►
Description
Sensor-Motor Integration in the Neocortex (Jeff Hawkins, 2013 NuPIC Fall Hackathon)
A
All
right
so
this
is
this
is
a
talk.
That's
not
really
a
talk.
Okay,
it's
a
it's
a
more
of
a
discussion
and
I
I've
given
this
internally
once
and
I
gave
a
shorter
version
of
that
at
a
neuroscience
conference
once
but
other
than
it's
never
been
taped
or
anything
like
that.
A
Progress,
it's
not
like
this
is
a
great
answer.
It's
it's
like
there's
some
musings
about
sensory
motor
integration,
and
I'm
I'm
interested
in
that.
So
I'm
going
to
basically
try
to
define
what
the
problem
is
we're
trying
to
that.
I
want
to
solve,
and
then
I
will
define
some
of
the
biological
constraints
on
the
problem
and
then
I-
but
I
don't
know
the
answer.
Okay,
but
I
feel
like
we're
close.
I
feel
like
we're
making
pride.
I
feel
like
I
made
some
progress
on
it.
A
So
the
first
time
you
have
seen
before,
if
you've
seen
any
of
my
talks.
Just
to
remind
you,
I
talked
about
the
cortex
as
a
memory
system,
and
one
of
the
things
I
say
is
that
the
cortex,
of
course,
is
constantly
interacting
with
the
world.
It's
generating
a
lot
of
behavior
such
as
my
speech
and
so
on,
and
so
what
we
learn
is
we
learn
a
model
of
the
world:
that's
a
sensory
motor
model
of
the
world.
A
It's
not
just
how
the
world
behaves.
It's
how
the
how
the
changes
on
my
sensory
organs
are
occur
as
I
interact
with
the
world
and
most
of
the
changes
on
my
sensory
organ
are.
My
organs
are
due
to
my
own
behavior,
so
the
vast
majority
that
changes
on
my
retina,
because
I'm
moving
my
eyes
and
I'm
moving
my
head,
you
guys
are
moving
too,
but
not
nearly
as
much
as
me.
Moving
and
same
with
even
hearing
surprisingly,
I
hear
mostly
my
own
sounds
as
I
create
them
and,
of
course,
touch
is
like
this
too.
A
You
can't
you
can't
palpate
or
touch
anything
without
moving
your
hand,
so
it's
very
much
intimately
related
to
it.
So
the
idea
that
you
could
you
know
you
can
build
a
system
that
doesn't
have
behavior
is
is
questionable
and,
of
course
the
cla
today
has
no
behavior
and
when
we
started
numenting
we
had
to
make
a
decision.
This
is
eight
years
ago
we
said:
do
we
think
we
can
do
something
useful
without
behavior?
We
said
yes,
let's
try
it.
A
So
that's
why
the
clha
has
no
behavior
and
that's
why,
for
example,
our
product
rock,
which
I
will
demonstrate
tomorrow,
has
no
behavior.
It's
just
a
purely
a
sensory
type
of
thing,
but
that's
not
normal.
The
brain
is
brain,
has
motor
behavior
as
well.
So
then
I
go
through
our
six.
A
The
six
properties
of
I've
seen
you
guys
have
heard
some
of
my
talks
in
the
past,
the
six
properties
of
the
neocortex,
and
I
always
talk
about
number
five
being
regions
of
sensory
and
motor,
and
what
this
means
that
everywhere,
they've
looked
all
the
different
regions
in
the
hierarchy
have
neurons
that
project
some
place
that
do
something
motor
like
this
is
a
it
wasn't.
A
It
was
a
surprise
for
if
you
go
back
in
early
textbooks,
even
like
20
30
years
ago,
probably
even
today,
you'll
find
them
they'll
say
that
oh
there's,
the
motor
sections
of
the
near
cortex
and
the
sensory
sections
of
the
neocortex,
but
the
evidence
suggests
that's
not
true
what
they
really
meant.
There
was
a
sensory
mode.
I
mean
somatosensory
motor
museum
of
the
cortex,
meaning
the
parts
that
control
most
of
the
muscles
in
your
body,
but
even
vision.
A
The
vision,
cortex
controls,
my
eye
movements
and
the
auditory
cortex
control,
your
head
movements,
and
things
like
that.
So
everywhere
they
look
motor,
which
is
a
great
discovery,
and
we
can
just
try
what
a
single
region
does
and
how
it
does.
It's
solve
the
you
know
it's
like
a
separate
problem.
It's
part
of
the
inherent
property
of
a
region
in
the
cortex.
Every
region
of
the
cortex
is
learning
a
sensory
motor
model
of
its
world
and
we
can
start
with
there
and
don't
have
to
worry
about
building
the
entire
hierarchy.
A
Okay,
so
that's
just
a
reminder
of
that.
So
this
problem
names
different
spellings.
