↓H␈↓presented␈αare␈αillustrative␈αand␈αnot␈αpart␈αof␈αan␈αaxiomatic␈αsystem␈αproposed␈αas␈αa␈αserious␈αcandidate␈αfor
␈↓ ↓H␈↓AI␈α⊂or␈α⊂philosophical␈α⊂use.␈α⊂ This␈α⊂is␈α⊂regrettable␈α⊃for␈α⊂two␈α⊂reasons.␈α⊂ First,␈α⊂AI␈α⊂use␈α⊂of␈α⊃these␈α⊂concepts
␈↓ ↓H␈↓requires␈α
formal␈αaxiomatization.␈α
 Second,␈αthe␈α
lack␈α
of␈αformalism␈α
focusses␈αattention␈α
on␈α
whether␈αthe
␈↓ ↓H␈↓paper␈αcorrectly␈αcharacterizes␈αmental␈αqualities␈αrather␈αthan␈αon␈αthe␈αformal␈αproperties␈αof␈αthe␈αtheories
␈↓ ↓H␈↓proposed.␈α I␈αthink␈αwe␈αcan␈αattain␈αa␈αsituation␈αlike␈αthat␈αin␈αthe␈αfoundations␈αof␈αmathematics,␈αwherein
␈↓ ↓H␈↓the␈α
controversies␈αabout␈α
whether␈αto␈α
take␈αan␈α
intuitionist␈αor␈α
classical␈αpoint␈α
of␈αview␈α
have␈αbeen␈α
mainly
␈↓ ↓H␈↓replaced␈αby␈αtechnical␈αstudies␈αof␈αintuitionist␈αand␈αclassical␈αtheories␈αand␈αthe␈αrelations␈αbetween␈αthem.
␈↓ ↓H␈↓In␈α
future␈α
work,␈α
I␈α
hope␈α
to␈α
treat␈α∞these␈α
matters␈α
more␈α
formally␈α
along␈α
the␈α
lines␈α
of␈α∞(McCarthy␈α
1977a
␈↓ ↓H␈↓and␈α∂1977b).␈α∂ This␈α∂won't␈α∂eliminate␈α∂controversy␈α∂about␈α∂the␈α∂true␈α∂nature␈α∂of␈α∂mental␈α∂qualities,␈α∂but␈α∞I
␈↓ ↓H␈↓believe that their eventual resolution requires more technical knowledge than is now available.
␈↓ ↓H␈↓␈↓ εH␈↓ 93


␈↓ ↓H␈↓α␈↓ ∧2WHY ASCRIBE MENTAL QUALITIES?

␈↓ ↓H␈↓␈↓ α_␈↓αWhy␈α∩ should␈α∩we␈α∩want␈α∩to␈α∩ ascribe␈α∩beliefs␈α⊃to␈α∩ machines␈α∩ at␈α∩all?␈↓␈α∩This␈α∩is␈α∩the␈α⊃converse
␈↓ ↓H␈↓question␈α∩to␈α⊃that␈α∩of␈α∩␈↓↓reductionism␈↓.␈α⊃ Instead␈α∩of␈α∩asking␈α⊃how␈α∩mental␈α∩qualities␈α⊃can␈α∩be␈α∩␈↓αreduced␈↓␈α⊃to
␈↓ ↓H␈↓physical ones, we ask how to ␈↓αascribe␈↓ mental qualities to physical systems.

␈↓ ↓H␈↓␈↓ α_Our␈αgeneral␈α
motivation␈αfor␈α
ascribing␈αmental␈α
qualities␈αis␈α
the␈αsame␈α
as␈αfor␈α
ascribing␈αany␈α
other
␈↓ ↓H␈↓qualities␈α
-␈αnamely␈α
to␈α
express␈αavailable␈α
information␈α
about␈αthe␈α
machine␈α
and␈αits␈α
current␈α
state.␈α To
␈↓ ↓H␈↓have␈αinformation,␈αwe␈αmust␈αhave␈αa␈αspace␈αof␈αpossibilities␈αwhether␈αexplicitly␈αdescribed␈αor␈αnot.␈α The
␈↓ ↓H␈↓ascription␈αmust␈αtherefore␈αmust␈αserve␈αto␈αdistinguish␈αthe␈αpresent␈αstate␈αof␈αthe␈αmachine␈αfrom␈αpast␈αor
␈↓ ↓H␈↓future␈α
states␈α
or␈α
from␈α
the␈α
state␈α
the␈α
machine␈α
would␈α
have␈α
in␈α
other␈α
conditions␈α
or␈α
from␈α
the␈α∞state␈α
of
␈↓ ↓H␈↓other␈α∩machines.␈α∩ Therefore,␈α∩the␈α∩issue␈α∩is␈α∩whether␈α∩ascription␈α∩of␈α∩mental␈α∩qualities␈α∩is␈α∩helpful␈α⊃in
␈↓ ↓H␈↓making these discriminations in the case of machines.

␈↓ ↓H␈↓␈↓ α_To␈α
put␈α
 the␈α
issue␈α
 sharply,␈α
consider␈α a␈α
computer␈α
program␈α
 for␈α
which␈α
 we␈α
possess␈αcomplete
␈↓ ↓H␈↓listings.␈α  The␈αbehavior␈αof␈α the␈αprogram␈αin␈αany␈αenvironment␈αis␈αdetermined␈αfrom␈α the␈αstructure␈αof
␈↓ ↓H␈↓the␈α∪program␈α∪ and␈α∪can␈α∪be␈α∩ found␈α∪out␈α∪ by␈α∪simulating␈α∪ the␈α∩action␈α∪of␈α∪ the␈α∪program␈α∪ and␈α∩the
␈↓ ↓H␈↓environment␈α⊂ without␈α∂ having␈α⊂ to␈α∂deal␈α⊂ with␈α⊂ any␈α∂ concept␈α⊂ of␈α∂belief.␈α⊂ Nevertheless,␈α⊂there␈α∂ are
␈↓ ↓H␈↓several reasons  for  ascribing belief  and other mental qualities:

␈↓ ↓H␈↓␈↓ α_1.␈α
Although␈αwe␈α
may␈αknow␈α
the␈αprogram,␈α
its␈αstate␈α
at␈αa␈α
given␈αmoment␈α
is␈αusually␈α
not␈αdirectly
␈↓ ↓H␈↓observable,␈αand␈αthe␈αfacts␈αwe␈αcan␈αobtain␈αabout␈αits␈αcurrent␈αstate␈αmay␈αbe␈αmore␈αreadily␈αexpressed␈αby
␈↓ ↓H␈↓ascribing certain beliefs and goals than in any other way.

␈↓ ↓H␈↓␈↓ α_2.␈α∀Even␈α∀if␈α∀we␈α∀can␈α∀simulate␈α∀its␈α∀interaction␈α∀with␈α∀its␈α∀environment␈α∀using␈α∃another␈α∀more
␈↓ ↓H␈↓comprehensive␈α
program,␈α
the␈α
simulation␈α
may␈αbe␈α
a␈α
billion␈α
times␈α
too␈αslow.␈α
 We␈α
also␈α
may␈α
not␈αhave
␈↓ ↓H␈↓the␈αinitial␈αconditions␈αof␈αthe␈αenvironment␈αor␈αthe␈αenvironment's␈αlaws␈αof␈αmotion␈αin␈αa␈αsuitable␈αform,
␈↓ ↓H␈↓whereas␈α∂it␈α∞may␈α∂be␈α∞feasible␈α∂to␈α∞make␈α∂a␈α∞prediction␈α∂of␈α∞the␈α∂effects␈α∞of␈α∂the␈α∞beliefs␈α∂we␈α∞ascribe␈α∂to␈α∞the
␈↓ ↓H␈↓program without any computer at all.

␈↓ ↓H␈↓␈↓ α_3.␈α∂Ascribing␈α∂beliefs␈α∞may␈α∂allow␈α∂deriving␈α∂general␈α∞statements␈α∂about␈α∂the␈α∂program's␈α∞behavior
␈↓ ↓H␈↓that could not be obtained from any finite number of simulations.

␈↓ ↓H␈↓␈↓ α_4.␈α  The␈α
belief␈αand␈α
goal␈αstructures␈α
we␈αascribe␈α
to␈αthe␈α
program␈αmay␈α
be␈αeasier␈α
to␈αunderstand
␈↓ ↓H␈↓than the details of  program as expressed in its listing.

␈↓ ↓H␈↓␈↓ α_5.␈α∞The␈α∞belief␈α∞and␈α∞goal␈α∞structure␈α∞is␈α∞likely␈α∂to␈α∞be␈α∞close␈α∞to␈α∞the␈α∞structure␈α∞the␈α∞designer␈α∂ of␈α∞the
␈↓ ↓H␈↓program␈α
had␈α
in␈α
 mind,␈α∞and␈α
 it␈α
may␈α
 be␈α∞easier␈α
to␈α
debug␈α
the␈α∞program␈α
in␈α
terms␈α
 of␈α∞this␈α
structure
␈↓ ↓H␈↓than␈α
directly␈α
from␈α
 the␈α∞listing.␈α
  In␈α
 fact,␈α
it␈α∞ is␈α
often␈α
 possible␈α
for␈α∞ someone␈α
to␈α
correct␈α
a␈α∞fault␈α
by
␈↓ ↓H␈↓reasoning␈α in␈αgeneral␈αterms␈αabout␈αthe␈α information␈αin␈αa␈αprogram␈αor␈αmachine,␈αdiagnosing␈α what␈αis
␈↓ ↓H␈↓wrong␈α⊃as␈α⊂a␈α⊃false␈α⊃belief,␈α⊂and␈α⊃  looking␈α⊂ at␈α⊃ the␈α⊃ details␈α⊂ of␈α⊃  the␈α⊂ program␈α⊃ or␈α⊃ machine␈α⊂ only
␈↓ ↓H␈↓sufficiently␈α∞to␈α∞determine␈α∞how␈α∞ the␈α∞false␈α∞belief␈α∞is␈α∞represented␈α∞ and␈α∞what␈α∞mechanism␈α∞caused␈α∞it␈α∞to
␈↓ ↓H␈↓arise.

␈↓ ↓H␈↓␈↓ α_6.␈α∞The␈α∞difference␈α∞between␈α∞this␈α∞program␈α
and␈α∞another␈α∞actual␈α∞or␈α∞hypothetical␈α∞program␈α
may
␈↓ ↓H␈↓best be expressed as a difference in belief structure.

␈↓ ↓H␈↓␈↓ α_All␈α
the␈α
above␈α
reasons␈α
for␈α
ascribing␈α
beliefs␈α
are␈α
epistemological;␈α
i.e.␈α
ascribing␈α
beliefs␈α
is␈α
needed
␈↓ ↓H␈↓␈↓ εH␈↓ 94


␈↓ ↓H␈↓to␈α∞adapt␈α∞to␈α∂limitations␈α∞on␈α∞our␈α∞ability␈α∂to␈α∞acquire␈α∞knowledge,␈α∞use␈α∂it␈α∞for␈α∞prediction,␈α∂and␈α∞establish
␈↓ ↓H␈↓generalizations␈α
in␈α
terms␈α
of␈α
the␈α
elementary␈α
structure␈αof␈α
the␈α
program.␈α
 Perhaps␈α
this␈α
is␈α
the␈αgeneral
␈↓ ↓H␈↓reason for ascribing higher levels of organization to systems.

␈↓ ↓H␈↓␈↓ α_Computers␈αgive␈αrise␈αto␈αnumerous␈αexamples␈αof␈αbuilding␈αa␈αhigher␈αstructure␈αon␈αthe␈αbasis␈αof␈αa
␈↓ ↓H␈↓lower␈α
and␈α
conducting␈α
subsequent␈α
analyses␈α
using␈α
the␈α
higher␈α
structure.␈α
 The␈α
geometry␈α
of␈αthe␈α
electric
␈↓ ↓H␈↓fields␈α
in␈αa␈α
transistor␈α
and␈αits␈α
chemical␈α
composition␈αgive␈α
rise␈αto␈α
its␈α
properties␈αas␈α
an␈α
electric␈αcircuit
␈↓ ↓H␈↓element.␈α
 Transistors␈α
are␈αcombined␈α
in␈α
small␈αcircuits␈α
and␈α
powered␈α
in␈αstandard␈α
ways␈α
to␈αmake␈α
logical
␈↓ ↓H␈↓elements␈α
such␈α
as␈α
ANDs,␈α
ORs,␈α
NOTs␈α
and␈α
flip-flops.␈α
 Computers␈α
are␈α
designed␈α
with␈α∞these␈α
logical
␈↓ ↓H␈↓elements␈αto␈αobey␈αa␈αdesired␈αorder␈αcode;␈αthe␈αdesigner␈αusually␈αneedn't␈αconsider␈αthe␈αproperties␈α
of␈αthe
␈↓ ↓H␈↓transistors␈αas␈αcircuit␈αelements.␈α When␈αwriting␈αa␈αcompiler␈αfrom␈αa␈αhigher␈αlevel␈αlanguage,␈αone␈αworks
␈↓ ↓H␈↓with␈αthe␈αorder␈αcode␈αand␈αdoesn't␈αhave␈αto␈αknow␈αabout␈αthe␈αANDs␈αand␈αORs;␈αthe␈αuser␈αof␈αthe␈αhigher
␈↓ ↓H␈↓order language needn't know the computer's order code.

␈↓ ↓H␈↓␈↓ α_In␈αthe␈αabove␈αcases,␈αusers␈αof␈αthe␈α
 higher␈αlevel␈αcan␈αcompletely␈αignore␈α the␈αlower␈αlevel,␈α
because
␈↓ ↓H␈↓the␈α
 behavior␈α
of␈α
the␈α
higher␈α level␈α
system␈α
is␈α
completely␈α
determined␈αby␈α
 the␈α
values␈α
of␈α
 the␈αhigher
␈↓ ↓H␈↓level␈α∂ variables;␈α∞e.g.␈α∂  in␈α∞order␈α∂to␈α∂determine␈α∞ the␈α∂outcome␈α∞of␈α∂a␈α∞ computer␈α∂program,␈α∂one␈α∞needn't
␈↓ ↓H␈↓consider␈α
the␈αflip-flops.␈α
 However,␈α
when␈αwe␈α
ascribe␈α
mental␈αstructure␈α
to␈α
humans␈αor␈α
goals␈αto␈α
 society,
␈↓ ↓H␈↓we␈α∞always␈α∞get␈α∞highly␈α∂ incomplete␈α∞systems;␈α∞the␈α∞ higher␈α∂level␈α∞behavior␈α∞ cannot␈α∞be␈α∂fully␈α∞predicted
␈↓ ↓H␈↓from␈α higher␈α level␈αobservations␈α and␈α higher␈α level␈α"laws"␈αeven␈αwhen␈αthe␈αunderlying␈α
lower␈αlevel
␈↓ ↓H␈↓behavior␈αis␈αdeterminate.␈α Moreover,␈αat␈αa␈αgiven␈αstate␈αof␈αscience␈αand␈αtechnology,␈αdifferent␈αkinds␈αof
␈↓ ↓H␈↓information␈α⊃can␈α⊃be␈α⊃obtained␈α∩from␈α⊃experiment␈α⊃and␈α⊃theory␈α∩building␈α⊃at␈α⊃the␈α⊃different␈α∩levels␈α⊃of
␈↓ ↓H␈↓organization.

␈↓ ↓H␈↓␈↓ α_In␈α
order␈α
to␈α
program␈α
a␈α
computer␈α
to␈α
obtain␈α
information␈α
and␈α
co-operation␈α
from␈α∞people␈α
and
␈↓ ↓H␈↓other␈α
machines,␈α
we␈α
will␈α
have␈α
to␈α
make␈α
it␈α
ascribe␈α
knowledge,␈α
belief,␈α
and␈α
wants␈α
to␈α∞other␈α
machines
␈↓ ↓H␈↓and␈α∞people.␈α∂ For␈α∞example,␈α∂a␈α∞program␈α∂that␈α∞plans␈α∂trips␈α∞will␈α∂have␈α∞to␈α∂ascribe␈α∞knowledge␈α∂to␈α∞travel
␈↓ ↓H␈↓agents␈αand␈αto␈αthe␈αairline␈αreservation␈αcomputers.␈α It␈αmust␈αsomehow␈αtreat␈αthe␈αinformation␈αin␈αbooks,
␈↓ ↓H␈↓perhaps␈αby␈α
ascribing␈αto␈α
them␈αa␈αpassive␈α
form␈αof␈α
knowledge.␈α The␈αmore␈α
powerful␈αthe␈α
program␈αin
␈↓ ↓H␈↓interpreting␈α∞what␈α
it␈α∞is␈α
told,␈α∞the␈α
less␈α∞it␈α
has␈α∞to␈α
know␈α∞about␈α
how␈α∞the␈α
information␈α∞it␈α
can␈α∞receive␈α
is
␈↓ ↓H␈↓represented␈αinternally␈αin␈αthe␈αteller␈αand␈αthe␈αmore␈αits␈αascriptions␈αof␈αknowledge␈αwill␈αlook␈αlike␈αhuman
␈↓ ↓H␈↓ascriptions of knowledge to other humans.
␈↓ ↓H␈↓␈↓ εH␈↓ 95


␈↓ ↓H␈↓αTWO␈α METHODS␈α OF␈α DEFINITION␈α AND␈α THEIR␈α APPLICATION␈α TO␈α∨MENTAL
␈↓ ↓H␈↓αQUALITIES

␈↓ ↓H␈↓␈↓ α_In␈αour␈α
opinion,␈αa␈αmajor␈α
source␈αof␈αproblems␈α
in␈αdefining␈αmental␈α
and␈αintensional␈α
concepts␈αis
␈↓ ↓H␈↓the␈α
weakness␈αof␈α
the␈α
methods␈αof␈α
definition␈α
that␈αhave␈α
been␈α
␈↓↓explicitly␈↓␈αused.␈α
 We␈α
introduce␈αtwo␈α
kinds
␈↓ ↓H␈↓of␈α∞definition:␈α∞ ␈↓↓definition␈α∞relative␈α∞to␈α∞an␈α∞approximate␈α∞theory␈↓␈α∞and␈α∞␈↓↓second␈α∞order␈α∞structural␈α∞definition␈↓
␈↓ ↓H␈↓and apply them to defining mental qualities.

␈↓ ↓H␈↓1. ␈↓αDefinitions relative to an approximate theory␈↓.

␈↓ ↓H␈↓␈↓ α_It␈α⊂is␈α⊂commonplace␈α⊃that␈α⊂most␈α⊂scientific␈α⊃concepts␈α⊂are␈α⊂not␈α⊃defined␈α⊂by␈α⊂isolated␈α⊃sentences␈α⊂of
␈↓ ↓H␈↓natural␈αlanguages␈αbut␈αrather␈αas␈αparts␈αof␈α
theories,␈αand␈αthe␈αacceptance␈αof␈αthe␈αtheory␈α
is␈αdetermined
␈↓ ↓H␈↓by␈αits␈αfit␈αto␈α
a␈αlarge␈αcollection␈αof␈α
phenomena.␈α We␈αpropose␈αa␈α
similar␈αmethod␈αfor␈αexplicating␈α
mental
␈↓ ↓H␈↓and␈α∞other␈α
common␈α∞sense␈α∞concepts,␈α
but␈α∞a␈α∞certain␈α
phenomenon␈α∞plays␈α∞a␈α
more␈α∞important␈α∞role␈α
than
␈↓ ↓H␈↓with␈αscientific␈αtheories:␈α the␈αconcept␈αis␈αmeaningful␈αonly␈αin␈αthe␈αtheory,␈αand␈αcannot␈αbe␈αdefined␈αwith
␈↓ ↓H␈↓more precision than the theory permits.

␈↓ ↓H␈↓␈↓ α_The␈α∞notion␈α∞of␈α∞one␈α∂theory␈α∞approximating␈α∞another␈α∞needs␈α∞to␈α∂be␈α∞formalized.␈α∞ In␈α∞the␈α∂case␈α∞of
␈↓ ↓H␈↓physics,␈αone␈αcan␈αthink␈αof␈αvarious␈αkinds␈αof␈αnumerical␈αor␈αprobabilistic␈αapproximation.␈α I␈αthink␈αthis
␈↓ ↓H␈↓kind␈α⊂of␈α∂approximation␈α⊂is␈α⊂untypical␈α∂and␈α⊂misleading␈α∂and␈α⊂won't␈α⊂help␈α∂explicate␈α⊂such␈α⊂concepts␈α∂as
␈↓ ↓H␈↓␈↓↓intentional action␈↓ as meaningful in approximate theories.  Instead it may go something like this:

␈↓ ↓H␈↓␈↓ α_Consider␈αa␈αdetailed␈αtheory␈α␈↓↓T␈↓␈α
that␈αhas␈αa␈αstate␈αvariable␈α
␈↓↓s.␈↓␈α We␈αmay␈αimagine␈αthat␈α
␈↓↓s␈↓␈αchanges
␈↓ ↓H␈↓with␈αtime.␈α The␈αapproximating␈αtheory␈α␈↓↓T'␈↓␈αhas␈αa␈αstate␈αvariable␈α␈↓↓s'.␈↓␈α There␈αis␈αa␈αpredicate␈α␈↓↓atp(s,T')␈↓
␈↓ ↓H␈↓whose␈α⊃truth␈α⊃means␈α⊃that␈α⊃␈↓↓T'␈↓␈α⊃is␈α⊃applicable␈α⊃when␈α⊃the␈α⊃world␈α⊃is␈α⊃in␈α⊃state␈α⊃␈↓↓s.␈↓␈α⊃ There␈α⊃is␈α⊃a␈α⊃relation
␈↓ ↓H␈↓␈↓↓corr(s,s')␈↓ which asserts that ␈↓↓s'␈↓ corresponds to the state ␈↓↓s. We␈↓ have

␈↓ ↓H␈↓1)␈↓ α8  ␈↓↓∀s.(atp(s,T') ⊃ ∃s'.corr(s,s'))␈↓.

␈↓ ↓H␈↓Certain␈α∞functions␈α∞␈↓↓f1(s),␈α∞f2(s),␈↓␈α∞etc.␈α∞have␈α∞corresponding␈α∞functions␈α∞␈↓↓f1'(s'),␈α∞f2'(s')␈↓,␈α∞etc.␈α∞ We␈α∞have
␈↓ ↓H␈↓relations like

␈↓ ↓H␈↓2)␈↓ α8  ␈↓↓∀s s'.(corr(s,s') ⊃ f1(s) = f1'(s'))␈↓.

␈↓ ↓H␈↓However,␈α⊃the␈α⊃approximate␈α⊃theory␈α⊂␈↓↓T'␈↓␈α⊃may␈α⊃have␈α⊃additional␈α⊂functions␈α⊃␈↓↓g1'(s')␈↓,␈α⊃etc.␈α⊃that␈α⊃do␈α⊂not
␈↓ ↓H␈↓correspond␈α
to␈α
any␈α
functions␈αof␈α
␈↓↓s.␈↓␈α
 Even␈α
when␈αit␈α
is␈α
possible␈α
to␈αconstruct␈α
␈↓↓g␈↓s␈α
corresponding␈α
to␈αthe
␈↓ ↓H␈↓␈↓↓g'␈↓s,␈α∞their␈α∞definitions␈α∞will␈α∞often␈α∞seem␈α∞arbitrary,␈α∞because␈α∞the␈α∞common␈α∞sense␈α∞user␈α∞of␈α∞␈↓↓g1'␈↓␈α∞will␈α∞only
␈↓ ↓H␈↓have used it within the context of ␈↓↓T'␈↓

␈↓ ↓H␈↓Concepts whose definition involves counterfactuals provide examples.

␈↓ ↓H␈↓␈↓ α_Suppose␈α
we␈α
want␈α
to␈α
ascribe␈α
␈↓↓intentions␈↓␈α
and␈α
␈↓↓free␈α
will␈↓␈α
and␈α
to␈α
distinguish␈α
a␈α
␈↓↓deliberate␈α
 action␈↓
␈↓ ↓H␈↓from␈αan␈αoccurrence.␈α We␈αwant␈αto␈αcall␈αan␈αoutput␈αa␈α␈↓↓deliberate␈αaction␈↓␈αif␈αthe␈αoutput␈αwould␈αhave␈αbeen
␈↓ ↓H␈↓different␈α
if␈α
the␈α
machine's␈α
 intentions␈α
had␈α
been␈αdifferent.␈α
 This␈α
requires␈α
a␈α
criterion␈α
for␈α
the␈αtruth␈α
of
␈↓ ↓H␈↓the␈α∂counterfactual␈α∂conditional␈α∂sentence␈α∂␈↓↓If␈α⊂its␈α∂intentions␈α∂had␈α∂been␈α∂different␈α∂the␈α⊂output␈α∂wouldn't
␈↓ ↓H␈↓↓have occurred␈↓, and we require what seems to be a novel treatment of counterfactuals.

