►
From YouTube: Framework for Intelligence (Episode 15)
Description
Get ready for a breakthrough new framework for intelligence based on grid cells in the neocortex! We understand more about how your brain models reality than ever before. Watch as we explain how your brain represents objects in space. More info about our theory at https://numenta.com/neuroscience-research/research-publications/papers/a-framework-for-intelligence-and-cortical-function-based-on-grid-cells-in-the-neocortex/
A
A
In
1979,
Francis
Crick
wrote
an
essay
titled
thinking
about
the
brain.
In
it
he
wrote,
in
spite
of
the
steady
accumulation
of
detailed
knowledge
how
the
human
brain
works
is
still
profoundly
mysterious.
He
posited
that
over
the
coming
years,
we
would
undoubtedly
accumulate
much
more
data
about
the
brain,
but
it
may
not
matter
as
our
entire
way
of
thinking
about
such
problems
may
be
incorrect.
He
concluded
that
we
lacked
a
theoretical
framework,
a
framework
in
which
we
can
interpret
experimental
findings
and
to
which
detailed
theories
can
be
applied.
A
This
is
really
what
Numenta
has
been
working
towards
since
our
beginnings.
We
must
understand
how
this
framework
for
intelligence
works
and
since
the
discovery
of
grid
cells,
we
think
we
have
a
much
better
idea.
This
episode
builds
upon
ideas
presented
in
previous
shows:
it's
okay!
If
you
don't
completely
understand
the
temporal
memory
algorithm,
but
you
really
need
to
understand
a
few
important
things
about
grid
cells.
First
of
all,
if
you
haven't
seen
it
watch
this,
it
explains
everything
you
need
to
know
about
grid
cells.
A
A
grid
cells
in
place
cells,
live
in
structures
called
the
inter
Rhino
cortex
in
the
hippocampus
respectively.
So
here
is
half
of
your
neocortex
half
of
your
brain,
the
cerebellum
I'm
gonna
pop
out
this
whole
old
midbrain
part
of
the
brain
I'm,
going
to
show
you
where
the
in
Toronto
cortex
is.
It
is
right
here
and
if
you
pop
this
open,
you
can
even
see
the
hippocampus
right
along
inside
you've
got
two
of
these
one
on
either
side
of
your
brain.
A
Now
in
the
neocortex,
we
think
grid
cells
also
exist
and
they
help
create
detailed
mental
maps
of
every
object
that
you've
ever
learned
and
it
works
by
expanding
the
tricks
that
already
exist
in
the
internal
cortex,
think
about
the
room
or
the
cafe
or
the
outdoor
space.
Your
body
currently
inhabits
everything
you
can
sense
right
now
is
your
model
of
reality
grid
cells
and
the
Intel
Rhino
cortex
are
used
to
represent
all
space
in
this
model
of
reality.
Your
environment
could
be
seen
like
a
reference
frame
with
you
moving
within
it.
A
A
You
can
feel
shag
carpet
under
your
feet,
so
you
know
you're,
probably
inside
someone's
house,
but
whose
house
is
it
we
need
to
move
in
order
to
find
out
look
we
found
a
lamp,
doesn't
look
familiar,
but
at
least
it
tells
us
we're
in
the
type
of
room
that
has
a
lamps
and
shag
carpets,
but
we
need
more
data
before
we
can
say
where
we
are
for
sure,
aha,
so
we're
in
a
bedroom
with
a
bed
and
a
lamp
beside
it
with
shag
carpet.
Let's
keep
looking
for
clues.
A
Hey
now,
there's
there's
a
rug
with
tassels
at
the
end
of
the
bed
and
I've
only
seen
that
in
my
am
Virginia's
bedroom
and
we've
just
identified
what
room
were
in
by
moving
and
sensing
the
features
of
the
room
and
Counting
out
all
the
rooms
that
didn't
match
the
arrangement
of
features
we
found
as
soon
as
we
sensed
enough
features
in
relation
to
each
other.
We
knew
exactly
where
we
were
now
imagine
if
there
were
five
copies
of
yourself
in
the
room
and
they
could
all
move
around
independently.
A
If
they
all
move
to
a
different
feature
and
shared
their
findings,
they
can
identify
the
by
observing
only
one
feature
each
similar
to
how
grid
cells
in
a
rhino
cortex
model,
your
environment.
We
think
that
grid
cells
in
the
neocortex
are
modeling
objects.
Both
things
have
features
which
might
be
a
sub
object,
think
of
the
handle
on
a
cup
or
the
rug
and
at
Virginia's
room
in
the
neocortex.
The
location
is
not
the
location
of
the
organism
in
an
environment
but
the
location
of
a
sensory
patch
in
space
to
really
dissect
this.
A
We
need
to
talk
about
sensory
patches
and
how
they
connect
to
cortical
columns.
Let's
talk
about
touch
and
the
part
of
the
brain
associated
with
sensory
info.
