►
From YouTube: The "Thousand Brains Theory" of AI | ZDNet
Description
Jeff Hawkins, co-founder of Numenta, talks about the process of making AI more flexible and generalized.
MORE TONYA HALL SHOW VIDEOS
- https://goo.gl/YbapCt
FOLLOW US
- Subscribe to ZDNet on YouTube: http://bit.ly/2HzQmyf
- Watch more ZDNet videos: http://zd.net/2Hzw9Zy
- Follow ZDNet on Twitter: https://twitter.com/ZDNet
- Follow ZDNet on Facebook: https://www.facebook.com/ZDNet
- Follow ZDNet on Instagram: https://www.instagram.com/ZDNet_CBSi
- Follow ZDNet on LinkedIn: https://www.linkedin.com/company/ZDNet-com
- Follow ZDNet on Snapchat: https://www.snapchat.com/add/zdnet_cbsi
A
B
Is
an
unusual
company
in
Silicon
Valley
we
have
two
missions
and
they're
both
related
to
intelligence
and
artificial
intelligence.
Our
primary
mission
is
actually
a
neuroscience
mission.
It's
reverse
engineering,
a
part
of
the
brain
called
the
neocortex,
and
so,
if
you
came
to
know,
Monta
you'd
see
a
bunch
of
scientists
working
on
as
if
we're.
B
A
B
There's
a
few
things
in
AI
today,
which
is
mostly
what's
a
deep
learning.
Networks
for
convolutional
neural
networks
are
called.
They
are
inspired
from
some
brain
ideas
many
many
years
ago,
4050
years
ago,
but
we've
learned
so
much
about
how
the
brain
works
now
and
they
really
the
brain
works
on
completely
different
principles
than
what
today's
current
AI
is.
So
we
can
talk
about
some
of
those
principles
but
they're
really
very
different.
Now
it's
another
question:
do
those
differences
matter
and
I'm?
B
That's
our
belief
is
that's
my
belief
that
those
differences
do
matter
and
if
you
think
about
the
world
of
AI
and
machine
learning
these
days,
they've
made
tremendous
progress,
but
it's
so
quite
limited.
No
one
thinks
that
you
know
today's
AI
is
really
intelligent
or
really
is,
like.
You
know
those,
that's
not
really
there.
It's
solving
interesting
problems,
but
it's
not
what
we
would
recognize
that
as
an
intelligent
entity
or
intelligent
agent,
and
so
it's
always
been
my
belief.
B
B
Well,
okay,
the
thousand
brain
Steven
tells
I'll
have
to
give
you
a
little
bit
of
a
background.
So
if
you
look
at
a
human
brain,
most
of
what
you
would
see
is
the
big
wrinkly
thing
around
the
outside.
That's
the
neocortex,
it's
about
75%
of
a
human
brain.
So
our
focus
is
on
that
organ.
It's
not
the
rest
of
the
brain
and
that's
the
organ
of
intelligence
and
the
thousand
brain
theory
is
as
a
scullery
real.
It's
not
a.
We
thought
this
we
discovered
this
is
that
most
people
thought
about
okay.
B
Well,
what
does
the
neocortex
do
and
one
way
to
look
at
that
as
it
builds
a
model
of
the
world?
Like
everything
you
know,
everything
has
to
be
stored
in
the
new,
your
cortex,
so
how
your
job
works
out,
microphones,
work,
how
computers
work,
how
language
way,
everything
and
and
the
typical
way
of
viewing
that
is?
Okay,
that's
it's
that
your
brain
has
this
one
model.
What
we
discovered
is
that
the
brain
has
thousands
and
thousands
of
models
of
the
world.
B
It's
it's
not
one
model
it's
distributed
actually
over
about
a
hundred
and
fifty
thousand
different
models.
So
then,
your
core
test
is
divided
in
these
little
columns
or
about
a
millimeter
in
diameter
and
each
one
of
those
there's
150,000
of
those
each
one
of
them
is
learning
a
complete
model
of
some
part
of
the
world.
It's
like
it's
like,
there's
a
whole
bunch
of
little
little
learning
energies
in
your
head,
so
the
thousand
brings
my
theory
of
intelligence.
Says:
hey.
B
B
We
were
just
starting
to
do
that.
There's
a
series
of
principles.
We've
learned
about
the
brain,
I'm
gonna,
just
tell
you
one
of
just
focus
on
one.
First,
everything
that
how
we
learn
about
the
world
and
the
models
we
build
in
our
head
are
all
based
on
movement.
That
is,
we
learn
by
moving.
We
if
I
want
to
learn
a
new
building
or
a
new
house.
