►
Description
What do we know about cortical columns? A neuroscience (and artificial intelligence) discussion with Jeff Hawkins of Numenta.
A
A
A
D
A
D
C
A
A
A
D
D
B
D
Is
just
like
a
rough
guideline?
It's
like
you
know.
There
are
more
cell
populations
than
there
are
layers,
so
then
each
layer,
you
can
subdivide
it
and
some
cell
populations
expand
later.
So
it's
like
the
layers
is
more
like
a
rough
guide
and
we
talk
about
them
as
if
they're
really
there.
In
some
cases
they
really
are,
but
it's
more
about
yeah.
This
is
all
time
is
down
than
this
layer.
So
often
people
look
at
layer,
2
3
and
you
don't
see
a
demarcation
within
layer,
2
or
3.
D
I
can
start
with
what
are
sort
of
the
general
big
ideas
that
people
talking
about
columns
and
then
we
can
divide
it
alright.
So
let's
do
that.
So
if
you
look
at
you
look
at
you
know
what
classically
most
people
believe
and
if
you
were
to
go
and
look
at
them
textbook
or
are
beginning
block
on
neurosyphilis.
D
D
Of
corner
there
another
so
the
general
view
goes
like
this
and
then,
when
you
talk
about
exceptions,
okay,
the
general
view
is
that
the
primary
input
layer
is
layer,
four
rights,
and
so
let
us
put
a
four
there.
And
that
means
that
if
you
were,
if
you're
getting
something
from
the
skin
or
from
the
eyes
or
from
the
ears-
and
these
are
primary
regions
of
the
cortex-
that
those
inputs
are
coming
in
too
late
for
and
I'll
just
maybe
I
should
draw
them
in
different
color,
I,
home
baby.
D
B
D
Than
ten
percent
yeah,
the
general
view
is
the
following:
that
information
now
goes
in
the
following
way.
People
say,
oh,
that
it's
information
flows
up
to
layers
to
three
evil,
there's
more
than
2/3
most
off.
This
is
what
you
see
in
most
books.
Yeah
most
introductory
stops
up.
It
goes
the
layer,
2
3,
so
afford
11
to
3
later
to
3,
actually
is
an
output
layer
on
the
column,
so
you
can
think
about
the
Oh.
Already.
You've
only
got
two
steps
and
your
output,
something.
C
D
But
then,
because
it
via
was
the
following
that
lady
layer
to
three
also
projects
I,
probably
didn't
draw
this
very
well
so
layer
5.
So
now
the
information
goes
like
this:
it
goes
down
to
layer,
5
and,
and
then
layer
6
its
so
in
the
classic
new
people
who
don't
really
think
about
it
too
much
because
it's
a
hard
to
look
at,
they
didn't
want
kids.
So
presumably
it
goes
down
to
layer,
6
and
another
output
of
the
column
is
later
fine.
This.
C
D
And
and
and
then
there's
some
other
things
going
on,
for
example,
it
is
generally
assumed
that
these
connections
going
back
to
layer
from
five
delay
or
to
see,
but
these
seem
to
be
non
drivers.
They
soon
be.
You
know
things
and
then
there's
and
then
there's
other
things.
There's
there's
an
in
layer
six.
We
can
do
subdivide
this
into
multiple
layers,
because.
D
D
D
I
didn't
mention
so
we
we
have
focused
on
a
couple
things
just
to
give
it
a
sense.
We
have
focused
in
some
of
our
papers.
We've
talked
about
how
these
two
layers
here.
Well,
we
showed
it.
We
saw
it
in
our
sequence.
Memory
pick,
that's
like
great
wines
and
dines
one
thousand
seven
to
zero.
We
basically
model
with
a
single
layer
of
cells.
So
in
that
paper
we'll
call
it
the
what
we
call
the
the
neuron.
We
call.
D
Paper,
that's
the
2016
paper
right
that
basically
plucked
a
one
layer
of
cells
for
learning
sequences,
it's
kind
of
input,
those
sequences.
We
don't
even
know
we
send
that
in
that
paper,
but
that
course
essentially
is
saying
one
of
those
places
that
could
be
occurring
was
layer
for
one
of
them.
We
also
think
it's
going
on
else
which
it
says
that
was
like
a
sequence
memory.
Then
then
we
did
the
paper
that
came
out
a
year
and
a
half
ago
which
talking
about
how
was
called
you
know.
We
call
the
commas
paper.
D
That's
the
first
time
he
said:
hey
and
there's
the
sequences
there's
you
can
do
it,
you
can
do
sensory
motor
inference
or
sensory
motor
prediction,
and
so
that
basically
said
I
was
the
first
time
we
said.
Okay,
you've
got
a
layer,
6a
you've
got
layer
four
and
you
got
layer,
2
3.
So
that
was
saying
in
that
paper.
6A
was
representing
a
location.
C
D
A
D
There
might
be
everywhere,
but
that
we
jump
this
model
in
Atlanta
for
yeah.
So
here's
in
this
columns
paper,
we
said:
oh
look,
there's
this
back
and
forth
between
6a
and
4
and
projections
of
2/3.
These
we
use
the
same
mechanism
and
therefore
is
in
the
neuron.
We
have
those
many
columns.
You
just
referred
to
it's
the
same
thing
so.
D
Is
the
this
is
now
going
to
represent
the
input
in
a
particular
context?
These
things
change
as
you
move
your
finger
or
moves
your
eyes,
but
this
thing
was
going
to
be
stable,
because
this
was
the
object
yeah.
