►
Description
Panel: The Future and Ethics of AI with Alex Housely of Seldon, Frederick Kautz of Doc.ai and Daniel Riek of Red Hat.
Filmed October 28th, 2019 in San Francisco.
B
Right
can
yep
come
on?
Okay,
oh
so
I'm
Aaron,
Boyd
I
spoke
a
little
bit
earlier
about
storage.
I'm
gonna,
be
your
moderator
if
you
are
just
joining
us,
I
work
for
Red,
Hat
and
office
of
the
CTO,
primarily
working
on
hybrid
cloud
and
multi
cluster
capabilities
for
storage.
But
enough
about
me
I'm,
not
the
smarties
on
this
panel.
C
A
Good
afternoon
so
I'm
Alex
Hasley,
founder
and
CEO
Sheldon
we're
an
open
source
machine
learning,
deployment
platform,
providing
models,
serving
model
management
and
governance,
so
yeah
good
question,
and
so
I
I.
My
most
controversial
view
is
really
that
this
kind
of
AGI
thing
that
people
are
talking
about
is
probably
unlikely
to
happen.
Okay,.
A
The
topic
quite
quickly
goes
on
to
this
kind
of
singularity,
and
you
know
the
world
in
super
intelligence
and
I
think
you
know.
That
is
quite
what
you
know
way
off,
if
not
unlikely
to
happen
at
all
and
there's
many
more
exciting
things
that
that
you
know
changing
the
world,
revolutionising
all
industries
transforming
with
the
technology
that
we
currently
have
available
today.
So
I
think
it's
really
good
to
focus
on
that
yeah.
D
I'm
Frederick
I'm,
head
of
edge
infrastructure
over
at
da
ki,
which
does
medical,
AI
I,
also
have
worked
in
the
open
source
community
extensively
and
networking,
and
one
of
the
things
that
I
focused
on
is
bringing
things
like
AI
into
into
the
infrastructure.
My
most
controversial
opinion
I
think
among
many
is
I.
Don't
think
that
we're
doing
enough
socially
to
try
to
work
out
what
to
do
when
AI
starts
to
replace
jobs,
and
we
need
to
start
focusing
on
that
now
and
not
wait
great.
B
Those
are
all
really
great
points.
Thank
you
guys.
So,
I'm
glad
you
guys
touched
on
how
AI
is
going
to
improve
our
lives.
It's
not
necessarily
this.
You
know
doom,
except
for
maybe
some
jobs,
but
for
the
most
part,
AI
seems
to
always
be
seen
as
a
bit
of
a
savior
in
enhancing
our
lives
and
a
new
technology
that's
going
to
revolutionize,
but
we're
also
seeing
how
that
technology
can
exploit
people
in
their
data
in
their
privacy.
So,
since
this
panel
is
about
AI
and
ethics,
tell
me
why
we
need
ethics
in
AI
Frederick.
D
So
I
think
you
framed
it
very
well,
so
like
AI
is
going
to
be
everywhere
and
it's
something
that
is
absolutely
going
to
improve
many
aspects
of
our
life
but
like
any
particular
tool.
The
AI
is
particularly
interesting
and
that,
as
we
start
to
build
more
AI
models,
the
type
of
things
that
we
decide
to
build
or
excuse
me
the
the
type
of
things
that
we
that
we
decide
to
train
it
on
the
the
types
of
biases
that
exist
in
the
data
centers
are
amplified.
D
It
says
it's
not
just
like
hey
I
have
this
tool:
I'm
gonna
use
a
tool,
I
used
it
once
it's
like
I
built.
This
tool
is
tool,
it's
fully
automated
and
it
learns-
and
it's
going
to
do
this
thing
over
and
over
and
over
again.
So
we
need
to
make
sure
that
the
type
of
things
that
we
build
that
when,
when
we
build
something,
that's
going
that
it's
going
to
reflect
on
our
on
our
TEKsystems
on
howie,
on
how
we
approach
things.
D
And
so
we
need
to
make
sure
that
we
that
we
form
these
type
of
thoughts,
that
we
have
these
type
of
discussions
that
are
absolutely
important
to
have
so
that
we
can
all
be
aware
of
them.
