Embodied artificial intelligence

gudgeonmaniacalIA et Robotique

23 févr. 2014 (il y a 3 années et 1 mois)

143 vue(s)

A
r
tifi
cial
Intelligence
149
(2003)
131–150
w
w
w
.
e
ls
e
v
ier
.
com
/locate/ar
t
i
nt
Embodied
artificial
intelligence
Ron
C
hrisle
y
Sc
hool
of
Computer
Science
,
Univer
s
ity
of
B
i
r
m
ingham,
B
ir
mingham,
U
K
1.
I
n
tr
od
u
c
ti
on
Mi
k
e
A
nders
on
1
has
g
i
v
en
us
a
t
hought
ful
a
nd
us
eful

e
l
d
gui
de:
N
ot
i
n
t
h
e
g
enre
of
a
bird-w
atcher’
s
guide
which
i
s
carried
in
th
e

eld
a
n
d
wh
ich
c
o
n
t
ain
s
d
e
tailed
d
escr
ip
tio
n
s
of
pos
s
i
bl
e
s
i
ght
i
ngs
,
b
ut
i
n
t
h
e
s
ens
e
of
a
gui
de
to
a

eld
(
in
this
cas
e
e
mbodied
cognition)
which
a
ims
t
o
i
dentify
t
hat

eld’
s
g
eneral
principles
and
p
roperties
.
I’
d
lik
e
t
o
m
ak
e
s
o
me
comment
s
t
hat
w
i
l
l
hopeful
l
y
compl
e
ment
A
nders
on’
s
w
ork,
hi
ghl
i
ght
i
n
g
poi
nt
s
o
f
agreement
a
nd
dis
a
greement
b
etween
his
v
ie
w
o
f
t
he
field
and
m
y
o
wn,
a
nd
acting
a
s
a
de
vil’
s
a
dv
ocate
i
n
p
laces
where
f
urther
dis
c
us
s
i
on
s
eems
t
o
b
e
r
equired.
G
i
v
e
n
t
he
v
e
nue
for
t
hi
s
gui
de,
w
e
can
s
a
fel
y
res
t
ri
ct
t
h
e
d
i
s
cus
s
i
on
t
o
embodi
ed
a
r
tifi
cia
l
in
tellig
en
ce
(EA
I),
e
v
en
i
f
s
u
ch
w
o
rk
dra
w
s
o
n
not
i
ons
of
embodi
ed
co
g
n
itio
n
from
t
he

e
l
d
s
o
f
phi
l
o
s
ophy
,
p
s
y
chol
ogy
and
l
i
ngui
s
t
i
c
s
.
In
part
i
c
ul
ar
,
I

l
l
r
es
t
r
i
c
t
m
y
di
s
c
us
s
i
on
t
o
t
h
e
i
mpact
t
h
at
embodi
ment
can
ha
v
e
on
t
h
e
t
as
k
o
f
c
reat
i
n
g
a
rt
i

ci
al
intelligent
a
gents
,
either
as
technological
ends
in
thems
e
lv
es
,
o
r
a
s
m
eans
t
o
unders
tanding
n
a
tu
r
a
l
i
n
t
ellig
en
t
s
y
s
tem
s
,
o
r
b
o
t
h
.
2.
I
f
I
s
ai
d
y
ou
r
r
ob
ot
h
a
d
a
b
o
d
y
w
o
u
l
d
y
ou
h
o
l
d
i
t
agai
n
s
t
m
e?
B
e
fore
mo
ving
on
to
the
d
etails
of
the
v
ariety
of
approaches
to
embodied
cognition,
t
r
aci
ng
i
t
s
hi
s
t
ori
cal
and
phi
l
o
s
ophi
cal
root
s
,
or
e
v
al
uat
i
n
g
t
he
ar
gument
s
for
a
nd
agai
ns
t
i
t
,
A
nders
on
ri
ght
l
y
deci
des
t
o
c
ons
i
d
er
e
x
act
l
y
w
h
at
i
s
meant
b
y
e
mbodi
ment
i
n
t
h
e
c
ont
e
x
t
of
A
I
.
M
uch
o
f
t
hi
s
i
s
done
i
n
a
n
e
g
at
i
v
e
m
anner
,
by
i
d
ent
i
f
yi
ng
t
h
e
c
ruci
al
pl
anks
i
n
s
y
mbolic
or
good
old-f
a
s
h
ioned
a
rtificial
intelligence
(GOF
AI,
[
25]),
and
pointing
out
ho
w
an
embodi
ed
approach
di
f
f
ers
.
Thi
s
ki
nd
of
charact
eri
s
at
i
o
n
o
f
E
A
I
i
s
,
t
o
a
l
a
r
g
e
e
xt
ent
,
jus
tified,
s
ince
much
of
the
w
ork
i
n
E
AI
has
i
dentified
its
elf
i
n
oppos
ition
t
o
a
GOF
AI
tar
g
et,
r
eal
or
imagined.
B
ut
it
does
h
a
v
e
t
he
s
t
andard
problem
of
definition-in-oppos
ition:
E
-
mail
addr
es
s
:
rlc@cs
.
bham
.
ac.
uk
(R.
C
hris
le
y).
UR
L:
http://w
w
w
.
c
s
.
bham
.
ac.
uk/~
r
lc
(
R
.
C
hr
is
le
y)
.
1
“E
m
bodied
Cognition:
A

eld
guide”,
this
is
s
u
e
[
4].
0004-3702/$

s
e
e
front
m
a
tter

2003
E
l
s
e
vier
B.
V
.
All
r
ights
r
es
erv
e
d.
doi:10.
1016/S0004-3702(03)00055-9
132
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
s
p
ending
v
a
luable
s
p
ace
on
telling
u
s
a
lot
a
bout
what
is
probably
a
lready
f
amiliar
(s
ymbol
s
,
kno
w
l
edge
bas
e
s
,
t
h
e
s
ens
e
-model
-
pl
an-act
c
y
cl
e,
model
i
n
g
t
he
w
o
rl
d,
cont
e
x
t
-
free
r
epres
e
nt
at
i
ons
,
c
ogni
t
i
vi
s
m
,
e
t
c
.),
s
pace
w
h
i
c
h
m
i
ght
ha
v
e
been
bet
t
e
r
s
pent
goi
ng
i
n
t
o
more
det
a
i
l
on
t
h
e
n
o
v
el
concept
s
t
h
at
are
d
efi
ni
t
i
v
e
of
EA
I
(
groundi
ng,
s
e
l
ect
i
v
i
t
y
,
s
i
t
u
at
ednes
s
,
act
i
v
i
t
y
,
a
nd
t
h
e
not
i
o
n
o
f
t
he
body
i
t
s
el
f).
F
o
r
e
xampl
e
,
d
i
f
ferent
approaches
t
o
EA
I
can
be
di
s
t
i
ngui
s
h
ed
by
t
h
e
not
i
o
n
o
f

body”
the
y
emplo
y
.
I
n
[
13]
a
d
is
tinction
i
s
m
ade
b
etween
four
ideas
of
what
is
required
f
or
a
s
y
s
t
em
t
o
be
embodi
ed,
from
t
he
great
er
t
o
t
h
e
l
es
s
i
ncl
u
s
i
v
e
:

P
hys
i
cal
real
i
s
at
i
on:
The
s
ys
t
e
m
m
us
t
m
erel
y
b
e
r
eal
i
s
ed
i
n
s
o
me
phys
i
cal
s
ubs
t
r
at
e
o
r
ot
her
.

P
hys
i
cal
embodi
ment
:
T
he
s
y
s
t
em
mus
t
be
real
i
s
ed
i
n
a
c
oherent
,
i
nt
e
g
ral
phys
i
cal
s
t
ructure.

O
r
gani
s
m
oi
d
e
mbodi
ment
:
T
he
phys
i
cal
real
i
s
at
i
o
n
o
f
t
he
s
y
s
t
em
mus
t
s
h
are
s
ome
(pos
s
i
bl
y
s
uperfi
ci
al
)
c
haract
eri
s
t
i
c
s
w
i
t
h
t
h
e
bodi
es
of
nat
u
ral
o
r
g
ani
s
ms
,
b
ut
need
not
be
al
i
v
e
i
n
a
n
y
s
e
ns
e.

O
r
gani
s
m
al
embodi
ment
:
T
he
phys
i
cal
real
i
s
at
i
o
n
o
f
t
he
s
y
s
t
em
mus
t
not
onl
y
b
e
or
gani
s
m
-l
i
k
e,
b
u
t
act
ual
l
y
or
gani
c
a
nd
al
i
v
e.
Fu
r
t
h
e
r
d
istin
ctio
n
s
can
b
e
m
a
d
e
o
n
th
e
b
asis
o
f
th
e
w
ay
th
at
a
p
ar
ticu
l
ar
n
o
tio
n
o
f
embodiment
is
emplo
y
ed:
I
s
i
t
c
laimed
that
embodiment
of
a
p
articular
kind
is
neces
s
a
ry
f
o
r
A
I
?
I
s
it
m
e
r
e
ly
n
ecessar
y
f
o
r
s
o
m
e
s
en
so
r
i
m
o
to
r
cap
acities,
o
r
is
it
r
e
q
u
i
r
e
d
f
o
r
in
tellig
en
ce
in
g
e
n
e
r
a
l?
Or
is
it
an
ap
p
r
o
ach
th
at
is
m
e
r
e
ly
pr
efer
r
e
d
,
not
requi
red,
for
t
h
e
e
xpl
anat
ory
o
r
t
echnol
ogi
cal
purpos
es
of
a
g
i
v
en
res
earch
group?
A
g
eneral
s
u
rv
e
y
of
th
ese
p
o
s
itio
n
s
with
in
EAI
w
o
u
l
d
b
e
u
sef
u
l.
Th
e
p
o
i
n
t
o
f
th
e
c
h
eap
g
a
g
i
n
t
h
e
title
o
f
th
is
sectio
n
?
On
e
can

t
b
e
g
i
n
t
o

g
u
r
e
o
u
t
h
o
w
to
react
to
claims
that
one’
s
AI
s
y
s
t
em
is
or
is
not
embodied
until
it
is
made
clear
what
not
i
o
n
o
f

body”
i
s
bei
n
g
e
mpl
o
yed
b
y
t
hos
e
c
l
a
i
m
s
.
3
.
The
t
r
o
uble
w
it
h
G
O
F
AI,
a
nd
t
h
e
e
mbo
died
so
lut
i
o
n
Per
h
ap
s
t
h
e
r
easo
n
wh
y
A
n
d
e
r
s
o
n
sp
en
d
s
little
o
r
n
o
tim
e
d
iscu
ssin
g
d
i
f
f
er
en
t
n
o
tio
n
s
of
t
h
e
body
2
i
s
t
h
at
ha
vi
ng
a
body
i
s
not
,
d
es
pi
t
e
t
h
e
n
ame
o
f
t
he

e
l
d
,
cent
r
al
t
o
t
h
e
E
A
I
w
o
rk
he
re
vie
w
s
.
T
o
s
e
e
w
hy
,
l
et’
s
look
at
ho
w
A
nders
on
characteris
es
the

eld
o
f
E
AI
in
oppos
ition
t
o
GOF
AI.
T
o
d
o
s
o,
Anders
on
mus
t
firs
t
characteris
e
GOF
AI,
w
hich
he
does
i
n
th
e
f
o
llo
win
g
ter
m
s:
(1)
A
n
e
mphas
i
s
o
n
e
xpl
i
c
i
t
,
s
e
nt
ent
i
a
l
r
epres
e
nt
at
i
on.
2
In
his
S
ection
3
(“E
m
bodim
e
nt
and
G
rounding”)
A
nders
on
does
d
is
cus
s
four
as
pects
of
our
bodies
(phys
iology
,
e
v
o
lutionary
his
t
ory
,
[in
v
olv
e
m
e
nt
in]
p
ractical
acti
v
ity
,
a
nd
s
o
cio-cultural
s
ituatednes
s
)
,
b
ut
that
dis
c
us
s
i
on
as
s
u
m
e
s
a
n
i
m
p
licit
notion
o
f,
rather
than
dis
c
us
s
e
s
,
what
a
body
actually
is
.
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
133
(2)
A
n
e
mphas
i
s
o
n
operat
i
ons
on
repres
ent
a
t
i
ons
(rul
e
s
)
,
w
hi
ch
are
a
l
s
o
e
xpl
i
c
i
t
and
s
e
nt
ent
i
a
l
,
and
w
hi
ch
operat
e
on
repres
ent
a
t
i
ons
by
vi
rt
ue
of
t
h
e
l
at
t
e
r’
s
f
orm
o
r
s
ynt
ax,
not
their
m
eaning
o
r
s
emantics
.
(3)
A
s
e
ns
e-model-plan-act
(SMP
A)
c
y
cle.
(4)
A
n
e
mphas
i
s
o
n
h
i
gh-l
e
v
e
l
c
ompet
e
nces
s
u
ch
as
t
hought
,
r
eas
on,
pl
anni
ng
and
probl
em-s
ol
vi
ng.
(5)
A
l
ack
of
concern
w
i
t
h
ho
w
t
hes
e
hi
gher
c
ompet
e
nces
ari
s
e
out
of
more
pri
m
i
t
i
v
e
competences
.
Although
Anders
on
gi
v
e
s
a
tentati
v
e
d
efinition
o
f
GOF
AI
in
terms
o
f
C
artes
i
anis
m
a
nd
Co
g
n
iti
v
i
sm
,
i
t
i
s
b
etter
,
as
we
sh
all
s
ee,
to
see
GOF
AI
as
in
v
o
lv
in
g
a
ll
o
f
th
e
a
b
o
v
e
str
a
n
d
s
t
h
at
A
nders
on
ment
i
ons
(e.g.,
t
h
e

C
+
C”
d
e
fin
itio
n
l
ea
v
e
s
m
y
s
ter
i
o
u
s
th
e
r
ele
v
an
ce
o
f
an
attack
on
SMP
A
to
an
analys
is
of
GOF
AI).
A
d
is
adv
a
ntage
o
f
c
haracteris
ing
E
AI
in
oppos
ition
t
o
GOF
AI
mak
e
s
its
elf
m
anifes
t
here.
T
hi
s
k
i
n
d
o
f
l
i
s
t
,
neces
s
a
ry
for
s
uch
a
n
a
pproach,
pol
ari
s
es
mos
t
readers
i
nt
o
a
n
unproduct
i
v
e
s
t
andof
f.
S
o
me
s
e
e
s
uch
a
l
i
s
t
a
nd
t
h
i
nk:
“Y
es
,
t
hat
s
ums
u
p
non-embodi
ed
A
I
,
a
nd
w
e
kno
w
w
hat
p
robl
ems
s
uch
a
pproaches
ha
v
e
.
3
S
o
EA
I
i
s
t
he
onl
y
w
ay
forw
ard”.
O
t
hers
s
e
e
t
he
l
i
s
t
a
s
t
hi
nk
“H
e
y
,
t
hat

s
s
t
r
a
w
-man
A
I
!
T
here’
s
l
o
t
s
of
A
I
w
o
rk
t
h
at
no
one
w
oul
d
cal
l

embodi
ed’
w
hi
ch
rej
ect
s
s
ome,
mos
t
or
al
l
o
f
t
hos
e
poi
nt
s
.
4
S
o
EA
I
i
s
not
needed;

non-embodi
ed’
A
I
i
s
a
l
r
eady
a
ddres
s
i
ng
t
h
e
l
i
m
i
t
a
t
i
ons
of
t
h
e
e
xt
reme
approach
pres
ented
a
bo
v
e”.
I
t
hi
nk
bot
h
o
f
t
hes
e
react
i
ons
t
o
A
nders
on’
s
l
i
s
t
a
re
bas
e
d
o
n
t
rut
h
s
,
b
u
t
t
hei
r
concl
u
s
i
ons
s
houl
d
b
e
a
v
o
i
d
ed.
T
rue,
much
non-embodi
ed
A
I
w
o
rk
does
not
f
a
l
l
under
t
h
e
a
bo
v
e
l
i
s
t
v
ery
w
el
l
,
s
o
EA
I
i
s
not
t
h
e
onl
y
w
ay
t
o
o
v
e
rcome
t
he
probl
ems
o
f
t
he
approach
defi
n
ed
by
t
h
at
l
i
s
t
.
B
u
t
i
t
i
s
a
l
s
o
t
rue
t
hat
E
A
I
s
houl
d
not
be
di
s
m
i
s
s
e
d
j
us
t
becaus
e
s
o
me
of
its
proponents
e
ngage
in
o
v
erzealous
rhetoric
and
unjus
tifi
ably
claim
i
t
i
s
t
h
e
onl
y
w
ay
of
deal
i
n
g
w
i
t
h
t
hos
e
p
robl
ems
.
Y
e
s
,
t
h
ere
m
i
ght
be
a
w
ay
of
o
v
e
rcomi
n
g
t
he
probl
ems
o
f
G
O
F
A
I
w
i
t
h
i
n
a
non-embodi
ed
approach.
B
ut
i
t
mi
ght
be
t
h
at
t
h
ere
i
s
a
w
a
y
of
o
v
e
rcomi
n
g
t
he
probl
ems
o
f
G
O
F
A
I
by
t
a
ki
ng
an
embodi
ed
approach.
L
et
a
t
hous
and

o
w
e
rs
bl
oom.
So
m
e
will
th
in
k
t
h
i
s
t
o
o
co
n
c
iliato
r
y
.
T
h
e
y
w
ill
n
o
t
e
t
h
a
t
i
n
s
o
m
e
s
itu
atio
n
s
we
can
u
s
e
our
po
w
e
rs
of
reas
oni
ng
t
o
s
e
e
i
n
a
dv
ance
t
h
at
a
p
art
i
c
ul
ar
res
earch
s
t
rat
e
gy
i
s
doomed,
a
nd
s
houl
d
t
hus
be
abandoned;
or
promi
s
i
ng,
and
s
houl
d
t
hus
be
encouraged.
H
o
w
e
v
er
,
I
don’
t
t
h
i
n
k
w
e
a
re
i
n
t
h
i
s
ki
nd
of
s
i
t
u
at
i
o
n
w
i
t
h
res
p
ect
t
o
ei
t
h
er
embodi
ed
or
non-embodi
ed
A
I
.
A
t
l
eas
t
not
on
an
y
s
t
r
ai
ght
forw
ard
r
eadi
n
g
o
f
“embodi
ment
”.
3
E
.
g.
,
t
he
problem
s
o
f
dynam
i
cs
and
r
ele
v
ance,
dis
c
us
s
e
d
b
elo
w
.
4
F
o
r
i
ns
tance,
connectionis
t
AI
rejects
m
os
t
o
f
t
he
item
s
on
the
lis
t;
procedural
AI
w
ould
r
eject
1
a
nd
2;
m
a
n
y
reacti
v
e
a
nd
teleo-reacti
v
e
s
ys
tem
s
reject
1,
2
a
nd
3;
m
u
ch
or
m
o
s
t
w
o
rk
on
A
I
vis
i
on
and
r
obotics
r
ejects
2
a
nd
4
(for
a
n
e
xam
p
le
of
a
GOF
AI
robot
(Freddy)
which
r
ejects
3
,
s
ee
[3]);
and
y
et
none
of
thes
e
a
pproaches
need
be
particularly
m
o
re
em
bodied
than
CYC
[
33]
is
,
i
n
a
n
y
s
ubs
tanti
v
e
s
ens
e
of
the
w
ord
“em
bodied”.
134
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
T
o
s
upport
t
hi
s
poi
nt
,
5
I’
ll
look
at
tw
o
a
rchitectural
features
that
are
purportedly
problems
w
ith
GOF
AI
which
d
emand
a
n
e
mbodied
approach:
F
ormality
,
a
nd
the
S
ens
e
-
Model
-
P
l
an-Act
frame
w
o
rk.
3
.
1
.
A
m
er
e
f
o
r
ma
lity
Anders
on
is
right:
E
AI
proponents
o
ften
criticis
e
non-embodied
AI
for
t
aking
t
he
vie
w
6
t
h
at
“j
us
t
a
s
i
s
t
he
cas
e
i
n
m
odern
l
ogi
c,
i
t
i
s
t
h
e
fo
rm
of
t
h
e
s
ymbol
...
and
not
its
meaning
th
at
is
th
e
b
asis
o
f
its
r
u
le-
b
ased
tr
an
sf
o
r
m
a
tio
n

