►
From YouTube: Machines Making Movies - GitHub Universe 2016
Description
Join Benjamin, AI screenwriter of Sunspring, along with his human chaperones Oscar Sharp & Ross Goodwin, as they explore the potential of writing machines and augmented creativity.
Watch Sunspring:
https://www.youtube.com/watch?v=LY7x2Ihqjmc
Find out more about Benjamin: https://benjamin.wtf
More info on the filmmakers' other projects:
Oscar Sharp - http://www.thereforefilms.com/
Ross Goodwin - http://rossgoodwin.com/
Actors:
Safiya Fredericks - http://safiyafredericks.com
Rinabeth Apostol - http://rinabeth.com
About GitHub Universe:
GitHub Universe is a two-day conference dedicated to the creativity and curiosity of the largest software community in the world. Sessions cover topics from team culture to open source software across industries and technologies.
For more information on GitHub Universe, check the website:
https://githubuniverse.com
A
C
C
C
So
that
was
our
film
and
yeah
we'll
get
to
the
explanation.
I
promise
you
of
what
Sofia
was
doing
at
the
start
very
soon,
but
yeah
we
put
this
up
online
and
it
it
got
quite
a
lot
of
hits
if
not
wasn't
huge,
but
her
Albertan
sort
of
about
a
million
or
so
across
a
few
versions.
The
main
riffing
was
the
press
response.
It
was
crazy.
We
had
something
like
a
hundred
really
long,
blog
articles
and
press
articles
rehab.
C
We
were
talking
to
French
TV
this
morning
on
skype,
and
it's
that's
three
months
ago
that
we
did
this
so
something
we
struck
some
kind
of
chord,
and
that
was
weird
and
what
they
all
asked
to
begin
with
is
how
do
you
guys
decide
to
do
this
like?
How
did
you
get
into
it
right
and
we
both
had
very
different
trajectories
coming
in.
C
For
me,
I
was
a
theater
director
in
London
many
many
years
ago,
like
decades
ago,
not
not
even
when
we
did
this
and
I
long
ago,
had
sort
of
had
a
real
interest
in
certain
strange
kinds
of
exercises
we
do
in
the
rehearsal
room.
In
fact,
I
will
show
you
one
of
those
exercises
now
and
we
sort
of
prepared
this
little
thing
earlier.
So
here
is
one
of
these
exercises.
C
C
Round
of
applause,
please,
for
our
amazing
performance,
so
what
you
just
saw
probably
seemed
to
be
something
that
sort
of
made
sense
right.
It
was
a
bit
weird
but
like
it
kind
of
made
sense,
there
was
a
person
interacting
with
the
thing
in
a
way
that
sort
of
made
sense
to
you.
But
the
thing
is:
it
was
created
completely
at
random.
I
had
a
big
long
list
of
possible
actions
and
I
and
I
am
Jose
from
them
at
random
and
chose
a
random
prop.
C
C
C
Not
good
enough
right,
so
I
needed
it.
I
needed
like
a
rule
system,
I
needed
somebody
help
somebody
who's
better
at
making
machines
and
every
time
I
met.
Someone
like
someone
in
this
room.
Anyone
who
was
like
oh
I
can
program
an
engineer.
I
would
grab
them
by
the
lapels
and
go
I
need
to
make
systematized
machine
dialogue.
Can
you
help
me
and
they
would
run
from
the
building
and
but
then
four
for
four
years,
I
find
isn't
going
to
get
anywhere
and
I
sort
of
become
a
filmmaker
and
I
get
a
Fulbright.
C
D
C
It's
this
place.
That's
full
of
all
these
kind
of
mad
scientists,
people
who
are
building
floating
robot
jellyfish
and
entire
walls
of
animatronic
penises
I
am
not
joking.
That
is
something
that
I
saw
there
and
wow.
This
is
a
cool
place.
Maybe
this
is
where
I'll
find
the
person
I
want
to
work
with
so
for
ages.
D
Natural
language
processing
in
natural
language
generation,
which
are
words
that
probably
make
sense
to
most
the
folks
in
this
audience,
but
working
with
spoken
or
written
language,
not
computer
code
using
computer
code
and
and
why
were
you
doing
that?