␈↓ ↓H␈↓␈↓ α_We␈α
treat␈α
the␈α
"relevant␈α
aspect␈α
of␈α
reality"␈α∞as␈α
a␈α
Cartesian␈α
product␈α
so␈α
that␈α
we␈α
can␈α∞talk␈α
about
␈↓ ↓H␈↓␈↓ εH␈↓ 96


␈↓ ↓H␈↓changing␈α
one␈α
component␈α
and␈α
leaving␈α
the␈α
others␈α
unchanged.␈α
 This␈α
would␈α
be␈α
straightforward␈α
if␈α
the
␈↓ ↓H␈↓Cartesian␈α∪product␈α∀structure␈α∪existed␈α∪in␈α∀the␈α∪world;␈α∀however,␈α∪it␈α∪usually␈α∀exists␈α∪only␈α∀in␈α∪certain
␈↓ ↓H␈↓approximate␈αmodels␈α
of␈αthe␈α
world.␈α Consequently␈αno␈α
single␈αdefinite␈α
state␈αof␈αthe␈α
world␈αas␈α
a␈αwhole
␈↓ ↓H␈↓corresponds␈αto␈αchanging␈αone␈αcomponent.␈α The␈α
following␈αparagraphs␈αpresent␈αthese␈αideas␈αin␈α
greater
␈↓ ↓H␈↓detail.

␈↓ ↓H␈↓␈↓ α_Suppose␈α
␈↓↓A␈↓␈α
is␈α
a␈αtheory␈α
in␈α
which␈α
some␈αaspect␈α
of␈α
reality␈α
is␈αcharacterized␈α
by␈α
the␈α
values␈αof␈α
three
␈↓ ↓H␈↓quantities␈α∂␈↓↓x,␈↓␈α∂␈↓↓y␈↓␈α⊂and␈α∂␈↓↓z␈↓.␈α∂ Let␈α∂␈↓↓f␈↓␈α⊂be␈α∂a␈α∂function␈α⊂of␈α∂three␈α∂arguments,␈α∂let␈α⊂␈↓↓u␈↓␈α∂be␈α∂a␈α⊂quantity␈α∂satisfying
␈↓ ↓H␈↓␈↓↓u = f(x,y,z)␈↓,␈αwhere␈α␈↓↓f(1,1,1) = 3␈↓␈αand␈α␈↓↓f(2,1,1) = 5␈↓.␈α Consider␈αa␈αstate␈αof␈αthe␈αmodel␈αin␈αwhich␈α␈↓↓x = 1␈↓,
␈↓ ↓H␈↓␈↓↓y = 1␈↓␈α∞and␈α∞␈↓↓z = 1␈↓.␈α∞ Within␈α∞the␈α∞theory␈α∞␈↓↓A,␈↓␈α∞the␈α∞counterfactual␈α∞conditional␈α∞sentence␈α∞␈↓↓"u = 3,␈α∞but␈α∞if␈α
x
␈↓ ↓H␈↓↓were␈α
2,␈α∞then␈α
u␈α
would␈α∞be␈α
5"␈↓␈α
is␈α∞true,␈α
because␈α
the␈α∞counterfactual␈α
condition␈α
means␈α∞changing␈α
␈↓↓x␈↓␈α∞to␈α
2
␈↓ ↓H␈↓and leaving the other variables unchanged.

␈↓ ↓H␈↓␈↓ α_Now␈α
let's␈α
go␈α
beyond␈α
the␈α
model␈α
and␈α
suppose␈αthat␈α
␈↓↓x,␈↓␈α
␈↓↓y␈↓␈α
and␈α
␈↓↓z␈↓␈α
are␈α
quantities␈α
depending␈αon␈α
the
␈↓ ↓H␈↓state␈α
of␈α
the␈α
world.␈α
 Even␈α
if␈α
␈↓↓u = f(x,y,z)␈↓␈α
is␈α
taken␈α
as␈α
a␈α
law␈α
of␈α
nature,␈α
the␈α
counterfactual␈α
need␈α
not␈α
be
␈↓ ↓H␈↓taken␈αas␈αtrue,␈α
because␈αsomeone␈αmight␈α
argue␈αthat␈αif␈α
␈↓↓x␈↓␈αwere␈α2,␈αthen␈α
␈↓↓y␈↓␈αwould␈αbe␈α
3␈αso␈αthat␈α
␈↓↓u␈↓␈αmight
␈↓ ↓H␈↓not␈α∂be␈α∂5.␈α∂ If␈α∂the␈α∂theory␈α∞␈↓↓A␈↓␈α∂has␈α∂a␈α∂sufficiently␈α∂preferred␈α∂status␈α∞we␈α∂may␈α∂take␈α∂the␈α∂meaning␈α∂of␈α∞the
␈↓ ↓H␈↓counterfactual␈α∂in␈α∂␈↓↓A␈↓␈α⊂to␈α∂be␈α∂its␈α∂general␈α⊂meaning,␈α∂but␈α∂it␈α∂may␈α⊂sometimes␈α∂be␈α∂better␈α∂to␈α⊂consider␈α∂the
␈↓ ↓H␈↓counterfactual as defined solely in the theory, i.e. as ␈↓↓syncategorematic␈↓.

␈↓ ↓H␈↓␈↓ α_A␈αcommon␈αsense␈αexample␈αmay␈αbe␈αhelpful:␈αSuppose␈αa␈αski␈αinstructor␈αsays,␈α␈↓↓"He␈αwouldn't␈αhave
␈↓ ↓H␈↓↓fallen␈αif␈αhe␈αhad␈αbent␈αhis␈αknees␈αwhen␈α
he␈αmade␈αthat␈αturn"␈↓,␈αand␈αanother␈αinstructor␈αreplies,␈α
␈↓↓"No,␈αthe
␈↓ ↓H␈↓↓reason␈α∞he␈α∞fell␈α∞was␈α∞that␈α∞he␈α∞didn't␈α∞put␈α∞his␈α∞weight␈α∞on␈α∞his␈α∞downhill␈α∞ski"␈↓.␈α∞ Suppose␈α∞further␈α∞that␈α∞on
␈↓ ↓H␈↓reviewing␈α⊂a␈α⊃film,␈α⊂they␈α⊃agree␈α⊂that␈α⊂the␈α⊃first␈α⊂instructor␈α⊃was␈α⊂correct␈α⊂and␈α⊃the␈α⊂second␈α⊃mistaken.␈α⊂ I
␈↓ ↓H␈↓contend␈αthat␈αthis␈αagreement␈αis␈αbased␈αon␈αtheir␈αcommon␈αacceptance␈αof␈αa␈αtheory␈αof␈αskiing,␈αand␈αthat
␈↓ ↓H␈↓␈↓↓within␈α
the␈α∞theory␈↓,␈α
the␈α∞decision␈α
may␈α∞well␈α
be␈α∞rigorous␈α
even␈α∞though␈α
no-one␈α∞bothers␈α
to␈α∞imagine␈α
an
␈↓ ↓H␈↓alternate␈α∂world␈α∂as␈α∂much␈α∂like␈α∂the␈α∂real␈α∂world␈α∂as␈α∂possible␈α∂but␈α∂in␈α∂which␈α∂the␈α∂student␈α∂had␈α∂put␈α∞his
␈↓ ↓H␈↓weight on his downhill ski.

␈↓ ↓H␈↓␈↓ α_We␈αsuggest␈αthat␈αthis␈αis␈αoften␈α(I␈αhaven't␈αyet␈αlooked␈αfor␈αcounter-examples)␈αthe␈αcommon␈αsense
␈↓ ↓H␈↓meaning␈αof␈αa␈αcounterfactual.␈α The␈αcounterfactual␈αhas␈αa␈αdefinite␈αmeaning␈αin␈αa␈αtheory,␈αbecause␈αthe
␈↓ ↓H␈↓theory␈αhas␈α
a␈αCartesian␈α
product␈αstructure,␈αand␈α
the␈αtheory␈α
is␈αsufficiently␈α
preferred␈αthat␈αthe␈α
meaning
␈↓ ↓H␈↓of␈α
the␈α
counterfactual␈α
in␈α
the␈α
world␈α
is␈α
taken␈αas␈α
its␈α
meaning␈α
in␈α
the␈α
theory.␈α
 This␈α
is␈α
especially␈αlikely␈α
to
␈↓ ↓H␈↓be␈αtrue␈αfor␈αconcepts␈αthat␈αhave␈αa␈αnatural␈αdefinition␈αin␈αterms␈αof␈αcounterfactuals,␈αe.g.␈αthe␈αconcept␈αof
␈↓ ↓H␈↓␈↓↓deliberate action␈↓ with which we started this section.

␈↓ ↓H␈↓␈↓ α_In␈α⊃all␈α⊃cases␈α⊃that␈α⊃we␈α⊃know␈α⊃about,␈α⊃the␈α⊃theory␈α⊃is␈α⊃approximate␈α⊃and␈α⊃incomplete.␈α⊂ Provided
␈↓ ↓H␈↓certain␈αpropositions␈αare␈αtrue,␈αa␈αcertain␈αquantity␈α
is␈αapproximately␈αa␈αgiven␈αfunction␈αof␈αcertain␈α
other
␈↓ ↓H␈↓quantities.␈α The␈α
incompleteness␈αlies␈αin␈α
the␈αfact␈αthat␈α
the␈αtheory␈αdoesn't␈α
predict␈αstates␈αof␈α
the␈αworld
␈↓ ↓H␈↓but␈α∂only␈α∂certain␈α⊂functions␈α∂of␈α∂them.␈α∂ Thus␈α⊂a␈α∂useful␈α∂concept␈α∂like␈α⊂deliberate␈α∂action␈α∂may␈α⊂seem␈α∂to
␈↓ ↓H␈↓vanish␈α
if␈α
examined␈α
too␈αclosely,␈α
e.g.␈α
when␈α
we␈α
try␈αto␈α
define␈α
it␈α
in␈α
terms␈αof␈α
states␈α
of␈α
the␈α
world␈αand␈α
not
␈↓ ↓H␈↓just in terms of certain functions of these states.

␈↓ ↓H␈↓Remarks:

␈↓ ↓H␈↓␈↓ αh1.1.␈α⊃The␈α⊃known␈α⊃cases␈α⊃in␈α⊃which␈α⊃ a␈α⊃concept␈α⊃is␈α⊃defined␈α⊃relative␈α⊃to␈α⊃an␈α⊂approximate
␈↓ ↓H␈↓theory involve counterfactuals.  This may not always be the case.

␈↓ ↓H␈↓␈↓ αh1.2. It is important to study the nature of the approximations.
␈↓ ↓H␈↓␈↓ εH␈↓ 97


␈↓ ↓H␈↓␈↓ αh1.3.␈α∞(McCarthy␈α∞and␈α∞Hayes␈α∞1969)␈α
treats␈α∞the␈α∞notion␈α∞of␈α∞␈↓↓X␈α
can␈α∞do␈α∞Y␈↓␈α∞using␈α∞a␈α∞theory␈α
in
␈↓ ↓H␈↓which␈α∞the␈α∞world␈α∞is␈α∞regarded␈α∞as␈α∞a␈α
collection␈α∞of␈α∞interacting␈α∞automata.␈α∞ That␈α∞paper␈α∞failed␈α∞to␈α
note
␈↓ ↓H␈↓that sentences using ␈↓↓can␈↓ cannot necessarily be translated into single assertions about the world.

␈↓ ↓H␈↓␈↓ αh1.4.␈α
The␈α
attempt␈α
by␈αold␈α
fashioned␈α
introspective␈α
psychology␈αto␈α
analyze␈α
the␈α
mind␈αinto
␈↓ ↓H␈↓an␈α⊂interacting␈α⊂␈↓↓will,␈↓␈α∂␈↓↓intellect␈↓␈α⊂and␈α⊂other␈α∂components␈α⊂cannot␈α⊂be␈α∂excluded␈α⊂on␈α⊂the␈α∂methodological
␈↓ ↓H␈↓grounds␈αused␈αby␈αbehaviorists␈αand␈αpostitivists␈αto␈αdeclare␈αthem␈αmeaningless␈αand␈αexclude␈αthem␈αfrom
␈↓ ↓H␈↓science.  These concepts might have precise definitions within a suitable approximate theory.␈↓∧5␈↓

␈↓ ↓H␈↓␈↓ αh1.5.␈αThe␈αabove␈αtreatment␈α
of␈αcounterfactuals␈αin␈αwhich␈αthey␈α
are␈αdefined␈αin␈αterms␈αof␈α
the
␈↓ ↓H␈↓Cartesian␈α
product␈αstructure␈α
of␈αan␈α
approximate␈αtheory␈α
may␈α
be␈αbetter␈α
than␈αthe␈α
␈↓↓closest␈αpossible␈α
world␈↓
␈↓ ↓H␈↓treatments␈αdiscussed␈α
in␈α(Lewis␈α
1973).␈α The␈αtruth-values␈α
are␈αwell␈α
defined␈αwithin␈α
the␈αapproximate
␈↓ ↓H␈↓theories,␈α∞and␈α∂the␈α∞theories␈α∂can␈α∞be␈α∂justified␈α∞by␈α∂evidence␈α∞involving␈α∂phenomena␈α∞not␈α∂mentioned␈α∞in
␈↓ ↓H␈↓isolated counterfactual assertions.

␈↓ ↓H␈↓␈↓ αh1.6.␈α
Definition␈α
relative␈αto␈α
approximate␈α
theories␈α
may␈αhelp␈α
separate␈α
questions,␈α
such␈αas
␈↓ ↓H␈↓some␈α∞of␈α
those␈α∞concerning␈α∞counterfactuals,␈α
into␈α∞␈↓↓internal␈↓␈α∞questions␈α
within␈α∞the␈α∞approximate␈α
theory
␈↓ ↓H␈↓and␈αthe␈α
␈↓↓external␈↓␈αquestion␈α
of␈αthe␈α
justification␈αof␈αthe␈α
theory␈αas␈α
a␈αwhole.␈α
 The␈αinternal␈αquestions␈α
are
␈↓ ↓H␈↓likely␈α∂to␈α∞be␈α∂technical␈α∞and␈α∂have␈α∞definite␈α∂answers␈α∂on␈α∞which␈α∂people␈α∞can␈α∂agree␈α∞even␈α∂if␈α∂they␈α∞have
␈↓ ↓H␈↓philosophical or scientific disagreements about the external questions.

␈↓ ↓H␈↓2. ␈↓αSecond Order Structural Definition.␈↓

␈↓ ↓H␈↓␈↓ α_Structural␈α⊃definitions␈α∩of␈α⊃qualities␈α⊃are␈α∩given␈α⊃in␈α⊃terms␈α∩of␈α⊃the␈α⊃state␈α∩of␈α⊃the␈α∩ system␈α⊃being
␈↓ ↓H␈↓described while behavioral definitions are given in terms of its actual or potential behavior␈↓∧6␈↓.

␈↓ ↓H␈↓␈↓ α_If␈α∞the␈α∞structure␈α∞of␈α∞the␈α∞machine␈α∞is␈α∞known,␈α∞one␈α∞can␈α∞give␈α∞an␈α∞ad␈α∞hoc␈α∞␈↓↓first␈α∞ order␈α∞ structural
␈↓ ↓H␈↓↓definition␈↓.␈α
   This␈α
is␈αa␈α
predicate␈α
␈↓↓B(s,p)␈↓␈αwhere␈α
␈↓↓s␈↓␈α
represents␈αa␈α
state␈α
of␈αthe␈α
machine␈α
and␈α␈↓↓p␈↓␈α
represents
␈↓ ↓H␈↓a␈αsentence␈αin␈α
a␈αsuitable␈αlanguage,␈αand␈α
␈↓↓B(s,p)␈↓␈αis␈αthe␈α
assertion␈αthat␈αwhen␈αthe␈α
machine␈αis␈αin␈αstate␈α
␈↓↓s,␈↓
␈↓ ↓H␈↓it␈α∞␈↓↓believes␈↓␈α∞the␈α∞sentence␈α∞␈↓↓p.␈↓␈α∞(The␈α∞considerations␈α
of␈α∞this␈α∞paper␈α∞are␈α∞neutral␈α∞in␈α∞deciding␈α∞whether␈α
to
␈↓ ↓H␈↓regard␈α
the␈α
object␈α
of␈α∞belief␈α
as␈α
a␈α
sentence␈α∞or␈α
to␈α
use␈α
a␈α
modal␈α∞operator␈α
or␈α
to␈α
admit␈α∞␈↓↓propositions␈↓␈α
as
␈↓ ↓H␈↓abstract␈αobjects␈αthat␈αcan␈αbe␈αbelieved.␈α The␈αpaper␈αis␈αwritten␈αas␈αthough␈αsentences␈αare␈αthe␈αobjects␈αof
␈↓ ↓H␈↓belief, but I have more recently come to favor propositions and discuss them in (McCarthy 1977a).

␈↓ ↓H␈↓␈↓ α_A␈α⊃general␈α⊃␈↓↓first␈↓␈α⊂␈↓↓order␈↓␈α⊃structural␈α⊃definition␈α⊂of␈α⊃belief␈α⊃would␈α⊂be␈α⊃a␈α⊃predicate␈α⊂␈↓↓B(W,M,s,p)␈↓
␈↓ ↓H␈↓where␈α␈↓↓W␈↓␈α
is␈αthe␈α"world"␈α
in␈αwhich␈α
the␈αmachine␈α␈↓↓M␈↓␈α
whose␈αbeliefs␈α
are␈αin␈αquestion␈α
is␈αsituated.␈α
 I␈αdo
␈↓ ↓H␈↓not␈αsee␈αhow␈αto␈αgive␈αsuch␈αa␈αdefinition␈αof␈αbelief,␈αand␈αI␈αthink␈αit␈αis␈αimpossible.␈α Therefore␈αwe␈αturn␈α
to
␈↓ ↓H␈↓second order definitions␈↓∧7␈↓.

␈↓ ↓H␈↓␈↓ α_A␈α∩second␈α⊃order␈α∩structural␈α⊃definition␈α∩of␈α⊃belief␈α∩is␈α⊃a␈α∩second␈α⊃order␈α∩predicate␈α⊃␈↓↓β(W,M,B).␈↓
␈↓ ↓H␈↓␈↓↓β(W,M,B)␈↓␈αasserts␈αthat␈αthe␈αfirst␈αorder␈αpredicate␈α␈↓↓B␈↓␈αis␈αa␈α"good"␈αnotion␈αof␈αbelief␈αfor␈αthe␈αmachine␈α␈↓↓M␈↓
␈↓ ↓H␈↓in␈αthe␈αworld␈α
␈↓↓W.␈↓␈α Here␈α"good"␈α
means␈αthat␈αthe␈α
beliefs␈αthat␈α␈↓↓B␈↓␈α
ascribes␈αto␈α␈↓↓M␈↓␈α
agree␈αwith␈αour␈αideas␈α
of
␈↓ ↓H␈↓what␈α⊂beliefs␈α∂␈↓↓M␈↓␈α⊂would␈α∂have,␈α⊂not␈α⊂that␈α∂the␈α⊂beliefs␈α∂themselves␈α⊂are␈α∂true.␈α⊂ The␈α⊂axiomatizations␈α∂of
␈↓ ↓H␈↓belief in the literature are partial second order definitions.

␈↓ ↓H␈↓␈↓ α_In␈α⊂general,␈α⊂␈↓αa␈α⊃second␈α⊂order␈α⊂definition␈α⊃gives␈α⊂criteria␈α⊂for␈α⊃criticizing␈α⊂an␈α⊂ascription␈α⊃of␈α⊂a
␈↓ ↓H␈↓αquality␈αto␈αa␈αsystem.␈↓␈αWe␈α
suggest␈αthat␈αboth␈αour␈αcommon␈α
sense␈αand␈αscientific␈αusage␈αof␈α
not-directly-
␈↓ ↓H␈↓observable␈α
qualities␈αcorresponds␈α
more␈α
losely␈αto␈α
second␈αorder␈α
structural␈α
definition␈αthan␈α
to␈αany␈α
kind
␈↓ ↓H␈↓␈↓ εH␈↓ 98


␈↓ ↓H␈↓of␈α∂behavioral␈α∂definition.␈α∂ Note␈α∂that␈α∂ a␈α∂ second␈α∂order␈α∂ definition␈α∂cannot␈α∂ guarantee␈α∂ that␈α∞ there
␈↓ ↓H␈↓exist␈α
 predicates␈α ␈↓↓B␈↓␈α
 meeting␈α
the␈αcriterion␈α
β␈α
or␈αthat␈α
such␈α
a␈α␈↓↓B␈↓␈α
is␈α
unique.␈α Some␈α
qualities␈α
are␈αbest
␈↓ ↓H␈↓defined jointly with related qualities, e.g. beliefs and goals may require joint treatment.

␈↓ ↓H␈↓␈↓ α_Second␈α∪  order␈α∪  definitions␈α∪criticize␈α∀  whole␈α∪  belief␈α∪structures␈α∪rather␈α∀than␈α∪ individual
␈↓ ↓H␈↓beliefs.␈α⊃ We␈α⊃can␈α⊃ treat␈α⊃individual␈α⊃beliefs␈α⊃ by␈α⊃ saying␈α⊃ that␈α⊃a␈α⊃ system␈α⊃ believes␈α⊃ ␈↓↓p␈↓␈α⊃ in␈α⊃state␈α⊃ ␈↓↓s␈↓
␈↓ ↓H␈↓provided␈αall␈α"reasonably␈αgood"␈α ␈↓↓B␈↓'s␈αsatisfy␈α␈↓↓B(s,p)␈↓.␈α Thus␈α we␈αare␈αdistinguishing␈αthe␈α"intersection"
␈↓ ↓H␈↓of the reasonably good ␈↓↓B␈↓'s.

␈↓ ↓H␈↓␈↓ α_(An␈α∀analogy␈α∀with␈α∀cryptography␈α∀may␈α∀be␈α∪helpful.␈α∀ We␈α∀solve␈α∀a␈α∀cryptogram␈α∀by␈α∪making
␈↓ ↓H␈↓hypotheses␈αabout␈αthe␈αstructure␈αof␈αthe␈αcipher␈αand␈αabout␈αthe␈αtranslation␈αof␈αparts␈αof␈αthe␈αcipher␈αtext.
␈↓ ↓H␈↓Our␈α
solution␈α
is␈α
complete␈α
when␈α
we␈α
have␈α"guessed"␈α
a␈α
cipher␈α
system␈α
that␈α
produces␈α
the␈αcryptogram
␈↓ ↓H␈↓from␈α∂a␈α∂plausible␈α∂plaintext␈α⊂message.␈α∂ Though␈α∂we␈α∂never␈α∂prove␈α⊂that␈α∂our␈α∂solution␈α∂is␈α⊂unique,␈α∂two
␈↓ ↓H␈↓different␈αsolutions␈αare␈αalmost␈αnever␈αfound␈αexcept␈αfor␈αvery␈αshort␈αcryptograms.␈α In␈αthe␈αanalogy,␈αthe
␈↓ ↓H␈↓second␈αorder␈αdefinition␈α
β␈αcorresponds␈αto␈α
the␈αgeneral␈αidea␈α
of␈αencipherment,␈αand␈α
␈↓↓B␈↓␈αis␈αthe␈α
particular
␈↓ ↓H␈↓system␈α
used.␈α∞ While␈α
we␈α∞will␈α
rarely␈α∞be␈α
able␈α∞to␈α
prove␈α∞uniqueness,␈α
we␈α∞don't␈α
expect␈α∞to␈α
find␈α∞two␈α
␈↓↓B␈↓s
␈↓ ↓H␈↓both satisfying β).

␈↓ ↓H␈↓␈↓ α_It␈α
 seems␈α
 to␈α
 me␈α
 that␈α
 there␈αshould␈α
 be␈α
 a␈α
 metatheorem␈α
 of␈α
mathematical␈α
 logic␈αasserting
␈↓ ↓H␈↓that␈α∞not␈α∞all␈α∞ second␈α∞order␈α∞definitions␈α∂can␈α∞ be␈α∞reduced␈α∞ to␈α∞ first␈α∞ order␈α∞definitions␈α∂ and␈α∞ further
␈↓ ↓H␈↓theorems␈α∂characterizing␈α∂ those␈α∂ second␈α⊂  order␈α∂ definitions␈α∂ that␈α∂ admit␈α∂ such␈α⊂reductions.␈α∂ Such
␈↓ ↓H␈↓technical␈α
results,␈α
if␈α
 they␈α
can␈α
be␈α
 found,␈α
may␈α
 be␈α
helpful␈α
in␈α
 philosophy␈α
and␈α
in␈α
 the␈α
construction␈α
of
␈↓ ↓H␈↓formal␈αscientific␈αtheories.␈α I␈αwould␈αconjecture␈αthat␈αmany␈αof␈αthe␈αinformal␈αphilosophical␈αarguments
␈↓ ↓H␈↓that␈α∩certain␈α∩mental␈α∩ concepts␈α∪cannot␈α∩be␈α∩reduced␈α∩to␈α∩ physics␈α∪will␈α∩turn␈α∩out␈α∩to␈α∩be␈α∪sketches␈α∩of
␈↓ ↓H␈↓arguments that these concepts require second (or higher) order definitions.