So
we
did.
We
did
this
temple
pooling
layer
here,
meaning
a
sable
representation
here.
His
map
is
all
these
different
patterns
down
here
are
mapped
to
a
single
pattern
up
here.
D
So
as
you
move
your
finger
over
an
object
or
in
lugares,
these
things
are
changing,
but
this
is
stable
as
long
on
the
same
object
right
and
so
that's
that
was
at
the
columns
paper.
We
include
corporate
three
of
those
and
then
in
the
the
paper
that
the
two
papers
are
came
from,
one
that
came
out
in
December
of
the
problem
to
take
us
and
their
framework.
Sitcoms
puzzles
just
can't
it's
not
even
I,
don't
think
it's
opposed
to
each
other.
It's
the
abstract
is
up
there,
yeah
except
it
yeah.
D
D
Hey
is
later
six
a
it's,
not
just
location,
we're
saying
its
grid
cells
four,
and
we
talked
a
lot
about
how
those
grid
cells
work
and
so
on.
So
we've
been
sort
of
up
to
now.
We've
been
modeling
three,
it's
sort
of
a
three
layers
of
many
of
these
layers
here
and
we
there's
a
lot
more
that
we
haven't
talked
about
yet
in
any
of
our
papers
and
a
lot
more
nuanced
than
what
we
have
here,
that
we
haven't
discussed
ability
publicly
right
so
so
now
we're
at
that
were
starting
to
talk
about.
D
You
know
what
what
things
that
we
talked
about
here
we
also
introduced
in
in
the
framework
a
per
we
introduced.
Another
idea
called
the
displacement
cells
and
we
speculated
that
those
were
mayor
farm,
so
actually
there's
sort
of
a
we've
got
an
in
the
frameworks
paper
we
added
so
another.
One
of
the
lay
of
five
cells
is
sort
of
speculating
on
its
role
as.
D
D
D
B
D
C
A
A
D
About
the
anti
so
now
people
have
started
using
the
term,
sometimes
I
use
layer,
five
thick
tufted
and
the
other
one
is
layer,
five
thin
tufted,
which
is
pretty
bad
because
the
both
two
teas
yeah
yeah
right,
but
that's
just
not
worried
about.
Let's
just
just
talk
about
the
two
different
cell
types
and
one
thing:
okay,
one
of
these
is
is
it
represents
behaviors.
It
goes
out
of
the
cortex
and
it's
it's
like
a
motor
output,
okay
and
every
column.
D
As
far
as
we
know
has
these
cells
and
they
project
subcortical
II,
to
generate
behavior
right.
So
even
the
parts
of
cortex
occasion,
probing
eyes
are
directing
how
to
ice,
to
move,
and
so
there's
a
part
of
the
neocortex
called
motor
cortex,
which
is
really
just
the
part
that
protects
your
spinal
cord,
but
every
part
seems
to
project
some
place
in
those
motor,
even
though
auditory
things
turn
your
head
and
things
like
that.
So
there's
a
motor
output
layer.
D
C
D
B
B
C
D
The
thalamus
everything
everything
that
comes
from
the
sensory
organs
to
the
to
the
cortex
goes
to
the
thalamus
and
when
I'm
on
region,
the
cortex
projects
to
a
high
region
from
layer
five
to
layer
four.
It
also
goes
to
the
thumb.
So
this
is
sort
of
like
this
is
parallel.
The
output,
a
one-column,
something
like
the
output
of
your
retina,
so.
A
D
You
know
every
single
guy
and
part
of
the
thousand
brains
theory
that
we
were
promoting
is
that
every
column
is
a
complete
centric
motor
system
right
every
column
on
its
own.
Is
there
any
sensory
motor
system
right?
So
every
column
has
input
by
should
sixth-century
everyone's
input.
Every
column
has
a
motor
output
every
column,
with
building
models
of
some
part
of
the
world,
based
on
its
input
in
its
motor
output
and
they're.
All
doing
this
everywhere
so
everywhere
should
have
a
motor
output,
because
somehow
it's
impacting
behavior
in
some
way
right.
D
So
that's
one
of
these
legs
will
they
find
the
other
layer,
five
actually
projects,
it
protects
elsewhere
and
it's
also
an
output
layer,
and
let
me
just
put
it
this
way.
It
goes
to
all.
It
goes
to
other
columns
also.
So
let's
call
this
five
eight.
For
the
moment,
the
five
video
would
project
to
five
a
and
other
columns
elsewhere.
It's
more
of
a
lateral
connection
is
like
five
Bay's
connect
to
the
five
eighty
everywhere.
Something
like
that.
D
A
D
Here's
the
really
weird
thing
about
it
is
that
the
two
major
outputs
layer,
two
three
and
let's
say
well
layer:
five.
They
both
can
project
library
like
layer
to
be
three
cells,
can
check
the
layer,
two
three
cells
yeah
and
they
can
a5h
cells
and
particularly
I,
find
ourselves.
Okay,
that's
our
Voting
concept
that
we
that
we've
described
in
these
papers,
but
it's
also.
C
D
D
D
Really
complicated
what
way
to
think
about
this?
The
way
the
way
I've
been
thinking
about
is
just
try
to
understand
what
a
column
does
on
its
own
and
then
you
can
ask.
Why
would
I
send
some
of
these
signals?
Someplace
and
some
of
these
signals
other
places,
and
you
can
understand
that
if
you
start
by
understanding
exactly
what
a
column
does
so
if
later,
two
three
is
sort
of
representing
an
object,
let's
say
well:
I
can
share
that
with
other
layer.
