Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:15
Pushkin. You're
0:25
listening to Brave New Planet, a
0:28
podcast about amazing new technologies
0:30
that could dramatically improve our world. Or
0:33
if we don't make wise choices, could
0:35
leave us a lot worse off. Utopia
0:38
or dystopia. It's up to us.
0:47
On September twenty sixth, nineteen
0:49
eighty three, the world almost
0:52
came to an end. Just
0:54
three weeks earlier, the Soviet Union
0:57
had shot down Korean Airlines
0:59
Flight Double O seven, a passenger
1:02
plane with two hundred and sixty nine
1:04
people aboard. I'm coming before you tonight
1:06
about the Korean Airline massacre. President
1:09
Ronald Reagan addressed the nation
1:11
the attack by the Soviet Union against
1:14
two hundred and sixty nine innocent men,
1:16
women, and children aboard an unarmed
1:18
Korean passenger plane. This
1:20
crime against humanity must never be forgotten
1:23
here or throughout the world. Cold
1:25
War tensions escalated, with the two
1:28
nuclear powers on high alert. World
1:31
War three felt frighteningly
1:33
possible. Then, on
1:35
September twenty sixth, in a command
1:38
center outside of Moscow, an
1:40
alarm sounded. The Soviet
1:42
Union's early warning system reported the
1:45
launch of multiple intercontinental
1:47
ballistic missiles from bases in the
1:49
United states. Statislav
1:52
Petrov, a forty four year
1:54
old member of the Soviet Air Defense
1:56
Forces, was the duty officer
1:59
that night. His role was
2:01
to alert Moscow that an attack
2:03
was under way, likely triggering
2:06
Soviet nuclear retaliation and
2:08
all out war. Petrov
2:12
spoke with BBC News in twenty
2:14
thirteen. The sirens sounded
2:17
very loudly, and I just sat there for
2:19
a few seconds, staring at the screen
2:21
with the word launch displayed
2:23
in bold red letters.
2:26
A minute later, the siren went off again.
2:29
The second missile was launched, then
2:31
the third, and the fourth, and the fifth.
2:34
The computers changed their alerts from
2:37
launch to missile strike. Petrov's
2:40
instructions were clear, report
2:42
the attack on the motherland, but something
2:45
didn't make sense. If
2:47
the US were attacking, why only
2:49
five missiles rather than an entire
2:52
fleet? And then I made
2:54
my decision. I would not trust
2:56
the computer. I picked up the
2:58
telephone handset, spoke to
3:00
my superiors and reported that
3:02
the alarm was false. But
3:05
I myself was not sure.
3:07
Until the very last moment. I
3:10
knew perfectly well that nobody
3:12
would be able to correct my mistake if
3:15
I had made one. Petrov,
3:20
of course, was right. The false alarm
3:22
was later found to be the result of a rare
3:24
and unanticipated coincidence sunlight
3:27
glinting off high altitude clouds over
3:29
North Dakota at just the
3:31
right angle to fool the Soviet satellites.
3:34
Statislav Petrov's story comes up
3:36
again and again in
3:38
discussions of how far we should go and
3:41
turning over important decisions, especially
3:44
life and death decisions, to artificial
3:47
intelligence. It's not an easy call.
3:50
Think about the split second decisions
3:52
and avoiding a highway collision.
3:55
Who will ultimately do better a
3:57
tire driver or a self driving
3:59
car? Nowhere
4:01
is the question more fraught than
4:04
on the battlefield. As technology
4:06
evolves, should weapons systems
4:08
be given the power to make life and death
4:10
decisions? Or do we need to
4:12
ensure there's always a human a
4:15
Stanislav Petrov in the loop.
4:18
Some people, including winners of the
4:20
Nobel Peace Prize, say that
4:22
weapons should never be allowed to make
4:25
their own decisions about who or
4:27
what to attack. They're
4:29
calling for a ban on what they
4:31
call killer robots. Others
4:35
think that idea is well meaning but
4:37
naive. Today's
4:40
big question lethal autonomous
4:43
weapons. Should they ever
4:46
be allowed? If so, when,
4:50
if not, can we stop them?
5:00
My name is Eric Lander. I'm a scientist who
5:02
works on ways to improve human health.
