All language subtitles for Secrets.Of.Big.Data.S01E02.WEBRip.x264-ION10

Narrator: Just east
of Tehran, Iran,

shots ring out as a brazen
daylight assassination

is carried out, killing top
Iranian nuclear scientist

Mohsen Fakhrizadeh.

As the gunfire ceases,
there is no one around

with their finger
on the trigger.

Ramona Pringle: Strangely the
bodyguards didn't find anybody

in the area, so who
killed Fakhrizadeh,

and how exactly did
they pull it off?

Narrator: On a quiet
night in Arizona USA,

one of the marvels of
modern technology,

a driverless car,
is out for a test drive,

when it fails to recognize a
pedestrian in its path.

Nikolas Badminton: Sensors only
recognized the bicycle

she was pushing

and then they could only
recommend emergency braking.

Narrator: At a
famous auction house,

a French Portrait created by
Artificial Intelligence

goes up for sale, stoking
controversy in the art world.

Pringle: Some in the
art establishment

weren't exactly thrilled to see
a computer generated painting

up for auction.

Mhairi Aitken: But the
suggestion that the piece

was created with no
human involvement

is overly simplistic.

Somebody had to write
the code that created it.

♪

Narrator: These are the stories
of the future,

that big data is bringing
to our doorsteps.

The real world impact of
predictions and surveillance.

The power of artificial
intelligence

and autonomous machines.

For better or for worse, these

are the Secrets of Big Data.

Mohsen Fakhrizadeh,

the head of Iran's
nuclear weapons program,

and the country's
deputy of defence,

is driving with his wife to
his vacation house in Absard,

around 80 kilometers
east of Tehran,

when a sudden burst
of gunfire rings out

across the quiet countryside.

The bullets tear through
the hood of his car.

As Fakhrizadeh swerves
and tries to stop,

another round of shots
shatters the windshield.

Now wounded, he exits the car
and tries in vain

to hide behind the open door.

A third torrent of shot tears
through Fakhrizadeh's body,

killing him instantly.

Anthony Morgan: This was a
brazen assassination

of a top government official.

Narrator: Fakhrizadeh's
security detail

that had been tailing his car
rushes to the scene, guns drawn,

ready to engage
with the attacker.

Pringle: Strangely, the
bodyguards didn't find anybody

in the area.
So who killed Fakhrizadeh?

And how exactly did
they pull it off?

Narrator: The answer
sits in an office

far away from the carnage,

staring at a computer screen as
though playing a video game.

But this is no video game.

This is a complex
military operation.

Morgan: The assassin
ended Fakhrizadeh's life

without ever setting foot
on Iranian soil.

Narrator: Unmanned military
weaponry is nothing new.

As far back as the 1930s,
the U.S. Navy

was experimenting with
unmanned aerial vehicles,

commonly known as drones.

But it wasn't until the
Vietnam War

that they were deployed
extensively.

Aitken: Early drone models
were used mainly

for reconnaissance and acted
as decoys during battles.

They also contributed to US
propaganda campaigns,

dropping leaflets in
enemy territory.

Narrator: Even though the US was
able to eventually succeed

in mass-producing drones for
military purposes,

they were often viewed as
costly and unreliable.

However, in 1982 this
attitude shifted

when Israel defeated
Syrian forces using

unmanned aircraft and suffered
very few casualties.

[explosion]

Pringle: The United States
sat up and took notice

of the success Israel had with
drones and decided to redouble

their efforts to further
develop their own technology.

Narrator: Weaponized drones
used by US Forces

made their debut in Afghanistan.

The first unmanned aerial
missile strike

happened in October 2001,

when the US attempted to
assassinate

Taliban leader Mullah Omar.

[explosion]

The missile failed to find
its intended target,

but several bodyguards
were killed

in a vehicle near
Omar's compound.

Aitken: These early
attack drones were reliant on a

remote human operator to control
them and launch their weapons,

but today the technology has
advanced to the point

where full autonomous
capabilities are possible.

Narrator: Armed with Artificial
intelligence systems,

the latest unmanned
weapons are self-guiding

and can even attack without
human intervention.

In the aftermath of the attack
on Fakhrizadeh in Iran,

there is much confusion,

his security detail tries to
figure out what happened.

Morgan: The immediate
assumption was that a drone

carried out the execution.

Drones are often used in
complex, high-risk operations.

Pringle: A drone assault
seems unlikely in this case

because they mostly
deploy missiles.

This was a high caliber
machine gun attack.

Narrator: The chaos is
heightened when moments later,

a pickup truck parked nearby
explodes,

raining debris down on the road.

In the smoldering wreckage,
Fakhrizadeh's bodyguards

notice what appears to be
a large machine gun

attached to a robotic apparatus.

Quite literally,
the smoking gun.

Morgan: The
assassination was carried out

with a remote-controlled
machine gun.

It was likely assisted
by some form of AI.

Narrator: The world's
superpowers are developing

Artificial Intelligence
applications for an array

of military functions
at breakneck speeds.

In recent simulated dogfights

tested by the United States
military,

AI fighter pilot
systems outperformed

their human counterparts
by a significant margin.

Morgan: AI pilot systems
use something called

deep reinforcement learning.

That's when they are repeatedly
put in combat situations

and are rewarded for
successful actions

and penalized for
unsuccessful ones.

Narrator: In the early stages,
AI systems are merely trying

to keep their aircraft
from crashing,

but after billions of
different scenarios,

they have become highly
proficient

in their techniques
in air combat.

Morgan: Although this
technology is advancing rapidly,

we are plausibly
years away before we see

autonomous warplanes in
real combat situations.

Narrator: But this
isn't the case

with other unmanned weaponry.