Given
this
widely
accepted
model,
you
can
imagine
what
a
section
of
neocortex
processing
a
hand
sensory
input
might
look
like
with
sections
of
cortical
columns,
mapping
across
the
hand
each
one
of
these
cortical
columns
is
modeling
space
and
the
location
of
its
associated
sensory
patch
and
processing
its
feed-forward
sensory
input
as
we
attend
to
an
object.
A
All
cortical
columns
receiving
sensory
input
model
the
same
object
each
with
a
different
position
and
sensory
experience
with
the
object.
Remember
that
touch
and
sight
are
two
very
different
senses.
The
retina
does
a
huge
amount
of
processing
before
sending
a
signal
to
your
brain,
but
you
can
still
think
of
it
as
representing
features
at
locations,
except
this
sensory
patch
is
more
of
a
projection
across
space
with
the
sensors
at
the
origin.
Each
one
of
these
cortical
columns
has
a
unique
representation
of
objects
because
they
each
get
different
sensory
input
through
movement.
A
Each
column
has
its
own
unique
experience
of
every
object.
It's
ever
taken
apart
in
learning.
We
believe
the
number
of
grid
cells
that
could
exist
within
one
layer
of
a
cortical
column
is
enough
to
represent
a
very
large
number
of
locations,
I
like
to
think
of
this
as
a
universe
of
space
inside
your
brain.
So
if
you
were
to
learn
a
new
object,
you
might
randomly
choose
a
point
within
this
non-physical
universe
to
start
off
at
and
then
move
your
sensors
through
the
object
space
to
learn
it,
since
the
universe
is
so
big.
A
This
object
is
kind
of
like
a
lone
planet
floating
in
space
light-years
away
from
the
nearest
object.
It
would
be
very
unlikely
that
our
randomly
chosen
starting
point
was
anywhere
near
other
objects.
We've
learned
so,
even
though
each
cortical
column
models
a
universe
of
space,
the
objects
are
so
far
away
from
each
other
within
that
universe
that
we
can
treat
them
as
being
within
their
own
reference
frames,
because
grid
cells
know
how
to
perform
path.
A
Integration
in
space,
given
movement,
we
can
predict
what
we
will
sense
as
we
move
our
senses
through
any
objects
reference
frame,
so
we
have
a
layer
of
cells
in
the
cortical
column
that
is
tracking
location
of
the
sensory
patch
through
space.
This
layer
is
grouped
into
grid
cell
modules
and
each
has
a
different
properties
regarding
how
it
interprets
space
as
the
location
updates.
A
phase
shift
is
applied
across
all
the
grid
cell
modules.
Remember
these
locations
are
represented
as
sparse,
binary
arrays
and
we
can
perform
operations
on
them.
A
A
This
displacement
must
come
from
somewhere
and
we
think
it
is
probably
represented
somewhere
else
in
the
cortical
column,
in
a
set
of
displacement
cells,
the
idea
is
simple:
the
displacement
cells
represent
a
movement
through
space
and
when
applied
to
the
current
grid
cell
module
activations
cause
a
phase
shift
and
will
represent
the
new
location.
There
are
two
really
important
things
you
can
do
with
these
displacement
cells
and
combination
with
red
cell
modules.
The
first
as
I've
shown
you
you
can
move
through
an
object's
reference
frame
and
associate
sensory
features
with
locations.
A
This
allows
you
to
learn
new
objects
through
movement
and
recognize
objects
based
upon
what
you've
already
learned,
number
two.
You
can
associate
objects
to
each
other
which
turns
out
to
be
super
important
because
it
enables
object,
composition
and
havior.
Remember
that
object.
Reference
frames
are
really
all
a
part
of
the
same
huge
universe
in
your
brain,
so
local
displacements
make
sense
when
associated
with
sensory
movements,
but
displacements
that
travel
between
objects
can't
be
associated
with
movements
anymore.
So
what
could
they
mean?
Well,
they
can
be
used
to
represent
the
exact
position
of
one
object.
A
A
Other
objects
seriously
think
about
that.
This
allows
object,
composition
so
that
any
object
can
be
efficiently
constructed
out
of
any
other
objects.
A
point
on
an
object
might
have
a
sensed
feature
or
a
displacement
that
projects
another
entire
object
instead
of
a
feature
so
moving
between
object.
Reference
frames
is
just
as
efficient
as
moving
within
an
object.
Reference
frame
and
I
didn't
even
have
to
use
the
wormhole
analogy.
A
Let's
take
this
displacement
idea,
one
step
farther.
If
a
displacement
can
place
a
sub
object
into
another
object's
reference
frame,
a
sequence
of
displacements
could
represent
an
object's
behavior
again.
These
sequences
can
be
learned
and
replayed
with
a
temporal
memory
algorithm,
as
defined
in
previous,
shows,
there's
more
but
FCC
regulations.
Only.
Allow
me
to
blow
your
mind
once
per
video
and
I
may
have
already
broken
that
law
in
the
next
video
we're.
Finally,
gonna
talk
about
hierarchy
when
I
introduce
something
we
like
to
call
the
thousand
brains.
Theory.