Sometimes
you
hear
called
the
sensory
motor
integration
problem.
Sometimes
it's
spelled
sensory
motor
one
word.
Sometimes
it's
sensory
motor
hyphenated,
sometimes
it's
called
contingencies
in
ai.
It's
called
embodied
ai,
sometimes
called
symbol
grounding
relatively
for
the
same
thing
all
about
we
learn
through
interaction
with
the
world
and
and
so
that's
what
it
is.
Everybody
said
that
I
already
said
that.
A
So
our
question
is:
how
does
the
cortex
learn
this
model
and
act
on
it
and
or
an
alternate
way
to
say,
is
how
do
I,
how
do
I
add
motor
behavior
to
grok
or
to
cla?
If
you
will
okay,
so
here's
some
attributes
of
the
solution
that
we
want
to
see,
I
thought
you
might
go
through
these
the
kind
of
things
I
think
about.
So
we
want
a
solution
that
is
not
particularly
not
tied
to
any
sensory
or
motor
modality.
A
How
does
the
codes
happen
in
your
eye
or
like
that?
We
were
looking
for
a
general
purpose
answer
to
like
how
the
system
or
the
sensory
motor
part
of
the
world.
So
if
you
go
read
the
literature
you
might
realize
about
cycads
sure
you're,
trying
to
capture
general
purpose
of
what's
going
on
here
and
not
the
specifics.
What's
the
cause,
it
should
work
with
biological
and
niob
non-biological
senses
and
motor
behavior.
A
So,
although
we
think
about
motor
behavior
like
a
robot
and
a
robot's
going
to
do
something
right,
well
sure
we
can
build
robots,
but
I
don't
see
why
we
have
to
re
constrain
to
that.
Why
can't
we
have
you
know
a
virtual
robot,
something
that
lives
in
cyberspace.
It's
moving
around
looking
at
different
sources
and
so
on.
A
So
you
don't
have
to
be
constrained
to
building
physical
things
that
move
well,
that
there's
nothing
wrong
with
that,
but
the
same
problem
should
apply
to
essentially
any
space
I'm
trying
to
explore
and
it
doesn't
have
to
be
a
physical
space
and,
as
I
said
earlier,
we
want
to.
We
want
to
find
a
solution
where
the
core
attributes
are
exhibited
in
a
single
region
of
the
cortex.
Just
like
the
cla
shows
the
core
method
for
learning
sequences
without
having
to
do
a
hierarchy
we
want
to.
A
We
want
to
figure
out
how
to
do
that
as
well
and,
of
course
it
must
work
in
a
hierarchy.
So
that's
the
complication
and
I
wanted
to
be
based
on
cla
principles,
because
that's
what
the
neuroscience
suggests
that
all
these
neurons
look
they're
connected
in
the
same
ways
that
the
cla
is
connected,
and
so
we
think
it's
going
to
be
based
on
something
like
that.
So
here's
how
I
go
through
it.
I
said:
here's
what
this
is,
what
the
cla
is
like
today
or
grok.
If
you
want,
I
sometimes
use
that
word.
A
A
So
it's
just
sitting
there
listening
and
the
only
way
it's
going
to
work
at
all
is
if
the
patterns
are
coming
to
it.
The
patterns
are
changing
coming
in
on
that
ear
and
it
better
be
an
active
world,
otherwise
nothing's
going
to
happen,
and
so
here's
another
situation.
Oh
that's,
that's
grock!
It
doesn't
have
to
be
physical.
That
grock
basically
looks
at
server
data
same
thing
right.
Server
data
is
changing
constantly
all
the
time.
So
we
can
just
look
at
that
data
and
look
at
the
patterns
in
the
server
data.
A
Here's
a
here's,
a
world
of
a
building
with
rooms,
and
I
could
have
my
body
and
my
sensor
in
this
building
and
if
it
can't
there's
nothing
moving
in
the
building
right,
but
there's
a
lot
of
structure
there,
and
so,
if
the,
if
this
thing
didn't
move,
there's
no
changes
on
the
sensory
organs.
So
if
you're
dead,
you're
flatlined
you're
just
sitting
there,
nothing
is
happening.
The
only
way
to
learn
about
this
world
is
to
move
through
the
world,
and
so
we
have
to
have
some
ability
to
move
through
this
structure.
A
Otherwise,
you're
going
to
learn
nothing.
That's
we
want
to
solve
that
problem.
How
would
a
system
learn
to
to
model
that
world
where
behavior
is,
is
what's
required?
Now,
of
course,
the
real
world
is
a
combination
of
me
moving
through
it
and
things
in
the
world
changing.
But,
as
I
said
earlier,
the
vast
majority
of
the
changes
on
your
own
sensory
organs
are
due
to
your
own
movements,
so
we
have
to
solve
both
of
those
problems
like
how
I
have
to
model
when
you
move
and
when
I
move,
but
it's
all
the
same.