What
was
the
well
it's
kind
of
a
long
trajectory
like
yours,
so
I
used
to
be
a
political
ghostwriter.
D
If
you
can
believe
that
this
is
me
about
seven
years
ago
and
my
first
job
after
college,
at
the
White
House
as
a
ghostwriter
for
Barack,
Obama
I
started
out
writing
letters
on
the
2008
presidential
campaign
and
at
the
White
House
I
read
presidential
proclamations
very
glamorous
job.
These
are
like
statements
of
national
days
weeks
and
months
of
things.
So
everything
from
Thanksgiving
to
you
know:
African
American,
History,
Month
to
lesser-known
observances
like
safe
boating
week.
Very
exciting
work
did.
C
D
At
this
point,
I'd
write
all
the
first
paragraphs,
then
all
the
second
paragraphs
and
so
on
in
a
spreadsheet
and
then
copy
the
columns
into
word
Docs
and
then
every
other
day,
I'd
run
a
simple
Excel
macro
to
randomly
rearrange
the
cells
and
then
edit,
those
that
handed
in
as
a
new
batch
of
letters.
So
that
allowed
me
to
take
an
eight-hour
a
day
job
and
her
didn't
do
it
two
hour
a
day.
Job
and
I
spent
the
rest
of
my
time
working
on
a
novel.
D
C
D
D
It's
kind
of
like
a
multiplayer
exquisite
corpse
game,
except
for
you
know,
spatially
configured
in
this
spreadsheet,
I
didn't
realize
it
when
I
was
doing
this,
but
I
thought
I
was
onto
something
really
new,
but
lots
of
other
folks
have
done
the
same
thing
in
the
past:
mixing
writing
and
algorithms.
This
is
Ramon
canoes
hundred
thousand
billion
poems
written
in
1961,
and
it's
designed
to
be
rearranged.
D
Canoe
was
part
of
Aleppo,
the
you've,
wadi
literature
potential
or
the
workshop
for
potential
literature,
and
that's
a
group
that
continues
to
explore
the
intersection
of
literature
and
computation
today,
and
it's
just
one
of
many
examples
of
folks
who
have
who
have
done
this
sort
of
work
in
the
past.
But.
C
D
C
Because
we
have
the
issue
that
you
know
before
my
lists
of
actions
work
for
theater,
but
they
don't
really
work
for
film,
so
I'm
like
well.
We
need
to
do
something
with
visuals
Ross
like
we
need
to
have,
and
so,
if
we
start
talking
about
the
rules
of
screenplay
and
in
screenplay,
there
is
some
action
description.
Then
there's
dialogue,
there's
the
only
two
things
and
in
the
action
description
it
tells
you
what
you
can
see.
Basically
yeah.
D
D
It
turns
images
into
strange
prose,
poems
uses
the
clarify,
API,
the
concept
net
database
and
a
template
system,
and
it's
really
about
redefining
the
photographic
experience,
but
really
I
made
it
because
I
was
thinking
about.
You
know
like
I,
said
how
screenplays
are
written.
Only
things
like
scene
gonna
be
described,
and
you
know,
starting
with
the
photos,
seem
like
the
right
place
to
go.
So
this
is
what
the
output
tended
to
look
like.
This
is
a
young
photo
of
vladimir
putin
and
we
have
meanwhile
a
history
group
in
an
outfit.
D
Undoubtedly,
the
history
may
repeat
itself.
The
group
may
reach
agreement
to
be
cooperative
and
the
outfit
is
a
set
of
clothing.
Whenever
the
history
is
for
remember,
passed
to
sum
up,
it
evokes
narrator
also
it
evokes
where
they're
gone
in
following
it
evokes
famous
figure.
So
it's
a
little
choppy,
but
it
is
descriptive
and
it's
basically
taking
words
that
it's
extracting
from
the
image
and
then
expanding
that
set
of
words
and
do
a
set
of
sentences
and
paragraphs
by
referencing
related
words
in
a
database.
So
I
started
to
make
a
physical
device.
D
Out
of
this
and
again
it's
about
redefining
the
photographic
experience,
but
that
was
sort
of
my
tag
at
the
time.