␈↓ ↓H␈↓␈↓ α_Here␈α is␈αan␈αapproximate␈αsecond␈αorder␈αdefinition␈α
of␈αbelief.␈α For␈αeach␈αstate␈α␈↓↓s␈↓␈αof␈αthe␈α
 machine
␈↓ ↓H␈↓and␈α∞each␈α∂ sentence␈α∞␈↓↓p␈↓␈α∞in␈α∂ a␈α∞suitable␈α∂ language␈α∞␈↓↓L,␈↓␈α∞we␈α∂assign␈α∞truth␈α∂to␈α∞␈↓↓B(s,p)␈↓␈α∞if␈α∂and␈α∞only␈α∂if␈α∞the
␈↓ ↓H␈↓machine␈α∞is␈α∞considered␈α
to␈α∞believe␈α∞␈↓↓p␈↓␈α
 when␈α∞it␈α∞ is␈α∞ in␈α
state␈α∞ ␈↓↓s␈↓.␈α∞   The␈α
 language␈α∞ ␈↓↓L␈↓␈α∞is␈α∞ chosen␈α
 for
␈↓ ↓H␈↓our␈αconvenience,␈αand␈α there␈α
is␈αno␈αassumption␈α that␈α
the␈αmachine␈αexplicitly␈αrepresents␈α
sentences␈αof
␈↓ ↓H␈↓␈↓↓L␈↓␈αin␈α
any␈αway.␈α
 Thus␈αwe␈αcan␈α
talk␈αabout␈α
the␈αbeliefs␈αof␈α
 Chinese,␈α dogs,␈α
corporations,␈α thermostats,
␈↓ ↓H␈↓and␈α computer␈αoperating␈α systems␈α without␈αassuming␈α that␈α they␈αuse␈α English␈α or␈αour␈αfavorite␈α
first
␈↓ ↓H␈↓order␈α
language.␈α∞ ␈↓↓L␈↓␈α
may␈α∞or␈α
may␈α∞not␈α
be␈α∞the␈α
 language␈α∞be␈α
the␈α∞language␈α
we␈α∞are␈α
using␈α∞for␈α
making
␈↓ ↓H␈↓other␈αassertions,␈αe.g.␈αwe␈αcould,␈αwriting␈αin␈αEnglish,␈αsystematically␈αuse␈αFrench␈αsentences␈αas␈αobjects␈αof
␈↓ ↓H␈↓belief.␈α However,␈α the␈αbest␈αchoice␈α for␈αartificial␈α
intelligence␈α work␈αmay␈αbe␈αto␈α make␈α␈↓↓L␈↓␈αa␈α
subset␈αof
␈↓ ↓H␈↓our  "outer" language restricted so as to avoid the paradoxical self-references of (Montague 1963).

␈↓ ↓H␈↓␈↓ α_We␈α⊃now␈α⊃subject␈α⊃␈↓↓B(s,p)␈↓␈α⊃to␈α⊃certain␈α⊃criteria;␈α⊃i.e.␈α⊃β␈↓↓(B,W)␈↓␈α⊃is␈α⊃considered␈α⊃true␈α⊃provided␈α⊃the
␈↓ ↓H␈↓following conditions are satisfied:

␈↓ ↓H␈↓␈↓ αh2.1.␈α∂The␈α∂set␈α∂␈↓↓Bel(s)␈↓␈α∂of␈α∂beliefs,␈α∂i.e.␈α∂the␈α∞set␈α∂of␈α∂␈↓↓p␈↓'s␈α∂for␈α∂which␈α∂ ␈↓↓B(s,p)␈↓␈α∂is␈α∂ assigned␈α∞true
␈↓ ↓H␈↓when ␈↓↓M␈↓ is in state ␈↓↓s␈↓ contains  sufficiently "obvious" consequences of some of its members.

␈↓ ↓H␈↓␈↓ αh2.2.␈α∂  ␈↓↓Bel(s)␈↓␈α⊂changes␈α∂in␈α⊂a␈α∂ reasonable␈α⊂ way␈α∂when␈α⊂the␈α∂ state␈α⊂changes␈α∂in␈α⊂time.␈α∂   We
␈↓ ↓H␈↓like␈α⊃ new␈α⊃ beliefs␈α⊃ to␈α⊃ be␈α⊃ logical␈α⊃ or␈α∩ "plausible"␈α⊃consequences␈α⊃of␈α⊃old␈α⊃ones␈α⊃or␈α⊃to␈α⊃come␈α∩in␈α⊃as
␈↓ ↓H␈↓␈↓↓communications␈↓␈α in␈α
some␈αlanguage␈α
on␈αthe␈αinput␈α
lines␈αor␈α
 to␈αbe␈α␈↓↓observations␈↓,␈α
i.e.␈α beliefs␈α
about␈αthe
␈↓ ↓H␈↓environment␈αthe␈α
information␈αfor␈αwhich␈α
comes␈αin␈αon␈α
the␈αinput␈α
lines.␈α The␈αset␈α
of␈αbeliefs␈αshould␈α
not
␈↓ ↓H␈↓change too rapidly as the state changes with time.
␈↓ ↓H␈↓␈↓ εH␈↓ 99


␈↓ ↓H␈↓␈↓ αh2.3.␈α∂  We␈α⊂prefer␈α∂ the␈α∂set␈α⊂ of␈α∂beliefs␈α⊂ to␈α∂be␈α∂ as␈α⊂consistent␈α∂ as␈α⊂possible.␈α∂ (Admittedly,
␈↓ ↓H␈↓consistency␈α is␈αnot␈αa␈αquantitative␈αconcept␈αin␈αmathematical␈αlogic␈α-␈α a␈αsystem␈αis␈α either␈αconsistent␈αor
␈↓ ↓H␈↓not,␈α∂but␈α∂ it␈α∂would␈α∂seem␈α∂that␈α∂we␈α⊂ will␈α∂sometimes␈α∂have␈α∂to␈α∂ascribe␈α∂ inconsistent␈α∂sets␈α∂of␈α⊂beliefs␈α∂to
␈↓ ↓H␈↓machines␈αand␈αpeople.␈α
 Our␈αintuition␈αsays␈αthat␈α
we␈αshould␈αbe␈αable␈α
to␈αmaintain␈αareas␈αof␈α
 consistency
␈↓ ↓H␈↓in␈α
our␈αbeliefs␈α
and␈αthat␈α
 it␈αmay␈α
be␈α especially␈α
important␈α to␈α
avoid␈α inconsistencies␈α
in␈α the␈α
machine's
␈↓ ↓H␈↓purely analytic beliefs).

␈↓ ↓H␈↓␈↓ αh2.4.␈α∞ Our␈α∞criteria␈α∂for␈α∞belief␈α∞systems␈α∂can␈α∞be␈α∞strengthened␈α∞if␈α∂we␈α∞identify␈α∞some␈α∂of␈α∞the
␈↓ ↓H␈↓machine's␈α
beliefs␈αas␈α
expressing␈αgoals,␈α
i.e.␈αif␈α
we␈αhave␈α
beliefs␈αof␈α
the␈αform␈α
"It␈αwould␈α
be␈αgood␈α
if␈α...".
␈↓ ↓H␈↓Then␈α
we␈α
can␈α
ask␈α
that␈α
the␈α
machine's␈α
behavior␈α
be␈α
somewhat␈α
␈↓↓rational␈↓,␈α
i.e.␈α
 ␈↓↓it␈α
does␈α
what␈α∞it␈α
believes
␈↓ ↓H␈↓↓will␈αachieve␈αits␈αgoals␈↓.␈αThe␈αmore␈αof␈αits␈αbehavior␈αwe␈αcan␈αaccount␈αfor␈αin␈αthis␈αway,␈αthe␈αbetter␈αwe␈αwill
␈↓ ↓H␈↓like␈αthe␈αfunction␈α␈↓↓B(s,p)␈↓.␈α We␈αalso␈αwould␈αlike␈αto␈αregard␈αinternal␈αstate␈αchanges␈αas␈αchanges␈αin␈αbelief
␈↓ ↓H␈↓in so far as this is reasonable.

␈↓ ↓H␈↓␈↓ αh2.5.␈α If␈αthe␈α
machine␈αcommunicates,␈αi.e.␈α
emits␈αsentences␈αin␈α
some␈αlanguage␈α that␈α can␈α
 be
␈↓ ↓H␈↓interpreted␈α as␈α assertions,␈α questions␈α
 and␈αcommands,␈αwe␈αwill␈αwant␈α
the␈αassertions␈αto␈αbe␈αamong␈α
 its
␈↓ ↓H␈↓beliefs␈α∞unless␈α∞we␈α∞are␈α∞ ascribing␈α∞to␈α∞it␈α∞ a␈α∞goal␈α∞or␈α∞ subgoal␈α∞that␈α∞involves␈α∞ lying.␈α∞ We␈α∞will␈α∂be␈α∞most
␈↓ ↓H␈↓satisfied␈α∂with␈α∞our␈α∂belief␈α∞ascription,␈α∂if␈α∂we␈α∞can␈α∂account␈α∞for␈α∂its␈α∞communications␈α∂as␈α∂furthering␈α∞the
␈↓ ↓H␈↓goals we are ascribing.

␈↓ ↓H␈↓␈↓ αh2.6.␈α Sometimes␈αwe␈αshall␈αwant␈αto␈αascribe␈αintrospective␈αbeliefs,␈αe.g.␈αa␈αbelief␈αthat␈αit␈αdoes
␈↓ ↓H␈↓not know how to fly to Boston or even that it doesn't know what it wants in a certain situation.

␈↓ ↓H␈↓␈↓ αh2.7.␈α
Finally,␈α∞we␈α
will␈α
prefer␈α∞a␈α
more␈α
economical␈α∞ascription␈α
␈↓↓B␈↓␈α
to␈α∞a␈α
less␈α∞economical␈α
one.
␈↓ ↓H␈↓The␈α
 fewer␈α
beliefs␈α
we␈α
ascribe␈α
and␈α
the␈α
less␈α
they␈α
change␈α
with␈α
state␈α
consistent␈α
with␈α
accounting␈α
for
␈↓ ↓H␈↓the␈α∃behavior␈α∃and␈α∃the␈α⊗internal␈α∃state␈α∃changes,␈α∃the␈α∃better␈α⊗we␈α∃will␈α∃like␈α∃it.␈α∃ In␈α⊗particular,␈α∃if
␈↓ ↓H␈↓␈↓↓∀s p.(B1(s,p) ⊃ B2(s,p))␈↓,␈α∩but␈α∪not␈α∩conversely,␈α∩and␈α∪␈↓↓B1␈↓␈α∩accounts␈α∩for␈α∪all␈α∩the␈α∩state␈α∪changes␈α∩and
␈↓ ↓H␈↓outputs␈α∂that␈α∂␈↓↓B2␈↓␈α∂does,␈α∂we␈α∂will␈α∂prefer␈α∂␈↓↓B1␈↓␈α∂to␈α∞␈↓↓B2.␈↓␈α∂ This␈α∂insures␈α∂that␈α∂we␈α∂will␈α∂prefer␈α∂to␈α∂assign␈α∞no
␈↓ ↓H␈↓beliefs␈α
to␈α
stones␈α
that␈α
don't␈α
change␈α
and␈α
don't␈αbehave.␈α
 A␈α
belief␈α
predicate␈α
that␈α
applies␈α
to␈α
a␈αfamily␈α
of
␈↓ ↓H␈↓machines is preferable to one that applies to a single machine.

␈↓ ↓H␈↓␈↓ α_The␈α∂above␈α⊂ criteria␈α∂have␈α⊂ been␈α∂ formulated␈α∂somewhat␈α⊂ vaguely.␈α∂ This␈α⊂ would␈α∂be␈α⊂bad␈α∂if
␈↓ ↓H␈↓there␈α
 were␈α
widely␈α
different␈α
ascriptions␈α
of␈α
 beliefs␈α
to␈α
a␈α
particular␈α
machine␈α
that␈α
all␈α
met␈αour␈α
 criteria
␈↓ ↓H␈↓or␈α⊂if␈α⊃the␈α⊂criteria␈α⊂allowed␈α⊃ ascriptions␈α⊂that␈α⊃ differed␈α⊂widely␈α⊂from␈α⊃ our␈α⊂intuitions.␈α⊃  My␈α⊂present
␈↓ ↓H␈↓opinion␈α∩is␈α⊃that␈α∩more␈α∩thought␈α⊃will␈α∩make␈α⊃the␈α∩ criteria␈α∩somewhat␈α⊃more␈α∩precise␈α⊃ at␈α∩no␈α∩cost␈α⊃ in
␈↓ ↓H␈↓applicability,␈αbut␈α
that␈α they␈α
␈↓↓should␈↓␈αstill␈α
remain␈α rather␈α
vague,␈αi.e.␈α
we␈αshall␈α
want␈αto␈α
ascribe␈αbelief␈α
in
␈↓ ↓H␈↓a␈α⊂␈↓↓family␈↓␈α⊃of␈α⊂cases.␈α⊂ However,␈α⊃even␈α⊂at␈α⊂the␈α⊃present␈α⊂level␈α⊂of␈α⊃vagueness,␈α⊂there␈α⊂probably␈α⊃won't␈α⊂be
␈↓ ↓H␈↓radically␈α∞different␈α∞equally␈α∂"good"␈α∞ascriptions␈α∞of␈α∂belief␈α∞for␈α∞systems␈α∂of␈α∞practical␈α∞interest.␈α∂ If␈α∞there
␈↓ ↓H␈↓were, we would notice unresolvable ambiguities in our ascriptions of belief to our acquaintances.

␈↓ ↓H␈↓␈↓ α_While␈αwe␈αmay␈α
not␈αwant␈αto␈α
pin␈αdown␈αour␈α
general␈αidea␈αof␈α
belief␈αto␈αa␈α
single␈αaxiomatization,
␈↓ ↓H␈↓we␈α
will␈αneed␈α
to␈αbuild␈α
precise␈α
axiomatizations␈αof␈α
belief␈αand␈α
other␈αmental␈α
qualities␈α
into␈αparticular
␈↓ ↓H␈↓intelligent computer programs.
␈↓ ↓H␈↓␈↓ εH␈↓ *10


␈↓ ↓H␈↓α␈↓ β<EXAMPLES OF SYSTEMS WITH MENTAL QUALITIES

␈↓ ↓H␈↓␈↓ α_Let␈α⊂us␈α⊂ consider␈α⊂some␈α⊂ examples␈α⊂of␈α⊃machines␈α⊂ and␈α⊂programs␈α⊂ to␈α⊂which␈α⊂we␈α⊃may␈α⊂ascribe
␈↓ ↓H␈↓belief and goal structures.

␈↓ ↓H␈↓␈↓ α_1.␈α∞ ␈↓αThermostats.␈↓␈α∞  Ascribing␈α∂beliefs␈α∞to␈α∞simple␈α∞thermostats␈α∂is␈α∞unnecessary␈α∞for␈α∞the␈α∂study␈α∞of
␈↓ ↓H␈↓thermostats,␈α⊃because␈α⊂their␈α⊃operation␈α⊂can␈α⊃be␈α⊃well␈α⊂ understood␈α⊃without␈α⊂it.␈α⊃ However,␈α⊃their␈α⊂very
␈↓ ↓H␈↓simplicity␈α∞ makes␈α∞it␈α∞clearer␈α∞what␈α∞is␈α∞involved␈α∞ in␈α∞the␈α∞ ascription,␈α∞and␈α∞ we␈α∞maintain␈α∞(partly␈α∞ as␈α
a
␈↓ ↓H␈↓provocation␈α∩to␈α∩ those␈α∩who␈α∩ regard␈α∩ attribution␈α⊃of␈α∩ beliefs␈α∩ to␈α∩machines␈α∩ as␈α∩mere␈α⊃ intellectual
␈↓ ↓H␈↓sloppiness) that  the ascription  is legitimate.␈↓∧8␈↓

␈↓ ↓H␈↓␈↓ α_First␈α⊂ consider␈α⊂a␈α⊂simple␈α⊂ thermostat␈α⊂that␈α⊃turns␈α⊂off␈α⊂the␈α⊂heat␈α⊂when␈α⊂the␈α⊂ temperature␈α⊃is␈α⊂a
␈↓ ↓H␈↓degree␈α
 above␈α
the␈αtemperature␈α
set␈α
 on␈α
the␈αthermostat,␈α
 turns␈α
on␈α
the␈αheat␈α
 when␈α
the␈αtemperature␈α
 is
␈↓ ↓H␈↓a␈αdegree␈α
below␈αthe␈αdesired␈α
 temperature,␈αand␈αleaves␈α
 the␈αheat␈α
as␈α is␈αwhen␈α
 the␈αtemperature␈α is␈α
  in
␈↓ ↓H␈↓the␈α two␈α  degree␈α range␈α around␈α
  the␈α desired␈αtemperature.␈αThe␈αsimplest␈αbelief␈α
predicate␈α␈↓↓B(s,p)␈↓
␈↓ ↓H␈↓ascribes␈αbelief␈αto␈αonly␈αthree␈αsentences:␈α "The␈αroom␈αis␈αtoo␈αcold",␈α"The␈αroom␈αis␈α too␈αhot",␈αand␈α"The
␈↓ ↓H␈↓room␈α∞is␈α∞OK"␈α∞-␈α∞the␈α∞beliefs␈α∞being␈α∞assigned␈α∞to␈α∞states␈α∞of␈α∞ the␈α∞thermostat␈α∞in␈α∞the␈α∞obvious␈α∞way.␈α
 We
␈↓ ↓H␈↓ascribe␈αto␈αit␈αthe␈αgoal,␈α"The␈αroom␈αshould␈αbe␈αok".␈α When␈αthe␈αthermostat␈αbelieves␈α the␈α room␈αis␈α too
␈↓ ↓H␈↓cold␈α∂ or␈α∞too␈α∂ hot,␈α∂it␈α∞sends␈α∂ a␈α∂message␈α∞saying␈α∂so␈α∂to␈α∞the␈α∂furnace.␈α∂A␈α∞slightly␈α∂more␈α∂complex␈α∞belief
␈↓ ↓H␈↓predicate␈αcould␈αalso␈α be␈αused␈αin␈α which␈αthe␈αthermostat␈α has␈αa␈α belief␈αabout␈αwhat␈αthe␈αtemperature
␈↓ ↓H␈↓should␈αbe␈αand␈αanother␈αbelief␈αabout␈αwhat␈αit␈αis.␈α  It␈αis␈αnot␈αclear␈α which␈αis␈αbetter,␈αbut␈αif␈α we␈αwished
␈↓ ↓H␈↓to␈α∂consider␈α∂possible␈α⊂errors␈α∂in␈α∂the␈α⊂thermometer,␈α∂then␈α∂we␈α∂would␈α⊂ascribe␈α∂beliefs␈α∂ about␈α⊂what␈α∂the
␈↓ ↓H␈↓temperature␈α
is.␈α
We␈α
 do␈α
not␈α
ascribe␈α
to␈α
 it␈α
any␈α
other␈α
 beliefs;␈α
it␈α
has␈α
 no␈α
opinion␈α
even␈α about␈α
whether
␈↓ ↓H␈↓the␈αheat␈αis␈αon␈αor␈αoff␈αor␈αabout␈αthe␈αweather␈αor␈αabout␈αwho␈αwon␈αthe␈αbattle␈αof␈αWaterloo.␈α   Moreover,
␈↓ ↓H␈↓it  has  no introspective  beliefs; i.e. it doesn't believe that it believes the room is too hot.

␈↓ ↓H␈↓␈↓ α_Let␈α⊂us␈α⊂compare␈α⊃the␈α⊂above␈α⊂␈↓↓B(s,p)␈↓␈α⊃with␈α⊂the␈α⊂criteria␈α⊂of␈α⊃the␈α⊂previous␈α⊂section.␈α⊃ The␈α⊂belief
␈↓ ↓H␈↓structure␈α∂is␈α∞consistent␈α∂(because␈α∂all␈α∞the␈α∂beliefs␈α∂are␈α∞independent␈α∂of␈α∂one␈α∞another),␈α∂they␈α∂arise␈α∞from
␈↓ ↓H␈↓observation,␈α
and␈αthey␈α
result␈α
in␈αaction␈α
in␈α
accordance␈αwith␈α
the␈α
ascribed␈αgoal.␈α
 There␈α
is␈αno␈α
reasoning
␈↓ ↓H␈↓and␈α
only␈α
commands␈α
(which␈α
we␈α
have␈αnot␈α
included␈α
in␈α
our␈α
discussion)␈α
are␈α
communicated.␈α Clearly
␈↓ ↓H␈↓assigning␈αbeliefs␈αis␈αof␈α
modest␈αintellectual␈αbenefit␈αin␈α
this␈αcase.␈α However,␈αif␈α
we␈αconsider␈αthe␈αclass␈α
of
␈↓ ↓H␈↓possible␈αthermostats,␈αthen␈αthe␈αascribed␈αbelief␈αstructure␈αhas␈αgreater␈αconstancy␈αthan␈αthe␈αmechanisms
␈↓ ↓H␈↓for actually measuring and representing the temperature.

␈↓ ↓H␈↓␈↓ α_The␈α∞temperature␈α∞control␈α∞system␈α∞in␈α∞ my␈α∞house␈α∞may␈α∞be␈α∞ described␈α∞as␈α∂ follows:␈α∞Thermostats
␈↓ ↓H␈↓upstairs␈αand␈α downstairs␈α
 tell␈αthe␈α central␈α
system␈α to␈αturn␈α on␈α
or␈αshut␈α off␈α
hot␈α water␈αflow␈α to␈α
these
␈↓ ↓H␈↓areas.␈α  A␈αcentral␈αwater-temperature␈αthermostat␈αtells␈αthe␈αfurnace␈αto␈αturn␈α on␈αor␈αoff␈α thus␈α keeping
␈↓ ↓H␈↓the␈α central␈α hot␈α
 water␈α reservoir␈αat␈α the␈α
 right␈αtemperature.␈α Recently␈αit␈α
was␈αtoo␈αhot␈αupstairs,␈α
and
␈↓ ↓H␈↓the␈α⊂question␈α⊂arose␈α⊃as␈α⊂to␈α⊂whether␈α⊃the␈α⊂upstairs␈α⊂thermostat␈α⊂mistakenly␈α⊃ ␈↓↓believed␈↓␈α⊂it␈α⊂was␈α⊃too␈α⊂ cold
␈↓ ↓H␈↓upstairs␈α∞ or␈α
 whether␈α∞ the␈α
furnace␈α∞ thermostat␈α∞ mistakenly␈α
␈↓↓believed␈α∞␈↓␈α
the␈α∞ water␈α
was␈α∞too␈α∞ cold.␈α
 It
␈↓ ↓H␈↓turned␈αout␈αthat␈α neither␈αmistake␈αwas␈αmade;␈αthe␈αdownstairs␈αcontroller␈α␈↓↓tried␈↓␈αto␈αturn␈αoff␈αthe␈αflow␈α of
␈↓ ↓H␈↓water␈α∂but␈α∂ ␈↓↓couldn't␈↓,␈α∂because␈α∂the␈α∂ valve␈α∂was␈α∂ stuck.␈α∂ The␈α∂plumber␈α∂came␈α∂ once␈α∂ and␈α⊂found␈α∂ the
␈↓ ↓H␈↓trouble,␈α∩ and␈α∩ came␈α∪again␈α∩ when␈α∩ a␈α∪replacement␈α∩valve␈α∩ was␈α∪ordered.␈α∩  Since␈α∩the␈α∪services␈α∩of
␈↓ ↓H␈↓plumbers␈α
are␈α
increasingly␈α
 expensive,␈α
and␈α
 microcomputers␈α
are␈α
 increasingly␈α
cheap,␈α
one␈α
is␈α
led␈αto
␈↓ ↓H␈↓design␈α
a␈α
temperature␈α
control␈α
system␈α
that␈α
would␈α
␈↓↓know␈↓␈α
a␈α
lot␈α
more␈α
about␈α
the␈α
thermal␈α
 state␈α∞of␈α
the
␈↓ ↓H␈↓house and its own state of health.