5:04
I helped lead the Human Genome Project, and
5:07
today I lead the Broad Institute of and
5:09
Harvard. In the twenty first
5:11
century, powerful technologies
5:13
have been appearing at a breathtaking pace,
5:16
related to the Internet, artificial intelligence,
5:19
genetic engineering, and more. They
5:21
have amazing potential upsides,
5:24
but we can't ignore the risks that come
5:26
with them. The decisions aren't just
5:28
up to scientists or politicians,
5:31
whether we like it or not, we all
5:33
of us are the stewards of a brave
5:36
New Planet. This generation's
5:38
choices will shape the future as
5:40
never before. Coming
5:47
up on today's episode of Brave New
5:49
Planet fully autonomous
5:52
lethal weapons or killer
5:55
robots, we
5:57
hear from a fighter pilot about
5:59
why it might make sense to have machines in
6:01
charge of some major battlefield
6:04
decisions. I know people who
6:06
have killed civilians,
6:09
and in all cases
6:11
where people made mistakes, it was
6:13
just too much information. Things were happening
6:16
too fast. I speak with one of the world's
6:18
leading robo ethesis. Robots
6:20
will make mistakes too, but hopefully, if
6:22
done correctly, they will make far far less mistakes
6:24
than human beings. We'll
6:27
hear about some of the possible consequences
6:29
of autonomous weapons. Algorithms
6:32
interacting at machine speed faster
6:35
than humans couldn't respond might result
6:37
in accidents, and that's something like a
6:39
flash war. I'll speak with a
6:42
leader from Human Rights Watch. The campaign
6:44
to stop Killer Robots is seeking new
6:46
international lure in the form of a new
6:49
treaty. And we'll
6:51
talk with former Secretary of Defense Ash
6:53
Carter. Because I'm the guy who asked to go out
6:56
the next morning after some women and children
6:58
have been accidentally killed. As
7:00
suppose I go out there, Eric and I say,
7:02
oh, I don't know how it happened. The machine
7:04
did it. I would be crucified. I should
7:07
be crucified. So
7:09
stay with us. Chapter
7:14
one, Stanley
7:17
the self driving Car. Not
7:21
long after the first general purpose computers
7:23
were invented in the nineteen forties, some
7:26
people began to dream about fully
7:28
autonomous robots, machines
7:31
that used their electronic brains to
7:33
navigate the world, make decisions,
7:35
and take actions. Not surprisingly,
7:38
some of those dreamers were in the US Department
7:40
of Defense, specifically the
7:43
Defense Advanced Research Projects
7:45
Agency or DARPA, the
7:47
visionary unit behind the creation of the
7:49
Internet. They saw a lot
7:51
of potential for automating battlefields,
7:54
but they knew it might take decades. In
7:57
the nineteen sixties, DARPA
7:59
funded the Stanford Research inst to
8:01
build Shaky the Robots,
8:04
a machine that used cameras to
8:06
move about a laboratory. In the nineteen
8:08
eight ease, it supported universities
8:11
to create vehicles that could follow lanes
8:13
on a road. By the early
8:16
two thousands, DARPA decided
8:18
that computers had reached the point that
8:20
fully autonomous vehicles able
8:23
to navigate the real world might
8:25
finally be feasible. To
8:27
find out, DARPA decided
8:30
to launch a race. I talked
8:32
to someone who knew a lot about
8:34
it. My name is Sebastian Thrun.
8:37
I mean the smartest person on the planet and the best looking.
8:39
That's kidding. Sebastian Thrun
8:42
gained recognition when his autonomous car,
8:44
a modified Volkswagon with a computer
8:47
in the trunk and sensors on the roof,
8:50
was the first to win the DARPA Grand
8:52
Challenge. A dun Challenge was his momentous
8:55
government sponsors robot raises epic
8:57
RaSE, can you bid a robot that can
9:00
navigate one hundred and thirty punishing
9:02
miles through the Mohabi Desert and
9:05
the best robot like seven miles
9:07
and then literally end up in many
9:10
many mini researchers had concluded can't be done.
9:12
In fact, many of my colleagues told
9:14
me I'm going to waste my time and my name
9:17
if I engaged in this kind of super hard race.
9:19
And that made you more interested in doing it, of course, and
9:21
so you built Stanley. Yeah,
9:24
So my students built Stanley and started as
9:26
a class And Stanford students
9:28
are great. If you tell them go to the moon
9:31
in two months, they're going to go to the moon. So
9:33
then two thousand or five, the actual
9:36
government sponsored race, how
9:38
did Stanley do. We came in first,
9:41
so we are focused
9:43
insanely strongly on software and specifically
9:45
on machine learning, and that differentiated as
9:48
from pretty much every other team that focused
9:50
on hardware. But the way I look at this is there
9:52
were five teams that finished this ruling
9:54
race within one year, and it's the
9:56
community of the people that build all these
9:58
machines that really won. So
10:00
nobody made it a mile in the first
10:03
race, and five different
10:05
teams made it more than one hundred and thirty
10:07
miles through the desert, just a year later, Yeah,
10:09
that's kind of amazing to me.
10:12
That just showed how fast this technology
10:14
can possibly evolve. And what's
10:17
happened since then? I worked
10:19
at Google for a vile and eventually
10:21
this guy, Larry Page came to
10:23
me and says, hey, Sebastian, I thought about this long and
10:26
heart. We should build a self driving car
10:28
they can drive on all streets in the world. And
10:30
with my entire authority, I said that
10:32
cannot be done. We just had driven
10:35
a desert raised there was never pedestrians
10:37
and bicycles and all the other people that we could kill
10:40
in the environment. And for me, just the sheer imagination
10:42
we would drive a self driving car to San Francisco
10:45
sounded always like a crime. So you
10:47
told Larry Page, one of the two
10:50
co founders of Google, that
10:52
the idea of building a self driving car that could
10:54
navigate anywhere was just not
10:57
Yeah, feelous. Later came back and said, he Sebastian,
10:59
look, I trust you, you're the expert,
11:01
but I want to explain Eric Schmidt, then the Google
11:03
CEO, and it's my co founder, surgery brain, why
11:06
it can be done. Can you give me the technical
11:09
reason? So I went home in agony,
11:12
thinking about what is
11:15
the technical reason why it can be done?
11:17
And I got back the next day and I said, so,
11:20
what is it? And I said, I
11:22
can't think of any and Lomi
11:24
Hoold. Eighteen months later, roughly ten engineers
11:27
we drove pretty much every street in California.
11:30
Today, autonomous technology is
11:33
changing the transportation industry. About
11:35
ten percent of cars sold in the US are
11:37
already capable of at least partly
11:40
guiding themselves down the highway.
11:42
In twenty eighteen, Google's
11:44
self driving car company Weymo
11:47
launched a self driving taxi service
11:49
in Phoenix, Arizona, initially
11:52
with human backup drivers behind every wheel,
11:55
but now sometimes even without.
11:57
I asked Sebastian why he thinks
11:59
this matters. We lose more
12:03
than a million people in traffic accidents
12:05
every year, almost exclusively
12:08
to us not pay attention. When
12:10
it was eighteen, my best friend died
12:13
in a traffic accident and it was
12:15
a split second poor decision
12:18
from his friend who was driving in who also died. To
12:21
me, this is just unacceptable. Beyond
12:23
safety, Sebastian sees
12:26
many other advantages for autonomy.