A
So
that's
the
thing
we
want
to
do.
It's
like
we
use
as
a
little
example
there.
So
I
mentioned
here.
You
know
you
behavior
is
needed,
so
we
do.
We
do
things
that
we
we
move
our
sensors,
like.
I
just
move
my
eyes
around.
I'm
doing
that
all
the
time
I
turn
my
head.
I
walk.
We
also
interact
with
the
world
like
I'll
push
things
and
change
it
and
make
the
world
move
and
change.
I
think
these
are
all
the
same
thing.
I'm
betting,
that
it's
all
the
same
problem
I
don't
have
to.
A
I
don't
make
a
differentiation
between
me
just
moving
my
sensors
and
me
actually
opening
a
door
to
me.
Those
are
similar
types
of
problems
I
want
to.
I
want
to
try
to
solve
this
when
one
fell
swoop,
we'll
see
if
we
can
okay.
So
now,
I
just
want
to
tell
you
these:
are
I'm
going
to
go
through
a
set
of
biological
principles?
So
these
are
not
speculation
at
the
moment
right.
This
is
all
principles
that
are
true.
A
All
animals,
including
yourself,
have
pre-wired
behaviors,
and
these
are
subcortical.
So
there's
I
don't
need
a
cortex
to
do
this.
We
have
lots
of
behaviors
that
you
do
not
need
to
learn
and
you
are
born
with
those
include
walking
running
eating.
Reflex
reactions
head
turning
coughing
sneezing
blinking,
avoiding
heat,
there's
lots
of
things
you
might
say.
Didn't
I
learn
the
walk.
Didn't
I
learned?
A
No,
you
actually
didn't
you're
pre-wired
for
it,
and
your
brain
wasn't
fully
developed
when
you
were
born
and
when
you
think
a
child
is
learning
to
walk
a
child's,
not
really
learning
to
walk.
The
child's
brain
is
finished,
developing
outside
of
the
womb,
and
it's
just
like
a
giraffe
comes
out
of
the
room
and
it's
up
and
running
in
the
day
or
an
hour.
A
If
we
went
full
gestation,
we
would
do
the
same
okay.
So
you
don't
really
learn
to
do
that.
You
have
all
these
behaviors
that
are
built
in
and-
and
you
can
be,
a
really
successful
animal
with
these
built-in
behaviors,
you
can
take
a
lizard
or
any
any
non-mammal
doesn't
have
a
cortex
and
they
have
successful
lives.
They
have
sensors
shown
there.
The
sensors
feed
information
to
some
pre-wired
behavior
generators,
one
or
more
bile
neural
systems
that
are
non-cortical.
A
Those
neural
systems
generate
behavior
and
we
these
are
evolved
systems.
We
don't
know
how
that
exactly
works
and
you'll
see
why,
in
a
moment,
we
don't
really
want
to
know
how
it
works.
It
just
does
so.
The
animal
does
something
already
without
any
additional
cortex
involved.
A
This
is,
and
by
the
way,
the
the
nervous
system
evolves
hierarchically.
So
every
if
you
go
back
starting
way
way
back
when
we're
just
like
worms
and
so
on.
The
nervous
system
is
always
adding
a
layer
on
top
a
layer
on
top
and
preserving
the
old
stuff.
So
you
got
the
spinal
cord
and
the
mid
and
then
like
the
the
what
they
call
like
the
basal
ganglia
stuff
and
then
every
time
the
evolution
adds
another
level
on
it.
That
level
has
to
basically
deal
with
the
levels
underneath
it.
A
It
doesn't
really
know
underneath
it
just
says,
that's
part,
so
the
cortex
comes
along
on
top
of
a
system
like
this
and
here's
how
it's
here's,
how
it
first
I'll
walk
you
through
a
way
of
thinking
about
it.
Is
it
working?
Okay
sure?
Are
we
having
a
mic
problem
for
the
for
the
people
online?
A
A
A
A
The
world
it's
got
some
behaviors,
those
behaviors
are
repeatable
and,
to
some
extent
predictions
about.
What's
going
to
happen
next,
based
on
just
what
it's
observing
that
the
organism
is
doing,
it
is
isn't
the
second
thing.
So
it's
received
something
input.
It
can
make
predictions
even
in
a
static
world,
not
useful.
A
So
the
next
thing
that
happens,
why
didn't
this
show
up
there?
Oh
there
we
go.
Let's
look
at
that.
This
is
all
classic
neuroscience
everywhere.
You
look.
It's
like
this.
This
is
facts
what's
happening
here.
Is
this
the
cells
that
actually
generate
the
behavior,
so
there's
some
part
brain
the
old
part
of
the
brain.