According to a
United Nations Report,

in March of 2020,

the first autonomous drone
attacks in history

took place on a battlefield
in Libya.

Soldiers loyal to warlord
Khalifa Haftar were retreating

from the Turkish backed forces
of the Libyan government

when they were hunted down
and dive-bombed

by munitions-packed drones

operating without human control.

Aitken: This marks
the first time ever,

that the decision to attack
humans was made by

a machine rather than a person.

These weapons are
what's known as

"loitering munitions"
or kamikaze drones

and are programmed to strike
without connectivity.

Narrator: The incident has
raised serious ethical concerns

among some observers, who feel
that it is morally reprehensible

for machines to make life or
death decisions

without any human input,

and there have been calls for a
ban on the technology.

Pringle: Some would claim that
these fears are exaggerated,

and that humans will ultimately
always be in control,

and the idea of autonomous
machines running amok

and hunting down humans
is a little far fetched.

But what's really scary
is these tools

ending up in the
wrong human hands.

Narrator: Defenders of
unmanned weaponry argue that

armed with the relevant data,

AI-enabled technology
is so advanced

that it can perform tasks on the
battlefield

with pinpoint accuracy,

minimizing collateral damage.

Such is the case in the
assassination of Fakhrizadeh.

Aitken: Fakhrizadeh's wife was
sitting right next to him,

in the passenger's seat
when the killing took place.

Considering that they are
in a moving car,

and the execution was
performed remotely,

it's remarkable that she
escaped without a scratch.

Narrator: The operation is a
master class in AI,

assisted military technology.

The first hurdle that the
assassination team has to clear,

is to positively identify
Fakhrizadeh as the driver,

in real time.

This was achieved by a decoy car
staged to look broken-down,

set up along the
scientist's route.

Morgan: Iranian
officials have speculated

that the decoy car
must have had a camera.

Images from that camera
would then fed

into a facial
recognition algorithm

to confirm that it was
indeed Fakhrizadeh.

Narrator: The robotic apparatus
and gun are constructed

to fit the back
of a pickup truck.

Several cameras aimed in
different directions

are then attached
to the vehicle,

giving the team a live view of
the whole area.

To top it all off, the truck is
loaded with explosives

so that it could be destroyed
after the assassination,

eliminating any
potential evidence.

Pringle: One of the
biggest challenges

the assassination team faces

is that a machine gun
recoils after every shot,

altering the course of
the bullets that follow.

Narrator: Also complicating
matters is that there is

a one and a half second delay

in what the command room sees
on their screens,

and what is happening
on the ground.

This may seem insignificant,

but the car is in motion
and it's enough of a lag

for even the best aimed shot
to miss its mark.

Amazingly, the AI system used
by the assassination team

is programmed to account
for the visual delay,

the gun's shake, and the speed
of Fakhrizadeh's car,

which results in the flawless
execution of the operation

with no collateral damage.

Fakhrizadeh's assassin
pulled the trigger

from the comfort of an office

thousands of miles away,
leading some experts to believe

that this is yet another example

that we are heading towards a
future of "armchair warfare"

where unmanned battlefields are
more and more commonplace.

Aitken: Those in favor
of autonomous weapons,

argue that delegating acts of
war to machines makes sense.

And by removing humans
from the firing line,

they might lead
to fewer casualties.

But that argument
largely only accounts

for casualties on one side.

Narrator: In the bigger picture,
there are those that fear

we are entering
a period of escalation

in AI-backed military technology

reminiscent of the
nuclear arms race

between the US and former Soviet
Union during the Cold War,

with China as a third party.

Aitken: We know very little
about what kind of AI-enabled

military technology Russia
or China may have developed.

That secrecy leads to
governments trying to

second guess
where others are at.

Narrator: Secrecy is indeed
vital to the success

of any sensitive military
operation

and as the burial of Mohsen
Fakhrizadeh took place,

no one had claimed
responsibility for his death,

but eventually
one nation confessed.

Morgan: In the months following
Fakhrizadeh's death,

the former head of the Mossad
all but acknowledged

Israel's responsibility
for the attack.

This isn't exactly a surprise
given that Israel's fear

of a nuclear-armed
Iran is well known.

Narrator: For many years,
the Mossad,

Israel's foreign
intelligence agency,

has been trying to derail
Iran's nuclear program.

So while the nationality of the
perpetrators is unsurprising,

the audacity and use of
technology in the operation,

leaves some international
observers taken aback.

And the Iranians still had no
idea how exactly the Israelis

got the gun and robotic
attachment into Iran.

Pringle: Using some old
fashioned military stealth,

the Israelis managed to sneak
the weapon across the border

by dismantling it, and then
smuggling the pieces into Iran

individually, at different
times and places.

The gun was then covertly
rebuilt at an unknown location.

Narrator: Curiously, on the
day of the assassination,

Fakhrizadeh ignored the
recommendation

of his security team
who suggested

they drive him to Absard
in an armoured vehicle.

For some reason, he insisted on
driving himself

and his wife in his own car.

And it ended up
costing him his life.

Pringle: It seems strange that
he would make this decision,

as the Israelis had
made numerous attempts

to kill him in the past.

Narrator: An assassination
squad had prepared an attack

on Fakhrizadeh in Tehran
twelve years earlier,

but the mission was cancelled
at the eleventh hour.

The Mossad believed
the plan was leaked,

and Iranian forces were
ready with an ambush.

This time, Fakhrizadeh
wasn't so lucky.

With the operation
a resounding success,

some experts wonder if the
assassination of Fakhrizadeh

is a glimpse into the future of
military operations,

where data driven Artificial
Intelligence systems

do most of the heavy lifting.