So
I
used
a
thermal
receipt
printer
to
output
the
text
so
that
it
would
be
like
a
Polaroid
and
I
built
them
inside
of
old
cameras,
because
I
wanted
to
provide
a
reference
that
users
would
immediately
understand
and
I
wanted
to
also
communicate.
This
is
a
primitive
form
of
something
that
I
felt
like
had
a
rich
artistic
potential,
so
there's
a
single
button
interface
on
the
back.
D
It's
just
a
it's
a
Raspberry
Pi
with
a
camera
module
inside
of
a
medium
format,
TLR
film
camera-
and
this
is
what
the
output
looked
like
at
the
top.
There
was
on
the
right.
The
left
is
just
a
Polaroid
on
the
top
generated
text
like
what
you
saw
before
and
then
on
the
a
passage
from
a
novel
from
an
onboard
database
that
was
algorithmically
matched
to
the
generated
text,
so
that
you
get
like
a
passage
that
relates
to
the
photo.
So.
C
Picture
if
you
camera
that
I
still
have
my
my
hands
around
Ross's
lapels
and
I'm
like
hooray,
we're
gonna
have
a
screenwriting
machine
and
then
he
builds
a
camera
that
writes
weird
poems,
which
isn't
exactly
a
screenwriting
machine.
And
then
we
try
and
do
dialogue
remember
the
thing
that
I
wanted
more
than
anything
else,
and
it
turns
out
that
Markov
chain
dialogue
sounds
like
this.
C
B
D
C
Very
impressed
with
everything
that
the
roster
tell
it
was
mind-blowing,
but
it
wasn't
really
the
screenplay
generators
or
is
a
bit
sad
at
which
point
my
life
slightly
exploded.
I
had
a
short
film
called
the
Carmen
mine,
which
you
can
look
up
on
the
New
Yorker.
It's
still
there.
If
you'd
like
to
watch
it
and
they
picked
it
up,
and
then
it
was
at
one
of,
though
the
prizes
got
nominated
for
a
BAFTA
and
then
after
the
nomination
I
get
this
phone
call
from
the
forum
of
spider-man
Tobey
Maguire.
C
C
To
write
a
science
fiction
screenplay
for
your
first
feature,
film,
I,
guess
that
sounds
really
fun,
having
never
done
it
before,
and
I
ship
off
to
Hollywood,
to
start
trying
to
write
my
first
science
fiction
screenplay,
which
turns
out
to
be
incredibly
hard,
but
while
I
am
doing
that,
Ross
has
made
a
discovery.
Yes,.
D
And
that
is
the
LST
M
recurrent
neural
networks
that
we
use
to
generate
the
screenplay
for
Sun
spring,
but
I
started
out.
You
know:
I
I,
like
I,
said,
with
the
first
version
of
word
camera
using
like
ap
eyes
and
just
pre-built
solutions,
but
in
December
of
last
year
and
while
you
gave
me
access
to
their
high
performance
computing
facilities,
so
I
was
able
to
start
training.
My
own
neural
nets
and
I
was
really
encouraged
by
the
early
results
from
that
on
the
left.
D
There's
some
image:
captioning
image,
captions,
sorry,
which
are
created
using
a
convolutional
neural
net
and
an
LS
TM
in
tandem
and
then
on
the
right.
We
have
some
poetry
that
was
generated
from
a
large
poetry,
corpus
or
a
large
collection
of
poetry,
texts
and
I
was
really
encouraged
by
these
early
results
and
I
I
kept
going,
and
you
know
the
text
doesn't
just
appear
by
itself.
It
has
to
be
seated
or
in
other
words
it
has
to
be
preceded
by
a
phrase
or
a
word,
or
even
just
a
letter
that
the
computer
then
completes.
D
So
you
know,
if
you
have
the
meaning
of
life.
Is
the
computer
completes
it
as
the
meaning
of
life
is
a
perfect
vision
of
the
proper
form,
the
concept
of
poetry,
referring
to
the
result
of
the
possession
of
life
and
strength
and
pride,
and
so
on,
so
that
seating
technique
you
know
or
that
that's
seating,
you
know
really
I
was
trying
to
find
ways
to
play
with
that
and
I
made.
