►
From YouTube: Numenta 2020: The Year in Review, The Decade Ahead
Description
A conversation with Jeff Hawkins and Subutai Ahmad.
Christy Maver, VP Marketing, sits down with Jeff Hawkins, Co-Founder, and Subutai Ahmad, VP Research, and asks them to reflect on the previous year while making predictions about the coming decade. Subutai shares the results of the effort he is leading to apply our neuroscience research to machine learning problems, and Jeff talks about continuing the neuroscience work and his upcoming book, which he spent most of 2019 writing.
The sparsity paper Subutai referenced is "How Can We Be So Dense? The Benefits of Using Highly Sparse Representations" - https://arxiv.org/abs/1903.11257
The "Frameworks paper Jeff and Subutai referenced is, "A Framework for Intelligence and Cortical Function Based on Grid Cells in the Neocortex" - https://numenta.com/neuroscience-research/research-publications/papers/a-framework-for-intelligence-and-cortical-function-based-on-grid-cells-in-the-neocortex/
Music: https://freemusicarchive.org/music/Ryan_Andersen/MORE/Telling_Me_Yes
A
Hi
I'm
Christy,
mayor
and
I'm
sitting
here
with
Jeff
Hawkins
and
some
time
on,
and
we
do
this
once
a
year
we
sit
down
we're
at
the
start
of
a
new
year,
it's
January
2020
and
every
year
we
sit
down
and
reflect
on
the
previous
year
and
look
ahead
in
the
coming
year.
What
lies
ahead
so
I
sat
down
with
the
two
of
you
about
a
year
ago,
and
it
was
a
pretty
exciting
time
for
Numenta
we
were
talking
about.
A
We
were
going
through
this
rebalancing
of
our
dual
missions
so
to
remind
people
of
our
dual
missions.
The
first
one
is
a
scientific
mission
which
is
reverse
engineering,
the
neocortex
figuring
out
how
the
brain
works.
That's
a
biological
scientific
mission
and
then
the
second
mission
is
really
applying
those
principles
to
the
field
of
machine
intelligence
and
so
sue.
Bataille.
A
A
year
ago,
you
were
lead,
kicking
off
an
effort
to
really
explore
that
mission
and
take
the
principles
of
our
theory
and
see
how
they
could
apply
to
practical
problems,
and
one
of
the
things
that
you
said
last
year
was
you
know
that
if
there's
pretty
comprehensive,
it
goes
all
the
way
down
to
synapses
and
dendrites,
and
so
it
was
kind
of
this
optimistic.
And
yet
who
knows
where
that
this
will
take
us
feeling.
So
here
we
are
year
later
yeah.
C
B
Think
you
know
the
frameworks
paper
had
been
published
and
I
just
started,
looking
to
see
how
we
can
take
all
of
the
neuroscience
principles
that
we
had
developed
actually
over
the
last
10
years
and
seeing
how
we
can
now
start
applying
them
to
practical
systems
and
machine
learning
and
I'd
say
over
the
year,
and
we
didn't
really
know
where
it
would
go.
But
during
the
year
I
think
we've
made
a
ton
of
progress.
B
We've
published
a
paper
on
archive
showing
that
sparsity
in
the
way
that
it's
done
in
the
neocortex
and
help
improve
robustness
even
in
deep
learning
systems,
we've
spoken
about
it
at
numerous
conferences
and
presentations
and
so
on
and
I
think
the
reception
is
being
really
great.
I
think
people
are
really
interested
in
seeing
that
neuroscience
can
actually
positively
impact
machine
learning,
which
is
a
rare
thing
these
days.
B
A
You
feel
like
one
of
the
things
you
noted
last
year,
was
that
there
there
were
increasingly
vocalizations
from
leaders
on
the
AI
side
of
things
saying
you
know
they
were
recognizing
the
limitations
of
existing
deep
learning
systems
and
it
seemed
like
more
of
them
we're
turning
towards
the
brain.
So
do
you
do
you
notice
that,
in
terms
of
the
reception
of
our
work
is
there
is
that
trying
to
continue.
B
Yeah
I
think
that
trend
is
definitely
continuing.
There
are
lots
of
keynote
talks
at
these
major
conferences
on
that
topic,
on
the
limitations
and
even
looking
at
how
neuroscience
can
help
so
I
think
that
trend
is
going
to
continue
and
that's
only
been
that's
been
very
helpful
from
our
standpoint
as
well
of.
A
A
B
A
That
kicked
off
2019
and
in
you
talked
about
that
paper
last
year,
because
we'd
released
it
on
archive
and
it's
it's
really
a
I
think
you,
you
said
it
was
a
rethinking
of
how
the
neocortex
worked.