12:28
During a commute, you can do something else
12:31
that means you're probably willing to commute
12:34
further distances. You could sleep, or watch the movie, or
12:36
do email. And then eventually people
12:38
can use cars that today can't operate
12:41
them blind people, old
12:43
people, children, babies.
12:45
I mean, there's an entire spectrum of people that are kindly
12:47
excluded. They would now be able to be mobile.
12:53
Chapter two the Tomahawk
12:56
so darpest efforts over the decades
12:58
helped give rise to the modern self
13:00
driving car industry, which promises
13:03
to make transportations safer, more
13:05
efficient, and more accessible. But
13:08
the agencies primary motivation was
13:10
to bring autonomy to a different challenge,
13:13
the battlefield. I traveled to Chapel
13:15
Hill, North Carolina, to meet with someone who
13:18
spends a lot of time thinking about the consequences
13:20
of autonomous technology. We
13:22
both serve on a civilian advisory
13:25
board for the Defense Department. My
13:27
name is Missy Cummings. I'm a
13:29
professor of electrical and computer engineering
13:32
at Duke University, and I
13:34
think one of the things that people find
13:36
most interesting about me is that
13:38
I was one of the US military's
13:40
first female fighter pilots in the Navy.
13:43
Did you always want to be a fighter
13:45
pilot? So when I was growing up,
13:47
I did not know that women could
13:50
be pilots, and indeed, when I was
13:52
growing up, women could it be pilots. And it
13:54
wasn't until the late
13:57
seventies that women actually became
13:59
pilots in the military.
14:02
So I went to college.
14:05
In nineteen eighty four, I was
14:07
at the Naval Academy, and then of course,
14:09
in nineteen eighty six Top Gun came out, and then
14:11
I know who doesn't want to be a pilot
14:13
After you see the movie Top Gun. Missy
14:16
is tremendously proud of the eleven years
14:18
she spent in the Navy, but she also
14:20
acknowledges the challenges of being part
14:22
of that first generation of woman
14:24
fighter pilots. It's no secret
14:27
that the reason I left the military was
14:29
because of the hostile
14:31
attitude towards women. None
14:33
of the women in that first group stayed in to make
14:36
it a career. The guys were very angry
14:38
that we were there, and
14:40
I decided to leave when they started sabotaging
14:43
my flight gear. I just thought, this
14:45
is too much. If something really bad happened,
14:47
you know, I would die. When
14:49
Missy Cummings left the Navy, she decided
14:52
to pursue a PhD in Human Machine
14:55
interaction. In my last three years
14:57
flying IF eighteens, there
14:59
were about thirty six people I knew
15:01
that died, about one person a month. They
15:04
were all training accidents. It just really
15:06
struck me how many people were dying
15:09
because the design of the airplane
15:11
just did not go with the human
15:13
tendencies. And so I
15:16
decided to go back to school to find out what
15:18
can be done about that. So I went
15:20
to finish my PhD at
15:22
the University of Virginia, and
15:24
then I spent the next ten years at MT
15:27
learning my craft. The
15:29
person I am today is half because
15:31
of the Navy and half because of MIGHT. Today,
15:34
Missy is a Duke University
15:36
where she runs the Humans an Autonomy
15:38
Lab, or for short HOW.
15:41
It's a nod to the sentient computer that
15:43
goes rogue in Stanley Kubrick's
15:46
film two thousand and one, A Space
15:48
Odyssey. This mission is too
15:50
important for me to allow you to jeopardize
15:53
it. I don't
15:55
know what you're talking about. How. I
15:58
know that you and Frank were planning to disconnect
16:00
me, and I'm afraid that's something
16:02
I cannot allow to happen. And
16:05
so I intentionally named my lab
16:08
how so that we
16:10
were there to stop that from happening. Right,
16:13
I had seen many friends die, not
16:15
because the robot became sentient, in
16:17
fact, because the designers of
16:19
the automation really had no
16:21
clue how people would or would not use this technology.
16:25
It is my life's mission statement
16:28
to develop human collaborative computer
16:31
systems that work with each other
16:33
to achieve something greater than either would alone.
16:35
The Humans and Autonomy Lab works on
16:37
the interactions between humans
16:40
and machines across many fields,
16:43
but given her background, Missy's
16:45
thought a lot about how technology
16:47
has changed the relationship between
16:50
humans and their weapons. There's
16:52
a long history of us distancing
16:55
ourselves from our actions. We
16:57
want to shoot somebody, we wanted
16:59
to shoot them with bows and arrows. We
17:01
wanted to drop bombs from five miles
17:03
over a target. We want cruise muscles
17:05
that can kill you from another country. Right, it
17:08
is human nature to back that
17:10
distance up, Missy
17:12
season inherent tension. On
17:15
one hand, technology distances
17:17
ourselves from killing. On
17:19
the other hand, technology is letting
17:21
us design weapons that are more accurate
17:24
and less indiscriminate in their
17:26
killing. Missy rotor PhD
17:29
thesis about the Tomahawk missile, an
17:31
early precursor of the autonomous
17:33
weapons systems being developed today.
17:36
The Tomahawk missile has these stored
17:38
maps in its brain, and
17:40
as its skimming along the nap of the
17:43
Earth, it compares the pictures that it's
17:45
taking with its pictures and its database to
17:47
decide how to get to its target. This
17:49
Tomahawk was kind of a set it and forget
17:52
it kind of thing. Once you launched it,
17:54
it would follow its map to the right place
17:56
and there was nobody looking over its shoulders.