Two
three
is
because
it's
an
object,
yeah
and
we
can
vote.
D
We
could
say
you're
thinking,
you're,
seeing
in
the
cropping
up
I
think
I'm,
seeing
the
we're
singing
the
coffee
now
right.
If,
however,
I
can
also
take
that
object
and
it
can
become
part
of
another
composite
object
and
another
reason
so
I
could
say
to
you
you're
nobody
now
I'm
gonna
talk
to
your
layer,
four
and
your
higher
up
in
the
cortex
I,
say:
Matt
you're,
building
a
more
composite
object,
the
tea
set
and
not
a
reality.
A
D
D
Either
that
or
it
doesn't
have
to
be
brought
a
few
it
just
it
could
be
usually
is,
but
more
it's
just
it's
a
composite.
It's
a
saying
that
this
thing
is
part
of
a
larger
thing
yeah,
and
so,
as
you
go,
if
you
think
about
the
cortex
is
a
hierarchy?
It's
it's
then
you're
gonna
be
building
objects,
compose
of
objects
and
objects.
If
you
think
about
it
as
a
whole,
such
columns
that
are
voting
together
then
they're
all
just
trying
to
agree
on
what
what
they're,
what
they're,
seeing
no.
D
B
D
C
D
D
Cells,
that's
the
thick
tuft
itself.
Okay,
those
are
the
ones
signs
of
movement.
No
one
knows
what
these
are,
but
I
respected.
I
did
what
they
are,
I
mean.
There's
you're
generally
be
talked
into
neuroscientists
they'll,
say
yes,
layer,
five,
a
cells
project;
broadly
they
go
all
over
the
cortex
to
other.
These
are
the
thin
tough
to
tell
yeah.
A
D
Now,
sort
of
waiting
for
we
would
argue
that
it's
a
voting
they're
voting
whatever
it
is
it
representing
you're
voting,
and
that
means
other
columns
that
this
thing
here
whatever
this
is,
is
going
to
be
stable,
volatile
vessel,
calm,
it's
like
it
sounds
like
the
object
is
stabilized.
You
move
your
finger
over
it
yeah.
D
The
stable
I,
don't
know
it
is
I,
don't
know
yeah,
but
I
mean
we
can
go.
We
can
go
down
this
path
because
we
think
the
system
is
we're
not
working
on
the
idea
that
we're
expanding
on
what
we've
already
published
and
it's
more
complicated
and
I
can
explain
where
we're
going
with
that
bad
enough,
I'm
losing
you
or
if
you
are,
if
you
want
to
get
us
something
specific,
okay,.
C
C
A
D
D
This
this,
this
basic
idea
that
you
have
movement,
input
stability,
right,
yeah
movement.
This
is
like
subset
of
movement,
but
location
representation.
You
have
some
sort
of
input.
We
are
not
saying.
Okay,
this
is
input
in
this
context,
and
now
I
can
learn
a
stable
representation
by
by
temple
pulling
over,
though
I
think
there's
two
of
these
going
on
in
each
column.
Yeah.
Let
me
just
redraw
this
picture
of
my
mind.
Yes,.
D
B
D
D
I'll
tell
you
why
I'll
tell
you
why
there's
a
very
least
6
a
cells
have
a
characteristic
look
to
them:
they're
very
the
kind
of
unusual,
the
way
they
were
the
way
their
dendritic
trees
look
and
how
they
make
injections
luck,
projections
late
for
it
and
go
into
them.
If
you
want
to
go
into
that,
but
the
kind
of
unique,
and
so
you
could
kind
of
identify
like
oh
yeah,
lots
of
cells,
they
look
different.
Okay,
6b
cells
look
very
similar
and
they
have
the
same
so
in
these
6h
cells.
D
B
D
D
Well,
well,
I,
there's!
Actually
we
have
a
need
for
doing
this
twice
and
I'll
talk
about
that
in
a
moment
and
then,
but
now
the
question
is
remember:
in
this
example:
the
input
to
to
this
system
right
here
was
going
into
labor
force.
So
we
can
think
of
this,
like
the
sensory
input
right
coming
in
and
driving
is
what's
driving
layer,
fine,
what
is
the
if
this
is
equivalent
to
layer?
Four?
What's
the
driver
to
layer
four
that
the
basic?
D
C
D
C
D
D
But
I'm
not
drawing
those
line
drives,
are
I'm
not
drunk
pretty
much
every
neuron
in
the
neocortex
every
pyramidal
cell,
when
he
makes
an
axon,
it
has
a
local
branch
that
goes
locally
and
then
it
gets
and
it
splits
and
and
often
they
go
someplace
else
too.
So,
just
because
there's
a
cell
here
and
I
can
say,
oh
those
cells
are
driving
there.
Five.
Those
cells
can
also
be
sending
their
output
elsewhere
right
so
same
cell,
the
same
axon
that
goes
laterally
splits
and
sends
it's
a
connection
down
here.
Okay,.
B
D
D
Try
your
problems,
but
also
with
it
so
I'm
like
I'm,
trying
to
stop
thinking
about
sharing
I'm,
just
like
every
single
column,
okay,
right,
I,
think
that's
the
way
to
do
it.
If
you
start
think
about
all
these
interconnections
between
these
columns,
it
gets
very
confusing.
If
you
can
understand
what
a
single
column
does,
then
you
can
explain
why
they
do
these
also.