It
introduced
a
new
framework.
It
had
some
really
big
ideas
in
it,
one
of
which
was
what
we
call
the
thousand
brains
theory
of
intelligence.
So.
B
A
B
C
Francis
Crick
wrote
this
about
about
the
brain
in
which
he
essentially
said.
We
don't
really
understand
how
the
brain
works
and
it's
not
because
they
don't
have
enough
data.
It's
because
we
haven't
organized
the
data
we
have
correctly
and
then
he
says
we
need
a
new
framework
in
which
to
his
exact
words,
we
need
a
framework
of
which
to
interpret
all
the
different
neuroscience
results
we
have,
and
that
was
my
personal
motivation
that
sa
and
that
particular
quest
to
find
that
framework.
C
What
how
do
we
take
all
this
nervous
on
stage
and
explain
how
the
brain
works?
That
is
what
I
set
out
in
my
life
to
do
yeah.
So
it
took
a
long
time.
The
frameworks
paper
is
in
many
ways
the
answer
to
that
question,
and
it's
not
coincidence
that
they
both
talk
about
a
framework
because
it
isn't
it
isn't
it's
a
basic
in
the
way
of
understanding
how
the
brain
works
as
a
whole
that
had
it's
really
not,
and
there
are
two
in
that
paper.
We
introduced
to
some
two
major
novelties
to
the
whole
thing.
C
So
we
explained
how
that
is
in
organizing
how
all
information
of
the
brain
is
is
organized
around
these
reference
frames,
which
are
good
for
good
cells.
But
one
of
the
consequences
that
came
out
of
that
which
we
did
not
anticipate
going
in,
was
a
sort
of
a
surprise
but
obvious
that,
in
hindsight,
was
that,
since
these
reference
names
exist
everywhere
in
the
New
York
cortex,
every
part
of
the
neocortex
is
building
models,
complete
models,
the
world.we
part
of
that
point.
C
Most
people
thought
that
the
brain
took
all
this
input
and
you
guys
to
come
up
to
one
point,
and
this
is
where
you
recognize
this,
and
this
is
where
you
recognize
that
kind
of
thing,
and
now
we
realize
that
that's
not
true
at
all.
There
are
thousands
and
thousands
of
models
of
everything.
Every
cortical
column
is
a
complete
modeling
system
and
and
how
they
work
together.
So
we
introduced
that
concept.
C
B
C
C
I
I've
made
this
analogies
multiple
times.
It's
like
the
brain
is
like
a
big
puzzle
and
you
have
thousands
of
puzzle
pieces
and
without
a
framework
they're
just
sitting
on
the
table,
you
don't
know
how
they
go
together.
You
don't
know
what
the
picture
is
and
say
all
you
do
the
truck
little
piece
here,
little
piece
there
and
it's
just
a
mess
and
that's
what
it's
been
like.
We
have
lots
and
lots
of
data,
but
what
the
framework
does
is
it
did
by
analogy
it's
like.
C
Oh
we've
now
put
in
the
border
pieces,
and
we
know
what
the
picture
looks
like
and
there's
lots
of
places
that
we
haven't
filled
in
yet,
and
we
have
lots
of
pieces
of
film
in,
but
we
kind
of
know
what's
gonna.
We
know
this
is
what
it
is.
This
is
the
size
of
it
this.
What
the
whole
picture
looks
like
and
now
we're
filling
in
these
pieces,
some
of
those
pieces.
There
are
more
interesting
to
us
than
others.
C
Yeah
I
mean
the
brain
is
still
very,
very
complicated,
but
there
were
some
things
that
that
I
personally
want
to
work
on
further.
But
with
that
framework
we
felt
that
was
this
a
good
time
to
really
take
this.
This
focus
and
go
back
just
sort
of
the
machine,
learning
side
of
things
or
the
AI
side
of
things,
so
we're
doing
somewhere
a
bit
more
neuroscience
here.
People
follow
our
work,
know
that,
but
we
really
shifted
this.
The
majority
of
the
work
here
has
bidden
to
move
to
the
machine.
Learning
worked
on
the
super
time.
A
C
Yeah
well,
the
book
is
trying
to
it.
Does
it
does
two
things?
Well,
one
could
argue
three
things,
but
first
of
all,
it
describes
all
the
neuroscience
we've
done
since
we
started
and
in
a
form
that
puts
it
in
one
big
picture
and
in
terms
that
any
person
can
understand
not
everyone
can
read
our
papers,
they're
hard,
the
details,
how
these
mechanisms
work
the
language
it
can
be
difficult,
but
many
many
people
are
interested
in
the
basic
idea.
How
does
the
brain
work
so?