17:58
Well, so the
18:00
Tomahawk missile that we saw in the Gulf
18:03
War, that was a fire and forget missile
18:05
that a target would be programmed
18:08
into the sole and then it would be fired
18:10
and that's where it would go. Later, around
18:13
two thousand, two thousand and three,
18:16
then GPS technology was coming
18:18
online, and that's when we got the tactical Tomahawk,
18:21
which had the ability to be redirected
18:23
in flight. That success
18:25
with GPS and the Tomahawk opened the military's
18:28
eyes to the
18:30
ability to use them in drones. Today's
18:33
precision guided weapons are far more accurate
18:36
than the widespread aerial bombing that
18:38
occurred on all sides in World War Two,
18:41
where some cities were almost entirely
18:44
leveled, resulting in huge numbers
18:46
of civilian casualties. In
18:48
the Gulf War, Tomahawk missile
18:50
attacks came to be called surgical
18:53
strikes. I
18:56
know people who have killed
18:59
civilians, and I
19:01
know people who have killed friendlies. They
19:04
have dropped bombs on our own forces and killed
19:06
our own people. And
19:10
in all cases where people made
19:12
mistakes, it was just too
19:14
much information. Things were happening too
19:16
fast. You've seen some pictures
19:18
that you've got in a brief hours ago, and you're
19:20
supposed to know that what you're
19:22
seeing now through this grainy image thirty
19:25
five thousand feet over a target is the
19:27
same image that you're being asked to bob.
19:29
The Tomahawk never missed its
19:31
target. It never made a
19:34
mistake unless it was programmed
19:36
as a mistake. And that's old
19:38
autonomy, and it's only gotten better
19:40
over time. Chapter
19:45
three, Kicking down Doors.
19:49
The Tomahawk was just a baby step
19:51
toward automation. With the ability
19:53
to read maps, it could correct its course,
19:56
but it couldn't make sophisticated decisions.
19:59
But what happens when you start adding modern
20:01
artificial intelligence? So
20:04
where do you see autonomous weapons
20:06
going? If you could kind of map out
20:08
where are we today and where do you think we'll be
20:10
ten twenty years from now. So,
20:13
in terms of autonomy and weapons, by
20:15
today's standards, the Tomahawk
20:17
missile is still one of the best ones that we
20:19
have, and it's also still one of the most advanced.
20:23
Certainly, there are research arms of the military
20:25
who are trying very hard to
20:29
come up with new forms of autonomy.
20:31
There was the Predicts that came out
20:33
of Lincoln Lab, and this was basically
20:36
a swarm of really tiny UAVs
20:39
that could coordinate together.
20:42
A ua V, an unmanned
20:45
aerial vehicle is military
20:47
speak for a drone. The Predicts
20:50
the drones that Missy was referring to.
20:53
They were commissioned by the Strategic Capabilities
20:55
Office of the US Department of Defense. These
20:58
tiny flying robots are able to communicate
21:00
with each other and make split second
21:02
decisions about how to move as a group.
21:05
Many researchers, have you been using bio inspired
21:07
methods? Be right? So
21:10
bees have local and global intelligence.
21:12
Like a group of bees, these drones
21:14
are called a swarm collective
21:17
intelligence on a shared mission.
21:19
A human can make the big picture decision
21:22
and the swarm of microdrones can then
21:24
collectively decide on the most efficient
21:26
way to carry out the order in the moment.
21:30
I wanted to know why exactly
21:32
this technology is necessary,
21:34
so I went to speak to someone who
21:37
I was pretty sure would know. I'm Ash
21:39
Carter. Most people will probably
21:42
have heard my name as the Secretary
21:44
of Defense who proceeded Gimatus.
21:48
You will know me in part from the fact
21:50
that we knew one another way back in Oxford
21:52
when we were both young scientists, and I guess
21:54
agents start there. I'm a physicist. When you were
21:57
doing your PhD in physics, I was doing
21:59
my PhD in mathematics at Oxford.
22:01
What was your thesis on? It was
22:03
on quantum chrominynamics.
22:05
That was the theory of quarks
22:08
and gluons. And how in the world
22:10
is somebody who's an expert
22:12
in quantum chromodynamics become
22:14
the Secretary of Defense. It's an
22:16
interesting story. The people
22:19
who were the seniors
22:22
in my field of physics, the
22:25
mentors, so to speak, were all
22:27
members of the Manhattan Project
22:30
generation. They had built
22:32
the bomb during World War Two, and
22:35
they were proud of
22:37
what they'd done because they
22:40
believed that it had ended the war with
22:42
fewer casualties than otherwise
22:44
there would have been in a full scale of invasion
22:47
of Japan, and also that it had
22:49
kept the peace through the Cold War, so they were proud
22:51
of it. However, they knew there was a
22:53
dark side, and they conveyed
22:55
to me that it was my
22:58
responsibility as
23:00
a scientist to be involved
23:02
in these matters. And the
23:06
technology doesn't determine what the balance
23:08
of good and bad is. We human beings
23:10
do. That was the lesson, and
23:13
so that's what got me started, and
23:15
then my very first Pentagon job, which was
23:17
in nineteen eighty one, right
23:19
through until the last time I walked
23:22
out of the Pentagon of Sectary Defense, which
23:24
was January of twenty seventeen.