(
[
4
,
S
ectio
n
1
]
,
o
r
ig
in
al
em
p
h
a
sis
7
).
B
u
t
i
t

s
not
cl
ear
t
h
at
EA
I
p
roponent
s
can
us
e
t
hi
s
a
s
a
s
t
i
c
k
w
i
t
h
w
h
i
c
h
t
o
b
eat
G
O
F
A
I
.
B
ri
an
C
a
nt
w
e
l
l
S
m
i
t
h

s
anal
ys
i
s
of
w
h
at
“formal
s
ymbol
mani
pul
at
i
on”
mi
ght
mean
concl
udes
t
hat
o
n
a
n
y
readi
n
g
o
f
t
hat
t
erm
t
hat
act
ual
l
y
hol
ds
t
r
ue
of
t
h
e
s
ys
t
e
ms
t
h
at
w
e
ha
v
e
t
a
k
e
n
t
o
p
aradi
g
mat
i
cal
l
y
f
a
l
l
under
i
t
,
“formal

cannot
mean
“mani
pul
at
i
o
n
i
ndependent
of
s
e
mant
i
c
s

:
F
a
r
from
lining
u
p
o
n
t
op
of
each
other
,
real-w
orld
computer
s
y
s
t
ems

phys
ical
[i.e.,
s
ynt
act
i
c

R
C
]
and
s
emant
i
c
boundari
es
cr
os
s
-
cut
,
i
n
r
i
c
h
a
nd
product
i
v
e
i
n
t
e
rpl
a
y
.
It
is
n
o
t
ju
st
th
at
co
m
p
u
t
er
s
a
r
e
in
v
o
lv
ed
in
an
en
g
a
g
e
d
,
p
a
r
ticip
ato
r
y
w
ay
with
e
x
t
e
r
nal
s
ubject
matters
,
i
n
o
ther
w
o
rds
,
as
s
ugges
t
ed
by
s
o
me
recent

s
ituated”
t
heoris
ts
.
T
he
y
are
p
art
i
c
i
p
at
ori
l
y
engaged
i
n
t
he
w
o
rl
d
as
a
w
hol
e

i
n
a
w
o
rl
d
t
hat
i
ndi
s
c
ri
mi
nat
e
l
y
includes
t
hems
elv
e
s
,
their
o
wn
internal
s
t
ates
and
p
roces
s
e
s
[
43,
p.
37].
H
e
goes
o
n
t
o
s
ay
t
h
at
t
h
i
s
i
n
t
e
rdependence
o
f
t
he
s
ynt
act
i
c
and
t
he
s
e
mant
i
c
i
s
“not
o
n
l
y
a
r
c
h
itectu
r
ally
essen
tial,
b
u
t
i
s
a
lso
c
r
itical,
w
h
e
n
t
h
e
tim
e
c
o
m
es,
i
n
e
stab
lish
i
n
g
and
g
rounding
a
s
ys
tem’
s
i
ntentional
capacities

.
W
hat’
s
i
mportant
here
is
that
Smith
is
s
p
eaking
o
f
r
eal-w
orld
computation
i
n
g
eneral;
i
nas
m
uch
a
s
a
rch-s
y
mbolic
AI
s
y
s
t
ems
ha
v
e
intentional
capacities
,
inas
much
as
the
y
can
be
unders
tood
as
ha
ving
s
e
mantics
o
r
e
v
en
s
yntax
(for
t
he
latter
s
urely
r
equires
t
he
former),
inas
much
as
the
y
are
c
omputers
a
t
all;
th
e
y
ar
e
ip
sis
f
a
c
tis
not
“formal

,
i
f

formal

i
mpl
i
e
s
a
n
i
ndependence
o
f
s
ynt
ax
and
s
e
mant
i
c
s
.
S
o
i
f
proces
s
i
ng
s
y
mbol
s
i
n
a
w
a
y
w
hi
ch
depends
on
t
h
ei
r
s
emant
i
c
s
a
s
w
el
l
a
s
t
h
ei
r
s
ynt
ax
i
s
a
h
al
l
m
ark
o
f
e
mbodi
ment
,
t
hen
a
l
l
real
-w
orl
d
comput
i
ng,
i
n
cl
udi
ng
mos
t
i
f
not
al
l
A
I
s
ys
t
e
ms
,
a
re
al
ready
e
mbodi
ed,
i
n
a
v
e
ry
non-t
r
i
v
i
a
l
s
ens
e
.
If
“formal
s
ymbol
mani
pul
at
i
on”
does
n

t
mean
“mani
pul
at
i
o
n
o
f
s
ymbol
s
i
ndependent
of
s
e
mant
i
c
s

,
t
hen
w
hat
does
i
t
mean?
I
t
h
i
n
k
t
he
bes
t
w
a
y
t
o
a
ns
w
e
r
t
hat
ques
t
i
o
n
i
s
t
o
l
ook
at
t
h
e
h
i
s
t
o
ri
cal
cont
e
x
t
.
W
h
en
beha
vi
ouri
s
m
h
el
d
s
w
a
y
,
an
yone
adv
e
rt
i
n
g
t
o
m
ean
in
g
f
u
l
in
ter
n
al
states
in
th
eir
e
x
p
l
an
atio
n
o
f
i
n
t
ellig
en
t
b
eh
a
v
io
u
r
w
a
s
s
u
s
p
ected
o
f
an
5
Actually
,
I
jus
t
m
a
de
four
points
:
(1)
E
AI
has
n

t
been
s
h
o
w
n
t
o
b
e
t
he
bes
t
approach;
(
2)
E
A
I
h
as
n’
t
b
een
s
h
o
w
n
t
o
b
e
doom
ed;
(
3)
non-E
A
I
h
as
n’
t
b
een
s
h
o
w
n
t
o
b
e
t
he
bes
t
approach;
(
4)
non-E
A
I
h
as
n’
t
b
een
s
h
o
w
n
t
o
be
doom
ed.
W
hat
f
ollo
ws
s
upports
point
(4)
a
nd,
to
a
l
es
s
e
r
e
xtent,
point
(1).
6
Incidentally
,
I’m
not
s
u
re
(
pace
[15])
t
hat

Cogniti
vis
m

i
s
a
good
nam
e
for
t
his
v
ie
w;
s
u
rely
a
C
ogniti
vis
t
is
s
o
m
e
one
w
h
o
e
m
phas
i
s
e
s
t
he
r
o
le
of
kno
w
l
edge
in
intelligence
and
m
entality
,
a
nd
this
is
quite
dis
tinct
f
r
o
m
a
p
er
s
o
n
(
a
F
or
m
a
lis
t?
Repr
es
entationalis
t?
S
e
ntentialis
t?)
w
ho
tak
e
s

r
e
pr
es
entation,
f
o
r
m
alis
m
,
and
r
ule-
bas
e
d
tr
ans
f
or
m
a
tion”
[
4
,
S
ection
1
]
t
o
b
e
centr
al.
7
I
n
this
paper
,
all
v
e
rbatim
quotations
are
r
eproduced
with
their
o
riginal
e
m
phas
i
s
,
e
x
cept
w
here
indicated
otherwis
e.
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
135
i
m
pl
i
c
i
t
dual
i
s
m,
of
rel
y
i
n
g
o
n
s
ome
k
i
n
d
o
f
ghos
t
i
n
t
he
machi
n
e.
The
c
omput
at
i
onal
i
s
t
i
n
s
i
ght
w
a
s
t
hat
a
machi
n
e
c
oul
d
b
e
a
s
e
mant
i
c
engi
ne
by
vi
rt
ue
of
bei
n
g
a
s
ynt
act
i
c
engi
ne,
and
t
hus
w
i
t
hout
vi
ol
at
i
n
g
a
n
y
pri
n
ci
pl
es
of
nat
u
ral
i
s
m.
The
poi
nt
of
modi
fyi
n
g

s
y
mbol
mani
pul
at
i
on”
w
i
t
h
“formal

w
a
s
t
o
i
ndi
cat
e
t
hat
t
hei
r
s
w
as
not
a
ghos
t
l
y
,
dual
i
s
t
i
c
,
ques
t
i
on-be
ggi
ng,
homuncul
ar
ki
nd
of
repres
ent
a
t
i
o
n
p
roces
s
i
ng,
b
u
t
r
at
her
a
ki
nd
of
manipulation
t
hat
w
as
naturalis
tically
admis
s
i
ble
b
ecaus
e
it
could
b
e
c
haracteris
ed
entirely
i
n
phys
i
cal
,
s
ynt
act
i
c
t
e
rms
.
8
The
poi
nt
w
a
s
n

t
t
h
at
s
e
mant
i
c
propert
i
e
s
p
l
a
yed
n
o
r
ol
e
i
n
proces
s
i
ng;
i
t
w
a
s
r
at
her
t
hat
i
nas
m
uch
a
s
t
he
y
d
i
d
,
t
he
y
d
i
d
s
o
by
vi
rt
ue
of
propert
i
e
s
t
hat
were
not
n
a
tu
r
a
listically
p
r
o
b
l
em
atic,
s
u
c
h
a
s
s
y
n
t
actic
p
r
o
p
e
r
ties,
p
l
ay
in
g
a
r
o
le.
9
S
een
t
h
i
s
w
a
y
,
i
t
i
s
a
m
i
s
t
a
k
e
for
E
A
I
t
o
oppos
e
i
t
s
el
f
t
o
f
ormal
i
s
m
per
s
e
,
s
in
ce
th
at
is
t
a
nt
amount
t
o
oppos
i
n
g
i
t
s
el
f
t
o
n
at
ural
i
s
m,
t
a
nt
amount
t
o
s
a
yi
ng
t
h
at
t
h
ere
i
s
s
omet
hi
ng,
in
addition
t
o
a
nd
irreconcilable
w
ith
phys
ical
properties
,
which
p
lays
a
caus
a
l
r
ole
i
n
t
he
product
i
o
n
o
f
b
eha
v
i
our
.
T
hat
w
oul
d
v
i
o
l
a
t
e
t
h
e
caus
a
l
c
l
o
s
u
re
of
phys
i
c
s
.
N
o
,
i
t
i
s
not
t
h
e
i
d
ea
t
h
at
one
can
ha
v
e
s
e
mant
i
c
proces
s
e
s
b
y
v
i
r
t
u
e
o
f
h
a
v
i
n
g
p
roces
s
e
s
w
hi
ch
operat
e
on
repres
entations
according
t
o
t
heir
form
which
E
AI
s
hould
b
e
t
ak
en
to
be
contes
ting.
W
h
at
is
contes
ted,
rather
,
i
s
GOF
AI’
s
pos
ition
o
n
w
hat
t
hat
f
orm
s
hould
b
e;
the
opponent
is
a
(particular)
fo
rma
lism
,
not
a
(
general
)
fo
rma
lity
.
T
hus
,
t
hi
s
E
A
I
cri
t
i
que
of
G
O
F
A
I
get
s
us
onl
y
a
s
f
ar
as
t
h
e
c
onnect
i
oni
s
t
cri
t
i
que
of
G
O
F
A
I
di
d.
Except
a
t
l
eas
t
t
he
connect
i
oni
s
t
s
had
a
concret
e
propos
al
for
w
hat
s
houl
d
s
uppl
ant
s
ent
e
nt
i
a
l
i
s
m;
t
h
e
E
A
I
al
t
e
rnat
i
v
e
i
s
not
s
o
clear
.
W
hat
is
clear
is
that
oppos
ing
s
ententialis
m
i
s
n
either
neces
s
a
ry
nor
s
u
f

cient
f
or
embodi
ment
,
i
n
a
n
y
s
e
ns
e.
3.2.
Shoot
fir
s
t
,
p
r
o
ve
pr
opos
itions
later
(
if
at
all)
The
b
ul
k
o
f
A
nders
on’
s
d
i
s
cus
s
i
on
of
t
h
e
p
robl
ems
o
f
G
O
F
A
I
w
h
i
c
h
E
A
I
ai
ms
t
o
redres
s
focus
e
s
o
n
t
he
S
e
ns
e-Model
-
P
l
an-A
ct
(S
MP
A
)
frame
w
o
rk.
A
nders
on’
s
s
ummary
of
t
h
e
cr
iticism
i
s
t
h
a
t
S
MP
A
i
s

to
o
e
x
p
e
n
s
i
v
e,
an
d
t
h
e
r
e
f
o
r
e
b
i
o
l
o
g
i
cally
im
p
l
au
sib
l
e
10
” [
4
,
Section
2
]
b
ecaus
e
of
tw
o
p
roblems
:

P
r
obl
em
of
dynami
cs
:
T
he
w
o
rl
d
can
change
aft
e
r
t
he
model
i
s
c
ons
t
r
uct
e
d
y
et
before
t
h
e
p
l
a
n
b
as
ed
on
t
h
at
model
i
s
e
x
ecut
e
d,
pos
s
i
bl
y
r
equi
ri
ng
a
c
hange
of
pl
an.
8
On
this
reading,
“form
a
l”
is
m
o
re
general
t
han,
s
a
y
“ef
fecti
v
e”,
in
that
a
f
orm
a
l
operation
n
eed
only
b
e
phys
ically
(non-oracularly)
pos
s
i
ble;
it
need
not,
e
.
g
.
,
cons
is
t
o
f
a
finite
num
ber
o
f
hum
anly-e
x
ecutable
s
teps
.
I
s
hould
point
out
that
I
a
m
not
of
fering
a
c
om
prehens
i
v
e
defi
n
ition
o
f

form
al”;
no
doubt
it
m
eans
d
if
ferent
things
in
dif
f
erent
c
onte
x
ts
.
I
only
s
eek
to
clarify
w
hat
i
t
m
eans
i
n
t
he
conte
x
t
o
f
t
he
phras
e

form
al
s
y
m
bol
m
a
nipulation”.
9
T
r
ue,
s
om
e
w
ho
em
braced
the
c
om
putationalis
t
a
pproach
m
i
s
unders
tood
thes
e
n
iceties
.
S
o
m
e
thought,
e
.
g
.
,
that
s
i
nce
t
here
w
a
s
a
s
yntactic
s
t
ory
a
bout
ho
w
a
beha
viour
w
a
s
p
roduced
there
c
ould
not
als
o
be
a
s
em
antic
s
t
ory
a
bout
ho
w
t
hat
b
eha
v
iour
(or
a
n
action
w
hich
it
realis
ed)
w
as
produced.
O
thers
t
hought
that
the
d
irectly
natur
a
lis
able
pr
oper
ties
b
y
v
ir
tue
o
f
w
hich
a
s
ys
tem
a
ls
o
h
ad
s
e
m
a
ntic
pr
oper
ties
h
ad
to
be
inter
n
al,
s
yntactic
ones
;
the
y
ignored
the
r
ole
t
hat,
e.
g.
,
e
xternal,
relational
p
roperties
can
play
in
fixing
s
em
antic
properties
.
Ne
v
e
rtheles
s
,
I
b
elie
v
e
the
h
is
tory
I
h
a
v
e
p
res
e
nted
illum
i
nates
t
he
role
that
“form
a
l
s
ym
bol
m
a
nipulation”
and
its
cognate
concepts
w
e
r
e
playing
i
n
t
he
theor
e
tical
dialectic
of
the
50s
,
60s
and
ear
ly
70s
.
10
F
o
r
a
dis
c
us
s
i
on
of
(
t
he
pos
s
i
ble
i
r
r
e
le
v
a
nce
o
f
)
biological
plaus
i
bility
,
s
ee
S
ection
6
.
136
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150

P
r
obl
em
of
rel
e
v
a
nce:
O
n
e
c
oul
d
p
erhaps
s
o
l
v
e
t
he
probl
em
of
dynami
cs
i
f
one
res
t
ri
ct
ed
ones
e
l
f
t
o
reas
oni
ng
about
change
onl
y
w
hen
t
he
change
i
s
“l
i
k
el
y
t
o
a
f
f
ect
th
e
ach
ie
v
a
b
ility
o
f
th
e
g
o
a
l”;
b
u
t
h
o
w
i
s
t
h
i
s
r
estr
ictio
n
t
o
b
e
ach
ie
v
e
d
?
[
4
,
S
ectio
n
2
]
.
The
p
robl
em
of
dynami
cs
can
be
addres
s
e
d
w
i
t
h
i
n
t
h
e
S
MP
A
frame
w
o
rk,
A
nders
on
not
es
,
b
y
e
i
t
h
er
(a)
m
odel
i
n
g
t
he
dynami
cs
,
a
nd
pl
anni
ng
“i
n
t
erms
of
t
h
e
e
xpect
ed
changes
i
n
t
h
e
e
n
v
i
r
onment

,
o
r
(
b)
ha
vi
ng
cont
i
ngenc
y
p
l
a
ns
.
The
c
ri
t
i
que
of
(a)
w
hi
ch
A
nders
on
pres
ent
s
i
s
:

N
a
t
u
ral
l
y
,
t
hi
s
onl
y
pus
hes
t
he
probl
em
back
one
s
t
ep,
f
or
no
w
w
e
h
a
v
e
t
o
m
oni
t
o
r
w
het
h
er
or
not
t
h
e
c
hang
es
are
t
he
e
xpected
ones
,
and
r
e-pl
an
w
h
en
t
h
e
y
are
not

[
4,
S
ect
i
o
n
2
].
Thi
s
i
s
i
ndeed
a
p
robl
em,
b
ut
i
t
does
n

t
feel
l
i
k
e
a
p
aradi
g
m-b
u
s
t
er
,
a
nd
i
t
cert
a
i
n
l
y
i
s
n’
t
t
he
be
gi
nni
ng
of
a
r
e
g
res
s
,
as
A
nders
on
s
eems
t
o
s
ugges
t
.
A
re
gres
s
w
oul
d
t
hreat
en
i
f
A
nders
on
had
s
ai
d
s
omet
hi
ng
l
i
k
e

for
t
he
w
a
y
t
hat
t
he
w
o
rl
d
c
hanges
m
ay
i
t
s
el
f
h
a
v
e
c
hanged
a
ft
er
t
h
e
t
i
m
e
a
t
w
hi
ch
w
e
l
a
s
t
model
e
d
t
h
e
w
orl
d

s
dynami
cs
yet
b
efore
t
he
t
i
m
e
a
t
w
hi
ch
our
pl
an
i
s
e
x
ecut
e
d”.
B
ut
he
di
dn’
t
s
ay
t
h
at
for
good
reas
on:
w
e
don’
t
b
el
i
e
v
e
(
pace
[41]!)
t
h
at
t
h
e
w
orl
d

s
dynami
cs
i
s
i
t
s
el
f
dynami
c.
So
in
ef
fect,
t
he
criticis
m
o
f
s
trate
g
y
(
a)
is
only:
SMP
A
may
g
et
the
dynamics
wrong.
Bu
t
t
h
i
s
i
sn

t
in
itself
d
e
v
astatin
g
;
11
i
t
onl
y
b
ecomes
s
o
i
f
i
t
i
s
p
romot
e
d
t
o
s
omet
hi
ng
l
i
k
e
“SMP
A
w
ill
al
w
a
ys
get
t
he
dynami
cs
w
r
ong”,
o
r
,
equi
v
a
l
e
nt
l
y
,

S
M
P
A
can
ne
ver
get
t
he
dynami
cs
ri
ght
”.
W
h
y
m
i
ght
w
e
t
h
i
n
k
t
hat
i
s
t
he
cas
e?
Co
m
p
le
x
ity
.
I
n
a
r
eal-
w
o
r
ld
situ
atio
n
,
th
er
e
a
r
e
ju
st
to
o
m
an
y
v
ar
iab
l
es
to
b
e
tak
e
n
i
n
t
o
account
,
a
nd
t
h
ei
r
i
nt
eract
i
o
n
y
i
e
l
d
s
a
combi
n
at
ori
a
l
e
xpl
os
i
on.
Thi
s
i
s
al
s
o
w
h
y
(
b)
w
on’
t
w
o
rk
ei
t
h
er:
I
t
w
oul
d
r
equi
re
t
o
o
m
an
y
c
ont
i
ngenc
y
p
l
a
ns
.