And
even
if
you
don't
have
all
the
answers-
or
we
don't
have
all
the
answers
today,
just
the
fact
that
we're
a
little
bit
more
aware
of
them
means
that
we
can
drive
in
the
in
the
right
direction
and
getting
come
up
with
a
more
fair
and
and
beneficial
outcome.
A
Say
it's
so
much
about
enforcement,
but
more
about
like
how
can
you
put
in
place
processes,
tools,
etc,
to
enable
an
organization
to
actually
deploy
machine
learning
models
in
a
fair
and
ethical
way?
You
know,
data
science
in
itself
was
and
still
is,
in
many
cases,
a
big
challenge
for
organizations.
A
So
you
know
that's
great
for
the
kind
of
challenge
between
data
science
and
DevOps
and
how
the
two
kind
of
teams
kind
of
collaborate
through
deployment,
but
the
things
which
really
matter
you
know
to
our
organization
as
a
whole
and
to
you
know,
execs
and
and
and
sort
of
you
know
people
at
board
levels
etc
are
around.
You
know,
will
my
organization
get
fined
by
regulators?
Will
we
get
reputational
damage
you
know?
Will
we
kill
people
by
accident?
A
B
B
C
Yeah,
so,
in
the
way
ice
used
today,
it's
basically
in
in
use
case
where
you
have
limited
liability
or
you
are
scapegoating
someone
else
with
a
liability
and
I
was
like
you
know,
if
I,
if
I
Drive
a
self-driving
car
famous
car
maker
from
California
I,
you
know
it
drives
itself
on
the
highway
about
in
messages.
Well,
I'm,
not
going
to
say
how
fast
it's
going.
C
It's
a
big
scandal
right
that
that
happened
and
it's
it's
a
whole
different
story
like
humans,
like
self-driving
cars,
have
a
much
better
track
record
than
humans,
as
in
like
they
kill
less
people
per
a
million
miles
driven,
but
it's
still
like
if
it
happens
once
it's
about
twice,
it's
a
it's,
a
big
big
scandal
and
the
way
they're
working
around
it.
Basically,
they're
telling
you,
as
a
driver,
you're
still
responsible
right,
even
though
everyone
knows
that
you're
not
living
up
to
the
responsibility.
C
The
whole
point
of
having
that
car
is
so
you
don't
so
that,
like
that
works
for
now,
but
that
doesn't
work
in
the
long
term.
Right
we
need,
and-
and
you
know,
if
you,
if
you
look
at
like
most
serious
applications
of
AI,
the
lack
of
explain
ability,
the
lack
of
control
around
is
the
biggest
inhibitor
to
the
actual
use
of
AI
in
many
very
beneficial
areas,
and
we
are
very
confining
it
to
this
kind
of
scapegoat
areas
or
you
can
or
or
confine
it
to
giving
advice.
B
And
so,
when
we
talk
about
as
far
as
liability
and
that
kind
of
also
enters
into
the
realm
of
privacy,
when
we
create
a
new
model
and
we're
training
that
model
we're
using
personal
data,
most
of
the
time
to
be
able
to
Train
that
so
what
is
being
done
within
the
community
to
help
protect
users,
data
or
randomize?
The
data
is,
as
it
learns,
so
that
we're
protecting
user
data
and
lowering
the
liability
to
those
models
that
are
being
created.
You
want
to
start
off
with
that,
frederick
sure.
D
So
there's
a
couple
things
that
you
can
start
off
with,
so
very
common
techniques
are
people
are
starting
with
things
like
anonymize,
datasets,
I
think
we
need
to
be
a
bit
careful
with
those,
though,
because
even
if
you
have
a
data
set,
that's
anonymized
in
isolation.
The
moment
you
started
to
pair
it
up
with
Twitter
data
Facebook
data,
then
you
can
often
do
naanum
eyes
many
of
these
data
sets,
and
so
in
terms
of
trying
to
protect
user
information.