D
D
There's
two
sensory
motor
systems
here:
right,
there's
this,
which
in
6a
and
for
one
between
six
being
fine
yeah,
they
worked
on
the
same
principle
right
movement.
Location
of
you
know,
though
the
pool
it's
happening
here,
but
in
this
case
is
you
know
this
is
like
a
grid
cell
type
lighter,
and
this
is
a
input,
layer
and
you're
forming
and
you
can
basically
model.
You
know
your
inputs
in
some
location,
space
right.
D
Okay,
it
turns
out
that
there
are
there's
a
need
in
the
brain
and
there's
evidence
that
this
is
happening
for
why
you
have
to
have
two
separate
sense.
You
go
to
inference
logical
systems,
but
it's
it's
a
little
different
than
you
think,
one
of
those
think
about
it's
all
about
movement,
movement
and
predicting
air
input,
movement
and
predicting
input
move
into
predicting
room.
That's
what
it's
all
about!
D
D
There's
another
type
of
movement
you
can
make
and
that's
orientation
right,
I
can
stay
in
one
location
and
I
can
move
left
and
right
up
and
down
my
inputs
changing
as
I
do
that
if
I'm
standing
here,
I,
look
up,
I,
look
down,
I,
look
left,
look
right,
that's
a
different
type
of
sensory
motor
inference
system
and
it
could
work
on
the
exact
same
principles.
What
I
think
is
going
on
my
best
guess
right
now
and
I
give
this.
D
Maybe
an
80%
chance
of
being
correct
is
that
this
system
here
this
one
here
684
to
3,
is
basically
doing
sensory
motor
inference
over
orientation
and
this
system
here
relay
6
and
ir5
is
doing
sensory
motor
inference
over
you
know:
grid
cell
location,
X,
Y,
Z,
X,
Y
type
of
location,
but
this
is
what
we
did.
Actually,
this
is
what
we
described
in
these
papers.
Okay,
why.
B
D
D
We
might
have
mentioned
the
paper,
but
we
didn't
really
understand
how
it's
done.
If
you
look
at
the
hippocampal
complex
and
around
the
cortex,
the
submit
column
and
the
hippocampus
you
find
these
head,
Direction
cells,
the
and
rat,
knows
which
way
tension.
Those
are
very
patient
yeah,
so
you've
got
to
tuck,
I
called
two
metric
spaces.
A
metric
space
is
something
you
can.
You
can
use,
you
can
move
through
and
you
changes
your
C.
A
D
C
D
Teams
like
XY
and
Z,
and
the
other
one
is
a
radio
system
which
is
like
orientation
there.
Those
are
two
basic
sensory
motor
inference.
Systems
that
have
to
be
compass.
I
can
I
can
model
in
my
space
around
me
if
I
move,
my
head
like
this
and
I
can
build
a
model
of
this
room
around
me
right
now,
based
on
one
location,
that
would
be
the
orientation
sensory
motor
system.
Okay
and
then
I
have
a
different
sensory
motor
system,
which
is
when
I
move
through
the
world
to
a
new
location.
D
B
D
Am
on
it
and
that
orientation
my
finger
I
get
a
different
sensation,
but
it's
the
same
location
in
space.
So
there's
a
gentleman.
This
is
the
general
principle
of
movement
that
there
are
these
two.
It's
two
for
your
body
is
true
for
your
fingers.
It's
true
for
your
hand,
your
eyes
that
there
was
these
two
metric
spaces,
there's
a
radio
lecture
space,
linear
magic.
Today's
and-
and
so
you
can
think
about
it.
D
This
way
this
system
is
modeling
the
the
place
you
are
right
now
and
instead
of
the
two
three
being
an
object
but
I
in
this
new
model.
There's
new
thinking
this
would
be
equivalent
to
place
cells
and
then
in
the
hippocampus
you
know
equivalent
to
place
cells,
meaning
this
is
a
stable,
represent
the
cells
and
hippocampus
that
are
stable,
meaning
they
stay
the
same
active
position.
If
the
animals
in
one
location,
independent
of
which
way
the
animals
face,
yeah.
B
D
Like
this
point
in
space,
right
is
an
object
and
it
object
is
independent
in
which
one
I'm
looking
it's
I'm,
defining
this
the
object
as
a
point,
is
surrounded
by
a
bunch
of
things,
and
so
this
is
stable
over
luminance
of
orientation
and
that's
what
I
need
to
project.
That
is
what
I
need
to
protect.
My
input
I
need
to
know
my
orientation
and
my
location,
okay,.
D
A
A
A
D
Okay,
now
I
want
to
learn
the
model,
this
room
and
I'm
going
to
do
that
by
observing
the
room.
So
the
way
we
described
it
in
our
paper
so
far
is
like
okay,
I
would
know
where
I
am
in
the
room.
I
would
have
some
sensory
feature
and
I
can
build
a
room
by
saying
this
century
feature
is
at
this
location
in
the
room
and
I
go.
D
Yeah
one
orientation
environment
ever
sitting
around
right
now
that
people
can
see
is
Lewis
and
Lois.
Would
he
modeled
there
is
when
Lewis
bother
to
this
in
the
same
demo,
yeah
Hearst.
We
didn't
understand
this
and
we
couldn't
get
it
to
work
because
to
predict
the
input
you
needed
to
know
the
orientation.
So
we
cheated,
we
created
sensors
that
were
orientation
was
right,
but
real
sensors
have
an
orientation
to
them.