B
u
t

,
c
ommon
s
ens
e
i
n
t
e
rj
ect
s
,
“m
o
s
t
o
f
t
h
e
v
a
r
i
ab
ility
o
f
th
e
w
o
r
ld
is
ir
r
e
le
v
a
n
t
to
m
o
st
task
s;
co
u
l
d
n

t
(
a)
an
d
(
b
)
h
a
v
e
a
c
hance
i
f
t
he
y
r
es
t
r
i
c
t
e
d
t
hems
el
v
e
s
onl
y
t
o
t
he
v
a
ri
abl
e
s
a
nd
cont
i
ngenci
e
s
w
hi
ch
mi
ght
af
f
ect
th
e
ach
ie
v
eab
ility
o
f
th
e
g
o
a
l?”
P
er
h
a
p
s
,
b
u
t
n
o
w
t
h
e
p
r
o
b
l
em
o
f
r
e
le
v
a
n
c
e
a
p
p
ear
s:
Ho
w
i
s
s
uch
a
res
t
riction
t
o
b
e
accomplis
hed?
Here
it
does
s
eem
that
a
t
horough-going
S
M
P
A
approach
encount
ers
a
re
gres
s
.
C
ont
rol
o
f
i
nference
i
n
general
,
and
d
et
ermi
nat
i
o
n
o
f
rel
e
v
a
nce
i
n
p
art
i
c
ul
ar
,
can
be
cons
i
d
ered
t
o
be
s
p
eci
es
of
“i
nt
ernal
act
i
on”.
T
herefore,
t
he
probl
ems
o
f
dynami
cs
and
r
el
e
v
ance
t
h
at
ari
s
e
f
or
S
M
P
A
-generat
ed
act
i
o
n
a
s
a
w
hol
e
a
l
s
o
ar
ise
f
o
r
th
e
actio
n
o
f

r
e
le
v
a
n
ce-
d
e
ter
m
in
in
g

wh
ich
w
as
m
ean
t
t
o
s
o
l
v
e
said
p
r
o
b
l
em
s.
12
Th
is
is
a
f
am
iliar
s
itu
atio
n
:
An
an
aly
s
is
o
f
so
m
e
asp
ect
A
o
f
r
a
tio
n
a
lity
is
o
f
f
e
r
e
d
,
an
d
i
t
s
prerequi
s
i
t
e
s
i
dent
i

ed.
I
t
i
s
t
hen
a
r
gued
t
hat
A
can
onl
y
b
e
r
at
i
onal
i
f
i
t
s
prerequi
s
i
t
e
s
are
r
at
i
onal
l
y
at
t
a
i
n
ed.
A
nd
yet
t
hi
s
,
ci
rcul
arl
y
,
r
equi
res
t
he
appl
i
cat
i
o
n
o
f
A
itself
,
wh
ich
requi
res
t
he
prerequi
s
i
t
e
s
f
or
A
,
ad
i
nfini
t
u
m
.
W
e
s
ee
t
h
i
s
ki
nd
of
paradox
i
n
“W
hat
t
he
T
o
r
t
o
i
se
Said
to
Ach
illes”
[
9
]
(
A
=
deduct
i
on),
H
ume’
s
[
29]
and
G
oodman’
s
[21]
ri
ddl
es
(
A
=
induction),
W
ittgens
t
ein’
s
[
47]
pri
v
ate
l
anguage
ar
gument(s
)
(
A
=
rul
e
-fol
l
o
wi
ng),
and
d
ebat
es
on
t
h
e
L
anguage
of
Thought
[17,20]
(
A
=
rul
e
-fol
l
o
wi
ng
agai
n).
11
I
n
f
act,
i
t
w
ould
b
e
a
de
v
a
s
t
ating
b
lo
w
a
gains
t
S
M
P
A
theor
y
as
a
m
odel
o
f
hum
an
cognition
i
f
s
uch
t
heor
y
im
plied
t
hat
S
MP
A
s
ys
tem
s
were
ne
v
e
r
w
rong,
s
i
nce
c
learly
we
hum
ans
o
ften
are!
12
N
o
te
that
it
is
n’
t
t
he
s
e
quential
d
etails
of
S
M
P
A
(
i
.
e
.
,
the
r
elati
v
e
o
r
d
er
of
the
c
ons
tituent
s
teps
)
t
hat
i
s
caus
i
ng
the
p
r
oblem
her
e
,
b
ut
r
a
ther
its
deliber
ati
v
e
n
atur
e.
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
137
There
s
eem
t
o
be
four
w
a
ys
out
of
t
h
i
s
conundrum:
(
1
)
T
h
e
ap
p
licatio
n
o
f
A
to
th
e
p
r
e
r
e
q
u
i
sites
o
f
A
is
a
d
if
f
e
r
e
n
t
task
th
an
th
e
a
p
p
licatio
n
s
of
A
to
e
x
ternal
actions
,
a
nd
therefore
m
ight
not
caus
e
a
r
e
g
res
s
.
In
the
cas
e
o
f
pl
anni
ng,
for
e
xampl
e
,
i
t
i
s
pos
s
i
bl
e
(
t
hough
perhaps
not
pl
aus
i
bl
e)
t
h
at
t
h
e
p
robl
ems
of
dynami
cs
and
r
el
e
v
ance
do
not
ari
s
e
f
or
t
h
e
act
i
o
n
o
f
d
et
ermi
ni
ng
rel
e
v
a
nce
i
t
s
el
f,
o
r
if
th
e
y
d
o
,
n
o
t
in
an
in
tr
actab
le
w
a
y
.
(2)
T
he
prerequi
s
i
t
e
s
o
f
A
may
b
e
j
us
t
i

ed
by
a
r
at
i
onal
p
rocedure
d
i
s
t
i
n
ct
from
A
.I
n
t
h
e
cas
e
o
f
p
l
a
nni
ng,
for
e
xampl
e
,
one
mi
ght
t
r
y
t
o
deduce
or
i
nduce
the
r
ele
v
anc
y
rel
a
t
i
ons
di
rect
l
y
,
r
at
her
t
han
p
l
a
nni
ng
“rel
e
v
ance-det
e
rmi
n
i
ng”
act
i
ons
,
t
hus
pro
v
i
d
i
n
g
a
r
at
i
onal
b
as
i
s
for
p
l
a
nni
ng
w
i
t
hout
re
gres
s
.
Thi
s
s
o
l
u
t
i
o
n
w
i
l
l
not
w
o
rk,
h
o
w
e
v
er
,
i
f,
as
s
eems
lik
ely
,
al
l
m
o
d
e
s
o
f
r
atio
n
a
lity
h
a
v
e
a
r
e
g
r
e
ss
p
r
o
b
l
em
.
(3)
O
ne
could
g
i
v
e
u
p
o
n
f
oundationalis
m
w
ith
res
p
ect
to
rationality
,
a
nd
ins
t
ead
hold
t
hat
A
can
be
rat
i
onal
e
v
e
n
i
f
i
t
i
s
b
as
ed
on
non-rat
i
onal
(
une
xami
ned)
as
s
u
mpt
i
ons
.
I
n
t
he
cas
e
o
f
p
l
a
nni
ng,
a
s
ys
t
e
m
m
i
ght
j
u
s
t
ha
v
e
s
o
me
arbi
t
r
ary
,
unj
us
t
i

ed,
unrefl
ect
ed-on
w
a
y
s
o
f
d
e
ter
m
in
in
g
r
ele
v
an
ce;
th
e
r
atio
n
a
lity
o
f
th
e
s
y
s
tem
i
s
r
atio
n
a
lity
r
e
lati
v
e
to
thos
e
p
re-gi
v
en,

x
e
d
p
arameters
.
13
On
th
is
v
i
e
w
,
r
atio
n
a
lity
is
a
m
atter
o
f
h
o
w
o
n
e
deal
s
w
i
t
h
w
h
at
one
i
s
gi
v
e
n.
14
(4
)
A
n
e
x
t
reme,

n
a
l
o
p
tio
n
i
s
t
o
g
i
v
e
u
p
o
n
r
atio
n
a
lity
alto
g
e
th
er:
A
ccep
t
t
h
a
t
A
i
s
not
str
i
ctly
r
a
tio
n
a
l,
y
e
t
d
en
y
t
h
a
t
s
u
c
h
s
tr
ict
r
atio
n
a
lity
is
a
r
eq
u
i
r
e
m
e
n
t
f
o
r
a
w
o
r
k
in
g
A
I
.
Th
is
o
p
tio
n
w
ill
b
e
attr
acti
v
e
t
o
t
h
o
s
e
A
I
w
o
r
k
e
r
s
wh
o
s
e
p
r
i
m
a
r
y
in
ter
e
st
is
in
b
u
ild
in
g
a
w
orki
ng
s
y
s
t
em,
r
at
her
t
han
i
n
a
dherence
t
o
a
rat
i
onal
i
s
t
i
deol
ogy
.
I
n
t
he
cas
e
o
f
pl
anni
ng,
w
h
at
w
oul
d
j
us
t
i
f
y
t
he
S
M
P
A
frame
w
o
rk
i
s
not
s
o
me
(w
ere
i
t
pos
s
i
bl
e)
a
pr
i
o
r
i
estab
lish
m
en
t
o
f
its
r
a
tio
n
a
lity
,
b
u
t
an
a
pos
t
e
r
i
or
i
estab
lish
m
en
t
o
f
its
u
tility
.
S
o
here
w
e
ha
v
e
four
w
a
ys
i
n
w
h
i
c
h
t
he
S
M
P
A
frame
w
o
rk
mi
ght
be
defended,
and
a
n
embodied
approach
made
unneces
s
a
ry
.
I
f
p
res
s
e
d
t
o
s
peculate
as
to
which
o
f
t
hes
e
four
pos
s
i
bl
e
s
ol
ut
i
ons
are
t
he
mos
t
promi
s
i
ng,
I’
d
h
a
v
e
t
o
p
l
u
mp
for
3
and
4
;
1
and
2
s
eem
im
p
l
au
sib
l
y
o
p
tim
istic
g
i
v
e
n
t
h
e
h
i
sto
r
y
o
f
a
ttem
p
ts
to
f
o
r
m
alise
r
easo
n
.
Y
e
t
t
h
e
r
e
is
an
i
n
t
e
res
t
i
n
g
s
ens
e
i
n
w
h
i
c
h
3
a
nd
4,
as
pat
c
hes
t
o
t
he
S
M
P
A
frame
w
o
rk,
t
hems
el
v
e
s
r
equi
re
a
k
i
n
d
o
f
e
mbodi
ment
.
A
di
s
t
i
n
ct
i
o
n
can
be
made
bet
w
een
t
hos
e
a
s
p
ect
s
o
f
ours
e
l
v
es
w
h
i
c
h
can
be
unders
t
ood
i
n
rat
i
onal
a
nd
concept
u
al
t
e
rms
,
and
t
hos
e
a
s
p
ect
s
w
hi
ch
cannot
.
F
or
e
x
ampl
e,
my
reas
ons
for
c
omi
n
g
t
o
w
ork
t
oday
,
and
m
y
b
el
i
e
f
t
hat
t
oday
i
s
M
onday
,
are
rational/conceptual
as
pects
o
f
m
ys
elf;
my
mas
s
,
v
o
lume,
h
eart
r
ate,
and
t
he
pos
itions
of
my
l
i
m
bs
are
not
.
T
he
t
horough-goi
ng
S
M
P
A
f
a
nt
as
y
i
s
t
hat
c
ompet
e
nt
real
-w
orl
d
act
i
o
n
can
be
achie
v
e
d
b
y
a
s
y
s
t
em
s
o
lely
by
virtue
of
its
conceptual
as
pects
.
B
u
t
i
f
3
or
4
i
s
t
h
e
c
orrect
res
pons
e
t
o
t
he
probl
ems
o
f
dynami
cs
and
r
el
e
v
ance,
a
t
horough-goi
ng
S
M
P
A
frame
w
o
rk
i
s
not
pos
s
i
bl
e:
t
h
e
S
MP
A
s
t
r
at
e
g
y
,
i
n
order
t
o
w
ork
i
n
t
he
real
w
o
rl
d,
mu
st
be
bas
e
d
o
n
a
s
p
ect
s
o
f
t
he
s
y
s
t
em
for
w
hi
ch
no
rat
i
onal
/
c
oncept
u
al
des
c
ri
pt
i
on,
anal
ys
i
s
,
o
r
13
A
n
d
i
s
t
hus
com
p
arable
to
the
notions
of
bounded
o
r
m
inim
al
rationality
put
forw
ard
b
y
S
im
on
[42]
and
Cherniak
[10],
inter
a
lia
.
14
T
hus
,
r
obots
c
ould
d
o
w
ors
e
than
pray
the
S
erenity
Prayer
(which
apparently
dates
b
ack
to
Boethius
),
s
u
itably
m
odifi
e
d:
“Mak
er
,
g
rant
m
e
rationality
to
change
the
t
hings
I
can
change,
e
m
bodim
e
nt
to
ef
fect
the
things
I
cannot,
a
nd
w
i
s
dom
to
kno
w
t
he
dif
f
erence”.
138
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
j
u
s
t
i

cat
i
o
n
i
s
a
v
a
i
l
a
bl
e.
If
a
r
obot
i
s
goi
ng
t
o
be
abl
e
t
o
act
i
n
t
h
e
r
eal
w
o
rl
d,
i
t
w
i
l
l
onl
y
partly
(if
a
t
a
ll;
cf.
s
olution
4
)
b
e
b
ecaus
e
of
the
r
ational
v
irtues
of
its
SMP
A
s
t
rate
gies
;
i
t
will
also
n
ecessar
ily
b
e
a
m
atter
o
f
ju
st
b
e
in
g
b
u
ilt
th
e
r
ig
h
t
wa
y
.
15
Th
at
is,
i
t
w
ill
p
a
r
tly
b
e
a
m
atter
o
f
h
a
v
in
g
a
n
u
n
r
eflecti
v
e
(
an
d
t
h
e
r
e
f
o
r
e
n
o
n
-
r
a
tio
n
a
l)
d
i
sp
o
s
itio
n
t
o
,
e.g
.
,
t
ak
e
cert
a
i
n
propert
i
e
s
a
nd
not
ot
hers
t
o
be
rel
e
v
a
nt
t
o
part
i
c
ul
ar
act
i
ons
.
I
f
w
e
r
econs
t
r
uct
t
h
e
c
l
a
i
m
s
o
f
t
he
embodi
ment
t
h
eori
s
t
s
t
o
b
e
c
l
a
i
m
s
a
bout
t
h
e
l
i
m
i
t
a
t
i
ons
of
t
h
e
pur
el
y
conceptual,
r
ational
s
ubject;
i
f
w
e
t
ak
e
“embodiment”
t
o
m
ean
thos
e
a
s
p
ects
a
nd
abilities
o
f
a
s
ys
t
e
m
w
hi
ch
cannot
be
anal
ys
ed
i
n
purel
y
r
at
i
onal
t
erms
,
b
ut
rat
h
er
mus
t
be
unders
t
ood
cau
sally
o
r
acco
rd
in
g
t
o
s
o
m
e
n
o
r
m
o
th
er
th
an
in
d
i
v
i
d
u
a
l
r
atio
n
a
lity
;
i
f
(
b
u
t
o
n
l
y
i
f)
we
tak
e
“embodi
ment

t
o
m
ean
t
h
e
a
s
p
ect
s
o
f
a
s
y
s
t
em
w
h
i
c
h
a
re
not
j
u
s
t
cogni
t
i
v
e
l
y
i
m
penet
r
abl
e
[39],
b
ut
w
h
i
c
h
a
re
not
t
h
e
out
comes
o
f
p
roces
s
e
s
w
hi
ch
can
be
cons
t
r
ued
a
s
r
at
i
onal
;
i
f
t
h
es
e
c
hanges
t
o
our
t
h
i
nki
ng
are
m
ade,
t
h
en
w
e
w
i
l
l
ha
v
e
reas
on
t
o
bel
i
e
v
e
t
h
at
real
w
o
rl
d
intelligence
mus
t
(in
t
his
s
ens
e
)
b
e
e
mbodied.
N
o
t
e
,
h
o
w
e
v
er
,
t
hat
t
hi
s
f
al
l
s
a
l
ong
w
a
y
s
hort
o
f
e
s
t
abl
i
s
hi
ng
t
h
e
m
ore
e
xt
reme
cl
ai
ms
of
EAI
p
roponents
.
In
particular
,
i
t
a
llo
ws
for
t
he
pos
s
i
bility
of
artificial
intelligence
being
grounded
i
n
a
s
ub-rat
i
onal
o
r
s
ub-concept
u
al
s
ubs
t
r
at
e
i
n
a
manner
qui
t
e
di
f
f
erent
from
t
he
w
a
y
t
hat
n
atural,
o
r
g
anic
intelligence
is
.
O
f
c
ours
e
,
A
I
can
benefit
from
a
n
unders
tanding
of
ho
w
t
he
body
accomplis
hes
t
his
g
rounding
in
the
n
atural
cas
e.
B
u
t
a
s
l
a
v
is
h
c
op
ying
of
nat
u
re
may
b
e
unneces
s
a
ry
and
i
n
s
ome
cas
es
unhel
p
ful
(
as
i
t
w
a
s
i
n
t
he
achi
e
v
e
ment
of
artificial
flight
[5,8,48];
s
ee
Section
6
).
So
th
e
o
n
l
y
t
ar
g
e
t
w
h
i
ch
m
o
st
EAI
-
t
alk
h
as
an
y
c
h
a
n
c
e
o
f
h
ittin
g
i
s
t
h
e
p
u
r
ely
r
atio
n
a
l,
t
horough-goi
ng
S
M
P
A
s
y
s
t
em.
B
ut
t
h
e
not
i
o
n
o
f
a
“purel
y”
rat
i
onal
A
I
s
ys
t
e
m
i
s
a
di
s
t
ract
i
on.
It

s
ques
t
i
onabl
e
w
het
h
er
an
y
i
mpl
e
ment
ed
A
I
s
y
s
t
em
has
e
v
e
r
b
een
purel
y
rat
i
onal
,
or
w
h
et
her
s
uch
w
oul
d
e
v
e
n
b
e
pos
s
i
bl
e.
16
A
phys
i
cal
s
y
s
t
em
i
n
w
h
i
c
h
e
v
e
ry
phys
i
cal
di
f
f
erence
mak
e
s
a
n
i
nt
ent
i
onal
d
i
f
ference
m
ay
be
lo
g
i
ca
lly
pos
s
i
bl
e,
b
u
t
i
t
b
e
g
g
a
r
s
th
e
i
m
a
g
i
n
a
tio
n
,
an
d
s
u
r
ely
h
as
little
to
d
o
with
th
e
b
est
c
o
u
r
se
f
o
r
A
I
r
esear
ch
.
Thi
s
recal
l
s
,
a
nd
perhaps
e
xpl
ai
ns
,
t
he
i
n
f
a
mous
remark
from
D
re
w
M
cD
ermot
t
[35]
w
h
i
c
h
A
nders
on
quot
es
:

no
w
o
r
k
i
n
g
A
I
p
r
o
gr
am
has
e
ver
been
bot
her
e
d
a
t
a
l
l
by
t
h
e
f
r
am
e
p
r
obl
em