A
A
For
behavioral
input
and
for
sensory
input
when
it
gets
to
the
cortex
they're,
just
all
together,
so
dla
now
is
looking
at
a
combination
of
sensory
and
from
the
world
and
motor,
and
now
it
can
better
behave,
a
better
prediction.
Imagine
I
had
some
pre-wired
behaviors.
This
is
I
walk
up
to
a
wall
and
when
I
get
close
to
the
wall,
I
turn
left
or
right.
I
don't
know
why,
but
sometimes
I
turn
left
sometimes
to
turn
right,
so
the
cla
on
its
own
would
say.
A
Well
when
I
see
a
wall,
I
know
to
expect
to
see
something
to
the
left,
to
see
something
to
the
right.
I'd,
maybe
see
a
dog
over
here
and
a
cat
over
there.
I've
learned
that,
however,
if
I'm
now
getting
a
motor
signal,
I
know
what
the
rest
of
the
body's
going
to
do
all
of
a
sudden
says.
Oh,
I
see
a
wall
and
I
know
that
I'm
going
to
turn
left
and
therefore
I
know
what
I'm
going
to
see.
A
A
A
Everywhere
they've
watched
they're
feminists,
so
we'll
go
for
it
right.
We
don't
in
fact
the
theory
we
want
doesn't
absolutely
require,
but
if
we're
gonna
have
a
motor
output,
this
is
what
it
works.
So
when
the
cortex
makes
now,
I
want
a
couple
things.
I
want
to
point
out
about
this.
So
in
there's
two
things
I
said:
well,
what's
the
relationship
between
layer,
five
and
then
three?
Well,
the
beauty
columns
go
between
them,
so
a
mini
com
and
is
as
fine
in
the
literature
spans.
A
B
A
I'm
not
going
to
let
it
fly,
it
might
have
been
my
four
yeah
so
so
we
know
that
they
have
similar
physiological
properties,
that
cells
in
layer
5
have
similar
feedback
response
properties
to
cells
when
they
accurate
now
they're,
not
the
same
they're,
not
exact
but
they're
similar.
So
and
we.
B
A
B
A
Between
layer,
3
and
layer,
5
is
unknown
almost
any
idea
how
what
life
5
cells
are
probably
going
to
do
what
they're
doing
so.
I
start
with
the
simplest
explanation.
The
simplest
explanation
is
that
the
layer
5
columns
are
activated
co-activated,
it's
the
same
column.
So
when
a
common,
where
three
is
out
there,
I'm
going
to
make
the
column
layer
five.
A
A
A
A
A
D
A
A
That's
an
auto
associative
memory
because
we
tie
the
cells
back
to
themselves
themselves.
The
same
process
can
work
right
apply
to
something
else.
So
imagine.
D
A
Of
patterns
of
cell
activities
going
on
this
pre-white
generated
behavior
that
that
mystery
system
has
neurons
that
are
going
to
be
sparsely
activated
and
they're,
going
through
some
sequence,
which
is
who.
A
A
A
Pattern,
I
will
form
a
link
and
I
will
be
able
to
say
the
leg.
5
cells
now
be
able
to
get
the
generator
cells
to
behave.
The
way
I
want
them
to
then
I
will
make
it
do
what
they
would
normally
do.
When
I
see
this
pattern,
it's
it's.
It's
just
basically
unfolding
the
cla,
instead
of
only
going
once
to
place
these
the
feet.
The
next
step
is
down
in
the
hidden
tube.
If
you
haven't.
D
A
B
A
Able
to
basically
reinforce
what's
already
going
on
down
there
and
what
can
happen
now
is
I
can
layer,
five
cells
and
go
through
a
sequence,
and
if
the
pre-wired
behavior
generator
is
not
going
through
sequence,
the
last
life
cells
can
make
it
both.
It
can
say:
do
this
now
now
I'm
going
to
stop
there
for
a
second,
because
this
is
really
important.
Does
everyone
understand
that?
Is
there
any
question
about
that?
Don't
forget
to
ask
me:
okay,
and
so
now,
all
of
a
sudden,
we
have
a
new
authority
in
the
book.
A
The
the
layer,
five
cells
are
saying.
I
can't
produce
this.
You
know
you
work
into
it.
There's
lots
of
examples
of
this
there's
lots
of
behaviors
that.
C
A
Subcortical
that
my
cortex
can
control
so
on
this
talk
with
some
real,
simple
ones
like
coughing.
I'm
able
to
coughing
is
an
autonomic
problem.
Then,
if
you
cough
because
you've
got
stuff
or
whatever
you,
but
but
the
cortex,
my
core
texture
said
look.
I
sense
that
I
know
when
you're
coughing,
I'm
now
sending
projections
down
to
the
coughing
generator,
and
now
I
can
make
you
cough
when
I
want
to
get
the
car.
So
I
go
right.
I
didn't
need
the
car
from
the
low
level
stuff.
