►
From YouTube: Education & Workforce WG: NSF National AI Research Institute for Adult Learning and Online..
Description
March 2022
Ashok Goel
Georgia Tech
A
The
institute
was
launched
on
november
1st,
so
it's
all
very,
very
recent
about
four
months
since
we
got
the
grant
since
we
launched
institute-
and
let
me
just
introduce
what
the
basic
problem
is,
that
we
are
trying
to
address
so
in
the
united
states
in
2021
about
25
america,
25
million
americans
left
their
jobs
november
alone,
about
4
million
americans
left
their
jobs
and
they're
all
kind
of
reasons
for
that.
But
one
of
the
reasons
certainly
is
that
people
are
not
necessarily
interested
in
continuing
their
jobs
as
they
were.
A
This
is
a
great
resignation
going
on
and
people
are
looking
for,
upskilling
and
reskilling.
These
are
adults,
they
have
families,
many
of
them.
Many
of
them
have
homes
and
houses,
so
they
can't
relocate
to
universities
and
colleges
they
can't
go
to
where
education
is.
Instead,
we
need
to
find
out
ways
of
taking
education
to
where
they
are
where
they
live
and
that
they
work
and
that
becomes
a
new
challenge.
A
The
world
economic
forum,
before
the
covert
19
happened
in
2019,
the
world
economic
forum
projected
that
there
will
be
133
million
new
jobs
in
the
united
states
by
2030..
This
number
is
almost
surely
a
lower
count
now
that
gold
kobe
does
happen.
So
what
do
we
do?
How
do
we
help
these
adult
learners,
millions
of
them
upskill
and
reschedule?
How
do
we
take
education
to
them?
So
that's
a
big
challenge
for
us
now.
A
This,
of
course,
is
a
very
big
problem,
but
what
we
want
to
do
is
to
see
if
we
can
use
ai
to
help
in
the
process,
so
we
want
to
connect
foundational,
ai
research
to
the
unique
circumstances
and
specific
signs
of
early
learning
and
online
education.
The
reason
for
online
education,
of
course,
is
because
these
are
remote
learners.
They
are
distant
learners.
A
Now,
what
are
the
difficulties
in
online
education
that
we
all
are
familiar
with,
because
we
all
have
taken
part
in
it
either
as
teachers,
the
students
and
sometimes
our
children
have
taken
part
in
online
education.
One
of
the
difficulties
with
online
education
is
the
quality
of
online
education
is
not
always
very
good.
It
certainly
is
not
as
as
good
as
in
person.
Education
can
often
is
so.
How
do
we
use
ai
to
improve
the
quality?
A
So
we
are
interested
not
just
in
quality,
but
in
availability.
We
want
it
to
be
education
to
be
available
to
everyone.
We
want
it
to
be
affordable
and
that's
what
online
education
promises
and
we
also
want
it
to
be
achievable,
achievable
in
the
sense
that
people
taking
that
should
be
able
to
succeed
in
it.
They
should
have
the
resources
available
that
they
can
succeed,
teaching
assistance,
for
instance.
A
A
We
want
to
optimize
humans,
so
ai
and
humans,
working
together,
but
it's
the
human
ai
here
is
in
the
service
of
humans
and
it's
the
humans
which
who
are
whom
we
are
trying
to
optimize
using
ai
in
some
way,
and
that
raises
issues
of
accessibility,
personalization,
scalability
and,
interestingly
enough.
Some
of
these
issues
occur
not
only
for
education,
but
also
for
ai.
How
do
we
make
ai
personalized
personalized?
A
How
can
anyone
in
the
world
whether
teacher
or
learner
create
his
or
her
own
ai's
in
a
manner
that
does
not
take
too
much
time
or
too
much
expertise
or
too
much
labor?
So
the
entire
approach
is
to
bring
education,
researchers
and
ai
researchers
together
and
think
in
terms
of
the
social
technical
system.
A
Now
here
are
some
of
the
innovations
that
we
are
aiming
for
now.
These
innovations
are
driven
by
needs
of
education,
so
they
are
innovations
in
ai
but
they're,
driven
by
education.