[
4,
S
ect
i
o
n
5
].
H
o
w
e
v
e
r
,
I
w
oul
d
r
epl
ace
“at
al
l

w
i
t
h
“i
ns
urmount
abl
y
”.
Thorough-goi
ng
S
M
P
A
i
s
j
u
s
t
as
perv
ers
e
as
t
horough-goi
ng
deduct
i
o
n
(
cf.
t
he
T
o
rt
oi
s
e
an
d
ach
illes
a
g
a
in
)
.
17
15
Of
cours
e
,
h
a
v
ing
a
n
e
f

cient
a
nd
ef
fecti
v
e
p
lanner
i
s
a
ls
o
a
m
a
tter
o
f

being
b
uilt
the
r
ight
w
a
y”,
b
ut
em
bodim
e
nt
is
only
that;
t
here
is
not,
i
n
a
ddition,
a
r
ational
c
haracteris
ation/jus
t
ific
a
t
i
on,
as
there
m
ay
be
for
a
planner’
s
d
ecis
i
ons
.
16
T
h
e
s
ys
tem
w
ould
h
a
v
e
t
o
b
e
s
uch
t
hat
a
ll
phys
ical
proces
s
e
s
a
nd
s
t
ates
w
ould,
at
all
tim
es
,
i
ns
tantiate
s
o
m
e
r
a
tional
p
r
o
ces
s
o
r
s
tate.
M
er
e
h
air
g
r
o
w
t
h
w
ould
b
e
out,
a
s
w
ould
m
er
e
m
as
s
i
ncr
eas
e/decr
eas
e,
m
e
r
e
batter
y
v
o
ltage
drop,
or
tw
o
m
olecules
m
erely
e
xchanging
pos
itions
.
T
hes
e
e
v
ents
could
only
b
e
a
llo
wed
i
n
a
“purely”
rational
s
ys
tem
i
f
s
om
ething
conceptually
or
rationally
norm
a
ti
v
e
s
uperv
ened
on
them
.
17
T
hus
,
non-thorough-going
SMP
A
is
,
i
n
s
om
e
s
ens
e
,
m
ore
r
ational
t
han
t
horough-going
SMP
A
—at
l
eas
t
you’
re
not
guar
anteed
to
get
eaten
by
the
tiger
.
B
ut
that
is
not
the
notion
o
f
r
ationality
I
h
a
v
e
b
een
em
plo
y
ing
in
this
dis
c
us
s
i
on.
A
p
roces
s
can
be
e
x
ter
nally
r
a
tional
(
i
.
e
.
,
is
pr
agm
a
tically
us
ef
ul,
g
i
v
es
the
r
ight
r
e
s
u
lts
f
o
r
s
u
rvi
v
al)
e
v
e
n
t
hough
it
is
not
inter
nally
r
a
tional
(bas
ed
on
a
p
rocedure
w
hich
can
be
unders
tood
to
further
t
he
agent’
s
d
es
ir
es
in
the
light
of
its
belief
s
)
.
T
h
e
i
m
p
licit
claim
w
hich
I
a
m
r
ejecting
i
s
t
hat
S
M
P
A
m
us
t
s
tr
i
v
e
t
o
b
e
thorough-goingly
i
nternally
rational.
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
139
H
a
vi
ng
a
body
i
n
an
y
r
ob
us
t
s
ens
e
,
t
hen,
i
s
not
a
r
equi
rement
for
d
eal
i
n
g
w
i
t
h
t
h
e
problems
o
f
S
MP
A
a
nd
pure
r
ationality
.
B
ut
e
v
en
if
there
i
s
n
o
knock-do
wn
ar
gument
i
n
f
a
v
our
of
EA
I,
an
y
d
i
f
ferent
,
n
o
v
e
l
w
ay
of
res
pondi
ng
t
o
t
h
e
p
l
a
nni
ng
chal
l
e
nge
i
s
w
o
rt
h
cons
i
d
eri
ng.
W
h
at
does
E
A
I
ha
v
e
t
o
of
fer?
The
E
A
I
res
pons
e
i
s
b
es
t
i
dent
i

ed
by
gi
vi
ng
e
x
ampl
es
.
D
on’
t
u
s
e
compl
e
x
g
eomet
r
i
c
reas
oni
ng
t
o
cal
cul
a
t
e
w
h
ere
t
he
cent
r
e
o
f
t
he
corri
dor
i
s

b
y
t
he
t
i
m
e
you

n
i
s
h,
your
pos
ition
w
ill
probably
h
a
v
e
c
hanged,
if
you
are
t
rying
t
o
act
in
the
w
orld
in
a
c
ontinuous
,
real
-t
i
m
e
w
ay
.
I
ns
t
ead,
m
erel
y
a
ppl
y
d
i
f
ferent
i
a
l
l
y
more
t
o
rque
t
o
your
ri
ght
w
h
eel
i
f
t
h
e
dark
s
pot
i
n
front
of
you
i
s
t
o
t
h
e
l
eft
o
f
your
vi
s
u
al

e
l
d
,
a
nd
t
h
e
c
on
v
e
rs
e
i
f
t
he
s
pot
i
s
on
t
h
e
r
i
ght
[28].
O
r
,
i
f
you’
re
a
c
ri
ck
et
:
D
on’
t
e
v
o
l
v
e
o
r
l
earn
c
ompl
e
x
pat
t
e
rn
det
ect
i
o
n
al
gori
t
h
ms
and
u
s
e
s
p
at
i
a
l
p
l
a
nners
/
d
el
i
b
erat
ors
t
o

gure
out
ho
w
t
o
m
o
v
e
t
o
w
ard
your
mat
e

your
mat
e
w
i
l
l
ha
v
e
mo
v
e
d
b
y
t
he
t
i
m
e
you

n
i
s
h
a
l
l
t
h
at
,
a
nd
t
h
e
p
robl
em
of
dynami
cs
s
e
t
s
i
n
.
I
ns
t
ead,
m
erel
y
h
a
v
e
a
n
ear
w
h
i
c
h
i
s
b
ui
l
t
s
o
t
h
at
w
h
en
your
mat
e

s
cal
l
s
ar
e
h
ear
d
,
th
e
s
h
a
p
e
o
f
th
e
ear
r
e
su
lts
in
th
e
p
r
o
p
e
r
s
ig
n
a
ls
b
e
in
g
s
en
t
t
o
t
h
e
le
g
s
to
r
e
su
lt
in
m
a
te-
d
ir
ected
m
o
v
e
m
e
n
t
[
4
6
]
.
Th
ese
d
o
s
eem
lik
e
f
r
e
sh
alter
n
ati
v
es
to
th
e
GOF
AI
ap
p
r
o
ach
.
A
n
d
d
e
sp
ite
th
e
w
ell-
kno
w
n
probl
em
of
ho
w
s
uch
s
ys
t
e
ms
coul
d
s
cal
e
u
p
t
o
h
andl
e
h
i
gher
-
l
e
v
e
l
c
ogni
t
i
v
e
t
a
s
k
s
(A
nders
on
gi
v
e
s
s
ome
good
references
for
bot
h
s
i
d
es
of
t
h
at
debat
e
),
i
t
cert
a
i
n
l
y
s
eems
w
o
rthwhile
for
a
t
l
eas
t
s
ome
A
I
r
es
earchers
t
o
i
n
v
es
tigate
e
mpirically
what
s
u
ch
s
y
s
t
ems
can
do.
B
u
t
i
t
i
s
a
mis
l
eading
m
is
nomer
to
tak
e
their
c
haracteris
tics
o
f
p
lanles
s
r
eaction,
cons
t
a
nt
at
t
e
nt
i
o
n
a
nd
s
e
l
ect
i
v
e
r
epres
e
nt
at
i
o
n
t
o
b
e
cent
r
al
t
o
“embodi
ed”
A
I.
H
a
vi
ng
a
body
is
neither
neces
s
a
ry
(e.g.,
cons
ider
Agre
and
C
hapman’
s
P
e
ngi
[2])
nor
s
u
f

cient
(unl
es
s
one
t
r
i
v
i
a
l
i
s
es
t
h
e
i
s
s
u
e
b
y
r
e-defi
n
i
n
g

body”
appropri
a
t
e
l
y
)
f
or
ha
vi
ng
t
hos
e
t
hree
features
.
So
it
turns
out
that
characteris
ing
E
AI
in
oppos
ition
t
o
f
ormality
,
o
r
i
n
oppos
ition
t
o
t
he
SMP
A
f
r
a
m
e
w
o
r
k
,
i
sn

t
th
at
h
e
lp
f
u
l.
Wh
er
e
e
lse
can
we
tu
r
n
?
4
.
Y
o
u’
r
e
g
r
o
unded!
O
n
e
o
f
A
nders
on’
s
i
mport
a
nt
poi
nt
s
i
s
t
hi
s
:
A
l
t
hough
t
h
e
G
O
F
A
I
vi
rt
ues
o
f
d
el
i
b
erat
i
o
n
and
r
eas
oni
ng
can
and
s
houl
d
b
e
r
et
ai
ned,
EA
I’
s
c
ont
ri
b
u
t
i
o
n
i
s
t
o
poi
nt
out
t
h
e
i
mport
a
nce
of
gr
oundi
ng
,
e
v
e
n
(
or
es
peci
al
l
y
)
f
or
s
u
ch
hi
gh-l
e
v
e
l
operat
i
ons
as
pl
anni
ng,
reas
oni
ng,
d
e
lib
er
atin
g
a
n
d
sy
m
b
o
l
p
r
o
cessin
g
.
I

m
in
clin
ed
to
ag
r
e
e
w
ith
th
is
statem
en
t,
with
th
e
pro
v
i
s
o
t
hat
I
mi
ght
di
s
a
gree
w
i
t
h
A
nders
on
and/
or
man
y
EA
I
p
roponent
s
o
n
w
hat
groundi
ng
amount
s
t
o;
more
on
t
h
at
l
a
t
e
r
.
F
i
rs
t
:
W
h
at
not
i
o
n
o
f
g
roundi
ng
i
s
EA
I
o
f
f
eri
ng,
according
t
o
A
nders
on?
N
e
i
t
h
er
A
nders
on
nor
H
a
rnad
[23],
w
ho
i
n
t
r
oduced
t
h
e
s
ymbol
groundi
ng
probl
em,
g
i
v
e
an
e
xpl
i
c
i
t
defi
n
i
t
i
on,
b
u
t
t
he
meani
n
g
o
f

groundi
ng”
s
eems
t
o
b
e
capt
u
red
b
y
s
omet
hi
ng
lik
e:
A
s
ymbol
i
s
grounded
i
f
i
t
h
as
i
t
s
meani
n
g
o
r
c
ont
ent
b
y
v
i
r
t
u
e
o
f
i
t
s
caus
a
l
p
ropert
i
e
s
an
d
r
elatio
n
s
to
th
e
r
ef
er
en
t
o
r
s
u
b
j
ect
m
a
tter
o
f
t
h
e
sy
m
b
o
l
;
o
r
i
s
a
p
p
r
o
p
r
iately
cau
sally
rel
a
t
e
d
t
o
(
defi
n
ed
i
n
t
e
rms
o
f)
grounded
s
ymbol
s
.
140
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
Thus
,
a
s
y
mbol
is
ungrounded
i
f
i
t
i
s
not
ultimately
c
ompletely
d
efinable
i
n
t
erms
of
s
y
mbol
s
w
hi
ch
ha
v
e
t
h
ei
r
c
ont
ent
b
y
v
i
r
t
u
e
o
f
t
hei
r
caus
a
l
r
el
at
i
ons
w
i
t
h
t
h
ei
r
s
ubj
ect
m
a
tter
.
On
e
w
ay
f
o
r
t
h
i
s
t
o
o
ccu
r
i
s
f
o
r
th
e
m
ean
in
g
o
f
t
h
e
sy
m
b
o
l
to
b
e
m
e
r
e
ly
a
m
atter
of
s
o
me
(ot
h
er?)
agent
a
s
c
ri
bi
ng
t
h
e
s
ymbol
t
h
at
meani
ng.
A
nders
on
gi
v
e
s
a
n
e
xampl
e
for
t
h
e
s
ymbol
“chai
r”
i
n
an
agent
:
i
t
i
s
grounded
onl
y
i
f
i
t
s
us
e
b
y
t
hat
a
gent
i
s
go
v
e
rned
by
an
ability
to
reliably
d
etect
(and
recognis
e?)
chairs
,
a
nd
if
it
gi
v
e
s
r
is
e
t
o
a
ppropriate
chair
beha
vi
our
.
Th
is
k
i
n
d
o
f
p
r
ag
m
a
tist
t
h
e
o
r
y
o
f
m
ean
in
g
h
as
f
a
m
iliar
p
r
o
b
l
em
s.
F
o
r
e
x
a
m
p
le,
i
t
seem
s
t
o
i
m
p
ly
th
at
if
I
act
in
ap
p
r
o
p
r
iately
to
w
a
r
d
so
m
e
th
in
g
,
I
ip
so
fa
cto
cannot
be
t
h
i
nki
ng
about
t
h
at
s
o
met
h
i
ng,
s
i
nce
a
ppropri
a
t
e
beha
vi
our
t
o
w
a
rd
i
t
i
s
a
r
equi
rement
for
repres
enting
it.
C
ons
ider
Anders
on’
s
e
xample
of
a
non-monarch
s
itting
o
n
a
king’
s
t
hrone;
A
nders
on
concl
udes
t
hat
s
uch
a
pers
on
mus
t
f
a
i
l
t
o
gras
p
e
i
t
h
er
t
h
e
c
oncept
“chai
r”
or
“t
hrone”.
B
u
t
i
f
w
e
g
eneral
i
s
e
t
hi
s
poi
nt
,
i
t
l
ooks
as
i
f
w
e
coul
d
n
e
v
er
be
w
r
ong,
s
a
y
,
i
n
m
a
th
em
atics.
I
f
I
s
ay

2
+
2
=
5

,
t
hen
o
n
t
he
pragmat
i
s
t
l
i
n
e,
I
m
us
t
not
gras
p
one
or
m
o
r
e
o
f
th
e
c
o
n
cep
ts
n
o
r
m
a
lly
in
v
o
lv
ed
in
th
at
e
x
p
r
essio
n
.
Bu
t
i
f
s
o
,
th
en
I
h
a
v
en

t
said
s
o
mething
f
als
e
about
2,
addition
a
nd
5;
rather
,
I
ha
v
e
either
s
a
id
s
o
mething
t
rue
a
bout
s
o
met
h
i
n
g
e
l
s
e,
or
I
h
a
v
e
s
ai
d
not
hi
ng
at
al
l
.
N
e
i
t
h
er
of
t
h
es
e
c
ons
equences
appeal
.
In
an
y
cas
e,
t
h
e
s
t
r
onges
t
form
of
t
h
e
E
A
I
groundi
ng
ar
gument
goes
s
omet
hi
ng
l
i
k
e:
P
e
opl
e
a
nd
EA
I
u
s
e
grounded
s
ymbol
s
;
G
O
F
A
I
does
not
.
T
herefore,
E
A
I
,
a
nd
not
G
O
F
A
I
,
_____
(fill
in
the
b
lank
as
required:
“can
pro
v
ide
good
models
human
cognition”,

is
pos
s
i
ble”,
“can
produce
t
rue
unders
tanding”,
e
tc.).
18
It
might
be
that
grounding
is
required
f
or
cognition,
unders
tanding,
intelligence,
etc.
b
u
t
t
he
abo
v
e
ar
gument
f
orm
i
s
not
a
good
one.
19
In
an
y
e
v
e
nt,
I’m
not
as
con
v
inced
as
A
nders
on
s
eems
t
o
b
e
t
hat
s
ymbol
groundi
ng
i
s
at
t
h
e
h
eart
o
f
E
A
I
.
E
A
I
i
s
much
more
radi
cal
:
I
t
i
s
not
a
t
hes
i
s
a
bout
ho
w
t
o
g
round
s
y
mbol
s
,
b
u
t
t
he
i
d
ea
t
h
at
s
y
mbol
s
,
grounded
o
r
not,
p
lay
only
a
s
m
all
r
ole
i
n
i
ntelligence
as
a
w
hole,
and
t
hat
t
here
are
m
a
n
y
asp
ects
o
f
m
en
tality
in
wh
ich
t
h
e
y
m
ay
p
l
ay
n
o
r
o
le
wh
atso
e
v
er
.
W
h
e
n
T
h
e
len
an
d
S
m
ith
[
4
4
]
tell
u
s
th
at
ch
an
g
e
s
i
n
t
h
e
m
a
ss
o
f
th
e
i
n
f
an
t’
s
l
e
g
p
l
ay
a
c
r
u
cial
r
o
le
in
the
acquis
ition
o
f
t
he
w
a
lking
s
kill,
when
W
e
bb
[46]
us
es
a
r
obot
to
s
h
o
w
us
ho
w
t
he
s
h
ape
o
f
t
he
cri
c
k
e
t
ear
al
l
o
w
s
t
h
em
t
o
achi
e
v
e
phonot
axi
s
,
w
hen
B
eer
[6]
a
nal
y
s
e
s
t
he
rob
u
s
t
gait
of
his
s
ix-le
gged
r
obots
u
s
i
ng
dynamical
s
y
s
t
ems
t
heory
,
when
B
r
eazeal
and
Scas
s
e
llati
[7]
s
ho
w
u
s
h
o
w
putting
e
yebro
w
s
o
n
K
is
met
e
ns
ures
the
p
roper
carer
-robot
dynamic
n
eces
s
a
ry
for
K
is
met
t
o
l
earn
t
o
t
rack
objects
v
is
ually;
w
hen
t
hes
e
res
earchers
mak
e
t
h
ei
r
c
ont
ri
b
u
t
i
ons
,
t
he
y
a
re
not
doi
ng
s
o
pri
m
ari
l
y
,
i
f
a
t
a
l
l
,
by
w
a
y
o
f
s
ho
w
i
ng
ho
w
s
ymbol
s
a
re
grounded.
The
i
s
s
u
e
j
us
t
does
n

t
come
up.
S
y
mbol
groundi
ng
may
b
e
an
i
m
port
a
nt
i
s
s
u
e,
e
v
en
t
o
s
o
me
EA
I
p
eopl
e,
b
u
t
i
t
j
us
t
does
n

t
s
eem
t
o
hi
t
t
he
nai
l
on
18
Anders
on
s
eem
s
t
o
a
llo
w
t
hat
p
erhaps
not
all
of
m
y
repres
entations
are
g
rounded
i
n
t
his
w
ay
,
b
ut
m
a
y
b
e
grounded
b
y
o
ther
hum
ans
.
E
.
g.
,
p
erhaps
m
a
n
y
of
m
y
s
y
m
bols
d
enoting
t
echnical
concepts
get
t
heir
m
eanings
from
t
he
content
a
s
s
i
gned
t
o
t
hem
b
y
t
he
e
xperts
t
o
w
hich
I
d
efer
.
B
ut
the
c
laim
of
the
p
roponent
of
em
bodim
e
nt
w
ould
b
e
t
hat
a
t
l
eas
t
so
me
of
m
y
s
y
m
bols
a
re
grounded
i
n
m
y
o
wn
caus
a
l
r
elations
to
things
in
the
w
orld
(other
than
e
xperts
)
,
a
nd
the
s
ym
bols
t
hat
d
o
h
a
v
e
d
eferential
c
ontent
m
us
t
b
e
s
o
g
rounded
f
or
the
e
xperts
(
either
indi
vidually
,
o
r
a
s
a
group)
to
which
I
defer
.
19
At
leas
t
not
without
s
o
m
e
kind
of
as
s
i
s
t
ance;
perhaps
t
hat
c
ould
b
e
p
ro
vided
b
y
t
he
points
D
re
yfus
m
a
k
e
s
concer
ning
intelligibility;
s
ee
S
ection
6
.
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
141
the
h
ead
when
attempting
t
o
c
haracteris
e
t
he
field,
o
r
a
t
l
eas
t
h
o
w
the