␈↓ ↓H␈↓␈↓ α_In␈α∩the␈α∪first␈α∩ place,␈α∪while␈α∩the␈α∩present␈α∪system␈α∩␈↓↓couldn't␈↓␈α∪turn␈α∩ off␈α∩the␈α∪flow␈α∩of␈α∪hot␈α∩water
␈↓ ↓H␈↓␈↓ εH␈↓ *11


␈↓ ↓H␈↓upstairs,␈α
there␈α
is␈α
no␈α
reason␈α
 to␈α
ascribe␈α
to␈α
it␈α
the␈α
 ␈↓↓knowledge␈↓␈α
that␈α
it␈α
couldn't,␈α
and␈α
 ␈↓↓a␈α
fortiori␈↓␈α
it␈α
had␈α
no
␈↓ ↓H␈↓ability␈α∂to␈α∂␈↓↓communicate␈↓␈α⊂this␈α∂␈↓↓fact␈↓␈α∂ or␈α∂to␈α⊂take␈α∂it␈α∂into␈α∂ account␈α⊂in␈α∂controlling␈α∂the␈α∂system.␈α⊂ A␈α∂more
␈↓ ↓H␈↓advanced␈α⊗system␈α↔would␈α⊗know␈α↔whether␈α⊗the␈α⊗ ␈↓↓actions␈↓␈α↔it␈α⊗␈↓↓attempted␈↓␈α↔succeeded,␈α⊗and␈α↔it␈α⊗would
␈↓ ↓H␈↓communicate␈αfailures␈αand␈αadapt␈α to␈αthem.␈α (We␈α adapted␈αto␈αthe␈αfailure␈α by␈αturning␈αoff␈α the␈αwhole
␈↓ ↓H␈↓system␈α∂ until␈α∂the␈α∞ whole␈α∂ house␈α∂cooled␈α∂ off␈α∞and␈α∂ then␈α∂letting␈α∂the␈α∞two␈α∂parts␈α∂warm␈α∂up␈α∞ together.
␈↓ ↓H␈↓The␈α∀present␈α∪system␈α∀has␈α∪ the␈α∀␈↓↓physical␈α∀ capability␈↓␈α∪ of␈α∀ doing␈α∪ this␈α∀ even␈α∪  if␈α∀ it␈α∀ hasn't␈α∪ the
␈↓ ↓H␈↓␈↓↓knowledge␈↓ or the ␈↓↓will␈↓.

␈↓ ↓H␈↓␈↓ α_While␈α⊃the␈α⊃thermostat␈α∩believes␈α⊃"The␈α⊃room␈α⊃is␈α∩too␈α⊃cold",␈α⊃there␈α⊃is␈α∩no␈α⊃need␈α⊃to␈α⊃say␈α∩that␈α⊃it
␈↓ ↓H␈↓understands␈αthe␈αconcept␈αof␈α"too␈αcold".␈α The␈αinternal␈αstructure␈αof␈α"The␈αroom␈αis␈αtoo␈αcold"␈αis␈αa␈αpart
␈↓ ↓H␈↓of our language, not its.

␈↓ ↓H␈↓␈↓ α_Consider␈αa␈α
thermostat␈αwhose␈αwires␈α
to␈αthe␈α
furnace␈αhave␈αbeen␈α
cut.␈α Shall␈αwe␈α
still␈αsay␈α
that␈αit
␈↓ ↓H␈↓knows␈α
whether␈α
the␈α
room␈α
is␈αtoo␈α
cold?␈α
 Since␈α
fixing␈α
the␈αthermostat␈α
might␈α
well␈α
be␈α
aided␈αby␈α
ascribing
␈↓ ↓H␈↓this␈αknowledge,␈αwe␈αwould␈αlike␈αto␈αdo␈αso.␈α Our␈αexcuse␈αis␈αthat␈αwe␈αare␈αentitled␈αto␈αdistinguish␈α-␈αin␈α
our
␈↓ ↓H␈↓language␈α⊂-␈α⊂the␈α∂concept␈α⊂of␈α⊂a␈α⊂broken␈α∂temperature␈α⊂control␈α⊂system␈α⊂from␈α∂the␈α⊂concept␈α⊂of␈α⊂a␈α∂certain
␈↓ ↓H␈↓collection of parts, i.e. to make intensional characterizations of physical objects.



␈↓ ↓H␈↓␈↓ α_2.␈α
␈↓αSelf-reproducing␈αintelligent␈α
configurations␈αin␈α
a␈αcellular␈α
automaton␈αworld␈↓.␈α
 A␈α␈↓↓cellular␈↓
␈↓ ↓H␈↓␈↓↓automaton␈↓␈α␈↓↓system␈↓␈αassigns␈αa␈αfinite␈αautomaton␈αto␈αeach␈αpoint␈αof␈αthe␈αplane␈αwith␈αinteger␈αco-ordinates.
␈↓ ↓H␈↓The␈α⊂state␈α⊂of␈α⊂each␈α⊂automaton␈α⊂at␈α⊂time␈α⊂␈↓↓t+1␈↓␈α⊂depends␈α⊂on␈α⊂its␈α⊂state␈α⊂at␈α⊂time␈α⊂␈↓↓t␈↓␈α⊂and␈α⊂the␈α⊂states␈α⊂of␈α⊂its
␈↓ ↓H␈↓neighbors␈αat␈αtime␈α␈↓↓t␈↓.␈α An␈αearly␈αuse␈αof␈αcellular␈αautomata␈αwas␈αby␈αvon␈αNeumann␈α(196?)␈αwho␈αfound␈αa
␈↓ ↓H␈↓27␈αstate␈αautomaton␈αwhose␈αcells␈αcould␈αbe␈αinitialized␈αinto␈αa␈αself-reproducing␈αconfiguration␈αthat␈αwas
␈↓ ↓H␈↓also␈α
a␈αuniversal␈α
computer.␈α The␈α
basic␈α
automaton␈αin␈α
von␈αNeumann's␈α
system␈α
had␈αa␈α
"resting"␈αstate␈α
0,
␈↓ ↓H␈↓and␈αa␈αpoint␈αin␈αstate␈α0␈αwhose␈αfour␈αneighbors␈αwere␈αalso␈αin␈αthat␈αstate␈αwould␈αremain␈αin␈αstate␈α0.␈α The
␈↓ ↓H␈↓initial␈αconfigurations␈α
considered␈αhad␈αall␈α
but␈αa␈αfinite␈α
number␈αof␈αcells␈α
in␈αstate␈α0,␈α
and,␈αof␈αcourse,␈α
this
␈↓ ↓H␈↓property would persist although the number of non-zero cells might grow indefinitely with time.

␈↓ ↓H␈↓␈↓ α_The␈α
 self-reproducing␈α∞ system␈α
used␈α
the␈α∞states␈α
of␈α
a␈α∞long␈α
strip␈α
of␈α∞non-zero␈α
cells␈α
as␈α∞a␈α
"tape"
␈↓ ↓H␈↓containing␈αinstructions␈αto␈αa␈α"universal␈αconstructor"␈α configuration␈α that␈α would␈α construct␈α a␈α copy
␈↓ ↓H␈↓of␈α
 the␈αconfiguration␈α
to␈αbe␈α
reproduced␈αbut␈α
with␈α
each␈αcell␈α
in␈αa␈α
passive␈α state␈α
that␈α
 would␈α persist
␈↓ ↓H␈↓as␈α long␈α
 as␈αits␈αneighbors␈α
were␈αalso␈αin␈α
passive␈αstates.␈α
 After␈αthe␈αconstruction␈α
phase,␈αthe␈αtape␈α
would
␈↓ ↓H␈↓be␈α∞ copied␈α∞ to␈α∞make␈α∞ the␈α∞ tape␈α∞for␈α∞the␈α∞new␈α∞machine,␈α∞and␈α∞then␈α∞the␈α∞new␈α∞system␈α∞would␈α∞be␈α∞set␈α∞in
␈↓ ↓H␈↓motion␈α
by␈αactivating␈α
one␈αof␈α
its␈αcells.␈α
The␈α
 new␈α system␈α
 would␈αthen␈α
move␈αaway␈α
from␈α
its␈αmother,
␈↓ ↓H␈↓and␈α
the␈α
process␈α
would␈α
start␈α
over.␈α
 The␈α
purpose␈α
of␈α
the␈α
design␈α
was␈α
to␈α
 demonstrate␈α
 that␈α
 arbitrarily
␈↓ ↓H␈↓complex␈α⊂configurations␈α⊂ could␈α⊃ be␈α⊂ self-reproducing␈α⊂ -␈α⊂ the␈α⊃ complexity␈α⊂being␈α⊂assured␈α⊃by␈α⊂also
␈↓ ↓H␈↓requiring that they be universal computers.

␈↓ ↓H␈↓␈↓ α_Since␈α_von␈α↔Neumann's␈α_time,␈α↔simpler␈α_basic␈α↔cells␈α_admitting␈α_self-reproducing␈α↔universal
␈↓ ↓H␈↓computers␈α∞have␈α∞been␈α∞discovered.␈α∂ The␈α∞simplest␈α∞so␈α∞far␈α∞is␈α∂the␈α∞two␈α∞state␈α∞Life␈α∞automaton␈α∂of␈α∞John
␈↓ ↓H␈↓Conway␈α
(Gosper␈α
1976).␈α
 The␈α
state␈α
of␈α
a␈α
cell␈α
at␈α
time␈α
␈↓↓t+1␈↓␈α
is␈α
determined␈α
its␈α
state␈α
at␈α
time␈α
␈↓↓t␈↓␈α∞and␈α
the
␈↓ ↓H␈↓states␈αof␈αits␈α
eight␈αneighbors␈αat␈α
time␈α␈↓↓t.␈↓␈α Namely,␈α
a␈αpoint␈αwhose␈αstate␈α
is␈α0␈αwill␈α
change␈αto␈αstate␈α
1␈αif
␈↓ ↓H␈↓exactly␈αthree␈αof␈αits␈αneighbors␈αare␈αin␈αstate␈α1.␈α A␈αpoint␈αwhose␈αstate␈αis␈α1␈αwill␈αremain␈αin␈αstate␈α1␈αif␈αtwo
␈↓ ↓H␈↓or three of its neighbors are in state 1.  In all other cases the state becomes or remains 0.

␈↓ ↓H␈↓␈↓ α_Although␈α∞this␈α∂was␈α∞not␈α∞Conway's␈α∂reason␈α∞for␈α∞introducing␈α∂them,␈α∞Conway␈α∞and␈α∂Gosper␈α∞have
␈↓ ↓H␈↓shown that self-reproducing universal computers could be built up as Life configurations.
␈↓ ↓H␈↓␈↓ εH␈↓ *12


␈↓ ↓H␈↓␈↓ α_Consider␈α⊃a␈α⊂number␈α⊃of␈α⊂such␈α⊃self-reproducing␈α⊂universal␈α⊃computers␈α⊂operating␈α⊃in␈α⊃the␈α⊂Life
␈↓ ↓H␈↓plane,␈αand␈α
suppose␈αthat␈α
they␈αhave␈α
been␈αprogrammed␈αto␈α
study␈αthe␈α
properties␈αof␈α
their␈αworld␈αand␈α
to
␈↓ ↓H␈↓communicate␈α≠among␈α≠themselves␈α≠about␈α≠it␈α≠and␈α≠pursue␈α≠various␈α≠goals␈α≤co-operatively␈α≠and
␈↓ ↓H␈↓competitively.␈α∪ Call␈α∪these␈α∪configurations␈α∪Life␈α∪robots.␈α∪ In␈α∪some␈α∪respects␈α∪their␈α∪intellectual␈α∪and
␈↓ ↓H␈↓scientific␈α
problems␈α
will␈α
be␈α
like␈α
ours,␈α
but␈α
in␈α
one␈α
major␈α
respect␈α
they␈α
live␈α
in␈α
a␈α
simpler␈α∞world␈α
than
␈↓ ↓H␈↓ours␈α
seems␈α
to␈α
be.␈α Namely,␈α
the␈α
fundamental␈α
physics␈αof␈α
their␈α
world␈α
is␈αthat␈α
of␈α
the␈α
life␈αautomaton,
␈↓ ↓H␈↓and␈α⊃there␈α⊃is␈α⊃no␈α⊃obstacle␈α⊂to␈α⊃each␈α⊃robot␈α⊃␈↓↓knowing␈↓␈α⊃this␈α⊂physics,␈α⊃and␈α⊃being␈α⊃able␈α⊃to␈α⊃simulate␈α⊂the
␈↓ ↓H␈↓evolution␈αof␈αa␈αlife␈αconfiguration␈αgiven␈αthe␈αinitial␈αstate.␈α Moreover,␈αif␈αthe␈αinitial␈αstate␈αof␈αthe␈αrobot
␈↓ ↓H␈↓world␈α∞is␈α∞finite␈α
it␈α∞can␈α∞have␈α
been␈α∞recorded␈α∞in␈α
each␈α∞robot␈α∞in␈α
the␈α∞beginning␈α∞or␈α
else␈α∞recorded␈α∞on␈α
a
␈↓ ↓H␈↓strip␈αof␈αcells␈αthat␈αthe␈αrobots␈αcan␈αread.␈α (The␈αinfinite␈αregress␈αof␈αhaving␈αto␈αdescribe␈αthe␈αdescription
␈↓ ↓H␈↓is␈αavoided␈αby␈αproviding␈αthat␈αthe␈αdescription␈αis␈αnot␈αseparately␈αdescribed,␈αbut␈αcan␈αbe␈αread␈α␈↓↓both␈↓␈αas␈α
a
␈↓ ↓H␈↓description of the world ␈↓↓and␈↓ as a description of itself.)

␈↓ ↓H␈↓␈↓ α_Since␈α∂these␈α∂robots␈α∞know␈α∂the␈α∂initial␈α∞state␈α∂of␈α∂their␈α∞world␈α∂and␈α∂its␈α∞laws␈α∂of␈α∂motion,␈α∂they␈α∞can
␈↓ ↓H␈↓simulate␈αas␈αmuch␈αof␈αits␈αhistory␈αas␈αthey␈αwant,␈αassuming␈αthat␈αeach␈αcan␈αgrow␈αinto␈αunoccupied␈αspace
␈↓ ↓H␈↓so␈α∩as␈α∩to␈α⊃have␈α∩memory␈α∩to␈α⊃store␈α∩the␈α∩states␈α⊃of␈α∩the␈α∩world␈α⊃being␈α∩simulated.␈α∩ This␈α∩simulation␈α⊃is
␈↓ ↓H␈↓necessarily␈αslower␈α
than␈αreal␈α
time,␈αso␈αthey␈α
can␈αnever␈α
catch␈αup␈αwith␈α
the␈αpresent␈α
-␈αlet␈α
alone␈αpredict
␈↓ ↓H␈↓the␈αfuture.␈α
 This␈αis␈αobvious␈α
if␈αthe␈αsimulation␈α
is␈αcarried␈α
out␈αstraightforwardly␈αby␈α
updating␈αa␈αlist␈α
of
␈↓ ↓H␈↓currently␈αactive␈α
cells␈αin␈αthe␈α
simulated␈αworld␈αaccording␈α
to␈αthe␈αLife␈α
rule,␈αbut␈αit␈α
also␈αapplies␈α
to␈αany
␈↓ ↓H␈↓clever␈αmathematical␈α
method␈αthat␈αmight␈α
predict␈αmillions␈αof␈α
steps␈αahead␈α
so␈αlong␈αas␈α
it␈αis␈αsupposed␈α
to
␈↓ ↓H␈↓be␈α∪applicable␈α∪to␈α∪all␈α∪Life␈α∪configurations.␈α∪ (Some␈α∪Life␈α∪configurations,␈α∪e.g.␈α∪static␈α∪ones␈α∪or␈α∩ones
␈↓ ↓H␈↓containing␈αsingle␈α␈↓↓gliders␈↓␈αor␈α␈↓↓cannon␈↓␈αcan␈αhave␈αtheir␈αdistant␈αfutures␈αpredicted␈αwith␈αlittle␈αcomputing.)
␈↓ ↓H␈↓Namely,␈αif␈αthere␈αwere␈αan␈αalgorithm␈αfor␈αsuch␈α
prediction,␈αa␈αrobot␈αcould␈αbe␈αmade␈αthat␈αwould␈α
predict
␈↓ ↓H␈↓its␈α∞own␈α∞future␈α
and␈α∞then␈α∞disobey␈α∞the␈α
prediction.␈α∞ The␈α∞detailed␈α∞proof␈α
would␈α∞be␈α∞analogous␈α∞to␈α
the
␈↓ ↓H␈↓proof of unsolvability of the halting problem for Turing machines.

␈↓ ↓H␈↓␈↓ α_Now␈α
we␈αcome␈α
to␈α
the␈αpoint␈α
of␈α
this␈αlong␈α
disquisition.␈α
 Suppose␈αwe␈α
wish␈α
to␈αprogram␈α
a␈αrobot␈α
to
␈↓ ↓H␈↓be␈αsuccessful␈αin␈αthe␈αLife␈αworld␈αin␈αcompetition␈αor␈αco-operation␈αwith␈αthe␈αothers.␈α Without␈αany␈αidea
␈↓ ↓H␈↓of␈αhow␈αto␈αgive␈αa␈αmathematical␈αproof,␈αI␈αwill␈αclaim␈αthat␈αour␈αrobot␈αwill␈αneed␈αprograms␈α
that␈αascribe
␈↓ ↓H␈↓purposes␈α∞and␈α
beliefs␈α∞to␈α
its␈α∞fellow␈α
robots␈α∞and␈α
 predict␈α∞how␈α
they␈α∞will␈α
react␈α∞to␈α
its␈α∞ own␈α∞actions␈α
by
␈↓ ↓H␈↓assuming␈α∂that␈α∂␈↓↓they␈α∞will␈α∂act␈α∂in␈α∞ways␈α∂that␈α∂they␈α∞believe␈α∂will␈α∂achieve␈α∞their␈α∂goals␈↓.␈α∂ Our␈α∂robot␈α∞might
␈↓ ↓H␈↓acquire␈α∞these␈α∞mental␈α∞theories␈α∞in␈α∞several␈α∞ways:␈α
 First,␈α∞we␈α∞might␈α∞design␈α∞the␈α∞universal␈α∞machine␈α
so
␈↓ ↓H␈↓that␈α
they␈α∞are␈α
present␈α∞in␈α
the␈α∞initial␈α
configuration␈α
of␈α∞the␈α
world.␈α∞ Second,␈α
we␈α∞might␈α
program␈α∞it␈α
to
␈↓ ↓H␈↓acquire␈αthese␈α
ideas␈αby␈αinduction␈α
from␈αits␈α
experience␈αand␈αeven␈α
transmit␈αthem␈α
to␈αothers␈αthrough␈α
an
␈↓ ↓H␈↓"educational␈αsystem".␈α Third,␈αit␈αmight␈αderive␈αthe␈αpsychological␈αlaws␈αfrom␈αthe␈αfundamental␈αphysics
␈↓ ↓H␈↓of␈αthe␈αworld␈αand␈αits␈αknowledge␈αof␈αthe␈αinitial␈αconfiguration.␈α Finally,␈αit␈αmight␈αdiscover␈αhow␈αrobots
␈↓ ↓H␈↓are built from Life cells by doing experimental "biology".

␈↓ ↓H␈↓␈↓ α_Knowing␈α⊂the␈α⊃Life␈α⊂ physics␈α⊂without␈α⊃some␈α⊂information␈α⊂ about␈α⊃the␈α⊂initial␈α⊃configuration␈α⊂is
␈↓ ↓H␈↓insufficient␈α⊂to␈α⊂derive␈α⊃the␈α⊂␈↓↓psychological␈↓␈α⊂laws,␈α⊃because␈α⊂ robots␈α⊂can␈α⊃ be␈α⊂constructed␈α⊂ in␈α⊃the␈α⊂ Life
␈↓ ↓H␈↓world␈α
in␈α
 an␈αinfinity␈α
of␈α
ways.␈α This␈α
follows␈α
from␈αthe␈α
"folk␈α
theorem"␈αthat␈α
the␈α
Life␈α
automaton␈α is
␈↓ ↓H␈↓universal␈αin␈αthe␈αsense␈α that␈αany␈αcellular␈αautomaton␈αcan␈αbe␈αconstructed␈αby␈αtaking␈αsufficiently␈αlarge
␈↓ ↓H␈↓squares of  Life cells as the basic cell of the other automaton.␈↓∧9␈↓

␈↓ ↓H␈↓␈↓ α_Men␈α⊂are␈α⊃in␈α⊂a␈α⊂more␈α⊃difficult␈α⊂intellectual␈α⊂position␈α⊃than␈α⊂Life␈α⊂robots.␈α⊃ We␈α⊂don't␈α⊃know␈α⊂the
␈↓ ↓H␈↓fundamental␈α⊂physics␈α⊂of␈α∂our␈α⊂world,␈α⊂and␈α⊂we␈α∂can't␈α⊂even␈α⊂be␈α⊂sure␈α∂that␈α⊂its␈α⊂fundamental␈α⊂physics␈α∂is
␈↓ ↓H␈↓describable␈α∂in␈α∂finite␈α⊂terms.␈α∂ Even␈α∂if␈α∂we␈α⊂knew␈α∂the␈α∂physical␈α∂laws,␈α⊂they␈α∂seem␈α∂to␈α⊂preclude␈α∂precise
␈↓ ↓H␈↓knowledge␈α∞of␈α
an␈α∞initial␈α∞state␈α
and␈α∞precise␈α∞calculation␈α
of␈α∞its␈α∞future␈α
both␈α∞for␈α∞quantum␈α
mechanical
␈↓ ↓H␈↓␈↓ εH␈↓ *13


␈↓ ↓H␈↓reasons␈α∩and␈α⊃because␈α∩the␈α∩continuous␈α⊃functions␈α∩needed␈α⊃to␈α∩represent␈α∩fields␈α⊃seem␈α∩to␈α∩involve␈α⊃an
␈↓ ↓H␈↓infinite amount of information.

␈↓ ↓H␈↓␈↓ α_This␈α∩example␈α∪suggests␈α∩that␈α∪much␈α∩ of␈α∪human␈α∩mental␈α∪ structure␈α∩is␈α∪ not␈α∩an␈α∪accident␈α∩of
␈↓ ↓H␈↓evolution␈α
 or␈α
even␈α
of␈α
 the␈α
physics␈α
of␈α
 our␈α
world,␈α
but␈α
 is␈α
required␈α
for␈α
successful␈α
problem␈αsolving
␈↓ ↓H␈↓behavior and must be designed into or evolved by any system that exhibits such behavior.



␈↓ ↓H␈↓␈↓ α_3.␈α_␈↓αComputer␈α↔time-sharing␈α_systems.␈↓␈α_ These␈α↔complicated␈α_computer␈α_programs␈α↔allocate
␈↓ ↓H␈↓computer␈αtime␈αand␈α
other␈αresources␈αamong␈α
users.␈α They␈αallow␈α
each␈αuser␈αof␈α
the␈αcomputer␈αto␈α
behave
␈↓ ↓H␈↓as␈αthough␈αhe␈α
had␈αa␈αcomputer␈αof␈α
his␈αown,␈αbut␈αalso␈α
allow␈αthem␈αto␈αshare␈α
files␈αof␈αdata␈αand␈α
programs
␈↓ ↓H␈↓and␈α
to␈α
communicate␈α
with␈α
each␈α
other.␈α
 They␈α
are␈α
often␈α
used␈α
for␈α
many␈α
years␈α
with␈α∞continual␈α
small
␈↓ ↓H␈↓changes,␈α
and␈α
and␈α∞the␈α
people␈α
making␈α∞the␈α
changes␈α
and␈α
correcting␈α∞errors␈α
are␈α
often␈α∞different␈α
from
␈↓ ↓H␈↓the␈αoriginal␈αauthors␈αof␈αthe␈αsystem.␈α A␈α
person␈αconfronted␈αwith␈αthe␈αtask␈αof␈αcorrecting␈αa␈α
malfunction
␈↓ ↓H␈↓or␈α
making␈α
a␈α
change␈α∞in␈α
a␈α
time-sharing␈α
system␈α
often␈α∞can␈α
conveniently␈α
use␈α
a␈α
mentalistic␈α∞model␈α
of
␈↓ ↓H␈↓the system.

␈↓ ↓H␈↓␈↓ α_Thus␈α⊂suppose␈α∂a␈α⊂user␈α∂complains␈α⊂that␈α⊂the␈α∂system␈α⊂will␈α∂not␈α⊂run␈α∂his␈α⊂program.␈α⊂ Perhaps␈α∂the
␈↓ ↓H␈↓system␈αbelieves␈α
that␈αhe␈αdoesn't␈α
want␈αto␈α
run,␈αperhaps␈αit␈α
persistently␈αbelieves␈αthat␈α
he␈αhas␈α
just␈αrun,
␈↓ ↓H␈↓perhaps␈αit␈αbelieves␈αthat␈αhis␈αquota␈αof␈αcomputer␈αresources␈αis␈αexhausted,␈αor␈αperhaps␈αit␈αbelieves␈αthat
␈↓ ↓H␈↓his␈αprogram␈αrequires␈αa␈αresource␈αthat␈αis␈αunavailable.␈α Testing␈αthese␈αhypotheses␈αcan␈αoften␈α
be␈αdone
␈↓ ↓H␈↓with surprisingly little understanding of the internal workings of the program.



␈↓ ↓H␈↓␈↓ α_4.␈α⊃␈↓αPrograms␈α⊃designed␈α⊃to␈α⊂reason.␈↓␈α⊃ Suppose␈α⊃we␈α⊃explicitly␈α⊂design␈α⊃a␈α⊃program␈α⊃to␈α⊂represent
␈↓ ↓H␈↓information␈αby␈αsentences␈αin␈αa␈αcertain␈αlanguage␈αstored␈αin␈αthe␈αmemory␈αof␈αthe␈αcomputer␈αand␈αdecide
␈↓ ↓H␈↓what␈αto␈α
do␈αby␈α
making␈αinferences,␈αand␈α
doing␈αwhat␈α
it␈αconcludes␈αwill␈α
advance␈αits␈α
goals.␈α Naturally,
␈↓ ↓H␈↓we␈αwould␈αhope␈α
that␈αour␈αprevious␈α
second␈αorder␈αdefinition␈α
of␈αbelief␈αwill␈α
"approve␈αof"␈αa␈α␈↓↓B(p,s)␈↓␈α
that
␈↓ ↓H␈↓ascribed␈α∩to␈α∩the␈α⊃program␈α∩believing␈α∩the␈α⊃sentences␈α∩explicitly␈α∩built␈α⊃in.␈α∩ We␈α∩would␈α∩be␈α⊃somewhat
␈↓ ↓H␈↓embarassed␈αif␈αsomeone␈αwere␈αto␈αshow␈αthat␈α
our␈αsecond␈αorder␈αdefinition␈αapproved␈αas␈αwell␈α
or␈αbetter
␈↓ ↓H␈↓of an entirely different set of beliefs.

