“These
must
indeed
be
splendid
clothes!
Had
I
such
a
suit
I
might
at
once…be
able
to
distinguish
the
wise
from
the
foolish.”
—
The
Emperor’s
New
Clothes,
Hans
Christian
Anderson
Remember
the
Hans
Christian
Andersen
story
The
Emperor’s
New
Clothes?
It’s
about
an
emperor
who
is
convinced
by
some
vendors’
BS
to
buy
a
set
of
what’s
described
as
a
beautiful
set
of
clothes.
There’s
only
one
problem;
the
clothes
are
imaginary.
When
the
emperor
wears
(or
actually
doesn’t
wear)
the
clothes
in
a
big
parade,
his
constituents
are
afraid
to
say
that
he’s
wearing
no
clothes.
Until
a
young
child
blurts
out
the
truth:
“The
emperor
is
wearing
no
clothes!”
I
can’t
help
thinking
a
bit
about
that
story
as
I
see
AI
tools
that
promise
everything
but
may
deliver
questionable
actual
value.
A
Siren
Song
for
Email?
Many
of
these
kinds
of
AI
tools
were
on
display
last
week
at
the
AI
Summit.
I
attended
a
presentation
by
Kyle
Miller,
Yahoo
Email
General
Manager.
Miller
described
an
AI-based
platform
for
managing,
summarizing,
and
acting
on
emails.
Let’s
face
it,
emails
are
the
bane
of
most
lawyers’
existence.
We
get
hundreds
of
them
every
day.
Many
are
routine.
Many
require
no
response.
Many
are
marketing
BS.
But
some
are
critically
important.
I
wanted
to
see
if
the
Yahoo
AI
email
management
was
something
useful
or
just
another
AI
tool
that
takes
more
time
than
it’s
worth.
The
goal
behind
the
Yahoo
platform
is
a
good
one:
Leverage
AI
to
create
a
more
personalized
email
inbox
experience.
The
idea
is
built
around
four
pillars:
The
tool
will
catch
what’s
important
in
the
email,
will
be
able
to
act
on
it,
will
adapt
as
it
goes
along,
and
will
evolve.
“The
goal,”
said
Miller
“is
that
the
platform
would
be
more
like
a
companion.”
It’s
similar
to
some
other
platforms
offered
by
email
providers.
I
took
a
look
at
the
present
state
of
the
Yahoo
mail
platform.
It
did
a
pretty
good
job
of
summarizing
emails
with
offers
and
deals.
It
did
not
do
quite
as
good
a
job
at
summarizing
substantive
emails.
The
Verification
Paradox
Miller
made
an
interesting
point:
If
it
takes
more
time
to
create
the
prompt
to
get
the
output
than
it
takes
to
get
the
output
some
other
way,
why
use
it?
I
was
skeptical
whether
the
platform
Miller
promoted
would
be
the
same
way.
Would
it
take
more
time
to
read
the
summary
and
what
the
AI
agent
planned
to
do
than
it
would
to
just
read
the
email
itself?
It’s
the
same
problem
Melissa
Rogozinski
and
I
have
written
about
with
respect
with
respect
to
verification
of
AI-generated
citations:
The
time
savings
are
offset
by
the
time
needed
to
verify
what
the
tool
provided.
In
our
rush
to
adopt
the
new
and
shiny
AI
toys,
we
forget
whether
they
really
do
what’s
promised.
The
ROI
of
AI
is
time
savings.
If
it
doesn’t
save
time,
why
use
it?
Why
buy
it?
ROI
and
Legal
Having
said
all
that,
I
wonder
about
the
use
of
these
kind
of
AI
email
platforms
in
legal.
I’m
not
sure
these
platforms
pass
the
ROI
test:
I
don’t
think
they
save
that
much
time
in
many
circumstances.
First,
any
substantive
email
(or
those
that
might
be
substantive)
will
need
to
be
reviewed
in
total,
irrespective
of
any
summary.
What
could
look
like
a
simple
reply-all
might
contain
something
the
AI
tool
might
miss.
More
complex
and
detailed
emails
often
require
an
understanding
of
context
and
nuance.
They
may
require
an
understanding
of
the
author,
the
business
from
which
they
come,
past
experiences,
and
the
like.
Perhaps
someday,
the
AI
platforms
may
be
able
to
do
this.
But
I
don’t
have
enough
confidence
in
them
today
to
cede
that
understanding
and
not
read
the
underlying
email.
Think
of
it
this
way:
You
may
know
your
client
in
ways
the
AI
tool
can’t.