D
Okay,
so
I
mean
cheating,
is
it
doesn't
sound
like
it
almost
say
it's
bad?
It's
what
you
do
when
you
you
build
up
to
absolutely
so
we
built
a
complete
lock.
We
said:
here's
a
model,
how
many
learning
objects
using
grid
cells
and
input
and
object
representations.
That's
great,
we
said,
might
be
six
eight,
four,
two
three,
but
we
did
that
without
the
company
or
you
can
accommodating
the
orientation
right
we
achieved
by
living
out
orientation.
Now,
if
I'm
going
to
leave
in
orientation,
I
think
it
change
a
sense.
D
I
think
that
what's
happening
here
is
on
modeling
us
a
place.
A
sensory
motor
inference
of
a
place,
which
is
really
the
movements,
are
related
to
orientate.
So
in
this
case,
I
would
say
this
is
my
orientation
at
this
moment
in
time,
which
you
know
like
three
degrees
of
freedom
like
this
and
I'm,
using
it
and
in
particular,
place
I
can
model
what
I'm
going
to
see
depending
on
my
orientation.
So
I
can
literally
it's
almost
inevitable.
D
B
D
Staying
in
one
place
and
I
move
it
in
my
head
in
my
orientation,
I
see
different
things
coming
in
then
I
could
say:
I
could
model
that
place
and
saying
give
it
all.
My
movements
I
recognize
where
I
am
I'm
at
this
place
in
the
world
in
a
particular
room.
It's
like
okay
I'm
at
this
place
in
this
particular
room.
I
know
where
I
am
I
haven't
moved
my
location
of
the
room
I.
D
Let
me
move
my
orientation
now
that
I
know
my
I
have
a
stable
representation
of
my
location,
which
is
really
what
play
cells
are
in
the
hippocampus
does
matter
the
place.
I'll
say
this
is
unique
representation
of
my
location
in
the
world
and
it's
independent
of
my
orientation.
That's
what
place
cells
do
right.
They
just
want
a
place
right
and
they're
independent.
If
the
animal
moves
its
head
place,
L
does
not
change.
Yeah.
B
D
Modeling
the
space
around
a
particular
place
in
the
world
and
the
other
one,
which
is
modeling
structures
based
on
linear
movements
in
the
world,
but
the
input
to
that
is
the
place
of
the
substance,
the
input
that
is
the
stable
representation
case
where
I
am
so.
This
is
this
gets
around
the
problem
that
Lewis
had
where
we
had,
where
the
you
know
where
we
didn't
want
the
input
to
them
to
the
modeling
system,
that
to
change
as
the
orientation
change
so.
D
We
described
is
six
eighty,
four
two
three
before
as
a
circuit,
the
circuit
is
correct,
but
we
but
I
think
what
we
were
using
it
to
model
the
XY.
These
relationships
of
a
traffic
cop
or
some
object,
I
think
that's
actually
going
on
here
and
what
this
circuits
action
modeling
the
same
circuitry
the
same
basic
function
is,
is
place,
and
so
you
have
two
of
these
systems.
One
is
modeling
place,
a
stable
representation
as
I
change,
my
orientation
and
then.
C
D
Can
now
move
through
space,
and
that's
why
you
know
going
forward
left
right
and
I
can
now
build
up
a
model
of
the
object.
So
this
would
be
the
model
of
the
room
where
each
point
in
the
room,
there's
a
stable
representation
of
a
place
and
I
can
break
down
that
place
into
different
predictions
based
on
my
orientation.
It's
a
two-stage,
but.
D
So
one
of
these,
if
I,
if
I,
if
I,
try
staying
warm
position
and
I
just
changed
my
head
direction,
this
one's
changing.
Alright,
if
I
keep
my
head
direction
straight
and
I
just
move
forward,
then
this
one's
changing
right
and
if
I,
do
both
if
I'm
like
walking
forward
and
turn
it
around
like
this
and
then
time
to
both
tenure
yeah.
D
D
And
I
never
would
have
anticipated
this,
but
there
seems
to
be
these
two
basic
needs
that
a
linear
movement
I
mean
a
linear
movement
and
us
radio
movement
I,
starting
out
my
head
around,
why
the
world
is
like
that,
but
it
seems
to
be,
and
we
see
these
type
of
cells
in
the
hippocampal
complex.
We
see
these
two
systems
and
parallel
systems
in
a
cortical
column.
D
You
know
what
I
think
we're
the
only
people
who
ever
try
to
explain
what
these
things
are
doing:
I'm
Hannibal
theories
that
explain
my
686
be
connected
this
one,
and-
and
so
you
now
you
see
this
nice
parallel
system
for
bollocks.
This
completes
the
picture
itself.
It
can
explain
how
you
can
model
rooms
and
objects
with
sensors
that
move
in
the
space
of
those
objects
and
there's
also
orientations
change
within
the
space
of
that
object.
Okay,.
A
D
A
D
A
D
All
right
all
right,
it's
a
lesson,
we're
going
to
add
another
level
of
complexity,
layer,
seven,
here's
a
little
section
of
the
thalamus
I'm
going
into
a
little
section,
that's
associated
with
this
column:
okay,
okay,
shaped
structures
and
but
each
column
is
getting
and
what's
going
on
here
is
that
when,
when
I
draw
when
I,
when
I
say
input
to
this
column,
it
goes
here.
These
relay
cells
in
the
thousands
of
dollars
and
and
then
it
comes
out
the
other
side
and
goes
like
this.