Aitken: It's an area
of significant concern.

But at least in the short term,
most militaries are looking

into Artificial
Intelligence to assist,

rather than replace,
humans in war.

Narrator: But there are
warning signs.

A 2018 research paper

by the non-profit Rand
Corporation declared

that the overuse of Artificial
Intelligence in militaries,

could result in an accidental

nuclear war by the year 2040.

Aitken: Inevitably
at some point,

mistakes will happen,
things will go wrong.

And we don't as yet have the
adequate legal or regulatory

frameworks in place to know
who is accountable, or

who should be held responsible
when mistakes happen.

Narrator: Are we moving
towards a future

where warfare will pit algorithm
against algorithm,

and military dominance will
shift from scale of force

and superior weaponry to
technological factors

like more advanced data
collection and A.I.?

Only time will tell, but if
recent developments

are any indication, it appears

that the future of warfare
is already upon us.

♪

♪ [show theme music]

Narrator: On a quiet
night in March, 2018,

the streets of Tempe, Arizona

bear witness to a marvel of
modern technology,

and one of Big Data's most
impressive accomplishments...

The autonomous car.

Morgan: It was an Uber
self-driving car

out for a test drive.

A Volvo XC90 SUV.

It was going just over
40 miles per hour

which was under the speed limit
for that four lane road.

Narrator: Sitting in the
driver's seat

is a safety backup driver,
Rafaela Vasquez,

a woman with a checkered past,

who was trying to turn her
life around,

and was able to take advantage
of Uber's hiring policy

that gives those in need
a second chance.

Vasquez's job is
to monitor the trip.

Her hands and feet are not
touching the controls

but she is ready to spring
into action if required.

Badminton: An autonomous
vehicle uses

a complex system of
sensor fusion.

It combines data from cameras,

from LiDar, from GPS,

and brings it all together to
build a picture of the world.

It can then determine how best
to drive that autonomous vehicle

through the spaces
that takes it.

Kris Alexander: So you have
massive amounts of information

being fed into an
onboard computer

with incredible processing
capability.

The system's autonomous
algorithms process everything

the vehicle encounters
and uses it to navigate.

Narrator: The algorithms
are programmed to use

machine-learning, and gain
intelligence in each moment,

as they process millions
of simulated

and real-world scenarios.

In 2016, Uber, envisioning a
future where it will replace

its human drivers with machines,
began test-driving

a fleet of nine self-driving
vehicles in Arizona.

Morgan: Uber
recognizes that human error

is the cause of many accidents.

And they dedicated themselves
to improving safety

for self-driving vehicles.

Narrator: Uber's self-driving
car is performing perfectly,

until a figure suddenly
appears in the darkness.

[distant sirens]

Morgan: A woman pushing
a bicycle was crossing the road

just in the Volvo's path
and for some reason,

the car didn't stop.

Narrator: The explanation
for this failure lies deep

in the state-of-the-art
technology that performs

thousands of operations
per second

onboard self-driving vehicles.

The car's sensor data reveals
a surprising fact:

for some reason, the
autonomous vehicle

didn't recognize Herzberg
as a pedestrian.

In the almost six seconds
before the crash,

she was classified and
reclassified

at least eight times.

Badminton: The radar and
LiDAR systems first mistook her

as a vehicle,
and then an unknown object,

and then a vehicle again,

before finally determining
that she was a bicycle,

coming into the path
of the oncoming Uber.

Narrator: As it turns out, there
was a mistake in the algorithms.

The engineers didn't program
their systems

to recognize people crossing the
street illegally.

Only those using crosswalks were
classified as pedestrians,

and Herzberg was
not at a crosswalk.

Badminton: The sensors
only recognized

the bicycle she was pushing,

and then they could only
recommend emergency braking.

Narrator: Braking has been
an ongoing issue with

Uber's self-driving cars, and
the problems stem directly

from one of their policies.

Morgan: Uber's vehicles
were experiencing

what they called "erratic
vehicle behavior".

Something as small as a bird
darting crossing the road

would be enough to trigger the
vehicle to slam on its brakes.

This caused some really
rough rides for passengers.

And so, they came
up with a solution.

Alexander: But Uber disabled
the auto-braking feature

in all of its vehicles
in autonomous mode,

so only the safety operator

could apply the brakes
in an emergency.

Morgan: Turning off the
auto-braking system

of an entire fleet of vehicles,

even if you have
good reason to do it,

it comes with a lot of risk.

Emergency personnel
were unable to revive her,

and Elaine Herzberg became the
world's first pedestrian

ever struck and killed
by a self-driving car.

[sirens]

Morgan: Vasquez who showed
no signs of impairment,

stayed on the scene.

She said she was "monitoring the
self-driving car's interface"

at the time of the impact and
didn't see the woman right away.

As soon as she did see her,

she slammed on the brakes,
but too late.

Narrator: But police estimate
Elaine Herzberg would have been

visible on the road nearly
six seconds before impact.

And the question was asked, why
didn't the vehicle's sensors

recognize she was crossing
in front of it,

and activate the braking system?

Alexander: When a
self-driving accident occurs,

assigning blame is very tricky.

It could be the
hardware, software,

or some other
technological malfunction.

And then there's also
the human element.

Narrator: As the investigation
into Elaine's death continued,

authorities discovered
another startling fact.

A former employee issued
a haunting warning

to his superiors just
five days before the crash.

Morgan: An Uber
operations manager sent a

resignation email to various
company executives,

saying that their vehicles were

"routinely in accidents" and

"hitting things
every 15,000 miles"

Narrator: The employee alleged
that "a car was damaged

nearly every other day"

In the month before
the Herzberg accident,

and also cited "poorly
vetted or trained"

safety drivers as
a cause of these accidents.