A
A
A
I'm
going
to
walk
over
here
anyway,
so
what's
happening,
is
that
the
the
cortex
learns
to
be
able
to
control
these
thought
behaviors
and
what
we
want
to
be
able
to
do
is
string
them
together
in
new
ways
that
the
old
brain
doesn't
know
how
different.
So
we
want
to
say.
Okay,
we
have
a
better
learning
system
with
the
learning
system.
Now
doctors
string
together
old
behaviors,
but
this
is
the
general
plan
that
working
in
the
brain,
okay,.
A
D
B
A
Are
very
similar
to
whether
these
cells
but
they're
different,
they
have
different
properties.
They
actually
have
a
slightly
different
firing
pattern.
Maybe
there's
a
little
bursting,
which
you
don't
see
there
there's
a
part
of
the
cells
which
we
never
talked
about
in
that
event,
which
is
these
cells?
Have
these
apical
dendrites
that
both
the
label
bond,
so
they
have
to
give
other
inputs,
so
random
five
cells
get
different
inputs
in
the
apocalyptics
and
then
three
cells,
so
they're
actually
getting
a
different
set
of
inputs
external
inputs.
A
So
if
you
think
about
layer,
five
is
the
sequence:
when
we
just
like
player
three
you're,
actually
getting
different
inputs
and
why
different
inputs?
Well,
for
example,
one
thing
we
talked
about
earlier
is
precise
timing.
When
I
generate
roller
behavior,
I
need
to
have
precise
timing.
I
need
to
be
able
to
time
my
actions
like
my
speech,
the
milliseconds
matter
between
the
weddings
and
what
muscles
follow.
What
and
so
on
and
it'd
be
like
playing
a
kind
of
melody
without
rhythm.
A
Definitely
now
that
about
rhythm,
you
honestly
never
recognize
that
nothing,
and
so
brains
have
to
be
able
to
generate
that
rhythm
to
generate
precise
timing.
I
don't
need
that
as
much.
I
do,
but
it's
slightly
different.
I
also
need
to
be
able
to
speed
up
and
slow
down.
I
can
talk
faster.
A
B
A
There's
this
issue
of
temple
pulling
and
all
I'm
saying
I
don't
forget
what
I
wrote
there,
but
but
there's
temple
pulling
is
necessary
some
of
the
time
and
maybe
not
something
just
except
for
that.
I
also
need
to
be
able
to
stop.
I
need
to
be
able
to
say
you
know
what
I
can
think
about
doing
something,
but
don't
do
it.
I
can
imagine
it
so
I.
A
Oh
come
imagine
I
have
some
cells
that
are
walking
through
these
secrets
to
sports
skill.
Imagine
them
doing
all
these
things.
You
see
I'm
going
down
the
slope,
and
this
makes
some
better
spears,
but
they
actually
don't
want
the
body
to
do
it
so
that'll
be
able
to
turn
off
the
volume
as
they're
playing
back
to
motor
behavior.
So
this
thing's
like
backbone
and
you'll,
get
a
sensor.
A
I
don't
really
know,
but
it
is
there's
some
plausible
reasons
why,
when
we
have
a
separate
layer,
five
for
line
three
books,
so
yeah
you're,
very
nervous.
Okay.
This
is
this
gets
a
little
more
complicated.
This
is
showing
that
this
this
actually
happens
in
ohio,
okay,
so
obviously
every
region
of
my
heart
is
doing
this
and
I'm
gonna
point
out
a
couple
of
things,
and
now
I'm
going
to
bring
you
to
this
game
more
and
more
involved.
A
This
is
really
details
that
that
that
you
may
not
really
care
about,
but
that
are
important
to
me
so
so,
first
of
all,
let's
look
at
the
first
little
region,
the
hierarchy,
one
we
had
before.
We
see
the
layer,
five
cell
projects
back
to
the
subcortical
regenerator
right.
You
see
all
that
but
notice.
A
Five
cell
goes
to
that
little
x
and
goes
up
to
the
next
region,
and
so
the
layer,
5
cell,
splits,
it's
the
actual,
literally
splits
in
half
the
action
potentials
go
both
ways
and.
A
Next
rejected
cortex,
essentially
we're
sending
a
motor
command
to
the
next
region,
which
is
what
we
want
to
do.
We
want
to
the
next
region
is
going
to
do
better
inference.
It
wants
to
know
hey
what
did
the
diagonal?
Only
just
do
right
I'll
give
a
better
prediction.
If
I
know
what
the
figure,
what
the
hyper
moment
is
notice,.
A
A
So
the
brain
is
able
to
get
that
forward
signal,
because
if
I'm
going
to
pass
it
up,
I'm
not
going
to
depend
and
that's
an
intentional
mechanism
notice.
The
other
thing
that's
going
on
here:
there's
an
output
from
layer
3
that
goes
up
to
the
next
feature
now
this
is
this
is
a
very
interesting
area
of
unknown.