23:26
Now, when you were secretary, there
23:29
was a Strategic Capabilities
23:31
Office that it's been
23:34
publicly reported, was experimenting
23:36
with using drones to
23:39
make swarms of drones that
23:41
could do things, communicate
23:43
with each other, make formations. Why
23:45
would you want such things? So it's a good question. Here's
23:48
what you do with the drone like that, You
23:50
put a jammer on it, a little radio
23:53
beacon, and you fly it right
23:55
into the eye of
23:58
a enemy radar. So
24:00
all that radar c's is
24:03
the energy emitted by that
24:05
little drone, and it's essentially dazzled
24:07
or blinded. If there's one
24:10
big drone, that radar
24:12
is precious enough that the defenders
24:14
going to shoot that drone down. But
24:17
if you have so many out there, the
24:19
enemy can't afford to shoot them all down. And
24:23
since they are flying right
24:25
up to the radar, they don't have to
24:27
be very powerful. So there's
24:29
an application where lots
24:31
of little drones can have
24:34
the effect of nullifying enemy
24:37
radar. That's a pretty big deal for a
24:39
few little, little microdrones. To
24:41
learn more, I went to speak with Paul Shari.
24:44
Paul's the director of the Technology and National
24:46
Security Program at the Center
24:49
for a New American Security. Before
24:51
that, he worked for Ash Carter at the Pentagon
24:53
studying autonomous weapons and
24:56
a recently authored a book called Army
24:59
of None, Autonomous Weapons
25:01
and the Future of War. Paul's
25:03
interest in autonomous weapons began when
25:05
he served in the Army. I enlisted
25:09
in the Army to become an Army ranger.
25:11
That was June of two thousand and one,
25:13
did a number of tours overseas and the
25:16
wars and a rock Afghanistan. So
25:18
I'll say one moment that stuck out for me
25:21
where I really sort of the light bulb
25:23
went on about the power of robotics in
25:25
warfare. I was in
25:27
a rock in two thousand and seven. We
25:29
were on a patrol, driving along in
25:31
a Striker armored vehicle. Came
25:33
across an ID improvised explosive
25:35
device makeshift road type bomb, and
25:38
so we called up bomb disposal folks. So
25:40
they show up and I'm expecting the
25:42
bomb tech to come out in that big bomb
25:45
suit that you might have seen in the movie The hurt Locker,
25:47
for example, and instead out
25:49
rolls out a little robot and
25:52
I kind of went, oh, that makes a lot
25:54
of sense. Have the robot diffused the bomb.
25:56
Well, it turns out there's a lot of things in war that are super
25:59
dangerous where it makes sense to have robots
26:01
out on the front lines, getting people
26:03
better stand off a little bit more separation
26:06
from potential threats. The bomb diffusing
26:08
robe are still remote controlled
26:11
by a technician, but ashe Carter
26:13
wants to take the idea of robots doing the
26:15
most dangerous work a step further
26:18
somewhere in the future. But I'm certain will
26:20
occur. Is I think there will
26:22
be robots who will be part
26:24
of infantry squads and that will
26:26
do some of the most dangerous jobs
26:29
in an infantry squad, like kicking down
26:32
the door of a building and being the first
26:34
one to run in and clear
26:36
the building of terrorists or whatever. That's
26:39
a job that doesn't sound like something
26:42
I would like to have a young American
26:44
man or woman doing if I could replace
26:46
them with a robot. Chapter
26:51
four Harpies,
26:55
Paul Shari gave me an overview of the sophisticated
26:58
unmanned systems currently used by
27:00
militaries. So I think it's worth separate running
27:02
out the value of robotics
27:05
versus autonomy removing
27:07
a person from decision making. So
27:09
what's so special about autonomy.
27:12
The advantages there are really about
27:15
speed. Machines can make decisions
27:17
faster than humans. That's why automatic
27:19
breaking and automobiles is valuable. But
27:21
you could have much faster reflexes
27:23
than a person might had. Pul separates the
27:25
technology into three baskets. First,
27:29
semi autonomous weapons. Semi
27:31
autonomous weapons that are widely
27:33
used around the globe today, where automation
27:35
is used to maybe help identify targets,
27:38
but humans are in the final decision
27:41
about which targets to attack. Second,
27:43
there are supervised autonomous
27:45
weapons. There are automatic
27:48
modes that can be activated on
27:50
air and missile defense systems
27:52
that allow these computers to defend
27:54
the ship or ground vehicle
27:57
or land base all on its own against
27:59
these incoming threats. But humans supervise
28:02
these systems in real time. They could,
28:04
at least in theory, intervene. Finally,
28:07
there are fully autonomous weapon There
28:09
are a few isolated
28:11
examples of what you might consider
28:14
fully autonomous weapons where there's no human
28:16
oversight and they're using an offensive capacity.
28:19
The clearest today that's an operation is the Israeli
28:21
Harpy drone that can
28:23
load over a wide area after
28:26
about two and a half hours at a time to search
28:28
for enemy radars, and then when it finds
28:30
one, it can attack it all on its
28:32
own without any further
28:35
human approval. Once it's launched,
28:37
that decision about which particular target to attack
28:39
that's delegated to the machine. It's
28:42
been sold to a handful of countries Turkey,
28:44
India, China, South Korea.
28:47
I asked Missy if she saw advantages to
28:49
having autonomy built into lethal weapons.
28:52
While she had reservations, she pointed
28:54
out that in some circumstances
28:56
it could prevent tragedies. A human
28:59
has something called the neuromuscular
29:01
lag in them. It's about a half second delay. So
29:04
you see something, you can execute
29:06
an action a half second later. So
29:10
let's say that that guided weapon fired
29:13
by a human is going into
29:15
a building, and then right before
29:17
it gets to the building, at a half second,
29:20
the door opens and a child walks out.
29:22
It's too late. That child is dead.
29:25
But a lethal autonomous weapon who
29:28
had a good enough perception
29:30
system could immediately
29:33
detect that and immediately guide itself
29:35
to a safe place to explode. That
29:38
is a possibility in the future. Chapter
29:43
five bounded morality.