The police then start
to look closely

at the role of the safety driver
Rafaella Vasquez,

and discover some
shocking evidence.

The SUV's dashcam footage

tells a disturbing tale.

Alexander:
After careful analysis,

authorities determined that for
roughly a third of the trip,

Vasquez was looking down to the
right instead of monitoring

the road and vehicle conditions
like she should have been doing.

It was likely she was
looking at her cellphone.

Badminton: During the
investigation,

Vasquez's phone told them
everything they needed to know.

She was watching
The Voice on Hulu

and wasn't paying attention
to the road at all.

Narrator: It's a stunning
discovery,

but some believe that
Rafaela Vasquez's behavior

is not solely responsible
for the accident,

and there may be some
underlying factors

within autonomous technology
that should shoulder

at least some of the blame.

Badminton: Her
distraction isn't surprising,

it's almost expected.

It's called
automation complacency.

And this happens when people
work with automated systems,

and they stop paying attention
to what's happening

on a second-by-second basis.

Alexander: These people work
tedious and repetitive shifts

and expect nothing to go wrong,
because nothing does go wrong

nearly 100% of the time.

So, naturally they lose interest
and don't pay attention.

Narrator: Authorities found that
Uber's self-driving division

failed to do the
necessary training

and supervision of
backup drivers.

Another factor in
the tragic outcome.

Badminton: It begs the question:
is the machine or the person

at fault in
Elaine Herzberg's death?

Narrator: As the media
stokes the controversy

around the tragedy, Arizona's
citizens lash out

at the driverless machines
on their roads.

Technology development
company Waymo,

part of Google, was targeted.

Alexander:
The publicity around the case

stirred up anger in people.

In nearby Chandler,
Waymo's autonomous white vans

were harassed repeatedly.

They had rocks thrown at them
and several attempts were made

to run them off the road.

Badminton: The anger
spilled out into action.

Tires were slashed,
and even on one occasion,

a backup driver was
threatened with a gun.

Narrator: But Waymo
representatives insist

their self-driving
vehicles are safe,

citing 20 million miles driven

on public roads by early 2020,

74,000 of those
without a driver.

Badminton: They've had multiple
accidents but zero fatalities,

which is pretty great
when you consider

there's an average
of one death per

160 million kilometers
driven by humans.

Narrator: Tesla has faced
lawsuits for two separate deaths

that resulted from crashes
while its drivers used

the car's Autopilot program.

Alexander: Autopilot is not a
fully-autonomous driving system.

It's a form of driver-assistance
that accelerates,

brakes and steers its vehicles
while the driver is present

but not actively driving.

The problem is, instead of
watching the road,

the drivers get bored and
careless and accidents happen.

Narrator: It's another case
of the dangers

of automation complacency.

Since 2016, there's been at
least ten deaths

in eight accidents where Tesla's
Autopilot was involved.

Morgan: The controversies
around self-driving cars

are extremely complex.

Just think of the
ethical decisions

you and I might have to
make while driving.

Imagine a self-driving vehicle
forced to choose between

driving onto the sidewalk and
hitting a crowd of pedestrians,

in order to save its passengers

from striking an object
on the road.

Narrator: Arizona authorities
clear Uber of any wrongdoing,

but suspend their program of

autonomous vehicle tests
on their roads.

Rafaela Vasquez is arrested

and charged with
negligent homicide.

Morgan: There's still no
denying the possible benefits

of autonomous vehicles
replacing cars

and transport trucks
on the road.

Some of the most significant
potential for growth

is in public transportation
where subways, trains and buses

could all be replaced by
autonomous vehicles.

Narrator: In fact, there have
already been many

pilot projects testing
driverless

public transit systems
around the world.

But some have been marred
by controversy.

At the time of the accident
that killed Elaine Herzberg,

confidence in autonomous
vehicles

was growing around the world.

But the incident cast a
shadow over the industry.

With many arguing that the
weaknesses in their

safety and reliability still
need to be addressed.

Badminton: Self-driving
vehicles are a new frontier,

and the market's
expected to grow

to $64 billion dollars by 2026,

a 22% growth year-on-year.

Narrator: And Uber is just
one of the companies

racing to be first in the global
autonomous vehicle market.

Traditional carmakers like GM,

Ford and Volvo are
trying to muscle in

on what was once the territory
of tech giants.

By March of 2018,

confidence in autonomous
vehicles

was at an all time high
around the world.

But in light of the accident
that killed Elaine Herzberg,

criticism of the technology
grows with many arguing

that the weaknesses in their
safety and reliability

still need to be addressed.

Alexander: The more
congested cities get,

the more safety issues
there will be

with self-driving vehicles.

And there's still many
unresolved dangers,

like adverse weather conditions.

How can a self-driving
vehicle navigate,

when snow or rain eliminates
its sensors' ability

to detect lane markers
and lane dividers?

Narrator: But one of the
biggest problems faced

by technology firms and
car manufacturers is:

will their autonomous vehicles
ever truly be able to

function safely without
any human interaction?

If the tragedy of Elaine
Herzberg is any indication,

we have a long road to travel

before we reach that
destination.

♪ [show theme music]

Narrator: New York City.

The epicenter of the
modern art world.

At the venerable Christie's
auction house in Manhattan,

a bizarre looking portrait is
about to go up for sale,

which will send shockwaves

through the global
art community.

Pringle: I think
it's an incredible piece.

It's strange because
you can tell it's a portrait

in the style of the old masters,

but the subject's facial
features are blurry and smudged.