A
A
A
There's
a
small
contingent
of
neuroscience
today
you
got
this
all
wrong.
The
major
feed-forward
pathway,
one,
that's
actually
the
biggest
most
influential,
is
the
one
coming
from
layer.
Five-
and
I
like
that
explanation,
and
so
this
is
something
that's
going
on
here,
because
that
one
makes
more
sense
and
yet
they
both
exist
to
some
extent.
B
A
That
that
each
feature
comes
from
the
hierarchy,
does
the
same
thing
goes
up?
Does
the
exact
same
thing
now
you.
A
That
higher
order
regions
are
also
projecting
down
to
the
subcortical.
Why
would
it
do
that?
Why
does
it
just
project
to
the
region
below
it?
Well
again,
this
is
slightly
controversial
as
far
as
they
look
they've
seen
this,
that
these
regions
are
projecting
out,
but
they
haven't
looked
everywhere.
They
generally
haven't
looked
very
high
level
in
particular
hierarchy,
so
we
don't
know
if
that's
a
property,
that's
always
true,
and
in.
A
A
Saying
look:
I'm
going
to
do
a
very
small
circuit,
very
small.
I
can
make
a
prediction
when
I'm
seeing
it
and
I
I
can
control
it,
but
if
you're
going
to
make
a
bigger,
sakai
and
further
away,
I
won't
be
able
to
do
it.
A
Something
called
the
basic
angle.
Basic
angle
is
part
of
the
whole
brain.
It
just
means
lower
regions,
that's
what
gangly
is
plural
for
gang
gang
beyond
and
kangaroos,
though
these
are
lower
collections,
not
very
precise,
it's
sort
of
in
the
right
assembly
of
the
following
year.
They
have
names,
there's
a
lot
known
about
different
types
of
terms
are
used
for
them.
There's
a
lot
of
things
that
they've
studied
about,
but
they're
very
sort
of
there's
lots
of
pieces
of
the
basic
angle
there,
the
connected
very
complex
phase.
This.
A
A
lot
of
your
emotional
centers
of
the
brain
are
so
it's
a
very
highly
evolved
during
the
brain
control
and
but
one
of
the
things
it
does
is
they
know
this
pretty
high
level
observation
is
the
basic
angular.
The
value
rates
different
behavior
somehow
tells
the
cortex,
which
are
multiple
behaviors.
It.
A
Seems
to
say:
well,
here's
what
I'm
not
here
for
things
I
could
do
with
the
raised
bed.
He
says
why
don't
you
do
that,
so
that's
going
on
there.
What
happens
is
there's
these
connections
from
these
regions
down
to
the
laser
ganglion
goes
around
and
around
bunch
of
all
this
stuff
goes
back
to
the
thousand
tone
comes
back
up
the
cortex
and.
A
Basic
angle,
I
don't
think
we
have
to
say
not
to
understand
how
it
works.
You
don't
have
to
have
human-like
emotions,
but
just
to
be
aware
that,
in
the
basic
ideas
that
someone
else
decides
what
the
objective
function
is
and
what
we
want.
A
A
Imagine
that
coming
they
actually
come
in
they
project
across
the
top
of
the
cortical
region
across
what's
called
layer,
one
right
where
those
type
of
cells
can
make
it's
just
this
accident,
so
layer,
6
projects,
these
cells
predict
wrongly
across
the
other
regions
below,
and
they
make
connections
to
the
cells
in
layer,
3
and
5,
because
the
layers
have
shaped
five,
they
should
they
have
these
equal
dendrites
that
go
to
what
you
want.
We
don't
talk
about
this
in
cla.
A
We
have
around
them,
but
these
cells
have
this
influence
where
they
want
to
look
what's
clearly
going
on
here.
If
a
region
such
as
the
middle
one
wants
to
tell
the
region
below
what
to
do,
this
isn't
signal
to
do
it
says
here's
what
I
want
you
to
do:
here's
the
sequence!
I
want
you
to
play
that
here's,
what
I'm
expecting
and
these
cells
in
the
same
way
that
a
region
can
control
the
sub-part
of
behavioral
generator.
A
A
Our
region
in
iraq
can
make
the
region
below
the
hierarchy
play
back
the
sequence
the
same
associated
linking
so
we
have
the
ability
to
region.
The
reason
reason
you
can
say
play
this
sequence
playing
the
same
thing
in
the
sequence.
Somehow
it
basically
is
all
this
stuff.
Well,
I
want
a
complex
message,
but
I
think
we
don't
have
to
do
all
that.
We
may
have
maybe
have
a
statement.
B
Question
sure,
six
to
three:
what's
the
inventory
versus
excite.
A
A
Splashed,
so
you
see
that
we
make
sure
we
give
ourselves
a.