So
in.
D
These
are
real
Excel,
which
are
kind
of
weird,
because
they
kind
of
imply
that
they
don't
do
anything.
It's
like
a
single
axon
comes
in
a
single
axon
comes
our
single
spike
comes
in
a
single
spike,
comes
on.
It's
not
like
that,
but
that's
how
they
talk
about
it,
because
it
does
look
like
that
at
name
time.
We
super
die
in
collaboration
with
the
Karma.
D
B
D
We
have
a
theory
that
we
worked
on,
which
talks
about
how
these
relay
cells
may
be
actually
be
able
to
multiplex
and
move
things
around.
Let's
not
go
there
for
a
moment.
Okay,
there's
another
thing
hung
on
the
thalamus
called
the
TR
n,
which
is
the
phonemic
reticular
nucleus,
which
I'm
going
to
talk
about
abyssion
and
what
happens
here.
This
is
really
interesting
so
from
layer
6a,
these
exact
same
cells
that
are
protective
layer
for
the
exact
same
cells.
They
said
their
axons
appear,
but
they
also
send
their
axons
back
to
the
thumb.
So.
B
D
This
is
representing
orientation.
This
physical
connection
is
well-documented
yeah,
so
our
interpretation,
what
6:8
is
doing
is
novel,
but
the
6a
Thomas
is
very
well-documented.
Now
these
these
things
coming
back.
There
are
numerous
of
these
connections,
so
there's
10
times
as
many
axons
going
this
way
as
there
are
going
this
way
right.
So
this
is.
D
Yeah
but
there's
a
sort
of
massive
feedback,
and
so
this
is
a
very
pointed
one,
and
this
is
more
of
a
diffusive
projection
back
okay,
whenever
we
see
a
lot
of
fibers
going
back,
what
it
tells
you
is
that
the
state
of
this
is
being
projected
down
here.
It's
like
it's
like
the
population
code
matters,
it's
not
like
a
one-on-one
fight.
Well,
here
it's
like
one
input.
One
output,
yeah.
D
D
Not
just
an
orientation
I
believe
it
talk
to
your
very
specific
orientation,
I
think
it's
saying
my
orientation
at
this
point
in
this
part
of
the
world
in
this
particular
object.
That's
the
way
these
cells
work
right
them
just
like
location
is
not
just
a
generic
location.
It's
a
location,
a
specific
object
orientation
would
be
my
orientation
specifically
related
to
the
tip
my
pivot
point
along.
B
D
It's
very
specific,
yes,
and-
and
so
why
what's
going
on
here,
clearly,
these
cells
seem
to
be
able
to
influence
these
cells.
They
they
projected
this
reticular
nucleus
which,
and
they
also
projected
these
inhibitory
cells
near
the
relay
cells.
There's
a
complex
circuitry
going
on
down
here,
because
this
is
what
might
you
be
doing?
One
of
the
things
that's
always
been
associated
with
is
attention.
D
This
is
goes
back
to
many
years
decades
that
the
thalamus
and
the
projection
from
the
cortex
Newtown
was
we
somehow
related
to
attention,
and
one
of
those
ideas
of
sort
of
the
searchlight
hypothesis
is
that
somehow
the
cortex
is
telling
the
thalamus
to
restrict
what
what
I'm
gonna
see
like
it
says.
Give
me
a
smaller
part
of
the
world.
Give
me
a
bigger
part
of
the
something
like
that.
You
know
like
yeah,
I'm
super
just
this
calm,
Center
saying
this
much
of
radial
space
and
the
vision
might
say
narrow
it
down.
Yeah.
B
B
D
D
A
D
D
This
is
getting
pretty
close
to
the
complete
thing
going
on,
I
mean
there's
other
cell
types
and
so
on,
but
yeah,
okay,
we're
getting
at
the
core
of
the
matter
or
the
con
was
doing
here
and
this
part
of
it.
It's
important.
It's
clearly
an
important
essential
element.
The
thoughts
every
column
looks
like
these
things.
Combined
every
single
column,
the
cortex
is
getting
input
from
the
thalamus
into
later,
for
and
in
conducting
back
to
the
thalamus
it
doesn't
matter.
D
If
it's
a
language
area
or
a
vision,
area
of
hearing
area,
it
doesn't
matter,
they
all
look
like
this
right.
So
this
is
part
of
the
big
you
know
when
the
traffic
or
what
a
cortical
colony
does
I
took
a
double
layer.
Different
system,
maybe
is
the
speculation,
orientation
and
location,
and-
and
why
would
the
orientation
be
passed
back
down
here
so.
B
B
D
Connect
the
same
way
as
excited
connects.
Doesn't
it's
different?
Okay,
just
leave
it
it's
different.
Let's
not
pay
attention
to
it
now.
Okay,
because
I
want
to
talk.
I
want
to
focus
on
what
a
single
column
does
and
if
6b
is
projecting
to
the
thalamus
and
some
one
other
part
of
the
cortex.
It's
not
related
exactly
what
this
columns
doing.
It's
like
another
voting
mechanism.
C
D
C
D
Does
not
this
will
protect
through
the
thalamus
and
it's
different
way,
they're
part
of
the
brain
and
it's
more
hierarchical,
and
so
it's
affecting
the
next
guy
in
the
hierarchy.
It's
not
affecting
this
guy.
Okay.
So
let's
leave
that
out
for
the
moment,
okay,
and
so
now,
this
is
sort
of
our
complete
picture
close
to
a
bit
of
a
single
column
and
again
remind
everyone
that
these
two
things
are
in
the
same
color
right.