It also kind of
looks unfinished,

with large sections
of the canvas left blank.

Narrator: The work is credited
to a Paris-based collective

who call themselves 'Obvious'.

The group consists of three
twenty-something French men,

Hugo Caselles-Dupré,

Pierre Fautrel and
Gauthier Vernier,

none of whom are well
known in the art world.

Alexander: One interesting thing
about the sale is that

the portrait was never
previously shown

at exhibitions or in galleries.

It's highly unusual for a
painting to come to market

at auction, sight unseen.

My guess is that Obvious were
intentionally trying to shock

the art world by keeping the
portrait under wraps,

until it was revealed
at Christie's.

Narrator: A tag on
the wall proclaims

that the subject of the portrait
is a man named Edmond de
Belamy.

He appears to be wearing
a black frock coat

with a white collar
showing at the neck.

This indicates that
he could be French,

and may be a member
of the clergy.

Aitken: I guess he was
somebody important,

but the name Edmond Belamy
doesn't ring a bell.

Narrator: There's a reason why
the name Edmond de Belamy

may be unfamiliar,
he never existed.

In the bottom right hand
corner of the portrait,

in lieu of a signature, there is
a mathematical equation

written in cursive
Gallic script.

Alexander: The signature is
actually a snippet

of computer code -
part of an algorithm.

Narrator: The Portrait of Edmond
de Belamy is about to

make history as the first work
of art produced

by artificial intelligence
to be sold at auction.

And in the process, it may just
redefine our notion of art

in the 21st century.

Pringle: Some in the art
establishment weren't exactly

thrilled to see a
computer-generated

painting up for auction.

Especially at Christie's, the
oldest and most prestigious

fine arts auction house
in the world.

Narrator: The inclusion of
Obvious certainly causes

some indignation in the high
brow art community,

and leaves some
asking the question,

"If the portrait was not
created by human hands,

is it even art?"

Aitken: It's an
interesting question.

But the suggestion that the
piece was created

with no human involvement
is overly simplistic.

Somebody had to write
the code that created it.

Narrator: Obvious believe that

"creativity isn't
just for humans."

And produce the portrait using
a machine-learning model

called a Generative
Adversarial Network

or GAN for short.

Alexander: GANs use
two neural networks

that are essentially competing
with one another to see

which one can become
better at its assigned job.

Narrator: These two neural
networks are known as

the generator
and the discriminator.

The purpose of the generator is
to produce outputs

that trick the discriminator
into mistaking them

for real data, while the
discriminator

attempts to flag which outputs
are artificial.

Pringle: It's a zero sum
game where one side loss

is another side's gain.

In this case, is it a face
or is it not a face?

Narrator: This back and forth
creates a feedback loop

and eventually the generator
will begin to create

superior output and the
discriminator will become more

adept at catching data that has
been artificially generated.

Morgan: Think of it
like a counterfeiter,

trying to fool a bank teller
with a fake note.

If the counterfeiter keeps
getting caught,

they're going to kind of
get better and better

at producing fake notes.

But at the same time,
the bank teller

will get better and better
at detecting them.

And there's this
evolutionary arms race.

Narrator: In time, the generator
produces synthetic data

that fools the discriminator
into believing it is genuine.

For Obvious, this meant
the creation of

the Portrait of
Edmond de Belamy.

Alexander:
Obvious fed 15,000 portraits,

from the 14th to the 20th
century into the GAN algorithm.

The generator network
then learned

the basic rules of
the images, for example,

they all have a mouth,
a nose and two eyes.

Based on those rules,
the network then begins

to generate new images.

Narrator: The discriminator
network reviews the images

and tries to guess which ones
are authentic portraits

from the dataset and which ones

were artificially created
by the generator.

When the generator succeeds in
tricking the discriminator,

the process is complete.

Aitken: At this point,
it's up to Obvious

to evaluate the outputs and
choose the best examples.

I guess they liked the Portrait
of Edmond de Belamy the most.

Narrator: Heading into the
auction, it was estimated

that the portrait would
likely fetch somewhere between

$7 and #10,000 dollars U.S.

But as the bidding begins,
it becomes abundantly clear

that interest in this new form
of artistic expression

has been grossly underestimated.

The creativity of AI has
also made headway

into other artistic disciplines.

In late 2021, Ai-Da the robot,

Robot: I am Ai-Da.

Narrator: ...made history by
becoming the first machine

to recite poetry
written exclusively

by its algorithms in front
of an audience.

Pringle: The robot was
fed Dante's epic poem

The Divine Comedy and then the
algorithms evaluated

the language patterns and used
her database of words

to create original poetry.

Narrator: Ai-Da is capable of
producing an astonishing

20,000 words in ten seconds,

generated by her
AI language model.

Ai-Da: I wept silently,

taking in the scene.

Aitken: Okay, now
that raises some concerns!

Narrator: And it's not just
in the written word

that AI's encroachment on the
arts is having an impact.

The music industry
is feeling it too.

♪ [gentle music]

Pringle: We're now seeing
artificial intelligence

that is capable of producing
music that is virtually

indistinguishable from
human compositions.

Narrator: In 2019,
Chinese tech giant Huawei

wanted to showcase the
technology in their smartphones,

so they decided to do
something bold,

finish one of the most notable,
incomplete symphonies

in music history
using artificial intelligence.

Alexander: Schubert's
symphony number 8

is one of his most
acclaimed works,

but for some reason,
he never finished it.

Narrator: With the help of
composer Lucas Cantor,

Huawei set out to
answer one question,

"if Schubert had finished the
last two movements,

what would they sound like?"

The composer's body of work,

some 2,000 pieces
of piano music,

was fed into the phone's
dual Neural Processing Unit

in the form of data.