D
This
bot
called
lexicon
drew
at
the
Twitter
bot
and
I
I,
made
it
by
training
an
LST
M
on
the
Oxford
English
Dictionary
and
by
seating
it
with
a
word
and
then
a
special
character
and
the
word
an
especial
character.
You
can
then
generate
a
plausible
definition
for
a
made-up
word
or
a
plausible
division
for
a
real
word.
If
any
one
you
want
to
do.
But
if
you
tweet
at
lexicon
juror
and
then
a
made-up
word,
you
will
receive
a
definition.
D
D
D
By
using
those
image
captions
to
seed
a
poetic
language
lsdm
I
was
able
to
create,
you
know
more
coherent
text,
a
little
bit
more
poetic
from
images,
and
that's
how
my
github,
it's
called
neural
snap,
and
so,
while
working
on
that
I
wanted
to
make
a
new
physical
device
and
a
documentation,
video
for
that
physical
device,
so
I
sent
the
poem
to
Oscar
because
oscar
has
a
wonderful
voice.
So
there
I
record.
C
It
for
voices
languishing
languishing
in
Los
Angeles,
struggling
desperately
to
write
my
first
science
fiction
screenplay
wishing
desperately
that
I
had
a
machine
to
do
it
for
me
and
feeling
a
bit
sad
and
I
realized
that
it
was
the
build-up
to
my.
It
was
my
birthday
and
my
birthday
falls
every
year
around
the
same
time
as
the
London
sci-fi
film
festival.
Short
film
challenge
right,
so
they
do
this
48
hour,
film
charmese.
They
give
you
48
hours,
they
give
you
a
line
of
dialogue
and
the
title
and
a
prop.
C
You
saw
this
at
the
beginning
and
exactly
40
hours
later,
you
upload
a
movie
that
you
have
written,
shot
and
cut
in
that
time,
which
is
tricky
but
kind
of
reboots.
Your
excitement
for
the
world
and
I'm
sort
of
thinking
about
this
wondering
what
I'm
gonna
do
for
it.
Wishing
that
how
to
machine
to
write
your
screenplays
and
I
get
this
email
from
Ross
with
a
pulley
minute
now,
as
you
know,
from
earlier
on,
in
the
story,
I've
seen,
Ross's
and
machine
generated
poems
before
and
I
have
a
slightly
bitter
memories
about
them.
C
C
So
the
thing
is
right:
this
was
better
than
the
previous
poems.
He
had
send
me.
It's
still
a
bit
weird,
but
it's
like
more
word,
salad,
weird
Markov
chain,
nonsense
like
make
it
before.
In
my
opinion,
so
I'm
like
this
is
strange.
I'm
experiencing
emotions,
I
can't
believe
it
so
I
phone
him
excitedly
and
go
Russ
Ross
I'm
experiencing
emotions.
You've
made
something
emotional.
What
can
we
do?
Maybe
I
should
perform
this
on
stage
no
wait.
Maybe
we
should
get
access
to
perform
it.
No
wait
hang
on.
D
C
C
It's
so
beautifully
performed
that's
from
that
model
and
it
was
cool,
but
the
problem
is
it
just
went
on
and
on
and
on
and
on
and
on
and
on
it
was
one
person
talking
infinitely,
and
this
was
a
bit
of
an
issue
because
that's
not
a
screenplay
looks
like
so
I
go
Ross,
Ross
I
think
we
could
use
this
for
this
film
competition.
But
can
you
make
it
look
more
like
a
screenplay?
Can
it
have
description
description
in
the
dialogue
be
of
a
normal
length
so.
D
I
took
a
very
obtuse
route
to
create
that
I.
Just
use
screenplays
variable
I
used
a
bunch
of
screenplays
further
than
raw
dialogue,
about
170
movies,
all
science
fiction
and
about
five
complete
TV
shows,
including
with
the
original
Star
Trek
and
Stargate
sg-1,
and
a
bunch
of
older
stuff.
And
the
cool
thing
is
that
the
ls
team
can
actually
replicate
the
formatting
of
the
screenplays.
D
Yeah
yeah,
the
character
names
are
tricky,
cuz,
proper
names,
LST
M's
have
a
lot
of
trouble
with
them,
because
if
you
think
about
it,
a
proper
name
is
a
word.