So
one
of
the
things
that
often
happens
in
online
education
is
that
there
is
not
enough
cognitive
engagement
or
there
is
not
enough
teacher
presence
in
an
in-person
class.
You
can
ask
questions
to
a
teacher
in
an
online
class
is
very
difficult
to
do.
A
We
want
to
collect
huge
amounts
of
learning
data,
build
generalizations
from
them
and
then
feed
them
back
to
learners
and
teachers,
as
well
as
ai
cognitive
assistants,
who
help
the
teachers
and
the
learners
now
in
the
right
side
of
the
column
are
some
innovations
in
ai
itself.
That
I'll
mention
very
briefly,
because
the
focus
of
this
particular
forum
is
on
education,
not
so
much
on
ai.
A
A
So
can
we
exploit
this
data
to
provide
very
personalized
feedback,
so
this
particular
student
is
having
a
difficulty
with
lesson
number
14
and
that
particular
concept
on
and
we
provide
feedback
when
needed
as
and
when
needed,
personalized
for
that
student,
and
that
capacity
can
be
done.
It's
not
becomes
available
with
online
education
and
is
typically
not
available
with
a
large
classes
in
person
classes.
A
Machine
teaching
is
the
idea
where
teachers
and
learners
can
teach
machines
themselves,
so
you
don't
have
to
be
an
expert
at
ai
in
order
to
be
able
to
make
ai
agents
work.
For
you
mutual
theory
of
mind
is
the
idea
that
humans
understand
how
the
ai
works
and
ai
understands
how
the
humans
work.
So
it
is
a
mutual
theory
of
mind
and
finally,
whenever
we
are
dealing
with
ai
for
humans,
there
is
an
enormous
need
for
responsible
ai.
A
A
So
one
way
of
thinking
about
this
because
ai
has
been
used
for
education
for
a
long
time,
the
notion
of
intelligent
tutoring
systems
have
been
around
for
a
very
long
time.
So
that's
shown
here
in
the
middle
close
to
the
sort
of
zero
zero
zero
axis
in
the
xy
dimension.
That's
its
intelligent
tutoring
systems
typically
deal
with
handcrafted
knowledge,
but
very
well
defined
problems.
For
example,
problems
in
arithmetic,
7,
plus
or
minus
5
is
equal
to
12..
There
was
this
well-defined
problems
which
have
a
single
answer.
That's
where
ai
has
previously
succeeded.
A
We
want
to
look
in
the
horizontal
axis
here
for
complex
problems,
ill-defined
problems,
for
example,
scientific
thinking
or
systems
in
another
dimension
in
the
y
dimension.
We
want
to
think
in
terms
of
not
just
one
learner,
interacting
with
a
machine,
because
that's
not
how
learning
occurs.
Typically,
learning
is
a
social
process.
So
how
do
we
enable
a
community
of
learners
to
engage
with
ai
and
with
each
other?
A
So
those
are
sort
of
big
picture
goals
here,
and
this
is
what
the
feedback
loop
that
I
was
talking
about.
So
usually
you
know
in
online
education,
teachers
begin
with
some
notion
of
how
learning
will
occur
and
the
constraints
and
affordances
of
online
learning
environment
and
to
teach
and
learners
learn.
But
now,
because
of
online
education,
we
can
collect
a
lot
of
data
and
do
feedback.
A
So
now
we
feed
it
back
into
teachers
and
learners
and
ai
agents
in
an
unprecedented
manner.
That
does
not
occur
in
in-person
classes,
and
our
hope
is.
Our
expectation
is
that
by
doing
this
kind
of
feedback
of
information
we
can
we
can
increase
the
quality
of
online
education,
improve
it
to
approximately
the
level
of
in-person
education,
perhaps
even
better.
A
One
of
the
things
that
connects
with
so
much
with
the
notion
of
nsf
south
big
data
hub
is
that
we'll
be
collecting
data
from
overall
over
the
entire
period
of
the
five-year
period
of
the
institute
from
over
2.5
to
3
million
learners.