The AI then analyzed the tones,
pitch and rhythms

that Schubert preferred
in his symphonies,

and created melodies
from that information.

Alexander: For those
who bristle at the idea

of AI impinging on something as
revered as classical music,

you could make the argument that
music is essentially just code.

Narrator: But algorithms can't
create art in a vacuum,

they still need to be
guided by humans.

For the Portrait of
Edmond de Belamy,

Obvious had to make creative
decisions and curate

the results generated by the
artificial intelligence.

Pringle: AI merely follows
a series of steps

laid out by a set of
prescribed rules.

It takes a human eye to discern
what may or may not have value.

Narrator: And back at the
auction house,

the Portrait of Edmond de Belamy
appears to have some value.

Alexander: There
are four competing bidders.

One online in France,
two others by phone,

and one in person
in the room in New York.

Narrator:
As the auction heats up,

eyebrows are being raised
in the art community.

Some onlookers are shocked that
a work created by a computer

is garnering so much attention.

Many still don't even consider
the portrait of Edmond de Belamy

to be art in the
traditional sense.

Aitken: Is it art?

I suppose that depends on
one's definition of the word.

Most creatives would say that

art is a method by which humans
express some concept or emotion.

So by that definition,
AI generated art

wouldn't be considered art.

Pringle: But isn't art,
like beauty,

in the eye of the beholder?

If it invokes some kind of
response in the audience,

then it by all means
should be considered art.

And who decides
what is and isn't art?

I would argue that the datasets
and algorithms are tools.

Just as paint and
canvas are tools.

The artist is the one
manipulating them

and making those
creative choices.

Narrator: Critics are
quick to point out

that Obvious' work lacks intent,

one of the cornerstones of
artistic expression.

Aitken: When human artists
distort the aspects

of a subject, they do it on
purpose, it has intent.

But in the case of the
Portrait of Edmond de Belamy,

the AI has reproduced
what it thinks

a human face might look like.
Simply put,

the AI hasn't exactly
been entirely successful.

Narrator: Doubters in the face
of new creative fields

are not unique to our times.

Photography as an art,
was derided in the early days

because it came from a machine,

but now, it's a vital
and respected

element in the art world.

Pringle:
Change is really hard.

People are afraid of new things.

Especially things they
don't fully understand.

Narrator: Acceptance into
the mainstream is not

the only challenge
facing AI artists,

there is also the
issue of credit.

Controversially, Obvious didn't
even write the algorithm

that they used to create the
Portrait of Edmond de Bellamy.

It was authored by
a pioneering artist

and programmer named
Robbie Barrat,

whose algorithm-generated
nudes and landscapes

were lauded by the AI community.

Barrat shared his code online,

as a free open source license.

Alexander: So if
Obvious didn't write the code

and the data sets they used
were old portraits

from hundreds of years ago,
not original works,

what exactly did they
contribute to the process?

Narrator: Obvious doesn't deny
using Barrat's algorithm,

but they claim that
they tweaked it

in order to produce
the desired outcomes.

In response to this, Tom White,

an AI artist from New Zealand

used Barrat's code to produce
his own set of portraits

and the results were strikingly
similar to the work of Obvious,

calling into question
just how much

they adjusted the algorithm.

Pringle: At the heart
of this controversy,

it's the definition of the
authorship and ownership,

which is complex in
a case like this.

If they used Barrat's code,
shouldn't he get some credit

and even a share in the
proceeds of the auction?

Alexander:
Appropriation happens

all the time in the art world.

Writers "borrow" styles
from other writers.

Rap artists sample other songs.

It's part of the game.

Narrator: Barrat bears no ill
will towards Obvious.

After all, he made the code
readily available

on the internet free of charge.

But he is somewhat critical of
the portrait itself,

and he is not alone.

Morgan: Many artists
would point out

technical flaws in the work.

The composition is skewed
to the upper left

and there are many gaps
in the composition

that make it feel incomplete.

Alexander: Maybe these
imperfections

are the whole point.

This is a new visual style
created by an artist

and a machine
collaborating in a way

that we really
haven't seen before.

Narrator: After seven minutes
of frenzied bidding,

when the hammer falls at
Christie's,

the Portrait of
Edmond de Bellamy

sells for an extraordinary

$432,500 USD

to an anonymous phone bidder,

more than 40 times
the estimated price.

Like it or not, AI generated art
has officially arrived

on the global stage.

Alexander: It's a
staggering amount of money,

considering that Obvious
were basically unknown

in the art world,
up until that point.

And they aren't even artists,

according to the traditional
definition of the word.

Narrator: The sale shocks
the establishment,

and may one day be
looked back on

as one of the seminal moments
for art in the 21st century,

redefining the very idea
of what art is.

There are two differing
arguments emerging

regarding this type
of creative AI.

Those that think it will
kill our creativity,

and those that think
it will be enhanced.

Alexander: Technology
has disrupted

pretty much every industry.

So why should
the art world be immune?

If artists learn to create in
conjunction with technology

and not use it as
some sort of crutch,

then I don't see it as
an existential threat.

Narrator: But there are
alarmists,

who fear that we are moving
towards a future,

where computers will
write our novels,

compose our songs
and paint our pictures,

and the result will be
an overall diminishment

in human creativity.
Others disagree.

Pringle: The bottom
line is that even the most

sophisticated AI systems can't
replicate the human instinct

to find inspiration
in our surroundings,

and to use our ingrained
creativity to produce art.

But it's an exciting
new tool for artists,

and in that way we are
right at the frontier

of what could be
a whole new medium.

Narrator: There will always
be doubters

who challenge the legitimacy of
AI-generated creative work,

questioning whether or not

it even qualifies as art.