I
have
to
move
through
it
if
I
want
to
learn
a
new
object
like
a
coffee
cup.
B
I,
don't
have
my
coffee
cup
here,
but
you
know
one
of
the
new
object,
like
a
coffee
cup
I
have
to
here's
a
here's.
A
thing
of
wieck's
I
have
to
like
put
my
fingers
over
it
and
move
it
or
I'll
move
my
eyes
over
it.
So
it's
we
call
that
a
sensory
motor
model
and,
if
you
think
about
how
could
you
learn
the
structure
of
the
world
without
moving,
you
really
can't
you
have
to
move
through
the
world
instruction
roll.
Today's
AI
basically
has
none
of
that.
B
Phase
III
is
like
here's,
some
pictures
or
some
images
or
some
patterns
that
are
static,
and
we
try
to
train
the
systems
to
recognize
that,
but
the
brain
works
by
movement
and
and
once
you
understand
how
it
uses
movement
to
build
the
model,
the
world.
You
feel
you
say:
oh
my
gosh.
If
we're
gonna
build
rolling
tolls
of
machines,
they
have
to
work
like
that.
It's
sort
of
a
fundament
the
aspect
of
what
it
means
to
learn
about
the
world.
So
do
we
have
a
series
of
things
like
that?
B
That's
one
really
big
one
that
we
say.
You
know
why.
If
we're
really
gonna
build,
truly
tells
machines,
they
have
to
incorporate
those
properties.
They
don't
have
to
necessarily
be
physically
embodies
like
a
robot
like
I
could
have
a
virtual
intelligence
that
lives
on
the
internet
and
it
moves
through
the
internet
right,
but
it
has
to
change
where
it
is.
It
has
to
change
its
sensory
input
through
some
sort
of
internal
movement
commands,
and
so
that's
like
those
are
like
some
breaks.
That's
one
very
fundamental,
a
difference
between
a
brain
and
today's
a
I
do.
A
B
Let's
break
that
apart,
you
don't
need
new
computing
hardware
to
do
this,
but
you
do
need
to
do
types
of
software
tools
and
nothing
dramatically.
Different
I
mean
we
are
just
starting.
We
do
similarly
we're
starting
to
apply
some
more
stuff
to
machine
learning
techniques
and
we
have
to
create
our
own
library
as
we
we
might.
We
might
need
to
accelerate
things.
You
might
need
some
non-standard
CPUs,
but
things
you
might
get
at
PJ's
or
GPUs
that
kind
of
stuff,
but
it's
not
like.
B
Oh,
we
have
to
create
for
now
to
do
what
we
need
to
do.
We
do
not
have
to
create
some
radically
new
architectures
I.
Imagine
in
the
future.
We
go
out
20
30
years
from
now
that
will
change
at
the
future.
When
we're
really
building
this
stuff,
it's
a
huge
business.
There
will
be
custom
hardware
designed
to
take
advantage
of
the
unusual
properties
of
how
brains
work.
It's
not
like
a
computer
at
all,
but
today
we're
not
we're
not
really
held
back
by
that
today.
B
B
That's
been
our
primary
mission
and
we're
just
starting
to
apply
it
to
the
machine
learning
stuff.
I'll.
Give
you
two
examples
that
we've
done,
one
that
we
did
a
few
years
ago,
which
is
the
brain
all
work.
It
works
on
temple
temple
time-based
patterns,
like
things,
are
always
changing
in
the
brain,
it's
never
static,
and
so
we
we
took
some
of
these
things.
We
learned
about
the
brain
about
how
they
learn
sequences
and
we've
made
some
interesting
anomaly:
detection
technology
and
so
anomaly
detection
is
in
streaming
data.
B
So
if
you
have
a
data
coming
off
a
machine
or
a
computer
or
a
windmill,
how
do
you
know
by
looking
at
the
time
based
patterns?
Is
it
changing?
Is
it's
a
healthy
or
not
so
there's
several
companies
have
licensed
that
technology.
There's
another
company
called
cortical
I/o,
which
is
based
doing
a
language
understanding
based
on
some
principles.
They
they
adopted
from
us,
that's
really:
cool
company,
cortical,
thio
and
and
now
we're
starting
to
work
on
things
in
like.
A
B
You
should
go
to
our
website
New
Mexico,
and
it's
really
simple,
but
we
have
a
lot
of
ways
you
can
engage
with
us.
We
have
an
active
forum
and
an
open
source
community.
We
have
a
lot
of
people,
there's
lots
of
discussions
every
day
we
have
all
of
our
papers,
both
scientific
papers
and
so
to
lay
versions.
Are
those
you
can
read,
there's
a
tremendous
amount
of
resources
we
have.