C
The
first
point
half
of
the
book
is,
is
to
explaining
basically
the
thousand
brains,
theory
of
intelligence.
It
explains
the
whole
grid
cell
idea
that
explains
the
early.
We
did
work
with
in
2016
about
how
neurons
make
predictions
it
puts
that
all
into
one
sort
of
bundle
that
interested
lay
reader
could
say.
Okay,
I
want
to
stand
how
my
brain
works.
What
is
happens
on
I'm
thinking
and
I
can
describe
that.
The
next
section
in
the
book
is
really
about
my
predictions.
C
What's
going
to
happen
in
the
field
of
AI
and
receive
intelligence,
and
although
we're
working
on
here
the
way
we're
going
about
and
how
is,
of
course,
of
our
incremental,
step-by-step
logical
but
I'm,
more
speculative
in
the
book
about
where
the
ultimate
goals
are,
where
we're
going,
where
is
it
gonna
go?
What
are
the
machines?
Are
gonna,
be
like
it's
almost
as
if
hey
once
we
build
machines
that
do
all
the
things
we
know
that
brains
do
all
the
things
subitize,
let
out
what
will
their
impact
be?
C
What
will
they
be
and
what
won't
they
be?
And
so
you
know
I
discuss
things.
That's
is
the
existential
risks
of
machine
intelligence
and
which
is
a
topic
many
people
are
concerned
about
and
how
would
impact
society
and
then
the
last
part
of
the
book
is
something
that
most
people
even
who've,
been
following
me
from
her
whole
life.
Don't
really
know
that
I
think
a
lot
about
which
is
sort
of
the
ultimate
impact
of
of
well
the
ultimate
destination
of
humanity.
C
C
We
tie
to
these
genetically
derived
old
systems
that
you
know
have
all
these
biological
needs
and
don't
always
behave
in
right,
very
good
ways,
and
so
what
are
the
implications
going
forward?
How
can
we
use
our
intelligence
to
for
the
first
time
craft,
a
futures-
that's
not
purely
driven
by
evolution?
I,
don't
give
away
the
answers
to
that.
But
there
are.
There
are
several
yeah
if
several
sort
of
it
takes
a
bit
of
thought
and
I
want
to
go
through
carefully
it's.
C
You
know
you
could
sound
a
little
bit
like
science
fiction,
but
I,
don't
really
mean
it
through
science
fiction.
I
really
want
to
be
thoughtful
about.
How
can
we
use
our
intelligence
to
craft
the
future?
That
makes
sense
that
provides
meaning
for
Humanity
and
for
our
lives
and
surprising
things
all
have
in
there?
C
C
A
Want
to
come
back
to
you
super-tight
what
one
of
the
changes
that
we
made
in
I
think
the
early
2019
the
first
time
for
2019.
Was
we
opened
up
the
research
meetings
to
become
to
be
live-streamed,
which
is
something
new
for
new
mentis?
So
how?
How
was
that
experience
for
you?
Did
it
change
the
research
meetings
at
all
or.
B
We've
tried
not
to
change
the
research
meetings
too
much
too
much,
so
we
try
to
keep
it
going,
as
it
always
has
done
so
I
think
from
the
community
standpoint.
They
can
really
see
it
fresh.
The
way
we
actually
have
our
research
meetings,
we're
not
trying
to
tailor
it
to
the
audience
or
anything
so
I.
A
B
We
keep
it
going
what's
been.
Interesting
is
sometimes
when
I
go
to
these
machine
learning
conferences,
I've
not
run
into
people.
We're
not
going
yeah
I've
watched
your
research
meeting
on
this
or
that
or
whatever
I'm
like
oh
really,
so
that's
quite
interesting
and,
and
you
know
so
I
think
it's
only
helped
increase
the
awareness
and
people
really
respect
us
for
doing
it.
It's
extremely
unusual
I,
don't
know
anyone
else.
Who
does
that
so
I
think
it's
it's
a
good
thing.
C
Yeah
really
kudos
to
Matt,
who
you
can't
tell
but
he's
standing
right
over
here,
but
kudos
to
Matt
for
doing
that.
I
think
it's
you
know
really.
You
know,
there's
such
a
hard
scientific
problems.
We
work
on
and
in
academia
is
still
very
isolated.
Towers
of
you
know,
academia
and
the
idea
that
publicly
sharing
your
dirty
laundry
and
your
mistakes
in
your
and
your
puzzlement
over
these
problems
was
a
very
clever
idea.
C
I'm
Matt's
part,
and
we
didn't
know
how
it
would
turn
out
and
I
guess,
there's
some
people
would
like
it
to
the
extent
that
people
like
it
and
get
something
out.