29:48
Some people think, and this point
29:50
is controversial, that robots
29:52
might turn out to be more humane
29:55
than humans. The history of
29:57
warfare has enough examples
29:59
of atrocities committed by soldiers
30:01
on all sides. For example,
30:03
in the middle of the Vietnam War in March
30:06
nineteen sixty eight, a company of American
30:08
soldiers attack the village in South Vietnam,
30:11
killing and estimated five hundred and four
30:14
unarmed Vietnamese men, women,
30:16
and children, all noncombatants.
30:19
The horrific event became known as
30:21
the Melai massacre in
30:24
nineteen sixty nine. Journalist
30:26
Mike Wallace of Sixty Minutes sat
30:29
down with Private Paul Medloe,
30:31
one of the soldiers involved in the massacre.
30:33
Well, I'm might a kill about ten
30:35
or fifteen of them men, women,
30:38
and children and babies
30:41
and babies. You're
30:43
married, right, children
30:46
too? How can a father of two young
30:48
children shoot babies? I
30:52
don't know when to sworn in things. Of
30:54
course, the vast majority of soldiers do
30:57
not behave this way. But humans
30:59
can be thoughtlessly violent. They
31:01
can act out of anger, out of fear, they
31:04
can seek revenge, they can murder
31:06
senselessly. Can robots
31:08
do better? After all, robots
31:11
don't get angry, They're not impulsive. I
31:13
spoke with someone who thinks that lethal
31:16
autonomous weapons could ultimately be
31:18
more humane. My name is Ronald Arkin.
31:21
I'm a regents professor at the Georgia
31:23
Institute of Technology in Atlanta, Georgia.
31:25
I am a roboticist for close
31:28
to thirty five years. I've
31:30
been in robot ethics for maybe the past
31:32
fifteen. Ron wanted to make it clear
31:35
that he doesn't think these robots are perfect,
31:37
but they could be better than our current
31:40
option. I am absolutely not pro
31:42
lethal autonomous weapons systems because
31:44
I'm not pro lethal weapons of
31:47
any sort. I am against killing
31:49
in all of its manifold forms. But
31:51
the problem is that humanity
31:53
persist in entering into warfare. As
31:55
such, we must better protect the
31:57
innocent in the battlespace, far better than
32:00
we currently do. So Ron
32:02
thinks that lethal autonomous weapons could
32:04
prevent some of the unnecessary violence
32:06
that occurs in war. Human being
32:09
don't do well in warfare in
32:11
general, and that's why there's
32:13
so much room for improvement. There's
32:16
unnamed fire, there's mistakes,
32:18
there's carelessness, and in the worst case,
32:20
there's the commission of atrocities, and
32:23
unfortunately, all those things lead
32:25
to the depths of noncombatants. And
32:28
robots will make mistakes too. They probably
32:30
will make different kinds of mistakes, but hopefully,
32:32
if done correctly, they will make far far less
32:35
mistakes than human beings do in certain
32:37
narrow circumstances where human beings are
32:39
prone to those errors. So how old the robots
32:42
follow these international humanitarian
32:45
standards? The way in which
32:47
we explored initially is
32:49
looking at something referred to as bounded morality,
32:52
which means we look at very narrow situations.
32:55
You are not allowed to drop bombs
32:57
on schools, on hospitals, mosques,
33:00
or churches. So the point
33:02
is, if you know the geographic location
33:05
of those, you can demarcate those
33:07
on a map, use GPS, and
33:10
you can prevent someone from pulling a trigger.
33:13
But keep in mind these systems are not only
33:15
going to decide when to engage it, but also
33:17
when not to engage a target. They
33:20
can be more conservative. I believe
33:22
the potential exists to reduce noncombatant
33:24
casualties and collateral damage in almost all
33:26
of its forms over what we currently have, so
33:33
autonomous weapons might operate more efficiently,
33:35
reduce risk to one's own troops, operate
33:38
faster than the enemy, decreased civilian
33:40
casualties, and perhaps avoid
33:43
atrocities. What
33:46
could possibly go wrong? Chapter
33:51
six? What could possibly go wrong?
33:55
Autonomous systems can do some pretty remarkable
33:58
things these days, but of
34:00
course, robots just do what their computer
34:02
code tells them to do. The
34:04
computer code is written by humans,
34:07
or, in the case of modern artificial intelligence,
34:10
automatically inferred from training
34:12
data. What happens
34:14
if a robot encounters a situation that
34:16
the human where the training data didn't
34:19
anticipate, well, things
34:22
could go wrong in a hurry. One
34:24
of the concerns with autonomous
34:27
weapons is that they might malfunction
34:29
in a way that leads them to begin
34:32
erroneously engaging targets.
34:34
Robots run amock, and this is
34:37
particularly a risk for
34:40
weapons that could target on their own. Now,
34:42
this builds on a
34:45
flaw and known malfunction of machine
34:47
guns today called a runaway gun.
34:50
A machine gun begins firing for one
34:52
reason another and because of the nature
34:54
of a machine gun where one
34:56
bullets firing cycles the automation
34:59
and brings in an ex bullet. Once it
35:01
starts firing, human doesn't have
35:03
to do it, and it will continue firing bullets. The
35:05
same sort of runaway behavior can
35:07
result from small in computer
35:10
code, and the problems
35:12
only multiply when autonomous
35:14
systems interact at high speed.
35:17
Paul Shara points to Wall Street
35:19
as a harbinger of what can go
35:21
wrong, and we end up some places like where
35:23
we are in stock trading today, where
35:26
many of the decisions are highly
35:28
automated, and we get things like flash
35:30
crashes. What
35:33
the pack is going on down here? I
35:35
don't know. There is fear. This is capitulation.