D
So
this
is
something
that
I
think
we
should
have
a
lot
of
training
and
focus
on
is
like
how
do
is
how
do
we
develop
and
use
techniques
that
are
designed
to
still
learn
the
signal
of
a
population
or
the
signal
of
your
data
set,
but
not
learn
any
individual
part
of
that
data
set,
so
there's
techniques
that
are
that
are
emerging.
So
we
have
things
like
fader
e,
federated
learning,
which
you
leave
the
data
where
it's
at
remotely
you
send
the
model
over
to
it.
You
train
on
it.
D
You
send
the
results
back,
so
you
never
have
to
centralize
the
data.
You
also
have
other
techniques
like
differential
privacy,
where
you
add
in
noise
in
certain
parts
of
the
while
you're
trained
in
the
model,
and
what
this
noise
does
is.
It
adds
in
plausible
deniability
into
the
model
itself
in
such
a
way
that
it
makes
it
very
difficult
to
extract
information
out
of
it
on
any
given
user,
but
the
noise
is
centered
around
it
centered
around
zero.
D
So
you
still
preserve
the
the
signal,
and
so
they
actually
use
this
technique
very
often
for
for
sensitive
questions
when
they
do
statistics,
so
they
might
ask
a
person
like
hey.
Have
you
tried
cocaine
in
the
past
year
and
if
you
just
ask
that
question
flat-out
people
will
say
no
for
a
variety
of
reasons
and
if,
but,
if
you
put
the
person
into
let's
say
a
into
a
box,
that's
isolated
and
you
put
a
queen
in
there
and
you
say:
okay!
Well,
flip
the
coin.
D
If
the
coin
comes
up,
heads
answer
the
question:
if
the
coin-
and
you
flip
the
coin
again,
just
do
it
to
erase
your
initial
coin
toss
if,
if
it
was
tails
on
the
first
time
you
flip
the
coin
and
then
if
it
comes
up
heads
you
write
yes,
if
its
tail,
it
comes
out.
No,
when
someone
says
oh,
you
answered
yes
to
this.
You
said
yeah
I
am
sir
the
the
coin
toss
question,
and
so
it
gives
them
plausible
deniability.
D
It
turns
out
those
same
techniques
work
in
the
while
you're
training
data,
while
you're
training
models.
So
we
can.
We
can
apply
these
type
of
techniques
in
such
a
way
to
help
reserve
them.
So
even
if
you
have
no
intention
of
even
sharing
the
model,
but
perhaps
the
model
is
is
is
stolen
by
some
group
of
attackers
or
so
on
and
ends
up
on
the
dark
web,
like
you
still
have
some
protection
for
those
users
that
you
train
the
model
on
so
I.
D
Think
it's
very
important,
like
these
type
of
techniques,
become
not
only
well
known
but
become
mature
and
and
standardized
through
the
industry,
and
they
do
require
more
data
to
train
on.
But
as
we
start
to
develop
as
an
industry,
we're
going
to
get
better
at
developing
on
large
quantities
of
data
and
also
develop
techniques
that
still
allow
us
to
to
train
on
smaller
sets
of
data,
but
still
maintain
these.
D
B
You
know
after
you've
developed
the
model
and
you've
you've
done
what
you
can
to
anonymize
the
data
or
add
noise,
so
it
makes
it
you
know
fair
quote-unquote.
You
also
have
to
be
able
to
say
how
do
we
get
that
result?
You
know
where
is
the
explained
ability
around
it
Alex
you
want
to
talk
about
that.
Yeah.
A
So
you
know
one
of
the
big
challenges
around
machine
learning
is
effectively
you're
pushing.
You
know.
Large
data
sets
through
complex
algorithms
and
producing
a
model
which
has
you
know
millions
of
features
and
and
rules
effectively
which
are
not
interpretable
by
by
people.
So
you
know
people
often
refer
to
them
as
like
a
black
box
and
there's
a
trade-off
really
between
kind
of
the
performance
or
accuracy
of
the
model
and
the
interpret
ability.
A
So
you
know
if
we
take
that
sort
of
self-driving
car
example,
the
car
will
crash
less
with
a
you
know,
neural
network
deep
learning
model,
which
is
you
know,
totally
uninterpretable
on
a
on
the
most
sort
of
you
know,
precise
basis,
and
so
the
challenge
there
is
really.