␈↓ ↓H␈↓␈↓ α_Such a program was first proposed in (McCarthy 1959), and here is how it might work:

␈↓ ↓H␈↓␈↓ α_Information␈αabout␈αthe␈αworld␈αis␈αstored␈αin␈αa␈αwide␈αvariety␈αof␈αdata␈αstructures.␈α For␈αexample,␈αa
␈↓ ↓H␈↓visual␈α∂scene␈α∂received␈α∂by␈α∂a␈α∂TV␈α∂camera␈α∂may␈α∂be␈α∂represented␈α∂by␈α∂a␈α∂512x512x3␈α∂array␈α⊂of␈α∂numbers
␈↓ ↓H␈↓representing␈αthe␈αintensities␈αof␈αthree␈α
colors␈αat␈αthe␈αpoints␈αof␈α
the␈αvisual␈αfield.␈α At␈αanother␈α
level,␈αthe
␈↓ ↓H␈↓same␈αscene␈α
may␈αbe␈α
represented␈αby␈α
a␈αlist␈α
of␈αregions,␈αand␈α
at␈αa␈α
further␈αlevel␈α
there␈αmay␈α
be␈αa␈α
list␈αof
␈↓ ↓H␈↓physical␈α
objects␈α
and␈αtheir␈α
parts␈α
together␈αwith␈α
other␈α
information␈αabout␈α
these␈α
objects␈αobtained␈α
from
␈↓ ↓H␈↓non-visual␈αsources.␈α Moreover,␈αinformation␈αabout␈αhow␈α
to␈αsolve␈αvarious␈αkinds␈αof␈αproblems␈αmay␈α
be
␈↓ ↓H␈↓represented by programs in some programming language.

␈↓ ↓H␈↓␈↓ α_However,␈α⊂all␈α⊂the␈α⊂above␈α∂representations␈α⊂are␈α⊂subordinate␈α⊂to␈α∂a␈α⊂collection␈α⊂of␈α⊂sentences␈α⊂in␈α∂a
␈↓ ↓H␈↓suitable␈α∂first␈α∂order␈α∂language␈α∞that␈α∂includes␈α∂set␈α∂theory.␈α∂ By␈α∞subordinate,␈α∂we␈α∂mean␈α∂that␈α∂there␈α∞are
␈↓ ↓H␈↓sentences␈α
that␈α
tell␈α
what␈α
the␈α
data␈α
structures␈α
represent␈α
and␈α
what␈α
the␈α
programs␈α
do.␈α
 New␈αsentences
␈↓ ↓H␈↓can␈αarise␈αby␈αa␈αvariety␈αof␈αprocesses:␈αinference␈αfrom␈αsentences␈αalready␈αpresent,␈αby␈αcomputation␈αfrom
␈↓ ↓H␈↓␈↓ εH␈↓ *14


␈↓ ↓H␈↓the␈α∨data␈α∨structures␈α∨representing␈α∨observations,␈α∨and␈α∨by␈α∨interpreting␈α∨certain␈α inputs␈α∨as
␈↓ ↓H␈↓communications in a one or more languages.

␈↓ ↓H␈↓␈↓ α_The␈αconstruction␈αof␈α
such␈αa␈αprogram␈αis␈α
one␈αof␈αthe␈αmajor␈α
approaches␈αto␈αachieving␈αhigh␈α
level
␈↓ ↓H␈↓artificial␈αintelligence,␈αand,␈αlike␈αevery␈αother␈αapproach,␈αit␈αfaces␈αnumerous␈αobstacles.␈α These␈αobstacles
␈↓ ↓H␈↓can␈αbe␈αdivided␈αinto␈αtwo␈αclasses␈α-␈α
␈↓↓epistemological␈↓␈αand␈α␈↓↓heuristic.␈↓␈α The␈αepistemological␈αproblem␈αis␈α
to
␈↓ ↓H␈↓determine␈αwhat␈αinformation␈αabout␈αthe␈αworld␈αis␈αto␈αbe␈αrepresented␈αin␈αthe␈αsentences␈αand␈αother␈αdata
␈↓ ↓H␈↓structures,␈α
and␈αthe␈α
heuristic␈αproblem␈α
is␈αto␈α
decide␈αhow␈α
the␈αinformation␈α
can␈αbe␈α
used␈α
effectively␈αto
␈↓ ↓H␈↓solve␈α∞problems.␈α∞ Naturally,␈α∞the␈α∞problems␈α∞interact,␈α
but␈α∞the␈α∞epistemological␈α∞problem␈α∞is␈α∞more␈α
basic
␈↓ ↓H␈↓and␈αalso␈αmore␈αrelevant␈αto␈αour␈αpresent␈αconcerns.␈α We␈αcould␈αregard␈αit␈αas␈αsolved␈αif␈αwe␈αknew␈αhow␈αto
␈↓ ↓H␈↓express␈αthe␈αinformation␈αneeded␈α
for␈αintelligent␈αbehavior␈αso␈α
that␈αthe␈αsolution␈αto␈α
problems␈αlogically
␈↓ ↓H␈↓followed from the data.  The heuristic problem of actually obtaining the solutions would remain.

␈↓ ↓H␈↓␈↓ α_The␈α
information␈αto␈α
be␈αrepresented␈α
can␈α
be␈αroughly␈α
divided␈αinto␈α
general␈α
information␈αabout
␈↓ ↓H␈↓the␈α⊗world␈α↔and␈α⊗information␈α⊗about␈α↔particular␈α⊗situations.␈α⊗ The␈α↔formalism␈α⊗used␈α↔to␈α⊗represent
␈↓ ↓H␈↓information␈α∪about␈α∪the␈α∪world␈α∪must␈α∪be␈α∪␈↓↓epistemologically␈↓␈α∪␈↓↓adequate,␈↓␈α∪i.e.␈α∪it␈α∪must␈α∪be␈α∪capable␈α∪of
␈↓ ↓H␈↓representing␈αthe␈αinformation␈αthat␈αis␈αactually␈αavailable␈αto␈αthe␈αprogram␈αfrom␈αits␈α
sensory␈αapparatus
␈↓ ↓H␈↓or␈αcan␈αbe␈αdeduced.␈α Thus␈αit␈αcouldn't␈αhandle␈α
available␈αinformation␈αabout␈αa␈αcup␈αof␈αhot␈αcoffee␈αif␈α
its
␈↓ ↓H␈↓only␈αway␈αof␈αrepresenting␈αinformation␈αabout␈αfluids␈αwas␈αin␈αterms␈αof␈αthe␈αpositions␈αand␈α
velocities␈αof
␈↓ ↓H␈↓the␈αmolecules.␈α Even␈αthe␈αhydrodynamicist's␈αEulerian␈αdistributions␈αof␈αdensity,␈αvelocity,␈αtemperature
␈↓ ↓H␈↓and␈α∀pressure␈α∪would␈α∀be␈α∪useless␈α∀for␈α∪representing␈α∀the␈α∪information␈α∀actually␈α∪obtainable␈α∀from␈α∪a
␈↓ ↓H␈↓television camera.  These considerations are further discussed in (McCarthy and Hayes 1969).

␈↓ ↓H␈↓␈↓ α_Here are some of the kinds of general information that will have to be represented:

␈↓ ↓H␈↓␈↓ α_1.␈α
Narrative.␈α
 Events␈αoccur␈α
in␈α
space␈α
and␈αtime.␈α
 Some␈α
events␈αare␈α
extended␈α
in␈α
time.␈α Partial
␈↓ ↓H␈↓information␈α
must␈α
be␈α
expressed␈α
about␈α
what␈α∞events␈α
begin␈α
or␈α
end␈α
during,␈α
before␈α
and␈α∞after␈α
others.
␈↓ ↓H␈↓Partial␈α∩information␈α∩about␈α∪places␈α∩and␈α∩their␈α∩spacial␈α∪relations␈α∩must␈α∩be␈α∪expressible.␈α∩ Sometimes
␈↓ ↓H␈↓dynamic␈α∂information␈α∞such␈α∂as␈α∂velocities␈α∞are␈α∂better␈α∂known␈α∞than␈α∂the␈α∂space-time␈α∞facts␈α∂in␈α∂terms␈α∞of
␈↓ ↓H␈↓which they are defined.

␈↓ ↓H␈↓␈↓ α_2.␈α∃Partial␈α∃information␈α∃about␈α∃causal␈α∃systems.␈α∃ Quantities␈α∃have␈α∃values␈α∃and␈α∃later␈α∃have
␈↓ ↓H␈↓different values.  Causal laws relate these values.

␈↓ ↓H␈↓␈↓ α_3.␈αSome␈αchanges␈αare␈αresults␈αof␈αactions␈αby␈αthe␈αprogram␈αand␈αother␈αactors.␈α Information␈αabout
␈↓ ↓H␈↓the effects of actions can be used to determine what goals can be achieved in given circumstances.

␈↓ ↓H␈↓␈↓ α_4.␈αObjects␈αand␈αsubstances␈αhave␈α
locations␈αin␈αspace.␈α It␈αmay␈α
be␈αthat␈αtemporal␈αand␈αcausal␈α
facts
␈↓ ↓H␈↓are prior to spatial facts in the formalism.

␈↓ ↓H␈↓␈↓ α_5. Some objects are actors with beliefs, purposes and intentions.

␈↓ ↓H␈↓␈↓ α_Of␈α∞course,␈α∞the␈α
above␈α∞English␈α∞description␈α
is␈α∞no␈α∞substitute␈α
for␈α∞an␈α∞axiomatized␈α∞formalism␈α
-
␈↓ ↓H␈↓not␈α⊂even␈α⊂for␈α⊃philosophy␈α⊂but␈α⊂␈↓↓a␈α⊃fortiori␈↓␈α⊂when␈α⊂computer␈α⊂programs␈α⊃must␈α⊂be␈α⊂written.␈α⊃ The␈α⊂main
␈↓ ↓H␈↓difficulties␈α∞in␈α∂designing␈α∞such␈α∞a␈α∂formalism␈α∞involve␈α∞deciding␈α∂how␈α∞to␈α∞express␈α∂partial␈α∞information.
␈↓ ↓H␈↓(McCarthy␈α∞and␈α∞Hayes␈α∞1969)␈α∞uses␈α∞a␈α∞notion␈α∂of␈α∞␈↓↓situation␈↓␈α∞wherein␈α∞the␈α∞situation␈α∞is␈α∞never␈α∂known␈α∞-
␈↓ ↓H␈↓only␈αfacts␈αabout␈αsituations␈αare␈αknown.␈α Unfortunately,␈αthe␈αformalism␈αis␈αnot␈αsuitable␈αfor␈α
expressing
␈↓ ↓H␈↓what␈αmight␈αbe␈αknown␈αwhen␈αevents␈αare␈αtaking␈αplace␈αin␈αparallel␈αwith␈αunknown␈αtemporal␈αrelations.
␈↓ ↓H␈↓␈↓ εH␈↓ *15


␈↓ ↓H␈↓It␈α⊃also␈α⊃only␈α⊃treats␈α⊃the␈α⊃case␈α⊃in␈α⊃which␈α⊃the␈α⊂result␈α⊃of␈α⊃an␈α⊃action␈α⊃is␈α⊃a␈α⊃definite␈α⊃new␈α⊃situation␈α⊂and
␈↓ ↓H␈↓therefore is isn't suitable for describing continuous processes.
␈↓ ↓H␈↓␈↓ εH␈↓ *16


␈↓ ↓H␈↓α␈↓ ∧."GLOSSARY" OF MENTAL QUALITIES

␈↓ ↓H␈↓␈↓ α_In␈α
this␈α
section␈α
we␈α∞give␈α
short␈α
"definitions"␈α
for␈α
machines␈α∞of␈α
a␈α
collection␈α
of␈α∞mental␈α
qualities.
␈↓ ↓H␈↓We␈αinclude␈αa␈αnumber␈αof␈α
terms␈αwhich␈αgive␈αus␈αdifficulty␈α
with␈αan␈αindication␈αof␈αwhat␈αthe␈α
difficulties
␈↓ ↓H␈↓seem to be.  We emphasize the place of these concepts in the design of intelligent robots.

␈↓ ↓H␈↓1.␈α∞␈↓αIntrospection␈α∂and␈α∞self-knowledge␈↓.␈α∂ We␈α∞say␈α∂that␈α∞ a␈α∂ machine␈α∞introspects␈α∂ when␈α∞it␈α∂ comes␈α∞ to
␈↓ ↓H␈↓have␈α
beliefs␈α
about␈α∞its␈α
own␈α
mental␈α
 state.␈α∞ A␈α
simple␈α
form␈α
of␈α∞ introspection␈α
takes␈α
 place␈α∞ when␈α
 a
␈↓ ↓H␈↓program␈α∂ determines␈α∂ whether␈α∂ it␈α⊂ has␈α∂ certain␈α∂information␈α∂ and␈α∂if␈α⊂not␈α∂asks␈α∂for␈α∂it.␈α⊂  Often␈α∂an
␈↓ ↓H␈↓operating␈αsystem␈α
will␈αcompute␈α
a␈αcheck␈α
 sum␈αof␈αitself␈α
 every␈αfew␈α
minutes␈α to␈α
verify␈αthat␈α
 it␈αhasn't
␈↓ ↓H␈↓been changed by a software or hardware malfunction.

␈↓ ↓H␈↓␈↓ α_In␈α∞principle,␈α∂ introspection␈α∞is␈α∞easier␈α∂ for␈α∞computer␈α∞programs␈α∂than␈α∞for␈α∞people,␈α∂because␈α∞the
␈↓ ↓H␈↓entire␈α∂memory␈α⊂in␈α∂which␈α⊂programs␈α∂and␈α⊂data␈α∂are␈α⊂stored␈α∂is␈α⊂available␈α∂for␈α⊂inspection.␈α∂  In␈α⊂fact,␈α∂a
␈↓ ↓H␈↓computer␈αprogram␈αcan␈α be␈α made␈αto␈α predict␈αhow␈α it␈α would␈αreact␈α to␈α particular␈αinputs␈αprovided
␈↓ ↓H␈↓it␈αhas␈αenough␈αfree␈αstorage␈α
to␈αperform␈αthe␈αcalculation.␈α This␈α
situation␈αsmells␈αof␈αparadox,␈αand␈α
there
␈↓ ↓H␈↓is␈αone.␈α Namely,␈αif␈αa␈αprogram␈αcould␈αpredict␈αits␈αown␈αactions␈αin␈αless␈αtime␈αthan␈αit␈αtakes␈αto␈αcarry␈αout
␈↓ ↓H␈↓the␈α∂action,␈α∞it␈α∂could␈α∂refuse␈α∞ to␈α∂do␈α∂what␈α∞it␈α∂has␈α∂predicted␈α∞ for␈α∂itself.␈α∂ This␈α∞only␈α∂ shows␈α∂that␈α∞self-
␈↓ ↓H␈↓simulation  is necessarily a  slow process, and this is not surprising.

␈↓ ↓H␈↓␈↓ α_However,␈α⊂  present␈α⊂   programs␈α⊂   do␈α⊂   little␈α∂   interesting␈α⊂introspection.␈α⊂  This␈α⊂is␈α⊂just␈α∂ a
␈↓ ↓H␈↓matter␈αof␈α the␈α
undeveloped␈α state␈αof␈α
artificial␈α intelligence;␈αprogrammers␈α
 don't␈αyet␈αknow␈α
 how␈αto
␈↓ ↓H␈↓make a computer program look at itself in a useful way.

␈↓ ↓H␈↓2.␈α
 ␈↓αConsciousness␈αand␈α
self-consciousness␈↓.␈α
 Suppose␈αwe␈α
wish␈α
to␈αdistinguish␈α
the␈α
self-awareness␈αof␈α
a
␈↓ ↓H␈↓machine,␈α
animal␈α
or␈α
person␈α
from␈α
its␈α
awareness␈α
of␈α
other␈α
things.␈α
 We␈α
explicate␈α
awareness␈α
as␈αbelief␈α
in
␈↓ ↓H␈↓certain␈αsentences,␈αso␈α
in␈αthis␈αcase␈α
we␈αare␈αwant␈αto␈α
distinguish␈αthose␈αsentences␈α
or␈αthose␈αterms␈α
in␈αthe
␈↓ ↓H␈↓sentences␈α
that␈α
may␈α
be␈α
considered␈α
to␈α
be␈α
about␈α
the␈α
self.␈α
 We␈α
also␈α
don't␈α
expect␈αthat␈α
self-consciousness
␈↓ ↓H␈↓will␈αbe␈αa␈αsingle␈αproperty␈αthat␈αsomething␈αeither␈αhas␈αor␈αhasn't␈αbut␈αrather␈αthere␈αwill␈αbe␈αmany␈αkinds
␈↓ ↓H␈↓of self-awareness with humans posessing many of the kinds we can imagine.

␈↓ ↓H␈↓␈↓ α_Here are some of the kinds of self-awareness:

␈↓ ↓H␈↓␈↓ αh2.1.␈α∂Certain␈α⊂predicates␈α∂of␈α∂the␈α⊂situation␈α∂(propositional␈α∂fluents␈α⊂in␈α∂the␈α⊂terminology␈α∂of
␈↓ ↓H␈↓(McCarthy␈α∂and␈α∞Hayes␈α∂1969))␈α∂are␈α∞directly␈α∂observable␈α∞in␈α∂almost␈α∂all␈α∞situations␈α∂while␈α∂others␈α∞often
␈↓ ↓H␈↓must␈α∞be␈α∞inferred.␈α∞ The␈α∞almost␈α∞always␈α∂observable␈α∞fluents␈α∞may␈α∞reasonably␈α∞be␈α∞identified␈α∂with␈α∞the
␈↓ ↓H␈↓senses.␈α
 Likewise␈α
the␈α
values␈α
of␈α
certain␈α
fluents␈α
are␈α
almost␈α
always␈α
under␈α
the␈α
control␈α
of␈α
the␈α
being␈α
and
␈↓ ↓H␈↓can␈α⊂be␈α⊂called␈α∂motor␈α⊂parameters␈α⊂for␈α∂lack␈α⊂of␈α⊂a␈α∂common␈α⊂language␈α⊂term.␈α∂ We␈α⊂have␈α⊂in␈α⊂mind␈α∂the
␈↓ ↓H␈↓positions␈α∩of␈α∩the␈α∩joints.␈α∩ Most␈α∩motor␈α⊃parameters␈α∩are␈α∩both␈α∩observable␈α∩and␈α∩controllable.␈α∩ I␈α⊃am
␈↓ ↓H␈↓inclined␈αto␈α
regard␈αthe␈α
posession␈αof␈α
a␈αsubstantial␈αset␈α
of␈αsuch␈α
constantly␈αobservable␈α
or␈αcontrollable
␈↓ ↓H␈↓fluents␈α
as␈αthe␈α
most␈αprimitive␈α
form␈α
of␈αself-consciousness,␈α
but␈αI␈α
have␈αno␈α
strong␈α
arguments␈αagainst
␈↓ ↓H␈↓someone who wished to require more.

␈↓ ↓H␈↓␈↓ αh2.2.␈αThe␈αsecond␈αlevel␈αof␈αself-consciousness␈αrequires␈αa␈αterm␈α␈↓↓I␈↓␈αin␈αthe␈αlanguage␈αdenoting
␈↓ ↓H␈↓the␈αself.␈α ␈↓↓I␈↓␈αshould␈αbelong␈αto␈αthe␈αclass␈αof␈αpersistent␈αobjects␈αand␈αsome␈αof␈αthe␈αsame␈αpredicates␈αshould
␈↓ ↓H␈↓be␈α∂applicable␈α∂to␈α∂it␈α∂as␈α∂are␈α∂applicable␈α∂to␈α∂other␈α∂objects.␈α∂ For␈α∂example,␈α∂like␈α∂other␈α∂objects␈α∂␈↓↓I␈↓␈α∂has␈α∂a
␈↓ ↓H␈↓location␈αthat␈αcan␈αchange␈αin␈αtime.␈α ␈↓↓I␈↓␈αis␈αalso␈αvisible␈αand␈αimpenetrable␈αlike␈αother␈αobjects.␈α However,
␈↓ ↓H␈↓we␈αdon't␈αwant␈αto␈αget␈αcarried␈αaway␈αin␈αregarding␈αa␈αphysical␈αbody␈αas␈αa␈αnecessary␈αcondition␈αfor␈αself-
␈↓ ↓H␈↓␈↓ εH␈↓ *17


␈↓ ↓H␈↓consciousness.␈α Imagine␈α
a␈αdistributed␈α
computer␈αwhose␈α
sense␈αand␈αmotor␈α
organs␈αcould␈α
also␈αbe␈α
in␈αa
␈↓ ↓H␈↓variety of places.  We don't want to exclude it from self-consciousness by definition.

␈↓ ↓H␈↓␈↓ αh2.3.␈α∪The␈α∪third␈α∪level␈α∪come␈α∪when␈α∪␈↓↓I␈↓␈α∪is␈α∪regarded␈α∪as␈α∪an␈α∪actor␈α∪among␈α∪others.␈α∪ The
␈↓ ↓H␈↓conditions␈αthat␈αpermit␈α␈↓↓I␈↓␈αto␈αdo␈αsomething␈αare␈αsimilar␈αto␈αthe␈αconditions␈αthat␈αpermit␈αother␈αactors␈αto
␈↓ ↓H␈↓do similar things.

␈↓ ↓H␈↓␈↓ αh2.4.␈α
The␈α
fourth␈α∞level␈α
requires␈α
the␈α
applicability␈α∞of␈α
predicates␈α
such␈α
as␈α∞␈↓↓believes,␈↓␈α
␈↓↓wants␈↓
␈↓ ↓H␈↓and␈α
␈↓↓can␈↓␈α
to␈α
␈↓↓I.␈↓␈α
 Beliefs␈α
about␈α
past␈α
situations␈α
and␈α
the␈α
ability␈α
to␈α
hypothesize␈α
future␈α
situations␈αare␈α
also
␈↓ ↓H␈↓required for this level.

␈↓ ↓H␈↓3.␈α␈↓αLanguage␈αand␈αthought␈↓.␈α Here␈αis␈αa␈αhypothesis␈αarising␈αfrom␈αartificial␈αintelligence␈αconcerning␈αthe
␈↓ ↓H␈↓relation␈αbetween␈α
language␈αand␈α
thought.␈α Imagine␈αa␈α
person␈αor␈α
machine␈αthat␈αrepresents␈α
information
␈↓ ↓H␈↓internally␈α
in␈α∞a␈α
huge␈α
network.␈α∞ Each␈α
node␈α
of␈α∞the␈α
network␈α
has␈α∞references␈α
to␈α
other␈α∞nodes␈α
through
␈↓ ↓H␈↓relations.␈α⊃ (If␈α⊃the␈α⊃system␈α⊃has␈α⊃a␈α⊃variable␈α⊃collection␈α⊃of␈α⊃relations,␈α⊃then␈α⊃the␈α⊃relations␈α⊃have␈α⊃to␈α⊃be
␈↓ ↓H␈↓represented␈αby␈αnodes,␈αand␈αwe␈αget␈αa␈αsymmetrical␈αtheory␈αif␈αwe␈αsuppose␈αthat␈αeach␈αnode␈αis␈αconnected
␈↓ ↓H␈↓to␈αa␈α
set␈αof␈αpairs␈α
of␈αother␈α
nodes).␈α We␈αcan␈α
imagine␈αthis␈α
structure␈αto␈αhave␈α
a␈αlong␈α
term␈αpart␈αand␈α
also
␈↓ ↓H␈↓extremely␈α⊂temporary␈α⊂parts␈α⊂representing␈α∂current␈α⊂␈↓↓thoughts␈↓.␈α⊂ Naturally,␈α⊂each␈α∂being␈α⊂has␈α⊂a␈α⊂its␈α∂own
␈↓ ↓H␈↓network␈α∞depending␈α
on␈α∞its␈α∞own␈α
experience.␈α∞A␈α∞thought␈α
is␈α∞then␈α∞a␈α
temporary␈α∞node␈α∞currently␈α
being
␈↓ ↓H␈↓referenced␈α⊂by␈α⊂the␈α∂mechanism␈α⊂of␈α⊂consciousness.␈α∂ Its␈α⊂meaning␈α⊂is␈α∂determined␈α⊂by␈α⊂its␈α⊂references␈α∂to
␈↓ ↓H␈↓other␈αnodes␈αwhich␈αin␈αturn␈αrefer␈αto␈αyet␈αother␈αnodes.␈α Now␈αconsider␈αthe␈αproblem␈αof␈αcommunicating
␈↓ ↓H␈↓a thought to another being.