You
know
columnar
inhibition
between
the
special
food
that
we
have
cellular
inhibitions,
but
these
cells
are
all
exciting
and
both
as
far
as
I
I
don't
know,
if
there
are
any
inhibitory
connections
anyway,
that's
the
way,
I
believe
it:
okay,
good!
You
like
that,
all
right,
good!
B
A
Want
to
say
that
one
reason
that
is
trying
to
get
a
goal
for
the
region
and
so
on
so
now
I
think
we
have
the
basis
for
learning
how
to
do
something.
He
and.
B
A
Talked
about
those
people,
I
now
have
a
system
like
this
and
I
put
the
system
in
some
environment
and
it
starts
navigating
your
mind
by
some
of
its
old
rules
right,
a
simple
rule
and
the
cortex.
What
is
what
it
has
the
advantages
you
can
learn.
The
old
part
of
the
dragon
may
have
to
really
do
not
it
doesn't
really
matter,
but
we're
getting
projects
better.
It's
going
to
throw
a
better
model
than
the
old
ones.
That's
what's.
B
A
A
In
the
sequence
and
says,
oh
I
recognize
this.
I
will
now
play
back
a
pattern
that
kind
of
could
be
longer.
A
Am
okay?
I
haven't
done
this.
I
haven't.
I
haven't
figured
out
how
to
make
the
goals
working.
We
haven't
done
the
reinforcement,
learning
you
have
to
figure
out.
You
know
evaluation
of
things
and
done
that,
but
we
have
a
framework
for
which
to
build
this.
What
I
like
about
it,
I
think
we
can
do
this
in
a
simple,
basically,
two
cla.
A
I
don't
need
to
have
a
hierarchy.
I
can
get
something
going
just
with
two
sailors:
twice:
the
complication.
Perhaps
what
we
have
today
and
that's
not
good.
A
Of
this
neuroscience
perspective
on
how
this
wiring
must
work
now,
this
is
this
is
just
a
picture
of.
This
is
another
illustration
of
the
same
thing.
A
All
right
this,
this
might
be
helpful
to
do
I'll.
Just
walk
you
through
it
real
quickly,
I'm
showing
connections
that
go
as
you
know
that
the
hierarchy
you
notice
here,
I'm
showing
layer,
one
they're,
two,
three
there's
four
or
five
or
six,
two
and
three
are
generally
rarely.
Are
they
separated
I'm
clear
that
it
makes
sense
layer
four
by
the
way,
is
part
of
the.
A
And
I
I
have
some
speculations
of
what
it's
doing,
but
we
can
ignore
it.
So
I
can
argue
why
that
is.
But
here's
your
feet,
this,
the
classic
feed
forward
pathway,
a
signal
comes
in,
you
could
be
from
the
sensory
organ
or
it
could
be
from
a
previous
region.
It
goes
to
layer
four
and
it
goes
to
layer
three
layer.
Three
is
the
primary
output
layer
it
actually
the
football
is.
If
you
go
out
the
layer
it
makes.
A
This
is
the
alternate
one
which
people
are
arguing
actually
is
the
main
one
again
you
have
that
goes
to
layer,
four
and
then
three,
the
output
cells
are
in
layer.
Five.
They
go
to
the
thalamus,
which
is
a
specific
relay.
They
go
up
to
the
next
region
and
they
simultaneously,
hopefully
simultaneously
the
big
green
line
they
project
down
to
the
central
pattern.
Generators,
that's
that's
the
terms,
so
it's
a
modal
region
and
then
I
show
that
orange
the
layer,
six
feedback
done
from
other
six
cells
and
you
notice.
A
A
A
Okay,
so
that's
the
picture
we're
dealing
with
and
and
that's
all
I
have
to
say
actually
so
I'll-
take
questions.
D
D
A
A
That
to
do
that,
I
have
to
be
changing
the
rate
of
many
cells
at
once,
not
one
cell,
because
this
is
a
it's
the
whole
cla,
the
whole
complex
pattern
generator
the
house,
hundreds
of
thousands
of
cells.
I
need
to
be
changing
the
rate
of
all
hundred
thousands
of
cells.
At
the
same
time,
it
has
to
be
a
shared
thing.
However,
it
doesn't
have
to
be
global
in
the
brain,
because
the
speed
of
which
I'm
hearing
something
maybe
difference
here
in
which
I'm
visualizing
something
so
I
wanted
to
I
want
to.
A
I
want
some
sort
of
control
mechanism
that
that
covers
a
lot
of
audition
or
really
covers
a
lot
of
vision.
It
comes
boxing
absentia,
but
doesn't
mix
between
them.
So
is
there
a
kind
of?
Is
there
something
that
looks
like
that?
There
is
something
exactly
like
that
and
and
if
it's
beautiful
it's,
I
mentioned
the
thalamus
before
mcdonald's.