35:38
Really. In May twenty
35:40
ten, computer algorithms drove
35:42
the Dow Jones down by nearly
35:45
one thousand points in thirteen
35:47
minutes, the steepest drop it
35:49
had ever seen in a day. The
35:51
concern is that a world where militaries
35:54
have these algorithms
35:57
interacting at machine speed, faster
36:00
than humans can respond, might
36:02
result in accidents. And that's something
36:04
like a flash war. By a flash
36:06
war, you mean this thing just cycling out of
36:08
control somehow, right, But
36:10
the algorithms are merely following their programming,
36:13
and they escalate a conflict into
36:15
a new area of warfare, a new
36:18
level of violence, in a way that might
36:20
make it harder for humans
36:22
to then dial things back and
36:24
bring things back under control. The system
36:26
only knows what it's been programmed or
36:28
been trained to know. The human
36:30
can bring together all of these other pieces
36:33
of information about context, and
36:35
human can understand what's at stake. So there's
36:37
no Stanislav Petrov on the loop.
36:40
That's the fear, right, is
36:42
that if there's no Petrov there
36:45
to say no, what might the machines
36:47
do on their own? Chapter
36:51
seven, slaughter Bots.
36:56
The history of weapons technology includes
36:58
well intentioned efforts to reduce violence
37:01
and suffering that end up backfiring.
37:04
I tell in the book the story of the Gatling
37:06
Gun, which was invented by Richard
37:08
Gatling during the American Civil War, and
37:11
he was motivated to invent this
37:13
weapon, which was a forerunner of the machine gun,
37:16
as an effort to reduce soldiers
37:18
deaths and more. He saw all of these soldiers coming
37:20
back maimed and injured from the Civil
37:22
War, and he said, would it be great if we needed
37:24
fewer people to fight? So we invented
37:27
a machine that could allow four
37:29
people to deliver the same lethal effects
37:31
in the battlefield as a hundred. Now, the
37:34
effect of this wasn't actually to reduce the number
37:36
of people fighting, and we got to World War
37:38
One, we saw massive
37:40
devastation and a whole generation
37:43
of young men in Europe killed because of this technology.
37:46
And so I think that's a good cautionary
37:48
tale as well, that sometimes
37:50
the way the technology evolves and how it's used
37:53
may not always be how we'd like it to
37:55
be used. And even if regular armies
37:58
can keep autonomous weapons within the confines
38:00
of international humanitarian law, what
38:03
about rogue actors? Remember
38:06
those autonomous swarms we discussed
38:08
with Ash Carter, those tiny drones
38:10
that work together to block enemy radar.
38:13
What happens if the technology spreads
38:15
beyond armies? What if a terrorist
38:17
adds a gun or an explosive and maybe
38:20
facial recognition technology to
38:22
those little flying bots. In
38:25
twenty seventeen, Berkeley professors
38:27
Stuart Russell and the Future of Life Institute
38:30
made a mock documentary called slaughter
38:32
Bots, is part of their campaign against
38:35
fully autonomous lethal drones. The
38:37
nation is still recovering from yesterday's
38:40
incident, which officials are describing as
38:42
some kind of automated attack which
38:44
killed eleven US senators at the Capitol
38:46
Building. They flew in from every rare,
38:48
but attack just one side of the aisle. It
38:51
was people were spreading. Unlike
38:54
nuclear weapons, which are difficult to build,
38:57
you know, it's not easy to obtain or work
38:59
with weapons grade uranium, the
39:01
technology to create and modify autonomous
39:03
drones is getting more and more accessible.
39:06
All of the technology you need from
39:08
the automation standpoint either exists
39:11
in the vehicle already or you can
39:13
download from GitHub. I
39:15
asked former Secretary of Defense As Carter,
39:18
if the US government is concerned about
39:20
this sort of attack, you're right to worry
39:22
about drones and Chris. It only
39:25
takes a depraved person who
39:27
can go to a store and buy a
39:30
drone to at least scare
39:32
people and quite possibly threaten people
39:35
hanging a gun off of it or
39:37
putting a bomb of some kind
39:39
on it, and then suddenly people don't feel safe
39:42
going to the super Bowl or landing
39:44
at the municipal
39:47
airport. And we can't have that. I mean
39:49
it, certainly. As your former secretary of Defense,
39:51
my job was to make sure that we didn't put up with
39:53
that kind of stuff. I'm supposed to protect
39:56
our people, and so how do
39:58
I protect people against drones? In
40:00
general? They can be shot down, but
40:03
they can put more drones up than it I can
40:05
conceivably shoot at. Not
40:08
to mention, shooting at things in a
40:10
Super Bowl stadium is an inherently
40:12
dangerous solution to this problem. And
40:15
so there's a more subtle way of dealing
40:17
with drones. I will either
40:20
jam or take
40:22
over the radio link, and
40:24
then you just tell it to fly away and
40:27
go off into the countryside
40:29
somewhere and crash into a field. All
40:31
right, So help me out if I have
40:34
enough autonomy, couldn't
40:36
I have drones without radio links
40:38
that just get their assignment and go off
40:40
and do things. Yes, and then
40:43
your mind as a defender goes
40:45
to something else. Now that they've got
40:47
their idea of what they're looking for a set
40:49
in their electronic mind. Let
40:52
me change what I look like, Let me
40:54
change what the stadium looks like to it,
40:56
let me change what the target looks like. And
40:59
for the Super Bowl, what do I do about that? Well,
41:02
once I know I'm being looked at, I
41:04
have the opponent in
41:07
a box. A few people
41:09
know how easy facial recognition
41:11
is to fool. Because
41:14
I can wear the right kind of goggles. Are
41:17
stick ping pong balls in my cheeks?