You
know
how
do
we
still
produce
an
explanation
that
you
know
is
interpretable
by
humans,
but
you
know
doesn't
require
you
to
use
a
sort
of
substandard
model.
So
you
know
the
variety
of
sort
of
techniques
emerging
most
through
open
research
and
open
source
projects.
A
You
know.
You're
then
able
to
present
back
to
whether
it's
a
data
scientist
looking
to
kind
of
debug
the
model
effectively
or
to
someone
who's
sitting
on
a
customer
service
desk
and
needs
to.
You
know,
speak
to
a
customer.
Then
you
know
these
are
that
these
are
it's
possible
to
explain
it
in
the
context
of
which
features
hadn't
had
that
impact
on
the
output,
and
it
could
be
very,
you
know,
easily
visualized
another
technique
which
was
seeing
a
lot
of
it's
very
helpful
and
interpretive
bullets
around.
A
It's
called
counterfactual
instances,
and
you
know
this
will
tell
you
what
you'd
need
to
change
on
the
input
feature
to
get
another
output.
So,
for
example,
if
you
have
been
declined
alone,
it
would
tell
you
what
you'd
need
to
change
on
the
loan
application
for
the
application
to
be
approved,
you
might
say,
get
a
higher
salary
or
whatever,
and
so
you
know
that's
so.
A
different
type
of
question
ask
the
explainer.
A
So
that's
kind
of
where
we
see
explanations
is
not
just
you
know
a
one
stop
or
one
type
of
question,
there's
lots
of
different
questions,
and
it's
only
just
starting
to
become
kind
of
accept,
accepted
and
understood,
but
from
the
work
that
we've
been
doing,
it's
Elden.
You
know
we
believed
a
lot
of
the
techniques
which
are
now
available
and
really
should
you
know,
are
at
the
standard
right
now
that
they
should
be
adopted
by
regulators
officially,
and
you
know
financial
services
another,
you
know,
industry,
regulated
industries
should
be
able
to
use
these
techniques.
B
B
C
C
Of
these
techniques
will
actually
use
machine
learning
themselves
to
you,
know,
watch
the
machine
learning
and
you
get
like
pretty
complex
things
where
you
know
you
have
to
input
your
code
and
you
have
you
have
training,
data
and
I.
You
know
I,
think
you
need
sufficient
transparency
on
both
of
that
to
actually
be
able
to
trust
us
sure
you
can
always
put
like
measurements
around
it,
but
that
only
Nick.
You
can
only
measure
what
you
it
gets
very,
very
complex,
right
and
gets
very
hard
to
trust.
It
I
think
I
think
it's.
It's
really
really.
C
C
You
know
arms
race
around
AI
that
you
know
Lydia.
You
cannot
prevent
AI
right
like
if
anyone
thinks
we
can
like
just
not
do
it
that
that's
ridiculous
right,
it's
it's!
Actually
that
would
be
unethical
in
itself,
because
we
can
prove
that
AI
saves
lives
in
we
had
a
dreaded
summit.
We
had
a
customer
case
of
like
detecting
sepsis
through
AI,
and
they
could
prove
that
they
saved
lives
with
that,
and
there
plenty
examples
like
that
right.
So
we
have
to
do
it
like
the
the
benefit
of
AI
is
so
clear.
C
So
this
is
not
about
like
limiting
AI,
it's
about
making
sure
that
is
beneficial
and
the
only
way
you
can
do,
that
is,
if
you
create
transparency
and
equal
access
for
everyone,
you
avoid
an
arms
race,
and
all
of
that,
like
the
only
way
to
really
do
that
is
with
open
source.
From
my
point
of
view,
okay,.
D
So
we
start
taking
a
look
at
bias,
so
there's
there's
a
couple
areas
where
more
than
a
few
areas
where
bias
can
can
come
in.
So
on
one
side
when
you
start
looking
at
bias
on
the
open
source
part.
So
you
start
looking
at
what
techniques
are
used
to
train
things,
and
so
we
want
to
make
sure
that
these
particular
techniques
are
well
understood,
well,
known
and
well
researched,
and
so
the
more
eyeballs
you
can
get
on
these
techniques,
the
the
better
that
you
are,
but
at
the
same
time
I.