eld
c
haracteris
es
itself
.
T
o
be
f
a
i
r
,
w
hen
A
nders
on
t
a
l
k
s
o
f
t
he
“phys
i
cal
groundi
ng
proj
ect

a
s
b
ei
ng
t
h
e
cent
r
al
proj
ect
of
t
h
e
e
mbodi
ed
approach,
h
e
m
i
ght
ha
v
e
s
o
met
h
i
n
g
m
ore
g
eneral
i
n
mi
nd
t
h
an
groundi
ng
s
y
mbol
s
i
n
t
he
manner
j
us
t
d
i
s
cus
s
e
d.
T
a
k
e
,
f
or
e
x
ampl
e,
hi
s
e
xpl
i
cat
i
o
n
o
f
t
he
project
as
“centrally
in
v
o
lving
unders
tanding
ho
w
c
ogniti
v
e
contents
(ho
w
e
v
er
thes
e
a
re
u
ltim
ately
c
h
a
r
acter
ised
,
s
y
m
b
o
lically
or
ot
her
w
i
s
e
)
m
us
t
u
ltimately
g
round
out
in
(terms
of)
t
he
agent

s
e
mbodi
ed
e
xperi
ence
and
phys
i
cal
charact
eri
s
t
i
c
s

(emphas
i
s
a
dded).
B
ut
th
is
still
p
l
aces
an
emp
h
a
sis
o
n
r
ep
resen
t
atio
n
a
l
c
o
n
t
en
t,
as
if
EAI
a
g
r
eed
with
GOF
AI
th
at
th
at
is
wh
er
e
t
h
e
actio
n
i
s.
20
There
i
s
a
ls
o
t
he
dra
w
back
that
s
y
mbol
grounding
is
neither
neces
s
a
ry
nor
s
u
f

cient
for
h
a
v
ing
a
rob
u
s
t
form
of
embodiment.
O
ne
could
b
e
doing
AI
that
placed
embodiment
cent
r
e
s
t
a
ge
and
y
et
not
be
produci
n
g
r
obot
s
w
i
t
h
grounded
s
ymbol
s
(
e.g.,
your
robot
mi
ght
not
ha
v
e
s
y
mbol
s
a
t
a
l
l
)
;
a
nd
con
v
ers
e
l
y
one
coul
d
b
e
doi
ng
non-embodi
ed,
y
et
grounded
(i
n
v
i
d
eo
cameras
and
d
i
s
embodi
ed
robot
arms
),
A
I
.
T
hat
m
i
ght
e
xpl
ai
n
w
hy
t
h
e
E
A
I
peopl
e,
w
h
o
p
res
u
mabl
y
a
re
i
n
t
e
res
t
ed
pri
m
ari
l
y
i
n
embodi
ment
,
don’
t
t
al
k
a
bout
s
y
mbol
groundi
ng
much.
A
nders
on
i
s
ri
ght
i
n
s
a
yi
ng
t
h
at
embodi
ment
s
houl
d
b
e
d
i
s
t
i
ngui
s
h
ed
from
s
i
t
u
at
ednes
s
.
Ho
we
v
e
r
,
I’m
not
s
o
s
u
re
that
“it
i
s
t
he
centrality
of
the
phys
ical
grounding
project
[to
embodied
cognition]
that
dif
f
erentiates
r
es
earch
in
embodied
cognition
from
r
es
earch
i
n
s
i
t
u
at
ed
cogni
t
i
on”
(emphas
i
s
r
emo
v
e
d).
I
nas
m
uch
a
s
I
unders
t
a
nd
w
h
at
i
s
meant
b
y
groundi
ng,
i
t
w
oul
d
s
eem
t
h
at
bei
n
g
e
mbedded
i
n
a
w
o
rl
d
i
n
t
he
proper
w
ay
has
a
s
m
uch
t
o
do
w
i
t
h
i
t
as
,
i
f
not
more
t
h
an,
h
a
v
i
n
g
a
body
does
.
(More
o
n
s
i
t
u
at
ednes
s
i
n
S
ect
i
o
n
5
.)
The
k
i
n
d
o
f
g
roundi
ng
t
h
at
I
t
hi
nk
i
s
i
m
port
a
nt
t
o
A
I
,
i
s
,
i
n
s
t
ead,
t
he
ki
nd
of
groundi
ng
w
e
s
a
w
i
n
S
ect
i
o
n
3
.2
t
o
be
es
s
e
nt
i
a
l
t
o
o
v
e
rcomi
n
g
t
he
(t
heoret
i
cal
)
p
robl
em
of
(pure)
rationality
.
T
his
k
ind
o
f
g
rounding
comes
f
or
free:
21
an
y
s
ymbol
i
c
pl
anner
that
does
n

t
get
i
nto
a
n
i
nfi
nite
loop
can
only
b
e
s
o
b
ecaus
e
it
is
grounded
i
n
t
he
non-
r
a
tio
n
a
l.
5.
Getti
n
g
s
i
tu
ated
A
s
I
j
us
t
s
t
a
t
e
d,
A
nders
on
i
s
ri
ght
t
o
di
s
t
i
ngui
s
h
embodi
ment
from
s
i
t
u
at
ednes
s
:
t
h
e
f
o
r
m
er
co
n
cer
n
s
th
e
w
ay
th
at
in
ten
tio
n
a
l
(
co
m
p
u
t
atio
n
a
l,
r
e
p
r
esen
tatio
n
a
l,
m
e
n
t
al,
e
tc.)
20
Perhaps
w
hat
n
eeds
t
o
b
e
g
rounded
i
s
s
om
ething
e
v
en
m
o
re
general
t
han
r
epres
e
ntational
c
ontent
(
s
u
ch
as
“beha
v
iour
”,
“acti
v
ity”,
“f
eatur
es
”,
“com
p
etences
”,
etc.
)
,
b
u
t
i
f
t
his
c
hange
is
m
a
de,
t
he
pr
oject
becom
e
s
s
o
m
e
thing
s
o
a
m
o
rphous
as
to
be
am
enable
to
the
s
taunches
t
s
upported
o
f
GOF
AI.
W
ho
b
u
t
a
dualis
t
w
ould
d
en
y
that
a
s
ys
tem

s
b
eha
v
iour
s
hould
b
e
g
rounded
i
n
its
phys
ical
characteris
tics
?
S
u
rely
it
is
the
phys
ical
cons
titution
of
a
P
C
w
hich
m
a
k
e
s
i
t
t
he
cas
e
t
hat
i
t
can
be
unders
tood
as
running
Open
Of
fice
under
L
inux?
W
h
at
is
ne
w
h
ere,
e
x
actly?
21
F
o
r
free?
Metaphys
ically
,
y
es
;
E
pis
t
em
ologically
,
no.
It
is
a
c
ons
iderable
intellectual
a
nd
engineering
achie
v
e
m
e
nt
to
tr
ans
f
or
m
,
e.
g.
,
a
f
o
r
m
al
s
p
ecifi
cation
o
f
a
com
putation
(
a
pur
ely
r
ational
s
ys
tem
i
f
a
n
y
thing
is
)
i
nto
a
w
o
rking,
phys
ical
im
plem
entation.
(T
.
S
.
E
liot
w
ould
n
o
doubt
s
a
y
I
unders
tate
the
cas
e:
“Between
the
i
dea
\\
A
n
d
t
he
r
eality
\\
...B
e
t
w
e
e
n
t
h
e
c
o
n
c
e
p
t
i
o
n
\\
A
n
d
t
h
e
c
r
e
a
t
i
o
n
...B
e
t
w
e
e
n
t
h
e
p
o
t
e
n
c
y
\\
And
t
he
e
x
is
tence
...
F
a
lls
the
S
hado
w


T
h
e
H
ollow
M
en
).
142
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
propert
i
e
s
a
re
dependent
on
(real
i
s
ed
i
n
,
m
ade
pos
s
i
bl
e
t
hrough,
cons
t
r
ai
ned
by)
a
non-
i
n
t
e
nt
i
onal
l
y
charact
eri
s
ed
phys
i
cal
s
ubs
t
r
at
e;
t
h
e
l
at
t
e
r
c
oncerns
ho
w
s
uch
p
ropert
i
e
s
a
re
dependent
on
a
(
pos
s
i
bl
y
i
nt
ent
i
onal
a
nd
non-phys
i
cal
l
y
charact
eri
s
ed)
s
urround.
D
e
s
p
i
t
e
this
dis
tinction,
much
of
Anders
on’
s
r
e
v
ie
w
o
f
t
he
field
of
embodied
cognition
m
entions
or
i
n
v
o
l
v
es
s
i
t
u
at
ednes
s
;
t
h
e
a
bo
v
e
s
ect
i
o
n
c
l
a
i
m
ed
t
h
at
A
nders
on’
s
not
i
o
n
o
f
g
roundi
ng
i
s
probabl
y
a
s
i
t
u
at
ednes
s
requi
rement
,
a
nd
hi
s
d
i
s
cus
s
i
ons
of
s
o
ci
al
embeddednes
s
and
e
xploitation
o
f
e
xternal
r
epres
e
ntations
and

props

(
tools
,
language,
e
tc.)
are
e
xplicitly
about
w
h
at
s
i
t
u
at
ed
A
I
mi
ght
l
ook
l
i
k
e.
Anders
on
looks
at
a
c
ouple
o
f
p
apers
from
t
he
traditional
A
I
camp
[
26,45]
that
attempt
t
o
as
s
i
mi
l
a
t
e
s
i
t
u
at
ednes
s
i
n
t
o
G
O
F
A
I
.
T
hat
i
s
,
t
h
e
y
cl
ai
m
t
hat
t
here
i
s
not
hi
ng
about
situ
ated
n
e
ss
th
at
is
in
co
m
p
atib
le
with
tr
ad
itio
n
a
l
A
I
t
h
e
o
r
y
(
e.g
.
,
t
h
e
Ph
y
s
ical
Sy
m
b
o
l
S
y
s
t
em
H
ypot
hes
i
s
)
and
m
et
hodol
ogy/
pract
i
ce.
(A
ft
er
al
l
,
w
e
are
o
ft
en
remi
nded,
t
h
e
S
i
mon
w
ho
ga
v
e
us
t
h
e
P
hys
i
cal
S
y
mbol
S
y
s
t
em
H
ypot
hes
i
s
i
s
t
he
s
a
me
one
w
h
o
g
a
v
e
u
s
S
i
mon’
s
a
nt
.)
W
h
i
l
e
s
y
mpat
het
i
c
t
o
(roughl
y)
a
hybri
d
of
G
O
F
A
I
w
i
t
h
embodi
ed/
s
i
t
u
at
ed
A
I
,
A
nders
on
i
s
not
i
m
pres
s
e
d
w
i
t
h
t
h
e
c
l
a
i
m
s
t
hat
s
uch
a
uni
on
i
s
b
u
s
i
nes
s
as
us
ual
f
or
th
e
GOF
AI
camp
.
In
part
i
c
ul
ar
,
A
nders
on
t
h
i
nks
t
h
at
H
a
yes
e
t
a
l
.
[26]
defend
G
O
F
A
I
agai
ns
t
t
he
s
i
t
u
at
ed
ons
l
a
ught
onl
y
b
y
r
edefi
ni
ng
G
O
F
A
I
(and
t
hus
,
b
y
c
ont
ras
t
,
s
i
t
u
at
ed
A
I
)
i
n
a
w
a
y
t
hat
tr
i
v
ialises
th
e
d
eb
ate.
Th
is
tr
i
v
ialisatio
n
i
s
m
ean
t
t
o
b
e
s
h
o
wn
b
y
th
e
f
act
th
at
th
e
r
esu
ltin
g
d
e
fin
itio
n
o
f
s
itu
ated
AI
is
so
im
p
l
au
sib
l
e
a
n
d
e
x
tr
em
e
a
s
t
o
b
e
n
o
o
n
e

s
p
o
s
itio
n
.
Hay
e
s
e
t
al
.
a
re
defendi
ng
t
h
ems
e
l
v
es
agai
ns
t
a
n
i
magi
nary
s
t
ra
w
m
an.
A
nders
on
w
r
i
t
e
s
:
W
ith
...
our
repres
ent
i
n
g
h
eads
s
ui
t
a
bl
y
e
xpanded
t
o
e
ncompas
s
t
h
e
r
equi
s
i
t
e
phys
i
cal
an
d
s
o
c
ial
t
er
r
ito
r
y
,
w
h
a
t
i
s
l
ef
t
f
o
r
SitNan
n
y
[
t
h
e
situ
ated
co
g
n
itio
n
t
h
e
o
r
ist—RC]
t
o
bel
i
e
v
e
i
s
‘t
hat
t
he
repres
ent
a
t
i
onal
t
ok
ens
t
hems
el
v
e
s
a
ren’
t
i
n
t
he
head
or
t
h
at
repres
entational
t
ok
en
can
only
h
a
v
e
a
n
e
xternal,
s
o
cial
e
x
is
tence,
or
e
v
en
that
there
i
s
n’
t
a
n
y
repres
ent
a
t
i
o
n
a
t
a
l
l

.
S
i
t
N
ann
y
may
b
el
i
e
v
e
t
h
i
s
,
b
ut
I
don’
t
kno
w
o
f
a
s
i
ngl
e
[s
i
t
u
at
ed
cogni
t
i
on]
res
earcher
w
h
o
does
.
I
c
oncl
ude
from
t
he
abs
e
nce
o
f
a
ci
t
a
t
i
o
n
f
or
this
claim
t
hat
t
he
authors
don’
t
kno
w
o
f
one,
e
ither
.
([4,
S
ection
5
],
citing
[
26,
p.
20])
B
u
t
H
ayes
et
al
.
a
re
not
hal
l
u
ci
nat
i
n
g
a
n
opponent
;
t
here
are
s
e
v
eral
peopl
e
w
ho
hol
d
vi
e
w
s
t
hat
t
he
y
a
s
c
ri
be
t
o
S
i
t
N
ann
y
.
I
f
M
i
c
hael
Morri
s
,
t
h
e
a
ut
hor
of
“W
hy
t
h
ere
i
s
n
o
s
uch
thing
a
s
m
ental
r
epres
e
ntations

[
36]
does
n

t
count
becaus
e
he
is
n’
t
a
s
ituated
cognition
res
earcher
,
one
can
i
n
s
t
ead
t
u
rn
t
o
[24],
[
34],
o
r
[
31],
a
mong
man
y
ot
hers
.
Anders
on
prefers
V
era
a
nd
Simon’
s
[
45]
comparati
v
e
a
nalys
i
s
o
f
GOF
AI
and
s
ituated
r
e
p
r
esen
tatio
n
,
b
u
t
s
till
th
in
k
s
th
e
y
o
v
er
state
t
h
e
d
i
f
f
er
en
ces:
[O
]n
t
h
i
s
unders
t
a
ndi
ng
t
h
e
s
ymbol
s
i
n
v
ol
v
e
d
m
ay
i
n
cl
ude
not
j
u
s
t
uncons
ci
ous
s
t
at
es
,
b
u
t
p
roces
s
e
s
t
aking
p
lace
primarily
in
the
central
nerv
ous
s
y
s
t
em
as
a
w
hole,
and
p
e
r
h
ap
s
o
n
l
y
m
in
im
ally
in
v
o
lv
in
g
t
h
e
b
r
ain
.
Wh
ate
v
er
th
e
u
ltim
ate
u
tility
o
f
callin
g
s
u
ch
proces
s
e
s
s
ymbol
i
c
,
w
e
s
houl
d
a
t
l
eas
t
b
e
a
w
a
re
of
t
h
e
g
reat
di
s
t
ance
bet
w
een
t
h
e
v
i
e
w
o
f
co
g
n
itio
n
p
u
t
f
o
r
w
ar
d
h
er
e
b
y
V
er
a
a
n
d
Sim
o
n
,
an
d
t
h
a
t
s
u
m
m
a
r
i
zed
in
th
e
cent
r
al
hypot
hes
i
s
o
f
G
O
F
A
I
,
a
s
d
efi
ned
i
n
S
ect
i
o
n
1
.
[
4,
S
ect
i
o
n
5
]
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
143
Bu
t
A
n
d
e
r
s
o
n

s

lo
o
s
e”
d
e
fin
itio
n
o
f
GOF
AI
in
ter
m
s
o
f
C
ar
tesian
ism
a
n
d
a

to
o
s
im
p
l
e
account

o
f
C
ogni
t
i
vi
s
m
s
eems
t
o
a
l
l
o
w
p
l
e
nt
y
o
f
r
oom
for
V
era
a
nd
S
i
mon’
s
v
i
e
w
.
22
It

s
th
e
phys
i
cal
s
y
mbol
s
y
s
t
em
hypot
hes
i
s
,
aft
e
r
a
l
l
,
not
t
h
e
cer
ebr
a
l
or
e
v
en
neur
al
s
y
mbol
s
y
s
t
em
hypot
hes
i
s
.
O
n
e
cannot
s
i
mul
t
a
neous
l
y
l
a
mbas
t
e
G
O
F
A
I
for
b
ei
ng
C
a
rt
es
i
a
n
i
n
t
hat
it
maintains
a
s
t
rong
multiple-realis
ation
t
hes
i
s
a
nd
an
autonomy
o
f
t
he
mental
from
t
he
phys
i
cal
,
w
hi
l
e
s
i
mul
t
a
neous
l
y
cl
ai
mi
ng
t
h
at
i
t
s
C
ogni
t
i
vi
s
m
as
s
u
mes
t
hat
t
he
s
y
mbol
s
o
f
mental
proces
s
i
ng
are
s
pecifically
cerebral,
and
not
merely
neural,
bodily
,
o
r
w
orldly
.
I
f
G
O
F
A
I
t
h
eori
s
t
s
a
re
as
bi
g
o
n
d
i
s
embodi
ment
as
t
h
e
C
art
e
s
i
an
epi
t
h
et
s
ugges
t
s
,
i
n
t
h
at
the
y
are
t
he
ones
t
hat
a
re
claiming
that
robots
,
nay
c
omputers
,
made
of
s
ilicon
can
be
in
tellig
en
t,
if
th
eir
v
ie
w
i
s
t
h
e
o
n
e
th
at
im
p
lies
t
h
a
t
p
r
o
p
e
r
l
y
s
tr
u
n
g
to
g
e
th
er
b
eer
can
s
[40]
can
real
i
s
e
t
hi
nki
ng,
t
h
en
s
u
rel
y
t
h
e
y
ha
v
e
no
s
t
rong
cl
ai
m
a
s
t
o
w
hat
phys
i
cal
s
t
uf
f
underl
i
e
s
hum
an
m
e
n
t
ality
.
T
h
e
y
cer
tain
ly
w
o
n

t
b
e
b
af
fled
b
y
s
u
c
h
q
u
e
stio
n
s
as
“Wh
a
t
does
i
t
m
ean
for
a
repres
entation
...
t
o
be
encoded
i
n
t
h
e
body
r
a
th
er
th
an
in
th
e
h
ead