Morgan: In the words
of Andy Warhol,

who, in the tradition of
many great artists,

may have stolen this phrase
from Marshall McLuhan,

"Art is what you can
get away with."

Ai-Da Robot: We looked up
from our verses

like blindfolded captives,

Sent out to seek the light;

but it never came,

A needle and thread
would be necessary,

For the completion
of the picture.

♪ [show theme music]

Narrator: A Japanese woman in
her 60s, is at an appointment

at the Institute
of Medical Science

at the University of Tokyo.

It is the foremost centre for
advanced medical

and bioscience research
in the country.

She has been recently diagnosed
with a rare form of cancer,

acute myeloid leukemia.

Which starts in the bone marrow
and can quickly spread

to the bloodstream and other
parts of the body.

If not treated, it can
be life-threatening.

Morgan: Unfortunately,
the treatments her previous

doctors administered, which
included chemotherapy

were largely unsuccessful.

Pringle: They didn't know
why she wasn't responding

to the treatments.

The woman's recovery was
unusually slow

and her doctors were confounded.

Narrator: If medical experts
at the Institute

can't find a solution,

it's likely the woman will
succumb to her illness.

For a time, the hospital
has been using AI

to help with the diagnosing
of patients.

An Artificial Intelligence
program developed by IBM,

nick-named "Watson".

Aitken: It's an exciting new
technology

they've been trying out.

They're hoping it can
successfully aid physicians

in the treatment of patients.

Watson has access to a
massive database,

containing 20 million
research papers.

This includes a huge
amount of information

on gene-related cancer.

Morgan: The hospital
has used the same AI

for other patients suffering
from hematological diseases,

with up to 80% success in

identifying the causes
of their illness.

Narrator: For AI's like 'Watson'
to work, it needs to

analyze a considerable
amount of information,

a patient's medical history,
current symptoms

and notes from their
previous doctors.

It will then cross-check it
against huge amounts

of clinical cancer case data,
where it will form

it's hypotheses and suggest
possible treatments.

Aitken: And it can do
this very, very quickly,

accessing and analyzing up to
200 million pages of information

on nearly 100 servers, in an
extremely short period of time.

Narrator: Over the past decade,

highly advanced machine
learning systems

have become an integral part
of healthcare systems

around the world - taking on
important medical tasks,

once only reserved for

highly-trained medical
specialists.

Aitken: These systems
don't just analyze

and interpret or cross-check
the raw data.

They actually "learn" from it
and they get smarter over time.

Pringle: It's really
incredible in that

it's been quite successful
in making these

reliable hypotheses
to help patients.

Narrator: AI has been deployed
in many ICUs,

where it assists in monitoring
and treating patients

with life-threatening
conditions.

It's also used in
Radiology departments

to aid technicians
in X-ray diagnostics,

and in planning appropriate
drug treatments.

In 2015, a Canadian woman,
Krista Jones

was facing a double mastectomy,

due to a rare form of
breast cancer.

But she was able to forgo
the invasive procedure

when machine-learning algorithms
were used to locate

previously imperceptible
tumors in her mammograms.

She is now cancer free.

While examples like
this are encouraging,

some still question
whether or not

there is concrete proof that AI
improves a patient's outcome.

Aitken: AI can analyze
and cross-check

literally millions of pieces
of medical data.

No doctor or even a team
of doctors can do that.

Narrator: Researchers at
Stanford University

found that in patients
they examined,

AI was much faster

than the average radiologist
when screening

and diagnosing for several
different pathologies.

A patient could obtain their
result in less than 2 minutes,

when normally, they would have
to wait up for up to 3 hours.

But there are still many critics
who are concerned about its use.

Pringle: The criticism is
that many of the studies

that have been released
are not conclusive.

And in fact, many argue that

there are actually
inherent risks to the patients.

Narrator: Researchers suggest a
potential contributing factor

are the limitations
AI encounters

in accessing various data sets.

Aitken: For AI to
function effectively, it needs

to scan large amounts of data
from many different sources.

But the problem is, not all
of these sources

are always included in the
AI's diagnostic decisions.

Sometimes that's a result of
human error,

sometimes not everything
gets transferred,

or there might be
inconsistencies

in how data is entered
into the system.

Narrator: A 2021 study by
the University of Washington

revealed an additional concern.

While examining various AI
models designed to detect Covid,

researchers found that instead
of 'learning' about pathologies

related to Covid, the technology
had a tendency

to search for potential
diagnostic shortcuts,

such as, making potentially
erroneous associations

with other irrelevant
medical factors,

like the age of the patient.

This increases its
chances of misdiagnosis.

Pringle: It's safe to say
that in their current state,

these AI programs
aren't perfect;

they can and sometimes do,
make mistakes.

Narrator: But AI advocates say
it's pros vastly outweigh

any potential cons.

Recently, St. Michael's hospital
in Toronto, Canada

began using machine learning
algorithms to help

with diagnoses and they saw
a 20-percent reduction

in mortality amongst
high-risk patients.

By using IBM's AI 'Watson',

medical experts at the
University of Tokyo

are hopeful it may be able to
help the woman

suffering from leukemia.

Watson will detail and
analyze gene mutations

in the female patient
and it turns out,

there are more than
one thousand of them.

Pringle: Watson
can do it in 10 minutes,

but it would take 2 weeks
for human doctors

to accomplish the same task.

Narrator: Doctors wait
for Watson's

diagnostic assessment on the
woman's condition.

But its use has led to both
ethical and legal dilemmas.

Aitken: Mistakes are
always a possibility,

and they could exacerbate a
patient's health problems,

or possibly even lead to death.