␈↓ ↓H␈↓␈↓ α_Its␈α∀full␈α∀communication␈α∀would␈α∀ involve␈α∀transmitting␈α∀the␈α∀entire␈α∀network␈α∀ that␈α∀ can␈α∪ be
␈↓ ↓H␈↓reached␈α from␈α the␈α given␈α node,␈αand␈αthis␈αwould␈αordinarily␈αconstitute␈αthe␈αentire␈αexperience␈αof␈αthe
␈↓ ↓H␈↓being.␈α  More␈α
than␈αthat,␈α
 it␈αwould␈α
be␈α necessary␈α
to␈α also␈α
communicate␈αthe␈α
 programs␈αthat␈αthat␈α
take
␈↓ ↓H␈↓action␈α⊂on␈α⊃the␈α⊂basis␈α⊂of␈α⊃encountering␈α⊂certain␈α⊂nodes.␈α⊃ Even␈α⊂if␈α⊂all␈α⊃this␈α⊂could␈α⊂be␈α⊃transmitted,␈α⊂the
␈↓ ↓H␈↓recipient␈α∩would␈α⊃still␈α∩have␈α⊃to␈α∩find␈α⊃equivalents␈α∩ for␈α⊃ the␈α∩ information␈α⊃ in␈α∩ terms␈α⊃ of␈α∩ its␈α⊃ own
␈↓ ↓H␈↓network.␈α Therefore,␈αthoughts␈α have␈α
to␈α be␈αtranslated␈α into␈α a␈α
public␈α language␈αbefore␈αthey␈αcan␈α
be
␈↓ ↓H␈↓communicated.

␈↓ ↓H␈↓␈↓ α_A␈α∂language␈α∂ is␈α∂also␈α⊂a␈α∂network␈α∂ of␈α∂associations␈α⊂and␈α∂ programs.␈α∂ However,␈α∂certain␈α⊂ of␈α∂the
␈↓ ↓H␈↓nodes␈α in␈αthis␈αnetwork␈α(more␈αaccurately␈αa␈α␈↓↓family␈↓␈αof␈αnetworks,␈αsince␈αno␈αtwo␈αpeople␈αspeak␈αprecisely
␈↓ ↓H␈↓the␈α∞same␈α∞language)␈α∂are␈α∞ associated␈α∞with␈α∂words␈α∞ or␈α∞set␈α∂phrases.␈α∞ Sometimes␈α∞the␈α∂translation␈α∞from
␈↓ ↓H␈↓thoughts␈α∂to␈α∂sentences␈α∞is␈α∂easy,␈α∂because␈α∞large␈α∂ parts␈α∂of␈α∞the␈α∂ private␈α∂networks␈α∞are␈α∂taken␈α∂from␈α∞the
␈↓ ↓H␈↓public␈α∩network,␈α∩and␈α∩there␈α∩is␈α∩an␈α∩advantage␈α∩in␈α∩preserving␈α∩the␈α∩correspondence.␈α∩ However,␈α∩the
␈↓ ↓H␈↓translation␈αis␈αalways␈αapproximate␈α(in␈αsense␈αthat␈αstill␈αlacks␈αa␈αtechnical␈αdefinition),␈αand␈αsome␈α areas
␈↓ ↓H␈↓of␈α⊂experience␈α∂ are␈α⊂ difficult␈α∂to␈α⊂ translate␈α∂ at␈α⊂ all.␈α∂ Sometimes␈α⊂this␈α∂is␈α⊂for␈α∂intrinsic␈α⊂ reasons,␈α∂and
␈↓ ↓H␈↓sometimes␈α
because␈α
 particular␈αcultures␈α
don't␈α
use␈α
language␈αin␈α
this␈α
area.␈α (It␈α
is␈α
my␈α
impression␈αthat
␈↓ ↓H␈↓cultures␈αdiffer␈αin␈αthe␈α extent␈αto␈αwhich␈αinformation␈αabout␈α facial␈αappearance␈αthat␈αcan␈αbe␈αused␈αfor
␈↓ ↓H␈↓recognition␈α
is␈α verbally␈α
transmitted).␈α
 According␈αto␈α
 this␈αscheme,␈α
the␈α
"deep␈αstructure"␈α
 of␈αa␈α
publicly
␈↓ ↓H␈↓expressible␈αthought␈α is␈αa␈αnode␈αin␈αthe␈αpublic␈αnetwork.␈α It␈αis␈αtranslated␈αinto␈αthe␈αdeep␈αstructure␈αof␈αa
␈↓ ↓H␈↓sentence␈α⊃as␈α⊃a␈α⊃tree␈α⊃ whose␈α⊃terminal␈α⊃nodes␈α⊃are␈α⊃the␈α⊃nodes␈α⊃to␈α⊃ which␈α⊃words␈α⊃or␈α⊃set␈α⊃ phrases␈α⊂are
␈↓ ↓H␈↓attached.␈α∂ This␈α⊂"deep␈α∂ structure"␈α⊂then␈α∂must␈α⊂be␈α∂translated␈α⊂into␈α∂a␈α⊂string␈α∂in␈α⊂a␈α∂spoken␈α⊂or␈α∂written
␈↓ ↓H␈↓language.

␈↓ ↓H␈↓␈↓ α_The␈α
need␈αto␈α
use␈αlanguage␈α
to␈αexpress␈α
thought␈αalso␈α
applies␈αwhen␈α
we␈αhave␈α
to␈αascribe␈α
thoughts
␈↓ ↓H␈↓to other beings, since  we cannot put the entire network into a single sentence.
␈↓ ↓H␈↓␈↓ εH␈↓ *18


␈↓ ↓H␈↓4.␈α⊂␈↓αIntentions␈↓.␈α⊃ We␈α⊂are␈α⊃tempted␈α⊂to␈α⊃say␈α⊂that␈α⊂a␈α⊃machine␈α⊂␈↓↓intends␈↓␈α⊃to␈α⊂perform␈α⊃an␈α⊂action␈α⊃when␈α⊂it
␈↓ ↓H␈↓believes␈α∀it␈α∀will␈α∪and␈α∀also␈α∀believes␈α∪that␈α∀it␈α∀could␈α∪do␈α∀otherwise.␈α∀ However,␈α∪we␈α∀will␈α∀resist␈α∪this
␈↓ ↓H␈↓temptation␈αand␈αpropose␈αthat␈αa␈αpredicate␈α␈↓↓intends(actor,action,state)␈↓␈αbe␈αsuitably␈αaxiomatized␈αwhere
␈↓ ↓H␈↓one␈αof␈αthe␈αaxioms␈αsay␈αthat␈αthe␈αmachine␈αintends␈αthe␈αaction␈αif␈αit␈αbelieves␈αit␈αwill␈αperform␈αthe␈αaction
␈↓ ↓H␈↓and␈α∞could␈α∞do␈α∞otherwise.␈α∞ Armstrong␈α∞(1968)␈α
wants␈α∞to␈α∞require␈α∞an␈α∞element␈α∞of␈α∞servo-mechanism␈α
in
␈↓ ↓H␈↓order␈αthat␈αa␈αbelief␈αthat␈αan␈αaction␈αwill␈αbe␈αperformed␈αbe␈αregarded␈αas␈αan␈αintention,␈αi.e.␈αthere␈αshould
␈↓ ↓H␈↓be␈α⊂a␈α⊂commitment␈α⊂to␈α⊂do␈α⊂it␈α⊂one␈α⊂way␈α∂or␈α⊂another.␈α⊂ There␈α⊂may␈α⊂be␈α⊂good␈α⊂reasons␈α⊂to␈α⊂allow␈α∂several
␈↓ ↓H␈↓versions of intention to co-exist in the same formalism.

␈↓ ↓H␈↓5.␈α␈↓αFree␈αwill␈↓.␈α When␈αwe␈αprogram␈αa␈αcomputer␈α to␈α make␈αchoices␈α intelligently␈α after␈αdetermining␈αits
␈↓ ↓H␈↓options,␈α
examining␈α
their␈αconsequences,␈α
and␈α
deciding␈αwhich␈α
is␈α
 most␈α favorable␈α
or␈α
most␈α
moral␈αor
␈↓ ↓H␈↓whatever,␈αwe␈αmust␈αprogram␈αit␈α to␈α take␈α an␈α attitude␈α towards␈α its␈α freedom␈α of␈α
 choice␈αessentially
␈↓ ↓H␈↓isomorphic␈α to␈α that␈α which␈α
a␈αhuman␈αmust␈αtake␈αto␈α
his␈αown.␈α A␈αprogram␈α
will␈αhave␈αto␈αtake␈αsuch␈α
an
␈↓ ↓H␈↓attitude towards another unless it knows the details of the other's construction and present state.

␈↓ ↓H␈↓␈↓ α_We␈α⊃can␈α⊃define␈α⊂whether␈α⊃a␈α⊃particular␈α⊃action␈α⊂was␈α⊃free␈α⊃or␈α⊃forced␈α⊂␈↓↓relative␈α⊃to␈α⊃a␈α⊃theory␈↓␈α⊂that
␈↓ ↓H␈↓ascribes␈αbeliefs␈αand␈αwithin␈αwhich␈α
beings␈αdo␈αwhat␈αthey␈αbelieve␈α
will␈αadvance␈αtheir␈αgoals.␈α In␈αsuch␈α
a
␈↓ ↓H␈↓theory,␈α∞action␈α∞is␈α∞precipitated␈α∞by␈α∞a␈α∞belief␈α∞of␈α∞the␈α∂form␈α∞␈↓↓I␈α∞should␈α∞do␈α∞X␈α∞now␈↓.␈α∞ We␈α∞will␈α∞say␈α∂that␈α∞the
␈↓ ↓H␈↓action␈αwas␈αfree␈αif␈αchanging␈αthe␈αbelief␈αto␈α␈↓↓I␈αshouldn't␈αdo␈αX␈αnow␈↓␈αwould␈αhave␈αresulted␈αin␈αthe␈αaction
␈↓ ↓H␈↓not␈α∞being␈α∞performed.␈α∞ This␈α∞requires␈α∞that␈α
the␈α∞theory␈α∞of␈α∞belief␈α∞have␈α∞sufficient␈α∞Cartesian␈α
product
␈↓ ↓H␈↓structure␈αso␈αthat␈αchanging␈αa␈αsingle␈αbelief␈αis␈αdefined,␈αbut␈αit␈αdoesn't␈αrequire␈αdefining␈αwhat␈αthe␈αstate
␈↓ ↓H␈↓of the world would be if a single belief were different.

␈↓ ↓H␈↓␈↓ α_It␈α⊃may␈α⊃be␈α⊃possible␈α⊃to␈α⊃separate␈α⊃the␈α⊃notion␈α⊂of␈α⊃a␈α⊃␈↓↓free␈α⊃action␈↓␈α⊃into␈α⊃a␈α⊃technical␈α⊃part␈α⊃and␈α⊂a
␈↓ ↓H␈↓controversial␈α⊃part.␈α⊃ The␈α∩technical␈α⊃part␈α⊃would␈α⊃define␈α∩freedom␈α⊃relative␈α⊃to␈α⊃an␈α∩approximate␈α⊃co-
␈↓ ↓H␈↓ordinate␈α~system␈α≠giving␈α~the␈α~necessary␈α≠Cartesian␈α~product␈α~structure.␈α≠ Relative␈α~to␈α≠the␈α~co-
␈↓ ↓H␈↓ordinatization,␈α∂the␈α∂freedom␈α∞of␈α∂a␈α∂particular␈α∂action␈α∞would␈α∂be␈α∂a␈α∞technical␈α∂issue,␈α∂but␈α∂people␈α∞could
␈↓ ↓H␈↓argue about whether to accept the whole co-ordinate system.

␈↓ ↓H␈↓␈↓ α_This␈α∂isn't␈α∞the␈α∂ whole␈α∂free␈α∞will␈α∂story,␈α∞because␈α∂ moralists␈α∂are␈α∞also␈α∂ concerned␈α∂with␈α∞whether
␈↓ ↓H␈↓praise␈α
 or␈α∞blame␈α
may␈α
be␈α∞ attributed␈α
to␈α
a␈α∞choice.␈α
  The␈α
following␈α∞ considerations␈α
would␈α∞ seem␈α
to
␈↓ ↓H␈↓apply to  any attempt to define  the morality of actions in a  way that would apply to machines:

␈↓ ↓H␈↓␈↓ αh5.1.␈αThere␈αis␈αunlikely␈α
 to␈αbe␈αa␈αsimple␈α
behavioral␈α definition.␈α Instead␈αthere␈α
would␈αbe
␈↓ ↓H␈↓a second order definition criticizing predicates that ascribe morality to actions.

␈↓ ↓H␈↓␈↓ αh5.2.␈α∩The␈α∪theory␈α∩must␈α∪contain␈α∩at␈α∩least␈α∪one␈α∩axiom␈α∪of␈α∩morality␈α∩that␈α∪is␈α∩not␈α∪just␈α∩a
␈↓ ↓H␈↓statement of physical fact.  Relative to this axiom, moral judgments of actions can be factual.

␈↓ ↓H␈↓␈↓ αh5.3.␈αThe␈αtheory␈αof␈αmorality␈αwill␈αpresuppose␈αa␈αtheory␈αof␈αbelief␈αin␈αwhich␈αstatements␈αof
␈↓ ↓H␈↓the␈αform␈α␈↓↓"It␈αbelieved␈αthe␈αaction␈αwould␈αharm␈αsomeone"␈↓␈αare␈αdefined.␈α The␈αtheory␈αmust␈αascribe␈αbeliefs
␈↓ ↓H␈↓about others' welfare and perhaps about the being's own welfare.

␈↓ ↓H␈↓␈↓ αh5.4.␈α⊂It␈α⊂might␈α∂be␈α⊂necessary␈α⊂to␈α⊂consider␈α∂the␈α⊂ machine␈α⊂as␈α∂imbedded␈α⊂in␈α⊂some␈α⊂kind␈α∂of
␈↓ ↓H␈↓society in order to ascribe morality to its actions.

␈↓ ↓H␈↓␈↓ αh5.5.␈αNo␈α
 present␈αmachines␈αadmit␈α
 such␈αa␈α
belief␈α structure,␈αand␈α
 no␈αsuch␈αstructure␈α
 may
␈↓ ↓H␈↓be␈α⊂required␈α⊂ to␈α⊂make␈α⊂ a␈α⊂machine␈α⊃with␈α⊂ arbitrarily␈α⊂high␈α⊂intelligence␈α⊂in␈α⊂the␈α⊂sense␈α⊃of␈α⊂problem-
␈↓ ↓H␈↓solving ability.
␈↓ ↓H␈↓␈↓ εH␈↓ *19


␈↓ ↓H␈↓␈↓ αh5.6.␈α∞It␈α
seems␈α∞unlikely␈α
that␈α∞morally␈α∞ judgable␈α
machines␈α∞or␈α
 machines␈α∞to␈α∞which␈α
rights
␈↓ ↓H␈↓might legitimately be ascribed should be made if and when it becomes possible to do so.

␈↓ ↓H␈↓6.␈α
␈↓αUnderstanding␈↓.␈α
It␈α
seems␈αto␈α
me␈α
that␈α
understanding␈αthe␈α
concept␈α
of␈α
understanding␈αis␈α
fundamental
␈↓ ↓H␈↓and␈αdifficult.␈α The␈α
first␈αdifficulty␈αlies␈α
in␈αdetermining␈αwhat␈α
the␈αoperand␈αis.␈α
 What␈αis␈αthe␈α"theory␈α
of
␈↓ ↓H␈↓relativity"␈α∂in␈α∂␈↓↓"Pat␈α∂understands␈α∂the␈α⊂theory␈α∂of␈α∂relativity"␈↓?␈α∂ What␈α∂does␈α∂"misunderstand"␈α⊂mean?␈α∂ It
␈↓ ↓H␈↓seems␈α∩that␈α∩understanding␈α∩should␈α∩involve␈α∩knowing␈α⊃a␈α∩certain␈α∩collection␈α∩of␈α∩facts␈α∩including␈α⊃the
␈↓ ↓H␈↓general␈α∩laws␈α∩that␈α∩permit␈α⊃deducing␈α∩the␈α∩answers␈α∩to␈α⊃questions.␈α∩ We␈α∩probably␈α∩want␈α∩to␈α⊃separate
␈↓ ↓H␈↓understanding from issues of cleverness and creativity.

␈↓ ↓H␈↓7.␈α∞␈↓αCreativity␈↓.␈α∂ This␈α∞may␈α∞be␈α∂easier␈α∞than␈α∞"understanding"␈α∂at␈α∞least␈α∞if␈α∂we␈α∞confine␈α∞our␈α∂attention␈α∞to
␈↓ ↓H␈↓reasoning␈α∞processes.␈α
 Many␈α∞problem␈α
solutions␈α∞involve␈α∞the␈α
introduction␈α∞of␈α
entities␈α∞not␈α∞present␈α
in
␈↓ ↓H␈↓the␈α⊃statement␈α⊂of␈α⊃the␈α⊂problem.␈α⊃ For␈α⊂example,␈α⊃proving␈α⊃that␈α⊂an␈α⊃8␈α⊂by␈α⊃8␈α⊂square␈α⊃board␈α⊃with␈α⊂two
␈↓ ↓H␈↓diagonally␈αopposite␈αsquares␈αremoved␈αcannot␈α
be␈αcovered␈αby␈αdominoes␈αeach␈αcovering␈α
two␈αadjacent
␈↓ ↓H␈↓squares␈α∞involves␈α∂introducing␈α∞the␈α∂colors␈α∞of␈α∞the␈α∂squares␈α∞and␈α∂the␈α∞fact␈α∞that␈α∂a␈α∞dominoe␈α∂covers␈α∞two
␈↓ ↓H␈↓squares␈α
of␈α∞opposite␈α
color.␈α
 We␈α∞want␈α
to␈α
regard␈α∞this␈α
as␈α
a␈α∞creative␈α
proof␈α
even␈α∞though␈α
it␈α∞might␈α
be
␈↓ ↓H␈↓quite easy for an experienced combinatorist.
␈↓ ↓H␈↓␈↓ εH␈↓ *20


␈↓ ↓H␈↓α␈↓ ∧lOTHER VIEWS ABOUT MIND

␈↓ ↓H␈↓␈↓ α_The␈α∞fundamental␈α∞difference␈α∞in␈α∞point␈α∞of␈α∞view␈α∞between␈α∞this␈α∞paper␈α∞and␈α∞most␈α∞philosophy␈α∞is
␈↓ ↓H␈↓that␈α⊃we␈α⊃are␈α⊃motivated␈α⊃by␈α⊃the␈α⊃problem␈α⊃of␈α⊃designing␈α⊃an␈α⊃artificial␈α⊃intelligence.␈α⊃ Therefore,␈α⊃our
␈↓ ↓H␈↓attitude␈α
towards␈α
a␈α
concept␈α
like␈α
␈↓↓belief␈↓␈α
is␈αdetermined␈α
by␈α
trying␈α
to␈α
decide␈α
what␈α
ways␈α
of␈αacquiring␈α
and
␈↓ ↓H␈↓using␈α
beliefs␈αwill␈α
lead␈αto␈α
intelligent␈αbehavior.␈α
 Then␈αwe␈α
discover␈αthat␈α
much␈αthat␈α
one␈αintelligence
␈↓ ↓H␈↓can find out about another can be expressed by ascribing beliefs to it.

␈↓ ↓H␈↓␈↓ α_A␈α
negative␈α
view␈α
of␈α
empiricism␈α
seems␈α
dictated␈α
from␈α
the␈α
apparent␈α
artificiality␈α
of␈α
designing␈α
an
␈↓ ↓H␈↓empiricist␈α⊃computer␈α⊃program␈α∩to␈α⊃operate␈α⊃in␈α∩the␈α⊃real␈α⊃world.␈α⊃ Namely,␈α∩we␈α⊃plan␈α⊃to␈α∩provide␈α⊃our
␈↓ ↓H␈↓program␈α
with␈α
certain␈α
senses,␈αbut␈α
we␈α
have␈α
no␈α
way␈αof␈α
being␈α
sure␈α
that␈αthe␈α
world␈α
in␈α
which␈α
we␈αare
␈↓ ↓H␈↓putting␈αthe␈αmachine␈α
is␈αconstructable␈αfrom␈α
the␈αsense␈αimpressions␈α
it␈αwill␈αhave.␈α
 Whether␈αit␈αwill␈α
ever
␈↓ ↓H␈↓know␈α
some␈α
fact␈α
about␈α
the␈α
world␈αis␈α
contingent,␈α
so␈α
we␈α
are␈α
not␈αinclined␈α
to␈α
build␈α
into␈α
it␈α
the␈αnotion
␈↓ ↓H␈↓that what it can't know about doesn't exist.

␈↓ ↓H␈↓␈↓ α_The␈αphilosophical␈α
views␈αmost␈αsympathetic␈α
to␈αour␈αapproach␈α
are␈αsome␈αexpressed␈α
by␈αCarnap
␈↓ ↓H␈↓in some of the discursive sections of (Carnap 1956).

␈↓ ↓H␈↓␈↓ α_Hilary␈α∞Putnam␈α∞(1961)␈α∞argues␈α∞that␈α∞the␈α∞classical␈α∞mind-body␈α∞problems␈α∞are␈α∞just␈α∞as␈α∂acute␈α∞for
␈↓ ↓H␈↓machines␈αas␈αfor␈αmen.␈α Some␈αof␈αhis␈αarguments␈αare␈αmore␈αexplicit␈αthan␈αany␈αgiven␈αhere,␈αbut␈α
in␈αthat
␈↓ ↓H␈↓paper, he doesn't try to solve the problems for machines.

␈↓ ↓H␈↓␈↓ α_D.M.␈α⊂Armstrong␈α∂(1968)␈α⊂␈↓↓"attempts␈α∂to␈α⊂show␈α∂that␈α⊂there␈α∂are␈α⊂no␈α∂valid␈α⊂philosophical␈α⊂or␈α∂logical
␈↓ ↓H␈↓↓reasons␈α
for␈α
rejecting␈α
the␈α
identification␈α
of␈α
mind␈α
and␈α
brain."␈↓␈α
He␈α
does␈α
this␈α
by␈α
proposing␈αdefinitions␈α
of
␈↓ ↓H␈↓mental␈α
concepts␈α∞in␈α
terms␈α∞of␈α
the␈α∞state␈α
of␈α
the␈α∞brain.␈α
 Fundamentally,␈α∞I␈α
agree␈α∞with␈α
him␈α∞and␈α
think
␈↓ ↓H␈↓that␈α∞such␈α∞a␈α∞program␈α∞of␈α∂definition␈α∞can␈α∞be␈α∞carried␈α∞out,␈α∞but␈α∂it␈α∞seems␈α∞to␈α∞me␈α∞that␈α∞his␈α∂methods␈α∞for
␈↓ ↓H␈↓defining␈α∞mental␈α
qualities␈α∞as␈α∞brain␈α
states␈α∞are␈α∞too␈α
weak␈α∞even␈α∞for␈α
defining␈α∞properties␈α∞of␈α
computer
␈↓ ↓H␈↓programs.  While he goes beyond behavioral definitions as such, he relies on dispositional states.

␈↓ ↓H␈↓␈↓ α_This␈αpaper␈αis␈αpartly␈αan␈αattempt␈αto␈αdo␈αwhat␈αRyle␈α(1949)␈αsays␈αcan't␈αbe␈αdone␈αand␈αshouldn't␈αbe
␈↓ ↓H␈↓attempted␈α∞-␈α∞namely␈α∞to␈α∞define␈α∂mental␈α∞qualities␈α∞in␈α∞terms␈α∞of␈α∂states␈α∞of␈α∞a␈α∞machine.␈α∞ The␈α∂attempt␈α∞is
␈↓ ↓H␈↓based␈α
on␈α
methods␈α
of␈α
which␈α
he␈α∞would␈α
not␈α
approve;␈α
he␈α
implicitly␈α
requires␈α
first␈α∞order␈α
definitions,
␈↓ ↓H␈↓and␈αhe␈αimplicitly␈αrequires␈αthat␈αdefinitions␈αbe␈αmade␈αin␈αterms␈αof␈αthe␈αstate␈αof␈αthe␈αworld␈αand␈α
not␈αin
␈↓ ↓H␈↓terms of approximate theories.