Is
this
little
egg-shaped
structure
right
in
the
middle
of
the
brain?
It
has
multiple
things
that
do
it's
these
gateways,
this
gated
tension
mechanism.
A
If
I
took
a
little
action
without
your
dominance,
those
are
called
the
specific
specific
climatic
relay
cells.
They
actually
they
work
like
little
gates,
but
there's
other
parts
of
the
thalamus
called
matrix
cells,
which
are
very
weird.
No
one
really
knows
what
they
do.
They're
diffuse,
they
don't
seem
what
they
do.
Is
they
project
broadly
to
layer,
one
in
regions
like
oh
over
a
bunch
of
auditory
speeches
or
a
whole
bunch
of
visual
pieces
over
a
bunch
of
prosthetic
sensors.
A
So
here
is
a
signal,
that's
generating
an
anomalous
which
could
affect
all
audition
or
could
affect
all
the
vision
it
could
affect
all
snap,
sensory
motors.
So
that's
beautiful.
It's
the
only
signal
that
exists
like
there's,
no
one.
I
don't
know
of
any
other
signal
that
could
play
that
role
of
going
there
and,
and
so
my
speculation
is
matrix
cells.
Then
you.
A
Kind
of
signal
could
the
matrix
cells
be
projecting
that
allows
these
guys
to
do
precise
timing,
and
I
don't
know
I
asked
the
neuroscientists
who
studied
the
thousands.
Does
anyone
measure
the
what
those
cells
look
like
when
they're
active
and
clearly
financially
measured,
so
there's
only
a
few
people
that
are
expecting
your
thoughts
to
ask
them
and
that's
two
to
three
questions,
and
they
neither
said
I
so
don't
think
anyone
any
evidence
about
what
those
cells
look
like
on
the
front.
A
Timing
can
only
work
up
to
about
750
milliseconds,
meaning,
if
you
want
to
private,
precise
timing
of
something
a
delay
of
more
than
that,
like
it's
more
than
a
second,
you
really
can't
do
it,
and
and
therefore
you
have
to
what
we
do
is
we
do
things
like
counting
I'll,
say
one
one
thousand,
two
of
them
so,
but
you
just
don't
but
anything
up
to
about
750
some
people
about
1000
seconds.
A
Won't
create
any
activity
right,
there's
a
null
sound.
Okay,
you
know
think
about
it.
Within
the
development
you
have,
you
have
attacks,
you
have
rhythms,
you
want
to
actually
have
a
beat
the
beat's
important,
because
it's
a
way
of
saying
start
to
walk.
You
know,
like
start
the
cascade,
and
because
I
mean
now
like
how
long.
A
A
B
A
C
A
A
C
A
A
A
B
A
A
A
These
columns
that
exist
everywhere,
so
we
got
all
these
things
going
on
once
we
have.
You
know:
shear
cells,
we
have
layers
everywhere.
We
have
hierarchy
connecting
parts
of
that
sheet.
Other
parts
of
the
city.
We
have
these
little
mini
columns
coming
across
everywhere
everywhere
the
cla
represents
one
layer
in
one
region,
actually,
one
layer
in
a
very,
very
small
it'd,
be
like
layer.
Three
now
I
talked
about
the
column
literally
because
the
column
does
slice
through
and
in
layer.
Three
there's
about
30
cells.
A
Many
people
are
not
careful
when
they
use
the
term
even
neurosciences.
The
minicom
is
a
physical
thing.
It
derives
from
the
the
growth
of
the
brain.
The
secret
progenitor
cell
comes
out.
You
can
see
it
physically.
You
want
to
see
I'll,
show
you
some
pictures
of
it.
These
cells
are
sort
of
bunched
together,
there's
a
little
gap,
there's
another
mini
column.
There's
these
hybrid
choice,
cells
that
seem
to
force
that
gap
between
the
main
columns.
This
is
a
physical
structure
there's
another
column
which
which
one
this
is
a.
B
A
A
Several
millimeters
sometimes
called
an
ice
cube
plot,
but
but
it's
generally-
and
this
is
a
column
like
in
visual
cortex,
humbling
reasonable
studies,
and
they
show
that
there
are
repetitive
patterns
across
visual
cortex,
repetitive
patterns
in
terms
of
cell
orientation.
A
So
if
you
remove
questions,
these
cells
respond
to
different
limitations,
they
get
back
to
10
locations,
ocular
dominance,
meaning
this
life
to
the
left
by
the
right,
there's
a
bunch
of
things
that
are
wrapped
up
to
be
one
and
those
things
repeat,
those
things
repeat
in
the
bigger
column
or
formula.
So
that
might
be
several
millimeters
across
where
I
get
all
my
ocular
dominance
pattern
before
I
get
all
of
my
orientation
codes,
that
larger
collar
structure
is
not
clear
that
it
exists,
and
so.