They
can
occur
in
the
same
context
and
two
completely
different
sentences
or
sorry,
two
identical
sentences
other
than
the
word
being
different
and
an
LS
TM
only
understands
a
character
to
character
sequences.
So
to
see
two
completely
different
words
in
the
same
context
is
very
I,
guess
confusing
for
the
machine.
D
So
there's
a
couple
solutions
that
I
found
to
this
problem,
one
of
which
was
to
reduce
the
sort
of
name
space
of
the
corpus
by
taking
the
large
set
of
names.
And
then
you
know,
regular
expression,
replacing
them
with
a
relatively
smaller
set
of
names,
but
we
did
for
some
spring
was
just
replace
those
names
with
the
first
letter
of
each
name.
So.
C
C
Finally,
had
a
machine
to
cover
ice
cream
space,
so
we
decided
we
gonna
do
it.
We
gather
a
room
together
on
the
on
this
particular
day.
My
birthday
as
it
happens,
we
have
our
48-hour
clock,
begin.
I've
got
this
one
thing
generated
by
by
the
by
the
machine
which,
by
the
way
at
the
time,
was
called
Jetson,
not
Benjamin
you'll
find
out
why
a
bit
later
and
so
Jetson
inspired
the
screenplay
and
I
gather
together.
All
the
actors
in
the
crew
and
I
didn't
want
to
interpret
it
beforehand.
C
I
put
everyone
in
one
room
and
I
gave
the
actors
the
thing
and
they
read
it
and
basically
I
want
you
to
see
just
how
cool
that
was
the
first
time.
We
did
it,
because
this
is
what
it's
all
about
the
fact
that
it
was
finally
something
an
actor
could
make
sense
of.
So
we
are
going
to
do
that
today.
In
the
same
way
with
our
two
glorious
actors.
Here,
Ross
I
would
like
to
generate
a
brand
new
piece
of
dialogue
for
us.
C
A
B
C
A
So
what
I'm
doing
a
play
or
doing
anything
else
like
you
generally,
have
like
a
whole
script
in
an
arc,
and
you
can
kind
of
tell
in
the
beginning
like
what
the
feel
of
this
character
is
like
what
their
voice
is
like
compared
to
what
this
other
voice
is
like.
But
with
this
it
was
like
whoo.
What
is
this
voice,
and
you
just
kind
of
have
to
put
your
experience
and
everything
else
kind
of
on
top
of
it,
rather
than
it
being
kind
of
generated
from
well.
C
You
were
about
to
do
that
interpretation
again.
Let's
give
you
one
of
each:
let's
assign
roles,
I,
guess
I
guess
you'll
speak
first,
let's
say
you
can
be.
The
character
is
called
a
and
D
a
and
D
are
here
and
let's
give
you
a
little
situation,
how
about
you're,
waiting
and
you're
waiting
and
waiting?
And
you
move
her
to
come
home
and
finally,
she
comes
home.
I
have
no
idea
what
this
is
like
I,
just
think
that
might
help
them,
let's
find
out
and
growl
around.
Please
roll
camera
action.
E
A
E
E
A
C
D
C
The
title
only
the
title
generator,
so
we
need
that
title
for
that
movie.
Right,
like
this
incredible
film
about
sex
changes
on
a
spaceship,
it's
definitely
called
ranch
boy.
The
next
generation
of
ranch
boy,
the
next
generation
that
is,
that
is
perfect,
holy
wood
Street.
This
is
so
these
are.
These
are
the
corpus.
The
input
corpus
for
this
are
all
of
the
movie
titles.
We
could
find
and
I
find
this
an
incredibly
useful
tool
as
a
filmmaker.
We
also,
we
tried
a
few
other
things.
D
C
The
dead
Peter
as
a
young
man
was
having
a
hard
time,
so
it's
interesting
I'm
forced
to
confront
this,
probably
the
most
common
word.
In
any
phrase,
in
any
movie
synopsis,
young
man
is
probably
the
second
most
common.
Actually,
we
ran
a
concordance
on
our
hundreds
and
hundreds
of
these,
and
we
found
out
that
men
I
mentioned
four
times
more
often
than
women,
and
that
a
small
town
is
definitely
the
most
common
place
that
movies
are
set.