␈↓ ↓H␈↓␈↓ α_His␈α∃final␈α∃view␈α∃of␈α∀the␈α∃proper␈α∃subject␈α∃matter␈α∃of␈α∀epistemology␈α∃is␈α∃too␈α∃narrow␈α∃to␈α∀help
␈↓ ↓H␈↓researchers␈α∂in␈α∂artificial␈α∂intelligence.␈α∞ Namely,␈α∂we␈α∂need␈α∂help␈α∞in␈α∂expressing␈α∂those␈α∂facts␈α∂about␈α∞the
␈↓ ↓H␈↓world␈αthat␈αcan␈αbe␈αobtained␈αin␈αan␈αordinary␈αsituation␈αby␈αan␈αordinary␈αperson␈αand␈αthe␈αgeneral␈αfacts
␈↓ ↓H␈↓about␈αthe␈αworld␈αwill␈αenable␈αour␈αprogram␈αto␈αdecide␈αto␈αcall␈αa␈αtravel␈αagent␈αto␈αfind␈αout␈αhow␈αto␈αget␈αto
␈↓ ↓H␈↓Boston.

␈↓ ↓H␈↓␈↓ α_Donald␈α∃Davidson␈α∀(1973)␈α∃undertakes␈α∀to␈α∃show,␈α∀␈↓↓"There␈α∃is␈α∀no␈α∃important␈α∀sense␈α∃in␈α∀which
␈↓ ↓H␈↓↓psychology␈α⊃can␈α∩be␈α⊃reduced␈α⊃to␈α∩the␈α⊃physical␈α⊃sciences"␈↓.␈α∩ He␈α⊃proceeds␈α⊃by␈α∩arguing␈α⊃that␈α∩the␈α⊃mental
␈↓ ↓H␈↓qualities␈α∂of␈α∂a␈α∂hypothetical␈α∂artificial␈α∂man␈α∂could␈α∂not␈α∂be␈α∂defined␈α∂physically␈α∂even␈α∂if␈α∂we␈α∂knew␈α∂the
␈↓ ↓H␈↓details of its physical structure.

␈↓ ↓H␈↓␈↓ α_One␈α∂sense␈α∂of␈α∂Davidson's␈α⊂statement␈α∂does␈α∂not␈α∂require␈α⊂the␈α∂arguments␈α∂he␈α∂gives.␈α⊂ There␈α∂are
␈↓ ↓H␈↓many␈α∂universal␈α∞computing␈α∂elements␈α∂-␈α∞relays,␈α∂neurons,␈α∞gates␈α∂and␈α∂flip-flops,␈α∞and␈α∂physics␈α∂tells␈α∞us
␈↓ ↓H␈↓␈↓ εH␈↓ *21


␈↓ ↓H␈↓many␈α
ways␈α
of␈α
constructing␈α
them.␈α
 Any␈α∞information␈α
processing␈α
system␈α
that␈α
can␈α
be␈α∞constructed␈α
of
␈↓ ↓H␈↓one␈α
kind␈α∞of␈α
element␈α∞can␈α
be␈α∞constructed␈α
of␈α∞any␈α
other.␈α∞ Therefore,␈α
physics␈α∞tells␈α
us␈α∞nothing␈α
about
␈↓ ↓H␈↓what␈α
information␈α∞processes␈α
exist␈α∞in␈α
nature␈α∞or␈α
can␈α
be␈α∞constructed.␈α
 Computer␈α∞science␈α
is␈α∞no␈α
more
␈↓ ↓H␈↓reducible to physics than is psychology.

␈↓ ↓H␈↓␈↓ α_However,␈αDavidson␈αalso␈α
argues␈αthat␈αthe␈αmental␈α
states␈αof␈αan␈α
organism␈αare␈αnot␈αdescribable␈α
in
␈↓ ↓H␈↓terms␈αof␈αits␈αphysical␈αstructure,␈αand␈αI␈αtake␈αthis␈αto␈αassert␈αalso␈αthat␈αthey␈αare␈αnot␈αdescribable␈αin␈αterms
␈↓ ↓H␈↓of␈α⊂its␈α⊂construction␈α∂from␈α⊂logical␈α⊂elements.␈α⊂ I␈α∂would␈α⊂take␈α⊂his␈α∂arguments␈α⊂as␈α⊂showing␈α⊂that␈α∂mental
␈↓ ↓H␈↓qualities␈αdon't␈αhave␈αwhat␈αI␈αhave␈αcalled␈αfirst␈αorder␈αstructural␈αdefinitions.␈α I␈αdon't␈αthink␈αthey␈αapply
␈↓ ↓H␈↓to second order definitions.

␈↓ ↓H␈↓␈↓ α_D.C.␈α∞Dennett␈α∞(1971)␈α∞expresses␈α∞views␈α∞very␈α∞similar␈α∞to␈α∞mine␈α∞about␈α∞the␈α∞reasons␈α∞for␈α∞ascribing
␈↓ ↓H␈↓mental␈α∪qualities␈α∪to␈α∪machines.␈α∪ However,␈α∪the␈α∪present␈α∪paper␈α∪emphasizes␈α∪criteria␈α∪for␈α∪ascribing
␈↓ ↓H␈↓particular␈αmental␈αqualities␈αto␈αparticular␈αmachines␈αrather␈αthan␈αthe␈αgeneral␈αproposition␈αthat␈αmental
␈↓ ↓H␈↓qualities␈α
may␈α
be␈α
ascribed.␈α
 I␈α
think␈α
that␈α
the␈α
chess␈α
programs␈α
Dennett␈α
discusses␈α
have␈α
more␈αlimited
␈↓ ↓H␈↓mental␈α⊂structures␈α⊂than␈α∂he␈α⊂seems␈α⊂to␈α⊂ascribe␈α∂to␈α⊂them.␈α⊂ Thus␈α∂their␈α⊂␈↓↓beliefs␈↓␈α⊂almost␈α⊂always␈α∂concern
␈↓ ↓H␈↓particular␈αpositions,␈αand␈αthey␈α
␈↓↓believe␈↓␈αalmost␈αno␈αgeneral␈α
propositions␈αabout␈αchess,␈αand␈αthis␈α
accounts
␈↓ ↓H␈↓for␈α
many␈α
of␈αtheir␈α
weaknesses.␈α
 Intuitively,␈αthis␈α
is␈α
well␈α
understood␈αby␈α
researchers␈α
in␈αcomputer␈α
game
␈↓ ↓H␈↓playing,␈α∂and␈α∂providing␈α∂the␈α∂program␈α∂with␈α∂a␈α∞way␈α∂of␈α∂representing␈α∂general␈α∂facts␈α∂about␈α∂chess␈α∞and
␈↓ ↓H␈↓even␈α⊃general␈α⊃facts␈α⊃about␈α⊂particular␈α⊃positions␈α⊃is␈α⊃a␈α⊂major␈α⊃unsolved␈α⊃problem.␈α⊃ For␈α⊃example,␈α⊂no
␈↓ ↓H␈↓present␈αprogram␈αcan␈αrepresent␈αthe␈α
assertion␈α␈↓↓"Black␈αhas␈αa␈αbackward␈α
pawn␈αon␈αhis␈αQ3␈αand␈αwhite␈α
may
␈↓ ↓H␈↓↓be␈αable␈α
to␈αcramp␈αblack's␈α
position␈αby␈αputting␈α
pressure␈αon␈αit"␈↓.␈α
 Such␈αa␈αrepresentation␈α
would␈αrequire
␈↓ ↓H␈↓rules␈α∞that␈α∞permit␈α∞such␈α∞a␈α∞statement␈α∞to␈α∞be␈α∞derived␈α∞in␈α∞appropriate␈α∞positions␈α∞and␈α∞would␈α∂guide␈α∞the
␈↓ ↓H␈↓examination of possible moves in accordance with it.

␈↓ ↓H␈↓␈↓ α_I␈α∂would␈α∂also␈α∂distinguish␈α∞between␈α∂believing␈α∂the␈α∂laws␈α∂of␈α∞logic␈α∂and␈α∂merely␈α∂using␈α∂them␈α∞(see
␈↓ ↓H␈↓Dennett,␈α
p.␈α95).␈α
 The␈α
former␈αrequires␈α
a␈α
language␈αthat␈α
can␈α
express␈αsentences␈α
about␈α
sentences␈αand
␈↓ ↓H␈↓which␈α
contains␈α
some␈α
kind␈αof␈α
reflexion␈α
principle.␈α
 Many␈αpresent␈α
problem␈α
solving␈α
programs␈αcan␈α
use
␈↓ ↓H␈↓␈↓↓modus␈α∀ponens␈↓␈α∀but␈α∀cannot␈α∀reason␈α∀about␈α∀their␈α∀own␈α∪ability␈α∀to␈α∀use␈α∀new␈α∀facts␈α∀in␈α∀a␈α∀way␈α∪that
␈↓ ↓H␈↓corresponds to believing ␈↓↓modus ponens␈↓.
␈↓ ↓H␈↓␈↓ εH␈↓ *22


␈↓ ↓H␈↓αNOTES

␈↓ ↓H␈↓1.␈α(McCarthy␈αand␈αHayes␈α1969)␈α
defines␈αan␈α␈↓↓epistemologically␈αadequate␈↓␈αrepresentation␈αof␈α
information
␈↓ ↓H␈↓as␈αone␈αthat␈αcan␈αexpress␈αthe␈αinformation␈αactually␈αavailable␈αto␈αa␈αsubject␈αunder␈αgiven␈αcircumstances.
␈↓ ↓H␈↓Thus␈αwhen␈αwe␈αsee␈αa␈αperson,␈α
parts␈αof␈αhim␈αare␈αoccluded,␈αand␈α
we␈αuse␈αour␈αmemory␈αof␈αprevious␈α
looks
␈↓ ↓H␈↓at␈α
him␈αand␈α
our␈αgeneral␈α
knowledge␈α
of␈αhumans␈α
to␈αfinish␈α
of␈α
a␈α"picture"␈α
of␈αhim␈α
that␈α
includes␈αboth
␈↓ ↓H␈↓two␈α_and␈α_three␈α→dimensional␈α_information.␈α_ We␈α→must␈α_also␈α_consider␈α→␈↓↓metaphysically␈α_adequate␈↓
␈↓ ↓H␈↓representations␈αthat␈αcan␈αrepresent␈αcomplete␈αfacts␈αignoring␈αthe␈αsubject's␈αability␈αto␈αacquire␈αthe␈αfacts
␈↓ ↓H␈↓in␈αgiven␈αcircumstances.␈α Thus␈αLaplace␈αthought␈αthat␈αthe␈αpositions␈αand␈αvelocities␈αof␈αthe␈αparticles␈αin
␈↓ ↓H␈↓the␈α∨universe␈α∨gave␈α∨a␈α∨metaphysically␈α∨adequate␈α∨representation.␈α∨ Metaphysically␈α≡adequate
␈↓ ↓H␈↓representations␈α∞are␈α∞needed␈α∞for␈α∞scientific␈α∞and␈α∞other␈α∞theories,␈α∞but␈α∞artificial␈α∞intelligence␈α∞and␈α∞a␈α∞full
␈↓ ↓H␈↓philosophical␈α∀treatment␈α∀of␈α∀common␈α∀sense␈α∀experience␈α∀also␈α∀require␈α∀epistemologically␈α∀adequate
␈↓ ↓H␈↓representations.␈α This␈αpaper␈αmight␈αbe␈αsummarized␈αas␈αcontending␈αthat␈αmental␈αconcepts␈αare␈αneeded
␈↓ ↓H␈↓for␈α⊗an␈α∃epistemologically␈α⊗adequate␈α⊗representation␈α∃of␈α⊗facts␈α∃about␈α⊗machines,␈α⊗especially␈α∃future
␈↓ ↓H␈↓intelligent machines.

␈↓ ↓H␈↓2.␈α
Work␈α
in␈α∞artificial␈α
intelligence␈α
is␈α∞still␈α
far␈α
from␈α
showing␈α∞how␈α
to␈α
reach␈α∞human-level␈α
intellectual
␈↓ ↓H␈↓performance.␈α∞Our␈α∞approach␈α∞to␈α∞the␈α∞AI␈α∞problem␈α∞involves␈α∞identifying␈α∞the␈α∞intellectual␈α
mechanisms
␈↓ ↓H␈↓required␈α
for␈α
problem␈α
solving␈α
and␈α
describing␈α
them␈αprecisely.␈α
Therefore␈α
we␈α
are␈α
at␈α
the␈α
end␈α
of␈αthe
␈↓ ↓H␈↓philosophical␈α⊃spectrum␈α⊂that␈α⊃requires␈α⊂everything␈α⊃to␈α⊂be␈α⊃formalized␈α⊂in␈α⊃mathematical␈α⊂logic.␈α⊃It␈α⊂ is
␈↓ ↓H␈↓sometimes␈α said␈α that␈α one␈αstudies␈αphilosophy␈α in␈αorder␈αto␈αadvance␈αbeyond␈α one's␈αuntutored␈αnaive
␈↓ ↓H␈↓world-view,␈α⊃but␈α∩unfortunately␈α⊃for␈α∩ artificial␈α⊃intelligence,␈α∩no-one␈α⊃has␈α∩yet␈α⊃been␈α∩able␈α⊃to␈α∩give␈α⊃a
␈↓ ↓H␈↓description␈α⊃of␈α⊃even␈α⊂a␈α⊃naive␈α⊃world-view,␈α⊂complete␈α⊃and␈α⊃precise␈α⊂enough␈α⊃to␈α⊃allow␈α⊃a␈α⊂knowledge-
␈↓ ↓H␈↓seeking program to be constructed in accordance with its tenets.

␈↓ ↓H␈↓3.␈α
Present␈α
AI␈α
programs␈α
operate␈α
in␈α
limited␈α
domains,␈α
e.g.␈α
play␈α
particular␈α
games,␈α
prove␈α
theorems␈αin␈α
a
␈↓ ↓H␈↓particular␈α
logical␈α∞system,␈α
or␈α
understand␈α∞natural␈α
language␈α
sentences␈α∞covering␈α
a␈α∞particular␈α
subject
␈↓ ↓H␈↓matter␈α∞and␈α∞with␈α∞other␈α∞semantic␈α∞restrictions.␈α∞ General␈α∞intelligence␈α∞will␈α∞require␈α∞general␈α∂models␈α∞of
␈↓ ↓H␈↓situations␈α
changing␈α∞in␈α
time,␈α∞actors␈α
with␈α∞goals␈α
and␈α∞strategies␈α
for␈α∞achieving␈α
them,␈α∞and␈α
knowledge
␈↓ ↓H␈↓about how information can be obtained.

␈↓ ↓H␈↓4.␈αOur␈αopinion␈α
is␈αthat␈αhuman␈α
intellectual␈αstructure␈αis␈α
substantially␈αdetermined␈αby␈α
the␈αintellectual
␈↓ ↓H␈↓problems␈αhumans␈αface.␈α
 Thus␈αa␈αMartian␈αor␈α
a␈αmachine␈αwill␈αneed␈α
similar␈αstructures␈αto␈αsolve␈α
similar
␈↓ ↓H␈↓problems.␈α∂Dennett␈α∂(1971)␈α∂expresses␈α∂similar␈α∂views.␈α∂On␈α∂the␈α∂other␈α∂hand,␈α∂the␈α∂human␈α∞motivational
␈↓ ↓H␈↓structure␈α
seems␈α
to␈α
have␈α
many␈α
accidental␈α
features␈α
that␈α
might␈α
not␈α
be␈α
found␈α
in␈α
Martians␈α
and␈α
that␈α
we
␈↓ ↓H␈↓would␈αnot␈αbe␈α
inclined␈αto␈αprogram␈αinto␈α
machines.␈α This␈αis␈αnot␈α
the␈αplace␈αto␈αpresent␈α
arguments␈αfor
␈↓ ↓H␈↓this viewpoint.

␈↓ ↓H␈↓5.␈αAfter␈αseveral␈α
versions␈αof␈αthis␈α
paper␈αwere␈αcompleted,␈α
I␈αcame␈αacross␈α
(Boden␈α1972)␈αwhich␈α
contains
␈↓ ↓H␈↓(among␈α⊂other␈α⊂things)␈α⊂an␈α⊂account␈α⊂of␈α⊂the␈α⊂psychology␈α⊂of␈α⊂William␈α⊂McDougall␈α⊂(1877-1938)␈α⊂and␈α⊂a
␈↓ ↓H␈↓discussion␈α⊂of␈α⊂a␈α⊂hypothetical␈α⊃program␈α⊂simulating␈α⊂it.␈α⊂ In␈α⊂my␈α⊃opinion,␈α⊂a␈α⊂psychology␈α⊂like␈α⊃that␈α⊂of
␈↓ ↓H␈↓McDougall␈α⊂is␈α⊂a␈α⊂better␈α⊂candidate␈α⊂for␈α∂simulation␈α⊂than␈α⊂many␈α⊂more␈α⊂recent␈α⊂psychological␈α∂theories,
␈↓ ↓H␈↓because␈αit␈αcomes␈αcloser␈αto␈αpresenting␈αa␈αtheory␈αof␈αthe␈αorganism␈αas␈αa␈αwhole,␈αproposing␈αmechanisms
␈↓ ↓H␈↓for␈αthoughts,␈αgoals,␈αand␈αemotions.␈αI␈αagree␈αwith␈αthe␈αways␈αin␈αwhich␈αBoden␈α
modernizes␈αMcDougall,
␈↓ ↓H␈↓but␈α∂even␈α∞with␈α∂her␈α∂improvements,␈α∞I␈α∂think␈α∂the␈α∞theory␈α∂is␈α∞a␈α∂long␈α∂way␈α∞from␈α∂being␈α∂simulatable,␈α∞let
␈↓ ↓H␈↓alone␈αcorrect.␈α One␈αmajor␈αproblem␈αis␈αthat␈αcompound␈α␈↓↓sentiments␈↓,␈αto␈αuse␈αMcDougall's␈αterm,␈αsuch␈αas
␈↓ ↓H␈↓reverence␈α↔are␈α↔diagrammed␈α↔in␈α↔Boden's␈α⊗book␈α↔as␈α↔essentially␈α↔Boolean␈α↔combinations␈α↔of␈α⊗their
␈↓ ↓H␈↓component␈α∀emotions.␈α∪ In␈α∀reality␈α∀they␈α∪must␈α∀at␈α∪least␈α∀be␈α∀complex␈α∪patterns␈α∀formed␈α∀from␈α∪their
␈↓ ↓H␈↓␈↓ εH␈↓ *23


␈↓ ↓H␈↓components␈α#and␈α$other␈α#entitities.␈α# Thus␈α$we␈α#must␈α#have␈α$sentences␈α#as␈α$complex␈α#as
␈↓ ↓H␈↓␈↓↓reveres(person1,␈αconcept1,␈αsituation)␈α≡␈αz.isbelief(z)␈α∧␈αascribes(person1,z,concept1,situation1)␈α∧␈αetc␈↓.
␈↓ ↓H␈↓If␈α∂I'm␈α∂right␈α∂about␈α∂this,␈α∂then␈α∂every␈α∂formulation␈α∂of␈α∂McDougall␈α∂will␈α∂have␈α∂to␈α∂be␈α∂taken␈α∂as␈α∞merely
␈↓ ↓H␈↓suggestive␈α⊗of␈α⊗what␈α⊗terms␈α⊗should␈α⊗be␈α⊗in␈α⊗the␈α⊗definitions␈α⊗and␈α⊗not␈α⊗as␈α⊗actually␈α⊗giving␈α∃them.
␈↓ ↓H␈↓Nevertheless,␈α∞it␈α∞seems␈α∞to␈α∂me␈α∞that␈α∞much␈α∞can␈α∞be␈α∂learned␈α∞from␈α∞contemplating␈α∞the␈α∞simulation␈α∂of␈α∞a
␈↓ ↓H␈↓McDougall man.  Axiomatizing the McDougall man should come before simulating it, however.

␈↓ ↓H␈↓6.␈α∂Behavioral␈α∞ definitions␈α∂are␈α∞often␈α∂ favored␈α∂in␈α∞ philosophy.␈α∂ A␈α∞system␈α∂is␈α∞ defined␈α∂to␈α∂ have␈α∞ a
␈↓ ↓H␈↓certain␈α∞ quality␈α∞if␈α∞ it␈α∞behaves␈α∞ in␈α∞ a␈α∞certain␈α∞way␈α∞ or␈α∞ is␈α∞ ␈↓↓disposed␈↓␈α∞to␈α∞behave␈α∞ in␈α∞a␈α∂ certain␈α∞way.
␈↓ ↓H␈↓Their␈α
virtue␈α∞ is␈α
 conservatism;␈α
  they␈α∞ don't␈α
 postulate␈α
internal␈α∞ states␈α
that␈α
 are␈α∞unobservable␈α
 to
␈↓ ↓H␈↓present␈α∩science␈α∩ and␈α∩may␈α∩remain␈α∩unobservable.␈α∩However,␈α∩such␈α∩ definitions␈α∩are␈α∪awkward␈α∩for
␈↓ ↓H␈↓mental␈α∩qualities,␈α∩because,␈α⊃as␈α∩common␈α∩sense␈α⊃ suggests,␈α∩a␈α∩ mental␈α⊃quality␈α∩ may␈α∩ not␈α∩ result␈α⊃ in
␈↓ ↓H␈↓behavior,␈α
because␈αanother␈α
mental␈α
 quality␈αmay␈α
prevent␈α
it;␈αe.g.␈α
 I␈α
may␈αthink␈α
you␈αare␈α
 thick-headed,
␈↓ ↓H␈↓but␈α⊂politeness␈α⊂may␈α⊂prevent␈α⊃my␈α⊂saying␈α⊂so.␈α⊂Particular␈α⊃ difficulties␈α⊂ can␈α⊂ be␈α⊂ overcome,␈α⊃ but␈α⊂ an
␈↓ ↓H␈↓impression␈α of␈αvagueness␈αremains.␈α The␈αliking␈α for␈αbehavioral␈αdefinitions␈αstems␈αfrom␈αcaution,␈αbut
␈↓ ↓H␈↓I␈αwould␈αinterpret␈αscientific␈αexperience␈αas␈αshowing␈αthat␈αboldness␈αin␈αpostulating␈α complex␈αstructures
␈↓ ↓H␈↓of␈αunobserved␈α entities␈α-␈αprovided␈αit␈αis␈α accompanied␈αby␈αa␈αwillingness␈αto␈α take␈αback␈αmistakes␈α-␈αis
␈↓ ↓H␈↓more␈αlikely␈α
to␈αbe␈α
rewarded␈αby␈αunderstanding␈α
of␈α and␈α
 control␈α over␈α
nature␈α  than␈α  is␈α
  positivistic
␈↓ ↓H␈↓timidity.␈α   It␈α is␈α particularly␈αinstructive␈αto␈αimagine␈αa␈αdetermined␈αbehaviorist␈αtrying␈αto␈αfigure␈α out
␈↓ ↓H␈↓an␈α⊂ electronic␈α⊂ computer.␈α⊂ Trying␈α⊂to␈α⊂define␈α∂each␈α⊂quality␈α⊂behaviorally␈α⊂would␈α⊂get␈α⊂him␈α∂nowhere;
␈↓ ↓H␈↓only␈α∃ simultaneously␈α∃ postulating␈α∀ a␈α∃ complex␈α∃structure␈α∀ including␈α∃ memory,␈α∃arithmetic␈α∀unit,
␈↓ ↓H␈↓control␈α∩structure,␈α∩and␈α∩input-output␈α∩would␈α⊃yield␈α∩predictions␈α∩ that␈α∩ could␈α∩ be␈α∩ compared␈α⊃ with
␈↓ ↓H␈↓experiment.␈α         There␈αis␈αa␈αsense␈αin␈αwhich␈αoperational␈αdefinitions␈αare␈αnot␈αtaken␈αseriously␈αeven
␈↓ ↓H␈↓by␈α
their␈α
proposers.␈α
 Suppose␈α∞someone␈α
gives␈α
an␈α
operational␈α∞definition␈α
of␈α
length␈α
(e.g.␈α∞involving␈α
a
␈↓ ↓H␈↓certain␈αplatinum␈α
bar),␈αand␈α
a␈αwhole␈α
school␈αof␈α
physicists␈αand␈α
philosophers␈αbecomes␈α
quite␈αattached
␈↓ ↓H␈↓to␈αit.␈α
 A␈αfew␈α
years␈αlater,␈αsomeone␈α
else␈αcriticizes␈α
the␈αdefinition␈αas␈α
lacking␈αsome␈α
desirable␈αproperty,
␈↓ ↓H␈↓proposes␈α∂a␈α∂change,␈α∂and␈α⊂the␈α∂change␈α∂is␈α∂accepted.␈α∂ This␈α⊂is␈α∂normal,␈α∂but␈α∂if␈α∂the␈α⊂original␈α∂definition
␈↓ ↓H␈↓expressed␈α
what␈α
they␈α
really␈α
meant␈αby␈α
the␈α
length,␈α
they␈α
would␈αrefuse␈α
to␈α
change,␈α
arguing␈α
that␈αthe␈α
new
␈↓ ↓H␈↓concept␈α
may␈α
have␈αits␈α
uses,␈α
but␈α
it␈αisn't␈α
what␈α
they␈αmean␈α
by␈α
"length".␈α
 This␈αshows␈α
that␈α
the␈αconcept␈α
of
␈↓ ↓H␈↓"length"␈α
as␈α
a␈α
property␈αof␈α
objects␈α
is␈α
more␈α
stable␈αthan␈α
any␈α
operational␈α
definition.␈α         Carnap␈α
has
␈↓ ↓H␈↓an␈αinteresting␈αsection␈αin␈α␈↓↓Meaning␈αand␈αNecessity␈↓␈αentitled␈α"The␈αConcept␈αof␈αIntension␈αfor␈αa␈αRobot"
␈↓ ↓H␈↓in␈α∞which␈α
he␈α∞makes␈α
a␈α∞similar␈α
point␈α∞saying,␈α
␈↓↓"It␈α∞is␈α
clear␈α∞that␈α
the␈α∞method␈α
of␈α∞structural␈α∞analysis,␈α
if
␈↓ ↓H␈↓↓applicable,␈α∪is␈α∩more␈α∪powerful␈α∪than␈α∩the␈α∪behavioristic␈α∩method,␈α∪because␈α∪it␈α∩can␈α∪supply␈α∪a␈α∩general
␈↓ ↓H␈↓↓answer,␈α∪and,␈α∪under␈α∪favorable␈α∪circumstances,␈α∪even␈α∪a␈α∪complete␈α∪answer␈α∪to␈α∪the␈α∪question␈α∪of␈α∪the
␈↓ ↓H␈↓↓intension␈α∩of␈α∩a␈α∩given␈α∩predicate."␈↓␈α∩         The␈α∩clincher␈α∩for␈α∩AI,␈α∩however,␈α∩is␈α∩an␈α∩"argument␈α⊃from
␈↓ ↓H␈↓design".␈α∞In␈α∞order␈α∞to␈α∞produce␈α∞desired␈α∂behavior␈α∞in␈α∞a␈α∞computer␈α∞program,␈α∞we␈α∞build␈α∂certain␈α∞mental
␈↓ ↓H␈↓qualities␈α⊃into␈α⊃its␈α⊃structure.␈α⊃ This␈α⊃doesn't␈α⊃lead␈α⊃to␈α⊃behavioral␈α⊃characterizations␈α⊃of␈α⊃the␈α⊃qualities,
␈↓ ↓H␈↓because␈α⊂the␈α⊂particular␈α⊂qualities␈α∂are␈α⊂only␈α⊂one␈α⊂of␈α∂many␈α⊂ways␈α⊂we␈α⊂might␈α∂used␈α⊂to␈α⊂get␈α⊂the␈α∂desired
␈↓ ↓H␈↓behavior, and anyway the desired behavior is not always realized.