D
B
D
D
Close
to
me
as
a
larger
object,
yeah,
it's
very
far
away,
yeah,
so
I'm
in
some
sense,
I
want
to
for
this
orientation
system
to
work.
Imagine
I'm
looking
into
the
visual
column
and
I've
seen
things
at
different
orientations
to
me.
Well,
I
may
see
the
same
things
if
I.
If
I
stood
here,
I
see
the
same
things
around
me
is
if
I
see
here
right.
The
difference
is:
is
that
they're
at
different
distances
and
different
scams?
Yeah.
B
D
D
D
Like
these,
it's
like
this
is
modeling
the
Stars
about
me
and
I'm.
Moving
to
some
point,
another
point
in
the
galaxy
this
started
with
shifting
positions,
it's
something
like
that.
It's
the
same
stars
yeah,
but
they're
kind
of
moving
around,
and
so
so
there's
that's
kind
of
this
weird
thing.
That's
going
on
yeah
I
want
to
know
where
I
am
this.
This
place,
you've
marked
where
I
am
the
actual
features,
I
see
and
the
orientations
I
see
anybody.
Let
me
slightly
different
depending
on
different
right.
A
A
D
D
There's
a
these
cells,
the
location
cells
like
the
grid
cells
that
there's
theories
about
how
they
work
and
one
of
those
theories
that
grid
cells
work
on
oscillations,
meaning
you
requires
these
oscillations
in
the
brain
to
to
for
them
to
know
where
you
are
and
the
update
where
you
are
so
when
you
move
what
you're
actually
doing
is
changing
frequency
of
the
oscillations
and
many
cells
know
how
to
change
their
state.
It's
a
complicated
theory
very
detailed.
But
it's
inherently
this.
That
theory
says
nord-ost.
A
D
From
here,
yes,
so
imagine,
I,
say
here's
my
I
can
write
my
name
right,
but
I
can
also
write
a
big
right.
It's
the
same
motor
plan,
I'm
playing
back
the
same
basic
motor
plan,
but
I
somehow
able
to
scale
it
up,
and
so
there
needs
to
be
a
sense
of
scale
right.
There's
also
we
do
scale.
We
do
this
for
physical
movements.
We
also
do
it
for
for
time.
It's
like
the
speed
of
a
melody
and
I
can
slow
down
the
melody,
yeah
and
even
think
about
radio
movements.
D
B
D
Never
know
I,
don't
even
notice
this
right.
You
don't
think
about
this.
It's
just
when
you're
closer
I
have
to
move
more
to
achieve
the
same
results.
I've
a
model
of
your
head
or
face
yeah
and
to
reckon
and
I
can
I
can
determine
how
far
away
you
are
by
how
much
it
takes
me
to
move
between
these
pages.
You.
D
D
I
gave
you
a
child's
coffee
cup,
and
literally
one
that's
half
this
size.
You
would
say:
that's
small,
but
you'd
still
be
able
to
use
it.
You'd
still
be
able
to
make
predictions
about.
Just
like
I
can
take
the
knowledge
of
the
coffee
cup
and
apply
to
a
scaled-down
version
there.
This
seems
to
be
an
inherent
property
of
time
and
space
in
the
brain
to
do
this
thing's.
D
So
that's,
what's
some
sense,
what's
happening
when
I
move
around
in
this
room,
I
see
the
same
objects,
but
their
scale
is
changing
and
I
need
to
know
how
far
away
they
are
and
in
one
of
the
ways
I
can
find
out
how
far
away
they
are
is
by
how
much
I
have
to
scale
to
make
the
model
work.
So
if
I
see
the
camera
up
there,
no
that's
not
good
there,
maybe
I'm,
just
looking
at
the
white
the
biggest
projection
they
might
be
screen
over
there
as
they
move
closer
to
it.
D
D
B
D
B
B
D
D
Thing
is
this
think
about
it.
This
way
this
thing
says
when
I'm
in
some
pointing
up
some
place
in
the
room
at
some
location.
There's
a
whiteboard
here
and
the
screen
there's
a
door
there
and
there's
a
projector
there.
Great
I've
learned
these
things
now,
I
moved
to
a
new
location
and
the
size
of
all
these
things
changed
and
and
as
I
look
at
the
size
changes.
So
if
I
were
if
I
could
say,
okay
in
this
new
location,
I'm
going
to
tell
you
to
scale
things
differently,
and
now
here's
where
this
contingent
thing,
oh.
D
D
There's
this
I'll
just
call
it
like
of
the
drawer
like
this.
Okay
there's
these
oscillations,
there
are
frequencies.
You
know
ten
Hertz,
something
like
that
and
and
the
theory
that
I'm
working
on
right
now
now
is
that
there's
that
this
oscillation
between
these
two
establishes
the
scale
of
things
at
this
point
in
time
and
at
this
location
in
disorientation
and
am
I
switched
to
a
different
orientation.
I
would
have
a
different
scale
for
the
object
of
every
at
every
point.
Memory
as
I
changed
my
orientation.
It's
always
unique
to
this
location.
D
D
The
projector,
what
is
the
scale
of
the
eraser?
What
is
the
skill
at
the
table,
these
change
as
I
that
for
any
particular
orientation,
the
scale
is
fixed
right,
but
if
I
move
to
new
orientation,
a
new
position,
a
new
location,
the
same
or
I'm,
looking
the
same
direction,
the
scale
is
different
right.