It's
great
I
feel
really
good
about
it
and
a
super
test
that
we
really
didn't
have
to
change
anything
other
than
occasionally
not
mention
something
proprietary
for
business
reasons
or
something
like
that,
but
some
a
scientific
point
of
view.
It's
been
it's
just
the
way.
We
are
it's
good
or
bad.
It's
what
it
is
so.
A
Okay,
so
let
me
end
with
asking
about
you
know
it's
the
start
of
a
new
decades.
Of
course,
everyone's
making
predictions
about.
What's
to
come
over
the
next
ten
years.
You
guys
have
both
been
with
Numenta
Numenta
from
the
very
beginning,
so
now
we're
entering
a
new
decade.
As
you
look
ahead,
I'll
start
with
you
super
tie.
What
what
lies
ahead
for
Numenta,
the
next
10
years,
yeah.
B
So
it's
it's
interesting!
You
know!
If
you
look
back
at
the
last
decade,
I
think
it's
really
about
10
years
ago
that
you
know
Jeff
first
started
thinking
about
active
dendrites
and
prediction
and
sparsity
and
how
that
can
all
come
together
into
the
temporal
memory
and
it
kind
of
started
this
whole.
You
know
really
in-depth
modeling
of
the
neuroscience
details
which
we
hadn't
done
before.
So
it's
kind
of
fitting
I
guess
that
in
2020,
we're
sort
of
taking
all
of
that
and
now
starting
to
look
in
how
to
apply
all
of
that
social.
B
And
I
think
I
think
that
our
pace
is
just
going
to
accelerate,
because
you
know
the
last
year
and
working
on
it
has
been
really
encouraging.
It's
it's
I
think
it's
there's
a
path
towards
putting
all
of
these
principles
into
practical
systems
and,
as
we
think,
about
active
dendrites
and
unsupervised
learning
and
continuous
learning
in
the
temporal
memory
and
and
location-based
representations
and
the
full
cortical
column
idea,
I
think
those
will
all
they
all
kind
of
build
one
on
top
of
one
another.
B
So
I
think
there
will
all
come
come
to
fruition
and
I,
don't
think
it'll
take
a
full
ten
years.
The
next
part
of
it.
You
know
looting
to
your
question
earlier.
I
think
the
machine
learning
community
is
very
open
now
to
neuroscience
information
and
I
predict
over
the
next
ten
years
that
there's
going
to
be
a
much
more
crosstalk
between
the
neuroscience
and
machine
learning
communities
than
ever
before.
So
I
think
that's
I'm,
very
hopeful
about
that
as
well.
C
A
C
Like
the
SS
super
Daisuke,
but
I'll
have
to
point
out
that
what
I
first
got
interested
in
this
Miller
40
years
ago,
right
and
I
had
this
first
of
all,
I
said
to
myself:
oh
man,
we're
gonna
solve
this
brain
problem.
I
thought
I
was
gonna,
be
too
late.
I
said
you
know,
this
must
be
people
who
are
more
senior
than
you're
working
on
this.
Now
they
must
be
going
on
it.
C
You
know
a
few
years
of
others
all
figured
out
I
wonder
if
I
could
do
any
of
this
because
I'm
too
late
to
come
to
the
party
when
I
turned
out
to
be
wrong
and
the
other
thing
I
thought
that
I
thought
it
would
be
totally
obvious
that
the
way
to
build
intelligent
machines
was
thanks.
I
mean
this
was
just
obvious
to
me
and
yet
I'm
surprised
how
long
it's
been
that
that
idea
still
hasn't
taken
off
now.
I'm
optimistic,
like
Shiba
thought.
C
Maybe
this
looks
really
like
the
decade
surprised
my
path,
but
it
doesn't
really
look
like
because
now
people,
other
people
are
really
saying
it,
but
I
do
think
it
will
probably
take
a
lot
of
work
on
our
part
to
make
it
happen
because
I
think
we
just
are
thinking
on
these
issues.
It's
at
least
different
and
and
if
I
want
to
be
more
optimistic,
I
say
we
know
where
the
leaders
in
this
kind
of
thinking,
but
it's
not
gonna,
be
easy.
So
that's
backing
up
super.
C
The
next
10
years,
some
of
those
ideas
I,
was
talking
about
in
in
the
book
that
I'm
writing
now.
I.
Don't
expect
him
to
come
to
fruition
in
the
next
10
years,
but
I
do
I
would
really
be
thrilled
if
those
ideas
became
discussed
broadly-
and
you
know
it's
not
the
way
to
talk
comes
out,
but
but
that
those
ideas
and
these
ideas
are
starting
to
permeate
through
people's
thoughts
and
discussions
about
the
future.
It's.