[4,
S
ect
i
o
n
5
].
If
one
unders
t
a
nds
ho
w
bodys
t
u
f
f
i
n
t
h
e
h
ead
can
i
n
s
t
ant
i
a
t
e
or
i
m
pl
ement
(“encode”)
a
r
epres
e
ntation,
one
ip
so
fa
cto
unders
t
a
nds
ho
w
bodys
t
u
f
f
not
i
n
t
h
e
h
ead
mi
ght
do
s
o
.
The
f
act
i
s
,
s
i
t
u
at
ednes
s
i
n
comput
at
i
onal
s
ys
t
e
ms
i
s
not
a
n
e
w
phenomenon.
W
h
en
I,
as
a
r
ecent
g
raduate,
w
as
programming
a
r
obot
to
na
vigate
the
c
orridors
o
f
X
erox
P
A
R
C
,
my
al
gori
t
h
ms
di
dn’
t
cal
cul
a
t
e
and
r
epres
e
nt
t
h
e
d
i
r
ect
i
o
n
t
hat
t
he
robot
needed
t
o
head
i
n
to
continue
after
t
aking
a
“s
ens
o
r
-readings
-collection”
break.
R
ather
,
becaus
e
the
r
obot’
s
heading
w
as
unlik
ely
t
o
c
hange
during
t
he
break,
t
he
robot’
s
pos
ition
its
elf
s
tored
t
he
information
a
s
t
o
t
he
heading
t
o
t
ak
e
a
fter
the
b
reak.
A
round
the
s
ame
time,
I
w
as
writing
programs
i
n
L
ISP
o
n
a
Macintos
h
t
o
g
enerate
m
us
ic
on
an
e
x
ternal
s
ynthes
i
zer
connected
by
a
s
eri
a
l
l
i
nk.
In
order
t
o
kno
w
w
hat
byt
es
t
o
s
e
nd
t
o
t
h
e
s
eri
a
l
port
t
o
achi
e
v
e
a
p
art
i
c
ul
ar
change
i
n
t
h
e
s
ynt
hes
i
zer
,
I
needed
t
o
kno
w
t
he
e
x
i
s
t
i
n
g
v
al
ue
at
t
h
e
a
ddres
s
I
w
a
s
a
bout
to
wr
ite
to
(
I
co
u
l
d
o
n
l
y
s
en
d
b
it
“m
ask
s
”,
n
o
t
actu
a
l
b
it
v
a
lu
es)
.
Af
ter
b
r
i
efly
c
o
n
s
id
er
in
g
th
e

GOF
AI

i
d
e
a
o
f
c
o
n
s
tan
tly
m
a
in
tain
in
g
a
m
o
d
e
l
o
f
t
h
e
en
tir
e
s
y
n
t
h
e
sizer
,
I
in
stead
decided
t
o
p
recede
a
ll
changes
w
ith
a
v
alue
query
that
w
ould
t
ell
m
e
t
he
current
v
a
lue.
I
n
b
o
t
h
cases,
I
w
a
s
u
sin
g
th
e
w
o
r
ld
as
its
o
w
n
b
est
m
o
d
e
l;
I
w
as
e
x
p
l
o
itin
g
t
h
e
r
e
latio
n
s
bet
w
een
t
h
e
c
omput
at
i
onal
s
ys
t
e
m
a
nd
i
t
s
en
vi
ronment
a
s
a
w
a
y
o
f
o
f

oadi
ng
(a
v
o
i
d
i
n
g
e
xpens
i
v
e
c
omput
at
i
onal
operat
i
ons
).
B
u
t
w
hat

s
i
mport
a
nt
about
t
h
i
s
i
s
not
t
h
at
I
w
as
doi
ng
an
yt
hi
ng
di
f
f
erent
t
han
a
n
ybody
el
s
e
before,
a
t
t
he
t
i
m
e,
or
s
i
nce.
W
h
at

s
i
m
port
a
nt
i
s
t
h
at
w
h
at
I
w
as
doi
ng
i
s
s
o
perv
as
i
v
e,
banal
e
v
e
n,
i
n
real
-w
orl
d
,
t
radi
t
i
onal
c
omput
at
i
on.
N
o
t
h
i
n
g
s
houl
d
b
e
r
ead
i
n
t
o
t
h
e
f
act
t
h
at
my
pers
onal
e
xampl
e
s
onl
y
c
i
t
e
e
xpl
oi
t
a
t
i
o
n
of
t
h
e
phys
i
cal
/
c
omput
at
i
onal
s
urround.
The
e
xampl
e
of
t
h
e
c
ompl
e
x
s
y
s
t
em
underl
yi
ng
s
h
i
p
na
vi
gat
i
o
n
[
30],
w
hi
ch
s
i
t
u
at
ed
t
h
eori
s
t
s
l
i
k
e
t
o
c
i
t
e
,
i
s
j
us
t
a
s
m
uch
a
model
f
or
ho
w
actu
a
l
c
o
m
p
u
t
atio
n
a
l
s
y
s
tem
s
f
u
n
c
tio
n
b
y
e
x
p
l
o
itin
g
a
co
m
p
le
x
w
eb
o
f
so
cially
-
m
ed
iated
rel
a
t
i
ons
,
a
s
i
t
i
s
a
model
f
or
ho
w
humans
d
o
s
o.
23
T
r
u
e
,
t
h
e
r
o
le
o
f
situ
ated
n
e
ss
in
co
g
n
itio
n
h
as
b
een
r
e
lati
v
e
ly
n
e
g
l
ected
b
y
tr
ad
itio
n
a
l
A
I
.
T
rue,
t
h
ere
i
s
a
l
o
t
t
o
b
e
done
t
o
i
m
pro
v
e
our
unders
t
a
ndi
ng
of
t
h
e
w
ays
t
hat
22
As
it
s
hould;
if
one’
s
cons
trual
o
f
GOF
AI
im
plied
t
hat
H
erbert
Sim
o
n
w
as
n’
t
doing
it,
that’
s
as
clos
e
a
s
a
r
e
ductio
ad
abs
ur
dum
of
that
defi
n
ition
a
s
one
m
i
ght
hope
for
.
23
F
o
r
a
s
t
udy
of
ho
w
t
he
s
u
cces
s
o
f
m
undane
com
putation
i
s
achie
v
e
d
only
t
hrough
an
e
x
tens
i
v
e
c
onte
x
t
o
f
carefully
re
gis
t
ered
and
m
aintained
s
ocial
e
m
b
edding,
s
e
e
[
1].
144
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
human
embodiment
f
acilitates
certain
forms
o
f
s
ituatednes
s
,
and
our
unders
tanding
of
the
means
b
y
w
hich
artificial
s
y
s
t
ems
can
e
xploit
t
heir
en
vironment
a
s
a
w
a
y
o
f
o
f

oading
comput
at
i
onal
t
as
ks
.
M
y
poi
nt
s
a
re
onl
y
t
hat
(
1)
s
i
t
u
at
ednes
s
i
s
or
can
be
an
i
m
port
a
nt
as
pect
of
non-rob
u
s
t
l
y
embodi
ed
s
y
s
t
ems
(
s
i
t
u
at
ednes
s
does
not
requi
re
embodi
ment
),
and
(
2)
accommodating
s
ituatednes
s
does
not
require
a
r
adical,
d
is
continuous
change
in
t
h
e
c
oncept
u
al
or
t
h
eoret
i
cal
frame
w
o
rk
of
repres
ent
a
t
i
onal
A
I
(
al
t
hough
i
t
mi
ght
requi
re
a
m
ark
e
dl
y
d
i
f
ferent
m
e
t
hodol
o
g
y
from,
s
a
y
,
t
h
at
of
C
Y
C
[
33]).
F
o
r
e
xampl
e
,
p
erhaps
unders
t
a
ndi
ng
and
d
es
i
gni
ng
s
y
s
t
ems
w
i
t
h
t
h
e
s
el
ect
i
v
e
r
epres
e
nt
at
i
ons
of
w
h
i
c
h
A
nders
on
s
p
eaks
(
and
w
hich
are
a
lready
p
res
e
nt
in
traditional
A
I
s
ys
tems
s
u
ch
as
C
a
s
s
a
ndra
[
38])
requi
res
a
dramat
i
cal
l
y
no
v
e
l
not
i
on,
t
h
at
of
non-concept
u
al
cont
ent
[
12,22].
B
ut
e
v
en
i
f
so
,
i
t
i
s
a
n
o
tio
n
w
h
i
ch
will
slo
t
in
to
an
d
m
o
d
i
f
y
an
e
x
istin
g
t
h
e
o
r
etical
f
r
a
m
e
w
o
r
k
o
f
repres
ent
a
t
i
ons
,
c
omput
at
i
on,
i
n
format
i
o
n
p
roces
s
i
ng,
et
c.,
not
demand
s
t
art
i
n
g
from
a
bl
ank
(
or
w
o
rs
e:
neurophys
i
o
l
ogi
cal
)
t
heoret
i
cal
s
l
at
e.
24
I
d
o
not
w
i
s
h
t
o
do
w
npl
ay
t
h
e
n
eed
t
o
e
xpl
ore
s
i
t
u
at
ednes
s
.
N
o
t
onl
y
d
o
w
e
n
eed
bet
t
e
r
t
o
unders
t
a
nd
ho
w
c
urrent
comput
ers
c
ruci
al
l
y
e
xpl
oi
t
r
el
at
i
ons
t
o
t
h
e
w
orl
d
i
n
order
t
o
g
et
t
h
ei
r
j
obs
done,
b
ut
w
e
w
i
l
l
benefi
t
from
unders
t
a
ndi
ng
ho
w
n
at
ural
s
y
s
t
ems
,
i
n
cl
udi
ng
humans
,
us
e
d
i
f
ferent
forms
o
f
s
i
t
u
at
ednes
s
t
o
achi
e
v
e
t
h
ei
r
goal
s
.
W
hat

s
m
os
t
s
t
ri
ki
ng
about
t
h
e
n
at
ural
cas
es
i
s
t
h
at
w
h
i
l
e
current
comput
at
i
onal
(
as
oppos
ed
t
o
robot
i
c
)
situ
ated
n
e
ss
is
(
a
lm
o
s
t?)
e
n
tir
ely
m
ed
iated
b
y
s
y
m
b
o
l
s,
th
e
m
o
s
t
b
asic
f
o
r
m
s
o
f
n
atu
r
al
s
i
t
u
at
ednes
s
e
xpl
oi
t
a
n
unconcept
u
al
i
s
ed
en
vi
ronment
d
i
r
ect
l
y
.
T
herefore,
d
i
s
co
v
e
ri
ng
w
a
ys
for
c
omput
at
i
onal
s
ys
t
e
ms
l
i
k
e
w
i
s
e
t
o
e
xpl
oi
t
t
hei
r
non-concept
u
al
s
u
rround
i
s
an
e
x
citin
g
p
r
o
ject.
B
u
t
wh
ile
it’
s
c
lear
th
at
to
d
o
so
,
s
u
c
h
s
y
s
tem
s
will
h
a
v
e
to
h
a
v
e
su
b
s
tan
tial
non-s
y
mbol
i
c
and
non-concept
u
al
phys
i
cal
as
pect
s
,
i
t
i
s
not
cl
ear
t
h
at
t
h
es
e
a
s
p
ect
s
w
i
l
l
s
u
m
u
p
t
o
a
n
y
t
h
i
n
g
t
hat
w
e
w
oul
d
r
ecogni
s
e
as
a
body
.
EA
I
p
roponent
s
r
i
ght
l
y
e
x
t
o
l
t
he
vi
rt
ues
o
f
s
i
t
u
at
ednes
s
,
b
u
t
a
re
l
e
s
s
fort
hcomi
n
g
i
n
th
eir
a
n
a
ly
sis
o
f
its
v
i
ces.
T
h
e
f
act
is,
t
h
e
r
e

s
a
t
r
a
d
e
-
o
f
f
b
e
tween
sp
eed
an
d
g
en
er
ality
.
T
h
e
more
one
opt
i
m
i
s
es
one’
s
met
hods
t
o
e
xpl
oi
t
c
ont
i
ngenci
e
s
o
f
t
he
en
vi
ronment
,
t
h
e
m
ore
one’
s
s
u
cces
s
i
s
bound
to
thos
e
c
ontingencies
;
change
the
c
onte
x
t
only
a
little
and
t
he
situ
ated
r
o
u
tin
e
f
ails,
u
su
ally
in
a
s
p
ectacu
lar
l
y
s
tu
p
i
d
w
ay
.
A
ch
allen
g
e
,
w
h
i
ch
An
d
e
r
s
o
n
ackno
w
l
edges
[
4,
S
ect
i
o
n
5
],
i
s
t
o
unders
t
a
nd
ho
w
a
s
y
s
t
em
can
e
xpl
oi
t
i
t
s
s
i
t
u
at
ed
rout
i
n
es
w
h
en
appropri
a
t
e
,
b
ut
res
o
rt
t
o
ot
her
m
eans
w
hen
t
he
cont
e
x
t
c
hanges
(
pos
s
i
bl
y
t
o
one
w
h
i
c
h
i
s
o
f
a
t
ype
w
h
i
c
h
h
as
not
been
encount
ered
before).
There
s
eem
t
o
be
at
l
eas
t

v
e
ki
nds
of
s
o
l
u
t
i
on:

Ha
v
e
a
sto
r
e
d
rout
i
n
e
f
or
t
h
e
n
e
w
cont
e
x
t
a
s
w
el
l
.
Thi
s
has
t
he
adv
a
nt
age
o
f
b
ei
ng
f
a
s
t
and
c
l
ear
ho
w
t
o
i
mpl
e
ment
,
b
ut
t
h
e
number
o
f
r
out
i
n
es
w
oul
d
b
e
c
ombi
nat
o
ri
al
l
y
prohi
bi
t
i
v
e
(compare
t
h
e
“cont
i
ngenc
y
p
l
a
ns

i
n
S
ect
i
o
n
3
.2).
24
E
v
en
Clance
y
[
14,
p.
113],
i
n
h
is
res
pons
e
t
o
V
era
a
nd
S
i
m
on’
s
p
aper
,
c
oncedes
:

Certainly
,
it
is
n’
t
n
eces
s
a
ry
(or
p
erhaps
pos
s
i
ble)
to
break
‘com
pletely
from
t
raditional
...
theor
i
es

(
p.
46)
b
u
t
i
ns
tead
to
r
econs
ider
the
relation
o
f
our
m
odels
to
the
c
ogniti
v
e
phenom
ena
w
e
s
ought
to
unders
tand.
Sym
bolic
m
odels
,
a
s
t
ools
,
will
al
w
a
ys
be
w
ith
us
.
Y
et,
a
lr
eady
t
he
s
h
if
t
h
as
be
gin
f
r
o
m
v
ie
w
i
ng
them
as
intelligent
b
eings
,
t
o
b
ut
the
s
hado
w
o
f
what
we
m
u
s
t
e
xplain.

R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
145

G
i
v
e
up
on
s
i
t
u
at
ednes
s
for
t
he
ne
w
c
ont
e
x
t
a
nd
empl
o
y
abs
t
ract
,
i
nt
ernal
d
el
i
b
erat
i
on.
Th
is
also
h
a
s
t
h
e
ad
v
a
n
t
ag
e
o
f
b
ein
g
clear
h
o
w
t
o
i
m
p
lem
e
n
t
,
b
u
t
it
h
a
s
t
h
e
f
a
m
iliar
s
p
eed
di
s
a
dv
ant
a
ge
of
del
i
b
erat
i
on.

O
n
e
i
mport
a
nt
i
d
ea
t
h
at
comes
out
of
t
h
e
s
t
udy
of
s
i
t
u
at
ednes
s
i
s
t
h
at
compl
e
x
probl
ems
(
s
u
ch
as
di
vi
di
ng
a
10-di
gi
t
number
b
y
a
5-di
gi
t
one)
can,
t
hrough
t
h
e
u
s
e
of
e
x
ter
n
al
sy
m
b
o
l
s,
b
e
r
e
d
u
ced
to
th
e
iter
a
ted
a
p
p
licatio
n
o
f
a
b
ilities
w
h
i
ch
wer
e
selected
for
i
n
our
e
v
ol
ut
i
onary
hi
s
t
ory
(
s
u
ch
as
pat
t
e
rn
mat
c
hi
ng
and
a
s
s
o
ci
at
i
on).
T
hus
t
h
ere
i
s
the
pos
s
i
bility
of
replacing
the
d
if
ficult
t
as
k
o
f
a
bs
tractly
reas
oning
(in
t
he
head)
a
bout
a
n
o
v
e
l
s
i
t
u
at
i
on,
w
i
t
h
abs
t
ract
l
y
reas
oni
ng
(us
i
ng
e
x
t
e
rnal
s
y
mbol
s
)
about
t
h
at
s
a
me
s
i
t
u
at
i
on.
B
u
t
t
he
s
a
me
compl
e
xi
t
y
cons
i
d
erat
i
ons
appl
y
t
o
bot
h;
i
f
i
n
t
e
rnal
reas
oni
ng
about
no
v
e
l
s
i
t
u
at
i
ons
i
s
i
n
t
r
act
abl
e
,
s
o
w
i
l
l
e
x
t
e
rnal
reas
oni
ng
be.

A
l
es
s
d
el
i
b
erat
i
v
e
a
ppl
i
cat
i
o
n
o
f
s
i
t
u
at
ednes
s
w
oul
d
b
e
s
omet
hi
ng
l
i
k
e:
S
uppos
e
w
e
manage
t
o
s
t
ruct
ure
our
w
o
rl
d
i
n
s
uch
a
w
a
y
t
hat
i
t
i
s
h
i
ghl
y
l
i
k
el
y
t
hat
a
no
v
e
l
c
ont
e
x
t
will
be
encountered
only
i
mmediately
a
fter
encountering
t
he
(f
amiliar)
conte
x
ts
to
which
i
t
i
s
m
os
t
s
imilar
.
In
s
u
ch
a
cas
e,
(fragments
of)
t
he
mos
t
recently
acti
v
e
r
outines
(or
t
heir
recombinations
)
w
ill
ha
v
e
a
good
probability
of
being
a
pplicable
in
the
n
e
w
situ
atio
n
.