So
this
has
been
an
incredible
ride
for
us.
C
We've
got
like
a
couple
of
minutes
left,
so
we
kind
of
should
wrap
things
up
here.
We
we
entered
the
contest
with
the
with
this.
With
this
film,
it
was
shortlisted
among
the
top
ten.
We
noticed
that
a
couple
of
the
other
contestants
in
the
online
poll
were
cheating.
They
were
click
and
we
were
annoyed
about
that,
and
so
our
machine
we
go
back
to.
C
Please
go
out
to
the
slide.
Our
machine
Jetson
as
it
was
called
at
the
time,
was
employed
to
vote
against
them
36,000
times
an
hour
and
constantly,
and
consequently
we
crushed
them,
but
we
phoned
the
people
who
ran
the
contest
the
next
morning.
I
went,
oh,
my
god.
Our
machine
went
rogue,
we're
so
sorry
we
didn't
mean
for
this
to
happen
and
they
thought
it
wasn't.
There
is
so
they
decided
to
interview
him
live
on
stage,
and
this
was
the
last
question
they
asked.
He
says
with
90
seconds
to
go
in
our
yeah.
C
Like
I
said,
you
may
have
been
wondering
how
it
is
that
it
came
to
be
called
Benjamin
right,
so
it's
runs
on
a
piece
of
software
called
jet.
A
piece
of
hardware
called
the
Nvidia
Jets
and
we'd
been
nicknaming
at
Jetson,
but
they
asked
it
what's
next
for
you
and
Jetson
as
it
was
called
then
replied
here
we
go.
The
staff
is
divided
by
the
Train
of
the
burning
machine
building
with
sweat.
No
one
will
see
your
face.
The
children
reach
into
the
furnace,
but
the
light
is
still
slipping
to
the
floor.
D
So,
just
back
to
the
film
for
a
second
I
really
love
this
scene,
the
the
line
from
the
screenplay-
and,
if
you
don't
know,
we
generated
the
action
descriptions
as
well
as
the
dialogue.
Obviously
the
line
from
the
screenplay
was,
he
pulls
his
eyes
from
his
mouth
and
I.
Think
this
really
shows
the
potential
of
a
learning
machine
for
creating
rich,
provocative,
abstract
imagery.
But
this
you
know
like
a
little
bit
more.
The
lion
is,
he
pulls
the
camera
toward
his
back.
D
He
is
on
the
phone,
but
then
the
angle
changes
and
he's
holding
nothing
and
the
first
part
was
machine
dictated.
The
second
part
was
human
interpretation
that
was
informed
by
years
of
Oscars,
filmmaking
experience
and
I.
Think
that
cycle
of
generation
and
interpretation
is
what
demonstrates
the
true
augmentative
capacity
of
these
learning
machines
and
how
they
can
help
us
tell
new
types
of
stories
that
we've
never
seen
or
heard
before,
and
that's.
C
Really
the
point
for
both
of
us
is
I
love
the
idea
of
these
collaborations
as
much
as
I
love,
collaborating
with
actors.
I
love
collaborating
with
all
of
my
collaborators
on
every
film
set
because
directors
don't
do
anything
they're
like
crowd-surfing
sloths.
Everyone
else
does
all
of
the
work,
but
but
now
I
have
a
new.
A
new
member
in
my
crowd.
I
have
this
incredible
machine
which
it's
not
the
machine?
It's
the
corpus,
it's
it's!
The
the
input
I
can
have
all
of
my
favorite
rom-com
writers.
Help
me
with
the
rom-com
in
writing.
C
I
can
have
all
of
my
favorite
cinematographers
as
we
build
Benjamin
out.
Help
me
choose
my
shots.
Help
me.
The
designers
help
me
to
help
me
design
the
colors
and
so
on,
and
this
is
a
completely
fascinating
thing
for
me,
and
it
means
that
between
us
we're
hoping
to
move
the
film
forward
so
I'm,
you
know,
if
any
of
you
guys
have
any
ways
we're
helping
us.
Please
come
and
have
a
chat,
we'll
be
around.
Ross
will
be
around
more
than
me,
but
that
is
the
plan
and
thank
you
so
much.
Thank.