Narrator: If something like this
does happen, many wonder,

who or even what would be
held responsible?

Pringle: It may be
difficult to establish

legal accountability in
these situations.

Is it the patient's doctor?
The programmer?

Or even the AI itself?

Morgan: It may sound
like science fiction

to put an AI on trial,
but it's not that far fetched

when you consider
the legal actions we've taken

against things like
corporations.

Narrator: Legal experts counter
that they currently have

more pressing concerns
regarding the technology.

Specifically, in how
personal medical data

is obtained and shared.

Aitken: AI primarily
learns about a patient

through medical records, and
if they share these record

with other institutions it may
be in violation

of a person's privacy.

Narrator: In 2020, the renowned
US medical centre,

The Mayo Clinic, acknowledged
for its speciality in

Cancer, Cardiology
and heart surgery,

was accused of allegedly sharing
private medical data

with several AI developers.

Pringle: It was for
research purposes,

but the patients were not
informed that they did this.

And understandably many say
this is highly unethical.

Narrator: Legal experts
also warn of another

potential breach of privacy.

A hospital's AI could obtain a
person's private data

from companies that are not
in the medical or tech industry,

such as banks or
insurance policies.

That could result in it making
inferences

to a patient's 'lifestyle'.

Morgan: It could change the way
that it does diagnostics,

the medications it prescribes,

the prognosis of the outcome.

This could have serious impacts

on whether we treat different
groups of people

the same way,
to treat them all fairly.

Narrator: There may be
other biases

entrenched in the AI algorithms.

There is evidence that in
the US, people of colour

are underrepresented in
the data it analyzes.

Aitken: Most of the data
used to train their algorithms,

is obtained from
only 3 U.S. states.

That's not a
diverse enough dataset,

and could result in
these AI systems

treating people
of colour less effectively.

Pringle: Some of these
algorithms allocate

less medical resources to
minority groups because often

they have less access
to medical care,

and therefore use less resources
than causcasian patients.

This results in the AI thinking
the minority patients

are healthier than they are,
and gives them less priority.

Narrator: Critics believe this
could lead to more incidents

of misdiagnosis in minority
patients and treatment errors,

such as prescribing
the wrong medications.

But AI developers and
hospitals counter

that many medical centers are
expanding

the diversity of their test
subjects,

and in the long run, people of
colour will greatly benefit

from AI's integration
into healthcare systems.

Morgan: The tech is
constantly improving

and according to its proponents,
it has greater and greater

access to minority populations
for research purposes.

Aitken: This is especially
important for people

living in less
developed countries.

For example, several
countries in Africa,

have hospitals with no
radiologists at all.

Thankfully AI is helping
to fill that gap,

analyzing images and x-rays
taken from patients.

Narrator: Asian countries
in particular,

have become leaders in
AI healthcare innovation,

with Japan leading the way.

And as Watson readies its
diagnosis

at the University of Tokyo,

we're about to see why.

Morgan: Her doctors
are stunned when Watson

determines that the original
diagnosis is in fact wrong.

It turns out she has a
rare form of leukemia.

And that's why the earlier
treatments weren't working.

They were treating
the wrong illness.

Narrator: Armed with
this new information,

doctors revise the woman's
treatment plan,

and her condition improves
significantly.

Morgan: What ended up saving
this woman's life...was data.

The AI was able to recognize
patterns in it

that her doctors were unable to.

Narrator: But the positive
outcome raises

an important question
that people working in

health and medicine
will have to deal with.

If AI can do this
more accurately and efficiently

than a medical professional,

what impact will this
have on the future

of Healthcare's workforce?

Pringle: It's a situation
that medical experts

are seriously concerned about.

The tech is fully automated,
it's self-reliant.

So, larger scale
integration of AI

could create a significant
amount of

unemployed, highly
skilled workers,

like doctors, nurses
and technicians.

Of course, what they've got
going for them

that the AI doesn't
have is empathy.

They've got, they've got
bedside manner.

Aitken: As AI plays
increasing roles

in all aspects of healthcare,

it will inevitably change the
ways we interact

with health services and lead to

different kinds of jobs
in healthcare.

But it's unlikely
ever to replace

medical professionals
altogether.

Narrator: Penetration
of AI into medicine,

will likely be very slow,

providing ample time
for healthcare institutions

to re-adjust and accommodate
factors such as staff training.

And according to recent
research, AI may actually

increase employment
opportunities in health care

by 15% in the upcoming years.

Morgan:
It's the old saying,

' when one door closes,
another opens,'

and a lot of people argue that
AI's integration

will produce a ton
of high- skilled

computer programming jobs.

But I'd say we should
be careful about this.

We want to make sure not to
automate the kinds of jobs

we would actually want to do.

Aitken: It's important to
note that AI's functionality

is most optimal when it's not
working autonomously,

but rather when it's used
as an additional tool

to assist medical practitioners.

Narrator: A perfect
example of that,

is the elderly Japanese woman,

whose life was saved
with the help of Watson.

She was discharged from the
hospital and began receiving

outpatient treatment at home.

Pringle: She is
managing her condition

and what's most important,
she is alive.

She can thank AI for that.

Aitken: There is no
denying that AI will play

a big part in the
future of healthcare.

Who knows how far the
technology will go?

Narrator: As AI continues
to be integrated

into our healthcare systems,

the moral and legal debate
surrounding its use

will surely intensify.

Is it really more dependable
than a human doctor?

Can it be ethical
and impartial

when dealing with
minority patients?

And if not, who or what
will be held accountable?

Or perhaps AI proponents are
correct, when they say

that there is something
more important,

all the lives it might
help save in the future.