␈↓ ↓H␈↓7.␈α∀Putnam␈α∀(1970)␈α∀also␈α∪proposes␈α∀what␈α∀amount␈α∀to␈α∪second␈α∀order␈α∀definitions␈α∀for␈α∪psychological
␈↓ ↓H␈↓properties.

␈↓ ↓H␈↓8.␈αWhether␈α
a␈αsystem␈αhas␈α
beliefs␈αand␈αother␈α
mental␈αqualities␈α
is␈αnot␈αprimarily␈α
a␈αmatter␈αof␈α
complexity
␈↓ ↓H␈↓of␈αthe␈α
system.␈α Although␈α
cars␈αare␈α
more␈αcomplex␈αthan␈α
thermostats,␈αit␈α
is␈αhard␈α
to␈αascribe␈α
beliefs␈αor
␈↓ ↓H␈↓goals␈αto␈αthem,␈αand␈αthe␈αsame␈αis␈αperhaps␈αtrue␈αof␈αthe␈αbasic␈αhardware␈αof␈αa␈αcomputer,␈αi.e.␈αthe␈αpart␈αof
␈↓ ↓H␈↓the computer that executes the program without the program itself.
␈↓ ↓H␈↓␈↓ εH␈↓ *24


␈↓ ↓H␈↓9.␈α Our␈αown␈αability␈αto␈αderive␈αthe␈αlaws␈αof␈αhigher␈αlevels␈αof␈αorganization␈αfrom␈αknowledge␈α of␈αlower
␈↓ ↓H␈↓level␈αlaws␈αis␈α also␈αlimited␈αby␈αuniversality.␈αWhile␈αthe␈αpresently␈αaccepted␈αlaws␈αof␈αphysics␈αallow␈αonly
␈↓ ↓H␈↓one␈αchemistry,␈αthe␈α laws␈αof␈αphysics␈αand␈αchemistry␈αallow␈αmany␈αbiologies,␈αand,␈α because␈αthe␈αneuron
␈↓ ↓H␈↓is␈α~a␈α~universal␈α~computing␈α~element,␈α~an␈α~arbitrary␈α~ mental␈α~structure␈α~is␈α~ allowed␈α~by␈α→basic
␈↓ ↓H␈↓neurophysiology.␈α→Therefore,␈α~ to␈α→ determine␈α→ human␈α~ mental␈α→ structure,␈α→ one␈α~ must␈α→ make
␈↓ ↓H␈↓psychological␈α experiments,␈α
 ␈↓↓or␈↓␈α determine␈α
 the␈α actual␈α  anatomical␈α
structure␈αof␈α
the␈α brain␈αand␈α
the
␈↓ ↓H␈↓information␈α
stored␈αin␈α
it␈α.␈α
One␈αcannot␈α
determine␈α
the␈αstructure␈α
of␈αthe␈α
brain␈αmerely␈α
from␈α
 the␈αfact
␈↓ ↓H␈↓that␈α
the␈α
 brain␈α
is␈α
capable␈α
of␈α
 certain␈α
problem␈α
solving␈α
performance.␈α
 In␈α
this␈α
 respect,␈α
our␈α
position␈α
is
␈↓ ↓H␈↓similar to that of the Life robot.

␈↓ ↓H␈↓␈↓ α_10.␈α∂Philosophy␈α∂and␈α∂artificial␈α∂intelligence.␈α∂ These␈α∂fields␈α∂overlap␈α∂in␈α∂the␈α∂following␈α∂way:␈α∞ In
␈↓ ↓H␈↓order␈αto␈αmake␈αa␈αcomputer␈αprogram␈αbehave␈αintelligently,␈αits␈αdesigner␈αmust␈αbuild␈αinto␈αit␈αa␈α
view␈αof
␈↓ ↓H␈↓the␈α
world␈α
in␈α
general,␈α∞apart␈α
from␈α
what␈α
they␈α
include␈α∞about␈α
particular␈α
sciences.␈α
 (The␈α∞skeptic␈α
who
␈↓ ↓H␈↓doubts␈αwhether␈αthere␈α
is␈αanything␈αto␈αsay␈α
about␈αthe␈αworld␈αapart␈α
from␈αthe␈αparticular␈αsciences␈α
should
␈↓ ↓H␈↓try␈αto␈αwrite␈α
a␈αcomputer␈αprogram␈α
that␈αcan␈αfigure␈αout␈α
how␈αto␈αget␈α
to␈αTimbuktoo,␈αtaking␈αinto␈α
account
␈↓ ↓H␈↓not␈αonly␈αthe␈αfacts␈αabout␈α
travel␈αin␈αgeneral␈αbut␈αalso␈α
facts␈αabout␈αwhat␈αpeople␈αand␈α
documents␈α have
␈↓ ↓H␈↓what␈αinformation,␈αand␈αwhat␈αinformation␈αwill␈αbe␈αrequired␈αat␈αdifferent␈αstages␈αof␈αthe␈αtrip␈αand␈αwhen
␈↓ ↓H␈↓and␈α
how␈α
it␈α
is␈α
to␈α
be␈α
obtained.␈α
 He␈α
will␈α
rapidly␈α
discover␈α
that␈α
he␈α
is␈α
lacking␈α
a␈α
␈↓↓science␈α
of␈αcommon␈α
sense␈↓,
␈↓ ↓H␈↓i.e.␈α∞he␈α∞will␈α∞be␈α∞unable␈α∞to␈α∞formally␈α∞express␈α∞and␈α∞build␈α∞into␈α∞his␈α∞program␈α∞"what␈α∞everybody␈α
knows".
␈↓ ↓H␈↓Maybe␈αphilosophy␈αcould␈αbe␈αdefined␈αas␈αan␈αattempted␈α␈↓↓science␈αof␈αcommon␈αsense␈↓,␈αor␈αelse␈αthe␈α␈↓↓science␈α
of
␈↓ ↓H␈↓↓common sense␈↓ should be a definite part of philosophy.)

␈↓ ↓H␈↓␈↓ α_Artificial␈α⊂intelligence␈α∂has␈α⊂a␈α∂another␈α⊂component␈α∂in␈α⊂which␈α∂philosophers␈α⊂have␈α⊂not␈α∂studied,
␈↓ ↓H␈↓namely␈α␈↓↓heuristics␈↓.␈α
 Heuristics␈αis␈α
 concerned␈α with:␈α
 given␈α the␈α facts␈α
 and␈α a␈α
 goal,␈α how␈α
should␈αit
␈↓ ↓H␈↓investigate␈α
the␈α
possibilities␈αand␈α
decide␈α
what␈α
to␈αdo.␈α
 On␈α
the␈α
other␈αhand,␈α
artificial␈α
intelligence␈αis␈α
not
␈↓ ↓H␈↓much concerned with aesthetics and ethics.

␈↓ ↓H␈↓␈↓ α_Not␈α∪all␈α∀approaches␈α∪to␈α∪philosophy␈α∀lead␈α∪to␈α∪results␈α∀relevant␈α∪to␈α∪the␈α∀artificial␈α∪intelligence
␈↓ ↓H␈↓problem.␈α∞ On␈α∂the␈α∞face␈α∞of␈α∂it,␈α∞a␈α∞philosophy␈α∂that␈α∞entailed␈α∞the␈α∂view␈α∞that␈α∞artificial␈α∂intelligence␈α∞was
␈↓ ↓H␈↓impossible␈α∂would␈α⊂be␈α∂unhelpful,␈α⊂but␈α∂besides␈α⊂that,␈α∂taking␈α⊂artificial␈α∂intelligence␈α⊂seriously␈α∂suggests
␈↓ ↓H␈↓some␈α
philosophical␈αpoints␈α
of␈αview.␈α
 I␈αam␈α
not␈αsure␈α
that␈α
all␈αI␈α
shall␈αlist␈α
are␈αrequired␈α
for␈αpursuing␈α
the
␈↓ ↓H␈↓AI goal - some of them may be just my prejudices - but here they are:

␈↓ ↓H␈↓␈↓ αh10.1.␈αThe␈αrelation␈αbetween␈α
a␈αworld␈αview␈αand␈αthe␈α
world␈αshould␈αbe␈αstudied␈αby␈α
methods
␈↓ ↓H␈↓akin␈αto␈αmetamathematics␈αin␈αwhich␈αsystems␈α
are␈αstudied␈αfrom␈αthe␈αoutside.␈α In␈α
metamathematics␈αwe
␈↓ ↓H␈↓study␈α∀the␈α∀relation␈α∀between␈α∪a␈α∀mathematical␈α∀system␈α∀and␈α∪its␈α∀models.␈α∀ Philosophy␈α∀(or␈α∪perhaps
␈↓ ↓H␈↓␈↓↓metaphilosophy␈↓)␈αshould␈αstudy␈αthe␈αrelation␈α
between␈αworld␈αstructures␈αand␈αsystems␈αwithin␈α
them␈αthat
␈↓ ↓H␈↓seek␈αknowledge.␈α Just␈αas␈αthe␈αmetamathematician␈α
can␈αuse␈αany␈αmathematical␈αmethods␈αin␈α
this␈αstudy
␈↓ ↓H␈↓and␈αdistinguishes␈αthe␈α
methods␈αhe␈αuses␈α
form␈αthose␈αbeing␈α
studied,␈αso␈αthe␈α
philosopher␈αshould␈αuse␈α
all
␈↓ ↓H␈↓his scientific knowledge in studying philosophical systems from the outside.

␈↓ ↓H␈↓␈↓ α_Thus␈α⊂the␈α∂question␈α⊂␈↓↓"How␈α⊂do␈α∂I␈α⊂know?"␈↓␈α∂is␈α⊂best␈α⊂answered␈α∂by␈α⊂studying␈α∂␈↓↓"How␈α⊂does␈α⊂it␈α∂know"␈↓,
␈↓ ↓H␈↓getting␈αthe␈α
best␈αanswer␈αthat␈α
the␈αcurrent␈αstate␈α
of␈αscience␈αand␈α
philosophy␈αpermits,␈αand␈α
then␈αseeing
␈↓ ↓H␈↓how this answer stands up to doubts about one's own sources of knowledge.

␈↓ ↓H␈↓␈↓ αh10.2.␈αWe␈α regard␈α ␈↓↓metaphysics␈↓␈αas␈α the␈αstudy␈α of␈αthe␈α general␈αstructure␈α of␈α  the␈α world
␈↓ ↓H␈↓and␈α
 ␈↓↓epistemology␈↓␈α
 as␈α
 studying␈α
 what␈α
knowledge␈α
of␈α
 the␈α
world␈α
 can␈α
be␈α
 had␈α
by␈α
 an␈α
 intelligence
␈↓ ↓H␈↓with␈α
 given␈α∞opportunities␈α
 to␈α
observe␈α∞ and␈α
experiment.␈α∞  We␈α
 need␈α
to␈α∞ distinguish␈α
what␈α∞can␈α
 be
␈↓ ↓H␈↓␈↓ εH␈↓ *25


␈↓ ↓H␈↓determined␈α⊃ about␈α⊃the␈α⊃ structure␈α⊃of␈α⊃ humans␈α⊃and␈α⊃machines␈α⊃ by␈α⊃ scientific␈α⊃  research␈α⊃ over␈α⊃ a
␈↓ ↓H␈↓period␈α
 of␈α
 time␈α
  and␈α
experimenting␈α
with␈α
many␈αindividuals␈α
from␈α
 what␈α
can␈α
be␈α
learned␈α
by␈α
in␈αa
␈↓ ↓H␈↓particular␈α∂situation␈α∂with␈α⊂particular␈α∂opportunities␈α∂to␈α⊂observe.␈α∂  From␈α∂the␈α⊂AI␈α∂point␈α∂of␈α⊂view,␈α∂the
␈↓ ↓H␈↓latter␈α∞is␈α∞as␈α∞important␈α∞as␈α∞the␈α∞ former,␈α∂and␈α∞we␈α∞suppose␈α∞that␈α∞ philosophers␈α∞would␈α∞also␈α∂consider␈α∞it
␈↓ ↓H␈↓part␈α∞of␈α∞epistemology.␈α∞ The␈α∞possibilities␈α∞of␈α∞reductionism␈α∞ are␈α∞also␈α∞different␈α∞ for␈α∞ theoretical␈α∞ and
␈↓ ↓H␈↓everyday␈α⊂epistemology.␈α∂   We␈α⊂ could␈α∂imagine␈α⊂that␈α∂the␈α⊂rules␈α∂of␈α⊂everyday␈α∂epistemology␈α⊂could␈α∂be
␈↓ ↓H␈↓deduced␈α
from␈α
a␈α
knowledge␈α
of␈α
physics␈α
and␈α
the␈αstructure␈α
of␈α
the␈α
 being␈α
and␈α
the␈α
world,␈α
but␈α we␈α
 can't
␈↓ ↓H␈↓see␈α how␈α one␈αcould␈α avoid␈α using␈α mental␈αconcepts␈α in␈αexpressing␈αknowledge␈αactually␈αobtained␈αby
␈↓ ↓H␈↓the senses.

␈↓ ↓H␈↓␈↓ αh10.3.␈α
 It␈α
 is␈α
 now␈α
 accepted␈α
that␈αthe␈α
basic␈α
concepts␈α
of␈α
physical␈α
theories␈α
are␈αfar␈α
removed
␈↓ ↓H␈↓from␈αobservation.␈α
 The␈α human␈α
 sense␈α organs␈αare␈α
 many␈α levels␈α
 of␈α organization␈α removed␈α
 from
␈↓ ↓H␈↓quantum␈α∪mechanical␈α∪states,␈α∪and␈α∀we␈α∪have␈α∪learned␈α∪to␈α∀accept␈α∪the␈α∪complication␈α∪this␈α∀causes␈α∪in
␈↓ ↓H␈↓verifying␈α
physical␈α∞theories.␈α
Experience␈α
in␈α∞trying␈α
to␈α
make␈α∞intelligent␈α
computer␈α∞programs␈α
suggests
␈↓ ↓H␈↓that␈αthe␈α basic␈α concepts␈α of␈α the␈α common␈αsense␈α world␈α are␈α also␈αcomplex␈αand␈αnot␈αalways␈αdirectly
␈↓ ↓H␈↓accessible␈α
to␈α
observation.␈α
  In␈α∞ particular,␈α
 the␈α
 common␈α
 sense␈α∞ world␈α
 is␈α
 not␈α
 a␈α∞construct␈α
 from
␈↓ ↓H␈↓sense␈α data,␈α but␈αsense␈α
data␈αplay␈αan␈αimportant␈α
role.␈α When␈αa␈αman␈α
or␈αa␈αcomputer␈αprogram␈α
sees␈αa
␈↓ ↓H␈↓dog,␈α⊃we␈α⊃will␈α⊃ need␈α⊃ both␈α⊃ the␈α⊃relation␈α⊃ between␈α⊃ the␈α⊃ observer␈α⊃and␈α⊃the␈α⊃dog␈α⊃and␈α⊃the␈α⊂relation
␈↓ ↓H␈↓between the observer and the brown patch in order to construct a good  theory of the event.

␈↓ ↓H␈↓␈↓ αh10.4.␈α⊂ In␈α⊃spirit␈α⊂this␈α⊂paper␈α⊃is␈α⊂ materialist,␈α⊃but␈α⊂it␈α⊂is␈α⊃logically␈α⊂compatible␈α⊃ with␈α⊂some
␈↓ ↓H␈↓other␈α≠philosophies.␈α≠  Thus␈α~ cellular␈α≠automaton␈α≠models␈α≠of␈α~the␈α≠ physical␈α≠world␈α≠may␈α~be
␈↓ ↓H␈↓supplemented␈α∀by␈α∀supposing␈α∀ that␈α∃certain␈α∀ complex␈α∀ configurations␈α∀ interact␈α∃with␈α∀ additional
␈↓ ↓H␈↓automata␈α
called␈α
  souls␈α
  that␈α
 also␈α
  interact␈α
  with␈α
 each␈α
  other.␈α
    Such␈α
␈↓↓interactionist␈α
 dualism␈↓
␈↓ ↓H␈↓won't␈α⊂ meet␈α∂  emotional␈α⊂ or␈α∂  spiritual␈α⊂objections␈α∂to␈α⊂ materialism,␈α∂but␈α⊂it␈α∂does␈α⊂provide␈α⊂a␈α∂logical
␈↓ ↓H␈↓niche␈α
for␈α
any␈α
empirically␈α
 argued␈α
 belief␈α∞in␈α
 telepathy,␈α
communication␈α
with␈α
the␈α
dead,␈α∞and␈α
 such
␈↓ ↓H␈↓other␈α
 psychic␈α∞phenomena␈α
as␈α
don't␈α∞require␈α
tampering␈α
with␈α∞causality.␈α
 (As␈α
does␈α∞precognition,␈α
for
␈↓ ↓H␈↓example).␈α⊂ A␈α⊂person␈α⊂who␈α⊂believed␈α⊂the␈α⊂alleged␈α⊂evidence␈α⊂for␈α⊂such␈α⊂phenomena␈α⊂and␈α⊂still␈α∂wanted
␈↓ ↓H␈↓scientific explanations could model his beliefs with auxiliary automata.
␈↓ ↓H␈↓␈↓ εH␈↓ *26


␈↓ ↓H␈↓αREFERENCES

␈↓ ↓H␈↓␈↓αArmstrong,␈αD.M.␈↓␈α
(1968),␈α␈↓↓A␈αMaterialist␈α
Theory␈αof␈α
the␈αMind␈↓,␈αRoutledge␈α
and␈αKegan␈α
Paul,␈αLondon
␈↓ ↓H␈↓and New York.

␈↓ ↓H␈↓␈↓αBoden, Margaret A.␈↓ (1972), ␈↓↓Purposive Explanation in Psychology␈↓, Harvard University Press.

␈↓ ↓H␈↓␈↓αCarnap, Rudolf␈↓ (1956), ␈↓↓Meaning and Necessity␈↓, University of Chicago Press.

␈↓ ↓H␈↓␈↓αDavidson,␈α
Donald␈↓␈α
(1973)␈α
The␈α
Material␈α
Mind.␈α␈↓↓Logic,␈α
Methodology␈α
and␈α
Philosophy␈α
of␈α
Science␈αIV␈↓,
␈↓ ↓H␈↓P. Suppes, L. Henkin, C. Moisil, and A. Joja (eds.), Amsterdam, North-Holland.

␈↓ ↓H␈↓␈↓αDennett, D.C.␈↓ (1971) Intentional Systems.  ␈↓↓Journal of Philosophy␈↓ vol. 68, No. 4, Feb. 25.

␈↓ ↓H␈↓␈↓αGosper,␈αR.W.␈↓␈α(1976)␈αPrivate␈αCommunication.␈α (Much␈αinformation␈αabout␈αLife␈αhas␈αbeen␈αprinted␈αin
␈↓ ↓H␈↓Martin Gardner's column in ␈↓↓Scientific American␈↓, and there is a magazine called ␈↓↓Lifeline␈↓).

␈↓ ↓H␈↓␈↓αLewis, David␈↓ (1973), ␈↓↓Counterfactuals␈↓, Harvard University Press.

␈↓ ↓H␈↓␈↓αMcCarthy,␈α⊃John␈↓␈α⊃(1959)␈α⊃Programs␈α∩with␈α⊃Common␈α⊃Sense.␈α⊃␈↓↓Mechanisation␈α⊃of␈α∩Thought␈α⊃Processes,
␈↓ ↓H␈↓↓Volume I␈↓.  London:HMSO.

␈↓ ↓H␈↓␈↓αMcCarthy,␈α∪J.␈α∩and␈α∪Hayes,␈α∩P.J.␈↓␈α∪(1969)␈α∪Some␈α∩Philosophical␈α∪Problems␈α∩from␈α∪the␈α∪Standpoint␈α∩of
␈↓ ↓H␈↓Artificial␈α∩Intelligence.␈α∩␈↓↓Machine␈α∩Intelligence␈α∩4␈↓,␈α∩pp.␈α⊃463-502␈α∩(eds.␈α∩Meltzer,␈α∩B.␈α∩and␈α∩Michie,␈α⊃D.).
␈↓ ↓H␈↓Edinburgh: Edinburgh University Press.

␈↓ ↓H␈↓␈↓αMcCarthy,␈α↔John␈↓␈α↔(1977a),␈α↔␈↓↓First␈α↔Order␈α⊗Theories␈α↔of␈α↔Individual␈α↔Concepts␈↓,␈α↔Stanford␈α⊗Artificial
␈↓ ↓H␈↓Intelligence Laboratory, (to be published).

␈↓ ↓H␈↓␈↓αMcCarthy,␈αJohn␈↓␈α(1977b),␈α␈↓↓Circumscription␈α
-␈αA␈αWay␈αof␈α
Jumping␈αto␈αConclusions␈↓,␈αStanford␈α
Artificial
␈↓ ↓H␈↓Intelligence Laboratory, (to be published).

␈↓ ↓H␈↓␈↓αMontague,␈α⊃Richard␈↓␈α⊂(1963),␈α⊃Syntactical␈α⊂Treatments␈α⊃of␈α⊂Modality,␈α⊃with␈α⊂Corollaries␈α⊃on␈α⊂Reflexion
␈↓ ↓H␈↓Principles and Finite Axiomatizability, ␈↓↓Acta Philosophica Fennica␈↓ ␈↓α16␈↓:153-167.

␈↓ ↓H␈↓␈↓αMoore,␈α↔E.F.␈↓␈α_(1956),␈α↔Gedanken␈α↔Experiments␈α_with␈α↔Sequential␈α↔Machines.␈α_ ␈↓↓Automata␈α↔Studies␈↓.
␈↓ ↓H␈↓Princeton University Press.

␈↓ ↓H␈↓␈↓αPutnam,␈αHilary␈↓␈α
(1961)␈αMinds␈αand␈α
Machines,␈αin␈α
␈↓↓Dimensions␈αof␈αMind␈↓,␈α
Sidney␈αHook␈α
(ed.),␈αCollier
␈↓ ↓H␈↓Books, New York.

␈↓ ↓H␈↓␈↓αPutnam,␈α∩Hilary␈↓␈α∪(1970),␈α∩On␈α∪Properties,␈α∩in␈α∪␈↓↓Essays␈α∩in␈α∩Honor␈α∪of␈α∩Carl␈α∪G.␈α∩ Hempel␈↓,␈α∪D.␈α∩Reidel
␈↓ ↓H␈↓Publishing Co., Dordrecht, Holland.

␈↓ ↓H␈↓␈↓αRyle, Gilbert␈↓ (1949), ␈↓↓The Concept of Mind␈↓, Hutchinson and Company, London.

␈↓ ↓H␈↓John McCarthy
␈↓ ↓H␈↓Artificial Intelligence Laboratory
␈↓ ↓H␈↓Stanford University
␈↓ ↓H␈↓␈↓ εH␈↓ *27


␈↓ ↓H␈↓Stanford, California 94305