D
Don't
think
that
open
source
alone
can
solve
many
of
the
bias
problems.
So,
for
example,
when
you're
working
in
the
medical
space,
you
have
HIPAA
data
that
you
may
want
to
train
certain
models
on
that
are
used
to
save
lives,
as
was
described.
If
that
data
is
isn't.
If
we
don't
account
for
bias
within
those
data
sets,
then
we
may
end
up
with
scenarios
where
people
from
from
minorities
or
people
in
poverty
may
end
up
with
worse
outcomes
than
then
people
who
have
currently
have
significant
resources.
D
And
so
so,
we
need
to
make
sure
that
we,
that
we
address
it
from
multiple
from
multiple
angles
but
being
having
an
open
source
model
or
having
an
open
source
thing
that
you
that
you
work
with
it
helps
along
a
variety
of
areas
and
also
even
interest
in
learning
how
to
do
some
other
stuff
like.
If
you
see
this
is
how
we,
this
is,
how
we
fixed
a
bias
issue
and
here's
an
open
source
example
of
how
of
how
we
solved
bias.
D
That
alone
means
that,
even
if
you
have
someone
do
it
in
close
source,
they've
learned
from
the
open
source
or
maybe
have
used
an
open
source
tool
in
order
to
make
that
happen,
and
so
I
do
think
that
open
source
plays
a
very
important
role
in
in
reducing
bias,
but
certainly
is
not
the
only
thing
we
need
to
do.
Ok,.
B
A
So
from
a
privacy
perspective,
well
you
I'm
from
the
UK
and
in
Europe
we
have
this
thing
called
the
gdpr
which
puts
lots
of
annoying
pop-ups
on
people's
websites,
and
you
know,
ultimately,
what
it's
trying
to
do
is
to
request
for
specific
opt-in
for
using
your
data.
You
know
I
think
over
the
last
you
know
couple
of
decades,
or
so
it's
kind
of
been
generally
accepted
that
you
know
you
can
opt
in
just
by
visiting
a
website
or
using
a
service
without
reading
the
long.
B
And
so
talking
a
little
bit
more
about
data
companies
like
pin
screen
that
can
create
realistic
videos
of
someone
talking
and
things
they
didn't
actually
say.
What
are
we
doing
in
open
source
to
you
know
create
data
provenance
knowing
where
the
data
is
coming
from
and
making
sure
that
what's
being
presented
is
actually
where
it
came
from
originally,
oh.
A
A
So
there's
there's
you
know
some
work
from
open
source
projects
like
model
DB,
which
is
it's
been
doing
a
good
job
on
this,
and
you
have
various
efforts
connected
to
you
know.
Various
open
source
ml
platforms
is
another
one
that
Selden's
connected
with
called
cube
flow.
That's
trying
to
figure
this
out
as
well
at
the
moment
so
I.
You
know
it
will
come
down
to
standards
and
metadata,
and
you
know
various
tools
that
are
part
of
that
pipeline.
A
C
Is
that
and
I
think
so?
You
gave
the
example
of
lip
deep,
fake
videos
right
like
where
we
learned
that,
like
video
evidence
is
actually
not
reliable
anymore
because
it
can
be
faked.
Very
you
know,
progressively
convincing
and
you're
part
of
the
problem.
There
is
the
any
solution
to
that
has
in
itself
privacy
implications,
I
like
when
you
start
like
source
signing
every
data
that
you
generate.
C
You
know
you,
video
camera,
basically
signs
all
the
videos,
so
Mithen
itself
becomes
a
problem
because
you
just
eliminated
the
ability
to
have
anonymous
videos,
and
things
like
that,
like
so
I
think
there
are
a
bunch
of
areas
where
we
have
to
find
broader
answer
in
society
and
maybe
start
thinking
about
reducing
the
stakes
a
little
bit
right,
because
some
of
these
things
are
problem
like.
Why
is
privacy
increasingly
like
we
had
a
face
where
no
one
cared
anymore?
C
Know
so
they
like
it,
they
are
there
some
deeper
question.