What
they
like.
The
latest
book
they
read.
Their
politics.
All
of
those
things
could
make
a
difference
in
how
you
respond
to
a
routine
email.
Or
how
about
this:
you
get
an
email
from
opposing
counsel
that
appears
to
be
routine
case
coordination
but
actually
contains
a
subtle
shift
in
their
settlement
position.
An
AI
summary
might
characterize
it
as
“standard
case
update”
and
miss
the
strategic
implications.
And
of
course,
if
I
have
to
read
the
underlying
email,
I
haven’t
saved
time.
I’ve
actually
spent
more
time.
Let’s
say
it
takes
a
minute
to
read
the
AI
summary
and
a
minute
to
read
the
email.
That’s
two
minutes
spent.
If
you
just
read
the
underlying
email,
it
takes
one
minute
total.
Add
that
up
over
100
emails
and
you’re
talking
real
time.
You’ve
spent
time
you
didn’t
need
to
and
may
have
even
billed
for
it,
much
to
your
client’s
dismay
and
state
law
ethics
updates.
But
that’s
the
paradox
we
find
ourselves
in
today:
Believing
without
question
in
the
power
of
AI.
Sometimes
the
emperor
indeed
has
no
clothes.
Unquestioned
Reliance
Leads
to
Compliancy
Without
doubt,
there
are
times
that
AI
platforms
save
time.
Significant
time.
They
save
time
by
separating
out
pure
marketing
materials.
Or
summarizing
offers.
Or
ferreting
out
promotional
materials
masquerading
as
“newsletters.”
But
the
danger
lurking
is
overreliance
and
complacency.
The
more
we
use
the
tool,
the
more
we
rely
on
it.
We
tend
to
get
complacent
and
forget
it
can
miss
things.
It
doesn’t
always
get
nuance.
It
doesn’t
know,
for
example,
the
person
sending
the
email
might
communicate
using
a
blunt
transactional
style
whereas
another
may
use
a
softer,
expressive
one.
Knowing
that
could
make
all
the
difference
in
how
you
react
and
respond.
I
have
worked
with
people
on
both
sides
of
these
styles
by
the
way.
I
can
read
an
email
from
someone
with
the
blunt
transactional
style
and
if
I’m
not
careful,
since
I’m
a
more
expressive
communicator,
get
really
agitated.
I
have
to
make
myself
slow
down
and
say,
“That’s
just
Sam’s
style,
don’t
take
offense.”
If
I
didn’t
do
that?
An
email
war
between
two
lawyers
could
erupt
costing
everyone
way
too
much
time
and
energy.
An
AI
tool
may
be
able
to
make
that
distinction
after
some
training
and
seeing
several
Sam
emails.
But
it
might
not
and
certainly
might
not
early
on.
At
the
very
least,
it’s
worth
hitting
pause
first
and
asking
the
question.
And
knowing
that
could
make
all
the
difference
in
the
end
result
and
your
peace
of
mind.
It’s
akin
to
getting
lazy
about
checking
citations:
The
AI
output
looks
good.
It
sounds
good.
To
save
time,
it
becomes
tempting
to
rely
on
it.
I’m
Not
a
Luddite
Don’t
get
me
wrong.
I
am
an
AI
fan.
But
in
the
age
of
hype
and
hyperbole,
where
vendors
and
pundits
beat
the
AI
drum
over
all
else,
it’s
right
to
apply
some
lawyerly
skepticism.
To
question
some
of
what
we
are
hearing.
To
take
with
a
grain
of
salt
that
AI
actually
saves
us
time
or
whether
we
just
think
it
does.
To
understand
that
just
because
a
vendor
says
the
clothes
for
the
emperor
are
beautiful
doesn’t
mean
they
are
or
even
that
they
exist.
In
legal,
where
the
margin
for
error
is
small
and
the
stakes
are
high,
we
can’t
afford
to
let
AI
vendor
promises
override
our
professional
judgment.
Before
implementing
any
AI
email
tool,
ask:
What
specific
problem
does
this
solve?
How
will
I
measure
the
time
savings?
What’s
my
fallback
if
the
tool
misses
something
critical?
Sometimes,
the
emperor
is
wearing
no
clothes.
Stephen
Embry
is
a
lawyer,
speaker,
blogger,
and
writer.
He
publishes TechLaw
Crossroads,
a
blog
devoted
to
the
examination
of
the
tension
between
technology,
the
law,
and
the
practice
of
law.