This
will
be,
we
believe,
the
largest
data
set
of
data
repository
of
earlier
learning
anywhere.
A
We
think
in
the
world,
but
certainly
in
the
country-
and
we
want
to
build
this
in
such
a
way
that,
within
the
bounds
of
privacy
and
anonymization
and
protecting
student
rights,
this
is
shareable
with
the
learning
community
as
a
whole,
not
just
within
our
institute,
so
we'll
build
safeguards
there.
But
within
those
safeguards
we
want
it
to
be
available
as
a
sort
of
public
resource.
A
We
will
be
following
this
sort
of
typical
design
based
research
methodology
for
this.
So
these
days
this
sometimes
some
people
call
learning
engineering.
So
you
do
interventions.
You
do
formative
and
summative
assessments.
Frequent
of
them-
and
they
sort
of
repeat
the
cycle
so
I'll
not
talk
a
lot
about
it
within
a
five
year
span.
We
think
we
can
do
about
13
to
15
such
cycles,
each
cycle
taking
about
a
year,
the
first
part
being
on
design,
the
second
part
being
on
deployment.
A
They
all
have
to
work
together
right,
so
we
can
focus
on
any
one
of
them
and,
for
instance,
ai
technologies,
and
we
can
make
it
work,
that's
very
interesting,
but
by
itself
not
of
much
use
unless
those
ai
technologies
are
working
together
with
learners
and
interaction
and
participatory
design
and
educational
context,
and
all
of
those
things
making
their
work
together.
Well
is
a
really
difficult
challenge.
That's
the
challenge
for
the
institute,
so
that,
of
course
requires
that
we
have
really
good
test
beds.
A
Usually,
when
people
do
this
kind
of
research
research,
they
do
it.
You
know
at
universities,
like
georgia,
tech,
but
we
are
deliberately
focusing
at
the
technical
college
system
of
georgia
as
one
of
our
test
bets.
We
have
several
testbeds,
but
we
are
beginning
with
the
technical
college
system
of
georgia
tcst,
which
has
about
300
000
students,
and
the
reason
for
that
is
that
the
demographics
there
is
dramatically
different
than
say
at
universities,
like
georgia,
tech
or
georgia,
state
or
emory
university
and
within
the
state
of
georgia.
A
A
B
Stop
sharing
the
screen
now
and
thank
you.
Are
there
any
questions?
I
mean
this
is
exciting.
You
know,
thanks
to
me,
I'm
excited
so
there's
any
questions
for
a
shook.
He
might
have
time
for
the
next
10
minutes.
I
think
he
might
have
to
leave
a
little
early.
So
if
we
want
to
ask
questions
now,
I
know
sometimes
we
do
at
the
end
we're
going
to
stop
and
do
it
now.
So
you
can
unmute
and
just
ask.
A
A
A
We
can
have
in-person
classes
and
online
classes,
use
the
same
instructor,
same
content,
same
assessments
and
see
whether
the
in-person
students
are
doing
as
well
as
the
online
students
and
the
other
way
around.
So
that's
one
way
of
pointing
out
to
showing
that,
in
fact,
we
this
our
technologies
help
improve
the
quality
of
online
education
to
make
it
on
par
with
in
person.
A
Another
way
of
doing
this,
of
course,
is
randomized
controlled
trials,
which
is
a
sort
of
standard
way
of
education,
but
there
are
other
ways
of
having
measuring
or
I
think
what
you're
saying
as
well.
So,
for
example,
how
many
people
can
we
reach?
Georgia?
Tech
has
a
omscs
program
which
you
know
has
become
quite
famous,
and
we
are
very
proud
of
the
fact
that
it
has
12
000
students
in
it.
It's
just
awesome.
A
A
Yeah,
so
that's
a
great
question,
not
something
that
we
have
a
full
answer
to
yet,
and
this
is
something
where
we
welcome
input.
So
we
would
love
to
hear
from
you
about
what
kind
of
feedback
would
you
like
to
see?
What
would
the
learning
science
and
education
science
community
like
to
see
so
here's
just
the
beginning
of
the
thought
process
that
we
have
been
engaged
in.