There
are
not
technology
questions
that
we
have
to
answer,
because
you
know
technology
will
force
us
right.
It
has
forced
us
here
who
is
the
technology
we
already
have
today
and
we
can
predict
where
this
is
going
to
go,
that
it's
going
to
increase
and
we
will
have
to
get
around
to
that.
You
can
go
into
things
like
citizens,
scores
and
stuff
like
that,
or
you
know
we
in
the
u.s..
C
B
D
No
and
that's
a
that's
a
tough
one,
I
think
in
terms
of
then
I'll
scope
this
around
a
high
and
ethics
as
opposed
to
just
like
hey.
How
do
you
do
AI
right,
so
I
think
part
of
it
is
take
a
look
at
what
it
is.
The
the
thing
that
you
want
to
do
bill
you're
a
on
take
a
look
at
the
impact
of
what
is
of
what
it's
going
to
do
have
on
I.
D
Do
do
a
there's,
there's
models
that
you
can
do
now,
where
that
don't
require
AI
to
develop
these
models
like
what
is
the?
What
is
the
risk
of
like
okay,
build
this
particular
system?
What
if
it
goes
wrong?
What,
if
something
breaks
what,
if
you
know
what
what
are?
What
are
the
reaction
risks
that
were
there
were
taking
on
with
this
and
use
that
to
help
develop
a
like
a
thread
model
see
about
developing
it
in
such
a
way
that
you
can
try
to
work
out?
Okay,
what
are,
if
I,
put
any
I
here?
D
What
are
the
risks
and
I'm
from
there?
Don't
skimp
on
on
the
time
necessary
in
order
to
try
to
well
number?
Why
do
you
should
it
should
be
something
you
even
build
in
the
first
place?
But
assuming
you
decide
yes,
it's
it's
worth
it
and
and
we're
gonna
build.
It
then
don't
skimp
on
on
working
on
the
efficacy
and
trying
to
work
out.
Is
this
thing
actually
doing
what
I
think
it
is
and
go
towards
the
explained
ability
and
fairness
and
so
on,
especially
if
it's
on
a
much
more
important
area?
D
You
know
and
it's
it's
really
a
mindset
and
in
a
scenario
it's
like
not
just
throwing
something
out
there,
because
you
saw
it
work
like
give
you
a
cake
example
when
I
was
very
first
starting
when
I
was
first
starting
to
her
in
AI
like
it
was
a
couple
weeks
in
I
was
super
excited.
My
model
had
like
a
seven
point:
five
percent
accuracy
and
then
I
looked
at
the
data
and
I
was
eighty
seven
point.
D
Five
percent
of
the
answers
were
no,
and
so
my
model
was
saying
no
to
everything
because
it
thought
okay,
this
is
great.
It's
working
and
I
was
like
super
excited
and
then
I
realized
I
had
to
spend
more
time
to
work
out.
Okay.
Well,
what
what
do
I
need
to
do
in
order
to
make
this
model
right?
I
know,
that's
an
extreme
example,
but
these
type
of
things
are
going
to
come
up.
D
A
Give
you
a
prompt
of
where
you
should
be
investigating,
so
you
know
there
are
kind
of
packs
of
information,
packs
and
checklists,
as
opposed
for
people
who
are
in
boards
and
running
projects
which
can
help
prevent
them
from
from
getting
something
wrong.
You
know
by
accident
as
well.
That's
one
of
the
biggest
problems
here
is
it's
a
complex
space
and
it's
very
easy
to
get
something
wrong
if
you're
not
looking
in
the
right
areas.
Okay,.
C
I'll,
pile
onto
that
or
it's
you
know,
understand,
problem
space
and-
and
you
know,
is
this
specific
right,
like
the
mentality
you
know
in
in
data
science,
you're
often
happy
when
you
get
99%
right,
but
then
you
know
if
you
apply
that
to
IT
security.
Like
you
know,
intrusion
just
needs
to
happen
once
and
then
you're
screwed
so
like.
So
there
is
a
difference
right.
What
we're
doing
today,
most
cases
are
these
areas
where,
like
you
know,
90
99
percent
is
great.