So
these
this
basically
says
at
a
particular
location.
I
can
specify
what
the
scale
should
be
for
everything
I'm.
Seeing
these.
A
D
B
D
B
B
D
D
Can
there
could
be
another
room
in
the
world
where
I
see
the
screen
and
this
and
this
and
you
and
I,
see
exactly
the
same,
but
what's
behind
me
is
different,
all
right,
so.
D
A
D
B
D
Is
a
temple
pulling?
This
is
temple
floor.
Okay,
and
this
is
outputting
someplace,
and
this
is
okay.
These
are
two
stable
layers.
That's
what
typically
means
are
stable
yeah.
This
is
stable
over
changes
in
my
orientation
at
a
particular
point:
its
base.
Okay,
these
are
stable
over
movement
in
a
particular
room
right.
D
And
this
is
stable:
okay,
right
I'm
at
this
simple
point:
I'm
at
this
table
room
right
if
I
move
through
the
room,
this
stay,
stable
or-
and
this
changes
right
right
could
my
point-
this
change?
Okay.
So
it's
just
a
nested
structure
of
sensory
motor
modeling
of
points
in
space
and
sensing
or
remodeling
of
objects.
If
you
will
okay
and
I
want
to
when
I
want
to
associate
with
the
location
on
the
object,
is
these
point
in
space?
It's
like
this
point
space
is
that
this
location
of
the
object
it
doesn't
matter.
B
D
Is
getting
good,
this
is
this
is
getting
very
close
to
a
complete
sensing
motor
theory
and
the
idea
that
and
I'm
sort
of
making
it
up
as
we
talk
and
I
haven't,
really
worked
in
three
the
ideas
that
is
that
at
any
point
in
space
you're
trying
to
figure
out
what
that
plan
is
and
you're
seeing
objects
around
you.
But
it's
the
the
scale
of
those
objects
is
very
important
to
know
where
you
are
yeah.
A
D
B
A
B
D
D
D
Our
idea
that
they're
going
to
be
most
people,
don't
think
of
them,
has
similar
to
grid
cells.
They
don't
think
of
as
a
this
idea
that
you
could
be
modeling
space
with
orientation,
I,
don't
think
I,
think
it's
new
and
that
these
guys
have
sort
of
similar
proper
to
these
guys
and
they
might
work
on
the
same
basic
mechanisms.
These
guys
I,
don't
think
I've
ever
read
that
so
we're
sort
of
taken
with
most
of
the
neuroscience
literature
considers
head
Direction
cells.
D
D
It's
almost
a
requirement:
I
mean
I,
want
you
get
into
the
you
go.
It
has
to
be
like
this
yeah,
so
people
get
annoyed
when
I
say
that,
but
that
waiting
so
so.
The
idea
here
is
that
you
can
also
scale
the
whole
room.
I
mean
I
could
say
here's
a
miniature
like
that,
but
we
don't
really
like
to
do
that.
D
B
D
D
A
D
D
Enough
you
go
pick
up
anyone
yeah,
so
the
same
thing
happens
in
the
Rattler
room,
but
here's
why
I
have
this
connection
to
need
6a
in
the
Thummim,
because
I
scale
this
all
the
time?
That's
normal
that
in
this
in
in
any
particular
room,
these
different
locations
I
have
to
change
my
scale
to
see
things
every.
D
So
that
I,
don't
what
I
see
instead
of
saying
I'm
in
a
different
room,
what
I
say
I'm
in
a
different
location?
It's
the
same.
It's
basically
the
same
thing
happening
here
if
I
stressed
the
size
of
the
room,
I
some
type
of
different
room
here,
if
I
spent
the
the
scale
of
the
orientation
I,
say:
I'm,
a
different
location
right,
so.
D
D
Ourselves
so
again,
it's
a
very
parallel
system,
but
he
run
constantly
in
different
location.
I
have
different
scale
for
the
things
I
observe.
By
doing
this
time,
you
like
this
is
helpful
because
I've
never
really
described
this.
Try
to
describe
it
like
this
I've
been
thinking
these
ideas,
but
I
tried
to
the
back
of
my
head
that
this
is
what's
going
on,
but.
A
D
A
B
A
D
Let's
bring
this
way,
imagine
the
world
is
composed
of
objects,
and
each
object
is
composed
of
other
objects
at
particular
locations
in
that
in
that
object,
okay,
yeah
this
room
has
objects
in
it.
The
table
the
chair
door,
whatever,
when
I
say
the
room
is
defined
by
this
other
set
of
objects
at
different
locations
in
the
room.
These
sub
objects
all
have
an
orientation
right.
If
the
if
the
table
was
oriented
different,
the
room
I
would
know
that
with.
D
D
D
An
orientation
no
but
does
is,
is
this
room?
It's
part
of
our
office
have
an
orientation.
Yes,
it
has
an
orientation
melted
to
the
office
because
it
could
have
a
different
orientation
knowledge
of
the
office.
Yeah
I
could
have
you
get
45
degrees.
That
would
be
different
orientation,
so
the
room
itself
has
no
orientation,
but
as
the
components
of
the
room
have
an
orientation
to
the
room,
the
room
has
an
orientation
to
the
office.
The
office
has
an
orientation
to
the
building.
The
building
has
an
orientation
to
this
to
the
town.
Things
like
that.
A
D
A
D
We'd,
you
know
my
motivation
is
basically
is
I
want
to
know
this
stuff
yeah.