A
r
elated
idea
is
to
com
put
e
a
n
e
w
rout
i
n
e
b
as
ed
on
e
x
i
s
t
i
n
g
ones
.
F
o
r
e
xampl
e
,
i
f
one’
s
rout
i
n
es
w
e
re
t
h
e
out
come
of
a
p
roces
s
w
i
t
h
a
f
e
w
cont
i
nuous
l
y
-v
aryi
ng
paramet
e
rs
,
and
t
he
s
y
s
t
em
coul
d
r
el
i
a
bl
y
a
dj
us
t
t
hes
e
paramet
e
rs
i
n
t
h
e
l
i
ght
of
changi
ng
cont
e
x
t
in
su
ch
a
w
ay
th
at
a
s
itu
ated
r
o
u
tin
e
a
p
p
r
o
p
r
iate
f
o
r
d
ealin
g
w
ith
th
e
n
e
w
co
n
t
e
x
t
w
a
s
t
he
res
u
l
t
,
t
h
en
perhaps
t
hi
s
c
oul
d
b
e
a
s
o
l
u
t
i
on.
O
r
,
i
f
r
out
i
n
es
can
be
rank
ed
i
n
t
e
rms
o
f
t
he
de
gree
t
o
w
h
i
c
h
t
he
y
a
re
s
i
t
u
at
ed
(roughl
y
,
t
h
e
number
o
f
c
ont
e
x
t
s
i
n
w
h
i
c
h
t
he
y
w
ork),
t
hen
p
erhaps
a

s
i
t
u
at
ednes
s
-reduci
ng”
operat
i
o
n
c
oul
d
b
e
a
ppl
i
e
d
t
o
t
h
e
r
out
i
n
e
t
hat
w
as
bei
n
g
u
s
e
d
i
n
t
he
pri
o
r
c
ont
e
x
t
,
yi
el
di
ng
one
of
s
l
i
ght
l
y
great
er
abs
t
ract
i
on,
enough
t
o
w
o
rk
i
n
t
h
e
n
e
w
s
i
t
u
at
i
on.
S
o
me
connect
i
oni
s
t
archi
t
ect
ures
s
eem
t
o
i
m
pl
ement
s
uch

pers
pect
i
v
e-dependence
r
educi
ng”
operat
i
ons
[11,16].
O
f
cours
e
,
t
here
s
houl
d
b
e
c
ompl
ement
a
ry
“s
i
t
u
at
ednes
s
-
i
n
creas
i
ng”
operat
i
ons
,
w
hi
ch
w
oul
d
a
l
l
o
w
one
t
o
maxi
mi
s
e
one’
s
of

o
adi
n
g
ont
o
t
he
en
vi
ronment
.
6.
I
n
ou
r
o
w
n
(b
od
y)
i
m
age?
Much
dis
c
us
s
i
on
of
EAI
centres
on
ho
w
i
mportant
embodiment
is
to
human
cognition,
and
t
hi
s
i
s
a
ppropri
a
t
e
l
y
refl
ect
ed
i
n
A
nders
on’
s
r
e
v
i
e
w
.
Lak
o
f
f
and
J
ohns
on
[32]
are
probabl
y
r
i
ght
t
h
at
t
h
e
n
at
ure
o
f
m
an
y
,
mos
t
or
al
l
o
f
our
concept
s
are
h
i
ghl
y
c
ons
t
r
ai
ned
b
y
our
modes
o
f
p
ercept
i
o
n
a
nd
act
i
on,
as
w
e
l
l
as
t
h
e
d
et
ai
l
s
of
our
neurophys
i
o
l
ogy
,
hormone
s
y
s
t
ems
,
et
c.
F
o
l
l
o
w
i
ng
t
h
e
d
i
s
cus
s
i
on
abo
v
e,
t
h
i
s
s
ugges
t
s
e
i
t
h
er:
1.
b
u
i
l
d
i
n
g
s
ys
t
e
ms
w
hos
e
c
oncept
u
al
repert
oi
res
a
re
cons
t
r
ai
ned
b
y
t
hei
r
non-concept
u
al
s
ubs
t
r
at
e
i
s
a
no
v
e
l
a
nd
promi
s
i
n
g
w
ay
t
o
do
A
I
;
o
r
2.
i
t
i
s
by
vi
rt
ue
of
ha
vi
ng
t
h
e
non-concept
u
al
s
ubs
t
r
at
e
t
hat
t
he
y
d
o
t
hat
i
nt
ent
i
onal
s
y
s
t
ems
,
be
t
h
e
y
nat
u
ral
(
w
i
t
h
ani
m
al
bodi
es
)
o
r
a
rt
i

ci
al
and
s
ymbol
i
c
,
a
re
abl
e
t
o
d
e
m
o
n
s
tr
ate
a
n
y
in
tellig
en
ce
o
r
in
ten
tio
n
a
lity
at
all.
146
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
B
u
t
s
ome
E
A
I
di
s
c
ours
e
goes
f
urt
h
er
,
c
l
a
i
m
i
n
g
t
hat
t
he
dependence
o
f
our
human
concepts
on
our
human
embodiment
demons
trates
the
n
eces
s
ity
of
a
body
,
e
v
e
n
a
human
one,
f
or
concepts
or
intelligence.
W
ithout
at
leas
t
one
further
p
remis
e
,
t
his
i
s
a
non-
seq
u
itu
r
;
i
s
t
here
an
ar
gument
t
o
s
upport
t
his
s
tronger
c
laim
concerning
the
n
eces
s
ity
of
a
rob
u
s
t
form
of
embodi
ment
for
A
I?
I
g
et
t
h
e
i
mpres
s
i
on
t
h
at
s
o
me
peopl
e
w
ho
es
pous
e
t
hi
s
f
orm
o
f
S
t
r
ong
EA
I
d
o
s
o
becaus
e
of
a
(
mis
t
ak
en,
i
n
m
y
v
ie
w)
belief
t
hat
s
omeone,
s
ome
w
here
has
a
lready
es
t
a
bl
i
s
hed
i
t
a
s
f
act
.
I
f
t
hei
r
bel
i
e
fs
are
m
ore
s
peci

c
t
h
an
t
h
at
,
t
hen,
t
ypi
cal
l
y
,
t
he
s
o
meone
i
s
J
ohn
S
earl
e
and
t
he
s
o
me
w
h
ere
i
s
t
he
C
h
i
n
es
e
r
oom
ar
gument
[
40].
T
he
y
tak
e
Searle
to
ha
v
e
s
h
o
w
n
t
hat
unders
tanding/cons
cious
nes
s
/intelligence
is
a
b
iological,
not
comput
at
i
onal
,
phenomenon.
N
e
v
e
r
m
i
n
d
t
hat
h
e
h
as
n’
t
s
ho
w
n
t
h
at
;
t
hat
t
he
bi
ol
ogi
cal
s
ource
he
has
i
n
m
i
n
d
i
s
onl
y
t
he
brai
n,
not
t
h
e
body;
and
t
hat
h
i
s
ar
gument
c
oncedes
t
h
at
a
non-embodi
ed
s
y
s
t
em
(e.g.,
t
h
e
C
hi
nes
e
room
i
t
s
el
f)
can
pas
s
t
h
e
T
uri
n
g
t
es
t
.
Als
o
,
a
lthough
pure
b
eha
v
iouris
m
s
hould
b
e
r
es
is
ted,
it
w
ould
b
e
t
he
height
of
political
i
n
correct
nes
s
and
a
nt
hropocent
r
i
s
m
t
o
d
en
y
f
ul
l
s
ubj
ect
hood
t
o
a
s
ys
t
e
m
t
hat
b
eha
v
es
in
tellig
en
tly
simp
ly
b
ecau
se
th
at
b
e
h
a
v
i
o
u
r
is
n
o
t
g
e
n
e
rated
b
y
t
h
e
same
b
i
o
l
o
g
i
cal
m
ech
an
ism
s
as
it
is
in
u
s
.
25
B
u
t
phi
l
o
s
ophi
cal
l
y
as
t
u
t
e
EA
I
p
roponent
s
can
do
bet
t
e
r
.
O
n
e
w
ay
of
ar
gui
ng
for
t
he
str
o
n
g
EAI
c
laim
is
to
in
sist
th
at:
(1)
w
hat
c
ounts
a
s
i
ntelligence
depends
on
what
we
can
unders
tand
or
recognis
e
to
be
in
tellig
en
t;
an
d
(2)
w
e
can
only
unders
tand
or
recognis
e
another
o
r
g
anis
m
t
o
b
e
i
ntelligent
i
f
t
he
y
s
hare
our
“w
ay
of
bei
ng”
(pos
s
i
bl
y
t
he
s
a
me
as
Da
sein
[
2
7
]
?)
o
r
wh
at
W
ittg
en
stein
called
our
“form
o
f
l
i
f
e”:
t
h
e
s
i
t
u
at
i
ons
,
goal
s
,
c
hal
l
e
nges
,
t
a
s
k
s
,
et
c.
t
h
at
mak
e
up
our
e
xperienced
e
x
is
tence;
(3)
t
w
o
or
gani
s
m
s
can
s
h
are
a
w
a
y
o
f
b
ei
ng
or
form
of
l
i
f
e
onl
y
i
ns
of
ar
as
t
h
e
y
are
phys
i
cal
l
y
(i
.e.,
bodi
l
y
)
s
i
m
i
l
a
r
.
26
A
f
ul
l
a
nal
y
s
i
s
o
f
t
hi
s
a
r
gument
can’
t
be
gi
v
e
n
h
ere,
b
u
t
I

l
l
m
ak
e
t
hree
remarks
:

Pr
em
ise
1
h
a
s
a
ll
th
e
s
tr
en
g
t
h
s
an
d
w
eak
n
e
sses
o
f
a
v
e
r
i
ficatio
n
i
st/an
ti-
r
ealist
m
eta-
phys
ical
pos
ition;
25
W
h
er
e

us
”,
chillingly
,
r
e
f
e
r
s
to
thos
e
i
n
p
o
w
er
.
W
hen
p
eople
s
ay
the
y
w
ould
d
en
y
a
s
c
r
i
ptions
of
m
e
ntality
,
cons
cious
nes
s
,
etc.
to
a
i
ntelligently-behaving
r
obot
becaus
e
it
is
not
b
u
ilt
the
s
am
e
w
ay
the
y
are,
I
f
ear
a
d
ay
when
that
philos
ophy
is
us
ed
to
den
y
m
e
m
y
o
w
n
s
ubjecti
v
ity
on
the
g
rounds
that
m
y
brain
i
s
d
if
ferent
from
t
he
brains
of
the
p
o
w
ers
t
hat
b
e!
L
e
s
t
one
think
t
his
pure
f
antas
y
,
c
ons
ider
that
this
form
of
reas
oning
already
f
orced
Anders
on
to
undertak
e
a
defens
e
o
f
t
he
intentional
p
roperties
o
f
t
he
handicapped.
T
h
is
is
a
g
rotes
que
s
ituation
t
o
be
in.
26
Cf
.
,
e.
g.
,
[
19]
.
T
he
s
i
m
ilar
ity
m
i
ght
ha
v
e
to
be
v
e
r
y
clos
e
i
ndeed,
accor
d
ing
t
o
W
ittgens
t
ein
(
“I
f
a
lion
c
ould
talk,
w
e
w
ould
not
unders
tand
him

[47,
II,
xi,
p
.
223])
a
nd
Nagel
(
“W
hat
i
s
i
t
lik
e
t
o
b
e
a
b
at?”
[37]).
It’
s
true
that
for
W
ittgens
t
ein,
m
e
re
s
a
m
e
nes
s
of
beha
viour
is
enough:
“O
nly
o
f
a
li
ving
hum
an
being
a
nd
w
h
at
res
e
m
b
les
(
behaves
lik
e
)
a
li
ving
hum
an
being
can
one
s
a
y:
it
has
s
ens
a
tions
;
i
t
s
ees
;
i
s
b
lind;
hears
;
is
deaf;
i
s
c
ons
cious
or
uncons
cious
.

[47,
Section
281].
B
ut
ho
w
s
im
ilar
can
the
b
eha
v
iour
of
the
C
hines
e
room
be
to
m
y
beha
viour
,
gi
v
e
n
t
hat
I
ha
v
e
a
body
and
i
t
does
not?
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
147

One
c
ould
accept
p
remis
e
3,
yet
i
ns
is
t
t
hat
phys
ical
s
i
milarity
can
be
meas
ured
in
more
abs
t
ract
t
e
rms

funct
i
onal
s
i
m
i
l
a
ri
t
y
,
s
ay
,
w
hi
ch
need
not
requi
re
s
a
menes
s
of
embodi
ment
;

It
w
oul
d
b
eho
v
e
t
h
e
E
A
I
communi
t
y
t
o
de
v
e
l
o
p
a
nd
refi
n
e
t
he
abo
v
e
ar
gument
,
e
xpl
or
-
i
n
g
t
he
man
y
pos
s
i
bl
e
i
nt
ermedi
at
e
pos
i
t
i
ons
bet
w
een
i
m
pl
ement
a
t
i
on-i
ndependence
and
c
hauvi
ni
s
m
,
l
aps
i
ng
i
n
t
o
neural
reduct
i
oni
s
m
concerni
ng
ment
al
s
t
at
es
onl
y
a
s
a
la
st
r
e
so
r
t
.
7
.
Ra
dica
l
e
mbo
diment
If
A
nders
on’
s
r
e
v
i
e
w
i
s
a
n
y
t
h
i
n
g
t
o
g
o
b
y
,
t
h
en
I
t
hi
nk
EA
I
p
roponent
s
unders
t
a
t
e
t
h
e
potential
i
mpact
that
embodiment
may
h
a
v
e
o
n
a
rtificial
intelligence
and
c
omputation
i
n
general
.
Ev
en
i
f
EA
I
i
s
a
n
e
nt
i
r
el
y
n
e
w
w
a
y
o
f
doi
ng
A
I
,
I
t
h
i
n
k
t
here
are
t
w
o
w
a
ys
i
n
w
h
i
c
h
E
A
I
can
be
more
radi
cal
t
h
an
addi
ng
ne
w
m
et
hodol
ogi
es
t
o
t
h
e
t
ool
box.
7.1.
C
oncept
ual
,
not
ont
i
c
,
n
o
vel
t
y
One
o
f
t
hes
e
w
a
ys
has
b
een
a
r
ecurring
theme
i
n
t
he
preceding
d
is
cus
s
i
on.
Specifically
,
th
e
p
r
o
m
i
se
o
f
EAI
i
s
n
o
t
ju
st
th
e
p
r
o
m
i
se
o
f
a
k
in
d
o
f
A
I
t
h
a
t
i
s
a
n
a
lter
n
ati
v
e
t
o
GOF
AI
,
b
u
t
t
he
promi
s
e
o
f
p
ro
vi
di
ng
t
h
e
c
oncept
u
al
t
ool
s
n
eces
s
a
ry
for
e
xpl
ai
ni
ng
man
y
or
mos
t
of
th
e
s
u
ccesses
o
f
t
r
a
d
itio
n
a
l,
sy
m
b
o
lic
co
m
p
u
t
atio
n
a
s
w
ell.
On
th
is
v
i
e
w
,
w
h
a
t
m
ak
es
e
v
en
m
u
ch
o
f
tr
ad
itio
n
a
l,
ar
ch
-
s
y
m
b
o
lic
AI
wo
rk
(w
hen
i
t
does
!
)
i
s
b
es
t
unders
t
ood
i
n
t
e
rms
o
f
t
h
e
w
ay
s
u
ch
s
y
s
t
ems
a
re
s
i
t
u
at
ed
i
n
t
h
e
phys
i
cal
and
s
oci
a
l
w
orl
d
,
t
hei
r
modes
o
f
act
i
v
i
t
y
,
and
t
hei
r
phys
i
cal
mani
fes
t
at
i
on.
R
e
s
t
ri
ct
i
n
g
ours
e
l
v
es
t
o
l
ooki
ng
onl
y
a
t
t
he
i
n
t
e
rnal
,
abs
t
ract
propert
i
e
s
o
f
c
omput
at
i
onal
s
ys
t
e
ms
,
a
l
t
hough
a
u
s
e
ful
f
ocus
i
n
s
o
me
s
i
t
u
at
i
ons
,
i
s
i
n
general
a
bl
i
ndi
ng
hi
ndrance
t
o
ful
l
e
xpl
anat
i
o
n
a
nd
des
i
gn.
O
n
t
h
i
s
vi
e
w
,
e
mbodi
ed
(and
s
i
t
u
at
ed)
c
omput
at
i
onal
s
ys
t
e
ms
aren’
t
s
o
much
o
n
tica
lly
n
ove
l
a
s
conceptually
n
ove
l
:
t
h
e
y

v
e
b
een
around
for
a
ges
,
b
u
t
w
e
j
us
t
d
i
dn’
t
kno
w
i
t
.
Seein
g
th
e
c
o
n
t
r
i
b
u
tio
n
o
f
E
AI
in
th
is
w
a
y
a
llo
ws
o
n
e
to
an
swer
th
e
q
u
e
stio
n
s
th
at
I
s
o
met
i
m
es
get
w
hen
t
al
ki
ng
about
EA
I
o
r
s
i
t
u
at
ed
robot
i
c
s
w
i
t
h
s
o
me
A
I
res
earchers
.
“Is
n

t
al
l
A
I
w
ork
e
mbodi
ed
i
n
s
o
me
s
e
ns
e?”
t
h
e
y
as
k.