A
So
within
our
ailo
institute,
there
are
sort
of
25
researchers
co-researchers
and
they
definitely
will
look
at
the
data
and
run
algorithms
and
build
generalizations,
for
instance,
learning
progression
or
learning
trajectory
or
trying
to
find
where
a
particular
student
needs
help
or
trying
to
find
what
kind
of
feedback
should
be
given
to
that
particular
student,
and
we
will
do
our
best.
But
I'm
sure
that
25
is
a
very
small
number
compared
to
the
learning
science
community
as
a
whole,
which
has
you
know,
orders
of
magnitude,
more
researchers
in
it.
A
So
we
are
very
open
to
learning
from
you
about
what
kind
of
algorithms
would
you
like
to
use
and
what
kind
of
generalizations
do
you
want
to
extract
out
of
them
and
we'll
try
our
best
to
support
it?
So
you
tell
us,
please
what
you
would
like
to
see
and
we
will
try
to
do
what
we
can
to
see
whether
we
can
help
you
run
those
algorithms
in
those
generalizations.
B
A
A
So
that's
another
way
and
right
now,
in
our
ailo
program
we
have
a
core
set
of
researchers
who
wrote
the
proposal
and
sent
it
to
nsf,
but
very
soon,
we'll
be
starting
a
affiliate
program,
we're
not
quite
there
yet
where
anyone
can
join
the
affiliate
program.
We'll
welcome
your
participation,
because
this
this
has
to
be
a
process
where
we
have
to
engage
a
larger
community
and
not
keep
it
limited
to
just
25
researchers.
B
B
A
Yeah
they're
not
a
great
question,
we're
especially
interested
in
teachers.
I
think
learners
may
come
later,
but
first
we
want
to
focus
on
teachers.
So
what
happens
is
that
we
can
create
an
ai
technology,
but
teachers
will
not
use
the
ai
technology
until
unless
it's
they
can
develop
their
own
ai
technology.
So
you
know
this
is
a
finding.
Ai
technologies
often
fail
because
they
can
do
something
interesting,
but
nobody
uses
them.
A
They
start
using
pen
and
paper
again
because
you
know
to
use
the
ai
technology
itself
is
an
upfront
investment
and
if
a
teacher
has
to
make
a
firm
investment
of
50
hours,
you
know
teachers
are
so
busy.
No
teacher
is
going
to
spend
50
hours
of
his
or
her
time
in
the
first
week
of
classes,
then
they
hope
that
this
ai
technology
will
help
them
later.
A
So
we
have
to
make
it
so
this
we
have
to
make
this
teachable
so
that
any
teacher
can
teach
the
air
technology
about
the
way
they
want
it
to
be
done,
and
we
have
to
do
it
in
such
a
way
that
a
teacher
can
do
it
in
not
in
50
hours
or
500
hours,
but
in
five
hours
or
half
an
hour
and
that's
the
challenge.
So
so
that
raises
a
different
set
of
issues
for
ai.
A
How
do
we
democratize
ai
and
how
do
we
make
it
teachable
and
tuneable
so
that
any
teacher
can
do
it
and
he
or
she
doesn't
have
to
be
an
expert
in
ai?
We
don't
know
how
to
do
that,
but
that's
one
of
the
challenges
when
I
say
we
don't
know
how
to
do
that.
We
have
research
projects
going
on
going
on
going
on,
of
course.
Otherwise
we
would
not
be
thinking
about
this,
but
we
haven't
completely
solved
the
problem.
Yet.
B
And
we
can
put
you
know,
I
think
vera.
This
is
a
different
project
that
you
worked
on
before.
That
was
very
much
focused
on
an
ai
to
help
researchers
and
and
kind
of
bringing
in
that
data.
So
we
could
put
that
link
in
there
too.
It's
a
example:
it's
not
in
the
education
space,
but
it's
definitely
ai
made
easy
for
researchers,
and
that
was
a
very
interesting
project
as
well
that
you
know
ashik
was
looking
at
in
our
in
his
spoke
project.