A
ren’
t
a
l
l
robot
s
s
i
t
u
at
ed?”.
Y
e
s
,
and
y
es
;
b
u
t
onl
y
a
n
A
I
t
heory
w
hi
ch
has
b
een
enri
ched
by
s
ubs
t
a
nt
i
v
e
c
oncept
s
of
embodi
ment
and
s
i
t
u
at
ednes
s
goes
b
e
yond
t
h
i
s
f
act
by
ac
know
l
e
dgi
ng
that
thes
e
a
s
p
ects
o
f
t
he
s
y
s
t
em
are
(oft
en)
e
s
s
e
nt
i
a
l
t
o
t
hei
r
e
xpl
anat
i
o
n
a
nd
des
i
gn,
and
u
s
i
ng
s
a
i
d
concept
s
t
o
t
hos
e
e
nds
.
A
n
y
ontic
ubiquity
of
embodiment
and
s
ituatednes
s
in
the
a
rea
o
f
i
ntelligent
s
ys
tems
s
hould
mak
e
us
s
u
s
p
ect
t
h
at
t
h
e
y
s
houl
d
b
e
m
ade
c
oncept
u
al
l
y
ubi
qui
t
ous
as
w
e
l
l
.
7
.
2
.
I
n
ter
a
ctivist
e
mp
iricism
A
n
e
v
en
more
radi
cal
rol
e
for
E
A
I
i
m
pact
s
o
n
t
he
de
v
e
l
opment
o
f
our
o
w
n
c
oncept
s
and
unders
tanding
of
computation
a
nd
the
m
ind.
Much
of
recent
c
ogniti
v
e
s
c
ience
h
as
emphas
i
s
e
d
t
he
rol
e
of
act
i
on,
percept
i
o
n
a
nd
e
xperi
ence,
as
oppos
ed
t
o
di
s
e
mbodi
ed
i
n
ference
a
nd
reas
oni
ng,
i
n
human
cogni
t
i
on.
B
u
t
s
i
n
ce
cogni
t
i
v
e
s
c
i
e
nt
i
s
t
s
and
A
I
148
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
res
earchers
a
re
thems
e
lv
es
humans
,
27
it
s
t
ands
to
reas
on
that
concept
l
earning
and
de
v
e
l
opment
,
as
w
e
l
l
as
t
h
eory
change,
i
n
c
ogni
t
i
v
e
s
c
i
e
nce
a
nd
A
I
s
houl
d
e
xpl
oi
t
t
he
e
xperiential
a
s
p
ect
of
cognition
w
hen
pos
s
i
ble.
Firs
t,
one
s
hould
ackno
wledge
that
the
goal
o
f
c
ogni
t
i
v
e
s
c
i
e
nce
i
s
a
n
e
xpl
anat
i
o
n
fo
r
e
xperi
enci
ng
agent
s
,
not
(pri
mari
l
y
)
t
he
appearance
of
a
s
et
of
marks
o
n
p
aper
i
n
s
o
me
j
ournal
.
S
econd,
one
can
as
k
w
hat
i
s
r
equi
red
for
s
uch
unders
t
a
ndi
ng;
i
t
appears
t
hat
s
ci
ence
i
n
general
h
as
been
o
v
e
rl
y
p
reoccupi
ed
w
i
t
h
t
h
eori
es
.
T
heori
e
s
doubt
l
e
s
s
l
y
p
l
a
y
a
cruci
a
l
r
ol
e,
b
u
t
t
here
may
b
e
m
odes
o
f
unders
t
a
ndi
ng
wh
ich
o
n
l
y
a
lter
n
ati
v
e
f
o
r
m
s
o
f
e
x
p
l
an
atio
n
,
su
ch
as
(
d
esig
n
o
f
a
n
d
in
ter
actio
n
w
ith
)
r
eal-
w
o
rld
A
I
a
rtef
acts
a
nd
implemented
v
irtual
machines
,
can
pro
v
ide.
Ev
en
if
all
f
orms
of
unders
tanding
can,
i
n
s
ome
s
ens
e
,
a
nd
in
principle,
be
written
d
o
w
n
(
a
c
onces
s
i
on
I
a
m
not
actu
a
lly
p
r
ep
ar
ed
to
m
a
k
e
)
,
it
still
seem
s
t
h
a
t
w
r
itin
g
t
h
e
m
d
o
w
n
i
s
n
o
t
al
w
a
y
s
an
ad
eq
u
a
te
means
o
f
t
rans
mitting
t
hat
unders
tanding.
Thus
,
i
t
m
ay
be
hel
p
ful
,
or
e
v
en
neces
s
a
ry
,
f
or
di
s
c
o
v
e
ri
ng
t
h
e
c
oncept
s
w
h
i
c
h
w
e
w
i
l
l
need
for
a
n
unders
tanding
of
intelligence
(and
therefore
w
hich
we
will
benefit
from
o
r
need
for
des
i
gni
ng
in
tellig
en
t
s
y
s
tem
s
)
,
th
at
we
in
ter
act
with
o
r
p
e
r
s
o
n
a
lly
tak
e
p
a
r
t
in
th
e
c
r
eatio
n
o
f
actu
a
l,
w
o
r
k
in
g
A
I
s
y
s
tem
s
.
T
o
c
o
m
e
u
p
w
ith
,
p
er
h
a
p
s
e
v
en
to
g
r
asp
,
t
h
es
e
c
oncept
s
may
r
equi
re
v
a
ri
ous
forms
o
f
e
xperi
ence
w
i
t
h
a
w
orki
ng
s
y
s
t
em,
m
i
ght
require
getting
one’
s
hands
dirty
,
rather
than
s
i
mply
reading
a
bout
the
s
ys
tem
(
or
e
v
en
about
e
xperi
ences
w
i
t
h
t
h
e
s
ys
t
e
m,
w
e
re
t
h
e
y
e
v
er
publ
i
s
hed)
i
n
a
j
ournal
a
rt
i
c
l
e
.
I
f
t
hi
s
i
s
s
o
,
i
t
m
ak
es
cent
r
al
f
act
ors
w
hi
ch
are
o
ft
en
t
hought
t
o
be
of
mar
g
i
n
al
i
n
t
e
res
t
t
o
A
I
(or
o
f
in
ter
e
st
f
o
r
s
o
m
e
o
th
er
r
easo
n
)
,
s
u
c
h
a
s
t
h
e
r
u
n
tim
e
d
etails
o
f
th
e
m
o
d
e
l/sim
u
l
atio
n
,
its
interf
ace/graphical
dis
p
lay
,
etc.
T
w
o
A
I
s
ys
tems
that
are
i
dentical
in
their
b
eha
v
iour
,
b
ut
di
f
f
er
i
n
,
e
.g.,
t
h
e
G
U
I
w
h
i
c
h
a
l
l
o
w
s
one
t
o
pert
urb
a
nd
obs
erv
e
t
h
e
s
t
a
t
e
s
o
f
t
hos
e
s
ys
t
e
ms
,
may
h
a
v
e
d
ras
t
i
cal
l
y
di
f
f
erent
i
mpact
s
o
n
t
he
unders
t
a
ndi
ng
and
c
oncept
u
al
de
v
e
l
opment
th
at
th
e
y
p
r
o
v
id
e.
28
Perhaps
h
ere
(
finally)
w
e
h
a
v
e
a
reas
on
why
A
I
m
us
t
b
e
m
ade
i
n
our
o
w
n
i
mage
(cf.
Section
6
):
B
ecaus
e
that
is
the
only
w
ay
that
we
will
be
able
to
gras
p
a
nd
refi
n
e
t
he
concept
s
neces
s
a
ry
for
A
I
d
e
v
el
opment
.
If
t
h
i
s
i
s
ri
ght
,
g
i
v
i
n
g
our
A
I
s
y
s
t
ems
a
r
ob
us
t
f
orm
o
f
e
mbodi
ment
may
h
a
v
e
a
s
m
uch
t
o
d
o
w
i
t
h
de
v
e
l
opi
ng
our
o
w
n
m
ent
a
l
ab
ilities
a
s
i
t
d
o
e
s
w
ith
d
e
v
e
lo
p
i
n
g
th
eir
s
.
Acknowledg
e
ment
s
I’
d
l
i
k
e
t
o
t
hank
D
a
vi
d
B
rooks
,
D
a
v
e
G
urnel
l
,
D
ean
P
e
t
t
e
rs
and
A
aron
S
l
oman,
a
l
l
members
o
f
t
he
C
ogA
f
f
res
earch
groupi
ng
of
t
h
e
S
chool
of
C
o
mput
er
S
c
i
e
nce
a
t
t
he
U
n
i
v
ers
i
t
y
of
B
i
rmi
ngham,
for
t
hei
r
hel
p
ful
c
omment
s
.
Thi
s
w
o
rk
w
a
s
s
upport
e
d
b
y
a
R
e
s
earch
F
e
l
l
o
w
s
hi
p
f
unded
b
y
g
rant
F
/
94/
B
W
from
t
he
Le
v
e
rhul
me
T
r
us
t
.
27
U
s
ually
.
28
Man
y
of
Douris
h’
s
[
18]
cons
iderations
,
a
nd
thus
Section
6
of
Anders
on’
s

eld
guide
[4],
becom
e
v
e
ry
germ
ane
h
ere,
though
perhaps
not
in
a
w
ay
either
anticipated.
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
149
Refer
e
nces
[1]
P
.
A
gre,
Com
putation
a
nd
Hum
a
n
E
xperience,
Cam
b
ridge
Uni
v
ers
ity
Pres
s
,
Cam
b
ridge,
1997.
[
2
]
P
.
A
gr
e,
D
.
Chapm
a
n,
P
e
ngi:
A
n
i
m
p
lem
e
ntation
o
f
a
theor
y
of
acti
v
ity,
i
n:
P
r
oceedings
of
the
S
ixth
N
a
tional
Conference
on
Artificial
Intelligence
(AAAI-87),
S
eattle,
W
A,
AAAI
Pres
s
,
1987,
pp.
268–272.
[3]
A
.
P
.
A
m
b
ler,
H
.
G
.
Barro
w
,
C.
M.
Bro
w
n,
R.
M.
Burs
tall,
R.
J
.
P
opples
tone,
A
v
e
rs
atile
com
puter
-controlled
as
s
e
m
b
ly
s
y
s
t
em
,
i
n:
T
h
e
P
r
o
ceedings
of
the
T
hir
d
I
n
ter
n
ational
J
oint
Conf
er
ence
on
A
r
tifi
cial
I
n
telligence,
(IJ
CAI-73),
S
tanford,
CA,
1973,
pp.
298–307.
[4]
M
.
A
nders
on,
E
m
bodied
Cognition:
A

eld
guide,
A
rtifi
cial
Intelligence
149
(1)
(
2003)
91–130,
this
is
s
u
e.
[
5
]
P
.
A
r
m
er
,
A
ttitudes
t
o
w
ar
ds
intelligent
m
achines
,
i
n:
R.
Chr
i
s
l
e
y
(
E
d.
)
,
A
r
tifi
cial
I
n
telligence:
Cr
itical
Concepts
,
V
ol.
I
V,
Routledge,
L
ondon,
2000,
pp.
325–342,
Originally
publis
hed
i
n
1962.
[6]
R
.
B
eer,
A
dynam
i
cal
s
y
s
t
em
s
p
ers
p
ecti
v
e
o
n
a
gent-en
v
ironm
ent
i
nteraction,
in:
R
.
C
hris
le
y
(
E
d
.
)
,
A
rtificial
Intelligence:
Critical
Concepts
,
V
ol.
III,
Routledge,
L
ondon,
2000,
pp.
210–255,
O
r
iginally
publis
hed
i
n
1995.
[7]
C
.
B
reazeal,
B
.
S
cas
s
e
llati,
Inf
a
nt-lik
e
s
ocial
i
nteractions
betw
een
a
r
obot
and
a
hum
an
care
g
i
v
er,
A
dapti
v
e
Beha
vior
8
(
2000)
49–74.
[8]
R
.
B
rooks
,
I
ntelligence
w
ithout
repres
entation,
in:
D
.
K
irs
h
(E
d.
),
F
oundations
of
A
r
tifi
cial
Intelligence,
MIT
P
res
s
,
Cam
b
ridge,
M
A,
1992,
pp.
139–159.
[9]
L
.
C
arroll,
W
h
at
the
t
ortois
e
s
aid
t
o
A
chilles
,
Mind
4
(
1895)
278–280.
[10]
C.
Cherniak,
M
inim
al
Rationality,
M
IT
P
r
es
s
,
Cam
b
ridge,
M
A
,
1986.
[11]
R.
Chris
l
e
y
,
A
.
H
olland,
Connectionis
t
s
ynthetic
epis
tem
o
logy:
Requirem
e
nts
f
or
the
d
e
v
elopm
ent
o
f
objecti
v
ity,
i
n:
L
.
N
i
klas
s
on,
M.
Bodén
(
E
d
s
.
),
Current
T
r
ends
in
Connectionis
m
:
P
roceedings
of
the
1995
S
w
edis
h
C
onference
on
Connectionis
m
,
L
a
w
rence
E
rlbaum
,
H
ills
dale,
N
J
,
1995,
pp.
283–309.
[12]
R.
Chris
l
e
y
,
T
aking
e
m
bodim
e
nt
s
e
rious
ly:
N
on-conceptual
content
a
nd
robotics
,
in:
A
ndroid
E
pis
t
em
ology,
MIT
P
res
s
,
Cam
b
ridge,
M
A,
1995,
pp.
141–166.
[13]
R.
Chris
l
e
y
,
T
.
Z
iem
k
e,
E
m
bodim
e
nt,
i
n:
L
.
N
a
del
(
E
d
.
)
,
E
nc
yclopedia
o
f
C
ogniti
v
e
S
c
ience,
Macm
illan,
L
ondon,
2002.
[14]
W
.
Clance
y,
Situated
action:
A
n
europs
ychological
interpretation
r
es
pons
e
t
o
V
era
a
nd
Sim
on,
Cogniti
v
e
Sci.
17
(1993)
87–116.
[15]
A.
Clark,
Being
T
here,
M
IT
Pres
s
,
Cam
b
ridge,
M
A,
1996.
[16]
A.
Cus
s
i
ns
,
T
he
connectionis
t
cons
truction
o
f
c
oncepts
,
i
n:
M.
Boden
(
E
d
.
)
,
T
he
Philos
ophy
of
Artificial
Intelligence,
O
x
ford
U
n
i
v
ers
ity
P
r
es
s
,
O
x
ford,
1990,
pp.
368–440.
[17]
D.
Dennett,
A
c
ure
f
or
the
c
om
m
o
n
c
ode?,
in:
D
.
D
ennett,
Brains
torm
s
,
Harv
es
ter
P
res
s
,
Brighton,
1978,
pp.
90–108.
[18]
P
.
Douris
h,
W
h
ere
t
he
Action
I
s
:
T
h
e
F
oundations
of
E
m
bodied
Interaction,
MIT
P
res
s
,
Cam
b
ridge,
M
A,
2001.
[19]
H
.
D
r
e
y
fus
,
Introduction
t
o
t
he
MIT
P
res
s
edition
o
f

W
h
at
Com
puters
C
an’
t
D
o
”,
in:
R
.
C
hris
le
y
(
E
d
.
)
,
A
r
tifi
cial
Intelligence:
Critical
Concepts
,
V
ol.
III,
Routledge,
L
ondon,
2000,
pp.
459–493,
O
r
iginally
publis
hed
i
n
1992.
[20]
J
.
F
odor,
P
s
y
chos
em
antics
:
T
h
e
P
roblem
of
Meaning
i
n
t
he
Philos
ophy
of
Mind,
MIT
P
res
s
,
Cam
b
ridge,
MA,
1987.
[21]
N
.
G
oodm
an,
F
act,
F
iction,
and
F
orecas
t,
H
a
rv
ard
U
ni
v
e
rs
ity
P
r
es
s
,
Cam
b
ridge,
M
A
,
1954.
[22]
Y
.
Gunther
(
E
d
.
)
,
E
s
s
a
ys
on
Nonconceptual
Content,
MIT
P
res
s
,
Cam
b
ridge,
M
A,
2003.
[23]
S.
Harnad,
T
he
s
y
m
bol
grounding
problem
,
P
hys
.
D
42
(1990)
335–346.
[24]
I.
Harv
e
y
,
U
ntim
ed
and
m
is
repres
ented:
Connectionis
m
and
t
he
com
puter
m
e
taphor
,
C
ogniti
v
e
Science
Res
earch
P
a
per
C
SRP
245,
School
of
Cogniti
v
e
and
C
om
puting
S
ciences
,
U
ni
v
e
rs
ity
of
Sus
s
e
x,
Brighton,
No
v
e
m
b
er
1992.
[25]
J
.
H
a
ugeland,
A
r
tifi
cial
Intelligence:
T
h
e
V
ery
I
dea,
MIT
P
res
s
,
Cam
b
ridge,
M
A
,
1985.
[26]
P
.
Hayes
,
K.
F
o
rd,
N
.
A
gne
w,
On
babies
and
b
athw
ater:
A
cautionary
tale,
A
I
M
agazine
14
(1994)
15–26.
[27]
M.
Heide
gger
,
Being
a
nd
T
i
m
e
,
H
arper
a
nd
Ro
w,
Ne
w
Y
ork,
1927.
[28]
I.
H
o
rs
w
ill,
Characterizing
a
daptation
b
y
c
ons
traint,
i
n:
T
o
w
a
rd
a
P
ractice
o
f
A
utonom
ous
S
y
s
t
em
s
:
P
r
oceedings
of
the
F
irs
t
E
u
ropean
Conference
on
A
r
tifi
cial
L
i
fe,
M
IT
P
r
es
s
,
Cam
b
ridge,
M
A
,
1992,
pp.
58–
64.
150
R
.
Chr
i
s
l
e
y
/
A
r
tificial
Intellig
ence
149
(2003)
131–150
[29]
D.
Hum
e
,
A
n
E
nquiry
Concerning
Hum
a
n
U
nders
tanding,
Oxford
Uni
v
ers
ity
Pres
s
,
Oxford,
1777.
E
d
ited
by
Nidditch
a
nd
Selby-Bigge.
[30]
E
.
H
u
tchins
,
C
ognition
i
n
t
he
W
ild,
M
IT
P
r
es
s
,
Cam
b
ridge,
M
A
,
1995.
[31]
F
.
K
e
ijzer,
R
epres
e
ntation
a
nd
Beha
vior,
M
IT
Pres
s
,
Cam
b
ridge,
M
A,
2001.
[32]
G.
L
a
k
o
f
f
,
M
.
J
ohns
on
(E
ds
.
)
,
P
hilos
ophy
in
the
F
les
h
:
T
he
E
m
bodied
Mind
and
its
Challenge
to
W
e
s
t
ern
T
hought,
B
as
ic
Books
,
N
e
w
Y
o
rk,
1999.
[33]
D.
L
e
nat,
E
.
A.
Feigenbaum
,
O
n
t
he
thres
holds
of
kno
wledge,
i
n:
D.
Kirs
h
(
E
d
.
)
,
F
oundations
of
Artificial
Intelligence,
MIT
P
res
s
,
Cam
b
ridge,
M
A
,
1992,
pp.
185–250.
[34]
H
.
Maturana,
F
.
V
arela,
A
u
topoies
is
and
C
ognition:
T
h
e
R
ealization
o
f
t
he
L
i
ving,
in:
B
os
ton
S
tudies
in
the
Philos
ophy
of
Science,
V
o
l.
42,
D.
Reidel,
D
ordrecht,
1980.
[35]
D.
McDerm
ott,
W
e

v
e
b
een
fram
e
d:
Or
,
w
hy
AI
is
innocent
o
f
t
he
fram
e
problem
,
i
n:
Z
.
Pylys
hyn
(E
d.
),
T
h
e
Robot’
s
D
ilem
m
a
:
T
he
F
r
am
e
P
roblem
in
A
r
tifi
cial
Intelligence,
A
b
le
x,
N
o
rw
ood,
N
J
,
1987.
[36]
M.
Morris
,
W
h
y
t
here
are
n
o
m
ental
r
epres
e
ntations
,
M
inds
and
M
achines
1
(
1991)
1–30.
[37]
T
.
Nagel,
W
h
at
is
it
lik
e
t
o
b
e
a
bat?,
i
n:
N.
Block
(
E
d
.
)
,
R
eadings
in
Philos
ophy
of
Ps
ychology
,
V
ol.
1
,
Harv
ard
U
ni
v
e
rs
ity
Pres
s
,
Cam
b
ridge,
M
A,
1980.
[
38]
L
.
P
r
yor
,
G
.
C
ollins
,
Cas
s
a
ndr
a:
P
l
anning
w
ith
contingencies
,
T
echnical
Repor
t
41,
I
n
s
titute
f
o
r
t
he
L
ear
ning
Sciences
,
N
orthwes
t
ern
U
ni
v
e
rs
ity
,
E
v
a
ns
ton,
IL
,
1993.
[39]
Z
.
P
y
lys
hyn,
Com
putation
a
nd
Cognition:
T
o
w
a
rds
a
F
oundation
f
or
Cogniti
v
e
S
c
ience,
MIT
P
res
s
,
Cam
b
ridge,
M
A,
1984.
[40]
J
.
Searle,
B
rains
,
m
i
nds
and
p
rogram
s
,
Beha
vioral
and
B
rain
Sciences
3
(
1980)
417–457.
[41]
R.
Sheldrak
e,
T
h
e
P
res
e
nce
o
f
t
he
P
a
s
t
:
M
orphic
R
es
onance
a
nd
the
H
abits
of
Nature,
G
rim
e
s
,
1988.
[42]
H
.
S
i
m
on,
T
h
eories
of
bounded
r
ationality,
i
n:
R.
Radner,
C.
Radner
(
E
d
s
.
),
D
ecis
i
on
and
O
r
g
anization,
North-Holland,
Am
s
t
erdam
,
1972,
pp.
161–176.
[43]
B.
Sm
ith,
T
he
foundations
of
com
puting,
in:
M
.
S
cheutz
(E
d.
),
Com
putationalis
m
:
Ne
w
D
irections
,
M
IT
Pres
s
,
Cam
b
ridge,
M
A,
2002.
[
44]
E
.
T
h
elen,
L
.
S
m
ith,
A
D
ynam
i
c
S
ys
tem
s
A
ppr
oach
to
the
D
e
v
elopm
ent
o
f
C
ognition
a
nd
A
c
tion,
M
I
T
Pres
s
,
Cam
b
ridge,
M
A,
1994.
[45]
A.
V
e
ra,
H
.
S
im
on,
Situated
action:
A
s
ym
bolic
interpretation,
Cogniti
v
e
Sci.
17
(1993)
7–48.
[46]
B.
W
e
bb,
Us
ing
r
obots
t
o
m
odel
a
nim
a
ls
:
A
crick
e
t
t
es
t,
Robotics
a
nd
Autonom
ous
Sys
t
em
s
1
6
(
1995)
117–134.
[47]
L
.
W
ittgens
t
ein,
P
h
ilos
ophical
In
v
e
s
tigations
,
B
lackw
e
ll,
O
x
ford,
1953.
E
d
ited
b
y
G
.
E
.
M
.
A
ns
com
b
e
a
nd
R.
Rhees
;
T
r
a
ns
lated
b
y
G
.
E
.
M
.
A
ns
com
b
e.
[48]
M.
Y
azdani,
B.
W
h
itby,
Building
b
irds
out
of
beer
cans
,
Robotica
5
(1987)
89–92.