41:20
There's always a stratagem
41:24
memo toself. Next time I go
41:26
to Gillette Stadium for a Patriots game, bring
41:29
ping pong balls? Really?
41:36
Chapter eight, The Moral Buffer.
41:39
So we have to worry about whether lethal autonomous
41:41
weapons might run them up or fall into
41:43
the wrong hands. But there
41:46
may be an even deeper question. Could
41:49
fully autonomous lethal weapons change
41:51
the way we think about war? I
41:54
brought this up with Army of non author Paul
41:56
Shari. So one of the concerns about autonomous
41:59
weapons is that it might lead to a breakdown
42:01
in human more responsibility
42:04
for killing and war. If the
42:06
weapons themselves are choosing targets, the
42:08
people no longer feel like they're the ones doing
42:10
the killing. Now, on the plus
42:13
side of things, that might mean
42:15
to less post traumatic stress in war.
42:17
These things have real burdens
42:20
that weigh on people, but some
42:22
argue that the burden of killing should
42:25
be a requirement of war. It's
42:27
worth also asking if nobody
42:29
slept uneasy at night, what
42:31
does that look like? Would there be less
42:34
restraint in war and more killing as
42:36
a result. Missy Cummings,
42:38
the former fighter pilot and current Duke
42:40
professor, wrote an influential paper
42:42
in two thousand and four about how increasing
42:44
the gap between a person and
42:46
their actions creates what
42:49
she called a moral buffer.
42:52
People ease the
42:55
psychological and emotional pain of
42:57
warfare by
43:00
basically superficially
43:03
layering in these other technologies to
43:05
kind of make them lose track of what they're doing.
43:07
And this is actually something that I do think it's a problem
43:10
for lethal autonomous weapons. If
43:13
we send a weapon and we'd tell it to
43:15
kill one person and it
43:17
kills the wrong person, then
43:19
it's very likely that people will
43:21
push off their sense of responsibility and accountability
43:24
onto the autonomous agent
43:26
because they say, well, it's not my fault,
43:28
it was the autonomous agent's fault. On
43:31
the other hand, Paul Scharide tells
43:33
a story about how when there's no
43:35
buffer, humans rely on an
43:37
implicit sense of morality that
43:39
might be hard to explain to a robot.
43:42
There was an incident early in the
43:44
war where I was part of an army ranger sniper
43:46
team up on the Afghanistan Pakistan
43:49
border and we were watching
43:51
for Taliban fighters infiltrating
43:53
across the border, and when dawn
43:55
came, we weren't nearly as concealed as we
43:57
had hoped to be, and very quickly
44:00
a farmer came out to relieve himself
44:02
in the fields and saw us, and we knew
44:04
that we were compromised. What I
44:07
did not expect was what they did next, which
44:09
was I sent a little girl to scout at our position. She
44:12
was maybe five or six, She was
44:14
not particularly sneaky. She stared
44:17
directly at us and we heard the
44:19
chirping of what we later realized was probably
44:21
a radio that you had on her, and she was reporting
44:23
back information about us, and
44:25
then she left it. Not long after, some fighters
44:27
did come and then The gun
44:30
fight that ensued brought out the whole valley, so we had
44:32
to leave. But later that day we were talking
44:34
about how it would treat a situation
44:36
like that. Something that just didn't come up in conversation
44:40
was the idea of shooting this little girl. Now,
44:42
what's interesting is that under the laws of war, that
44:45
would have been legal. The laws
44:47
of war don't set an age for combatants.
44:50
Your status as a combatant just based on your actions,
44:53
and by scouting for the enemy,
44:55
she was directly participating on hostilities.
44:58
If you had a robot that was programmed
45:00
to perfectly comply with the laws of war, it
45:03
would have shot this little girl. There
45:05
are sometimes very difficult decisions
45:07
that are forced on people in but I don't think
45:09
this was one of them. But I think it's
45:12
worth asking how would a robot know the difference between what's
45:14
legal and what's right, and
45:16
how would you even begin to prehend that into a machine.
45:22
Chapter nine, The Campaign
45:24
to stop killer robots. The
45:28
most fundamental moral objection
45:30
to fully autonomous lethal weapons
45:33
comes down to this, As
45:35
a matter of human dignity, only
45:38
a human should be able to make the decision
45:40
to kill another human. Some things
45:43
are just morally wrong, regardless
45:45
of the outcome, regardless of whether or not you know,
45:48
torturing one person saves a
45:50
thousand, its torture is wrong.
45:53
Slavery is wrong. And
45:55
from this point of view, one might say, well,
45:58
look, it's wrong to let a machine
46:00
decide whom to kill. Humans have to
46:02
make that decision. Some people have been working
46:04
hard to turn this moral view into
46:07
binding international law. So
46:09
my name is Mary Warem. I'm the advocacy
46:12
director of the Arms division of Human Rights
46:14
Watch. I also coordinate
46:16
this coalition of groups called
46:19
the Campaign to Stop Killer Robots,
46:21
and that's a coalition of one hundred
46:23
and twelve non governmental organizations
46:26
in about fifty six countries
46:28
that is working towards a single goal, which
46:31
is to create a prohibition
46:33
on fully autonomous weapons. The
46:36
campaign's argument is rooted in the
46:38
Geneva Conventions, a set
46:40
of treaties that establish humanitarian
46:43
standards for the conduct of war.
46:46
There's the principle of distinction, which
46:48
says that armed forces must recognize
46:51
civilians and may not target them.
46:53
And there's the principle of proportionality,
46:56
which says that incidental civilian deaths
46:59
can't be disproportionate to an
47:01
attack direct military advantage.
47:04
The campaign says killer robots
47:06
fail these tests. First,
47:09
they can't distinguish between combatants and noncombatants
47:12
or tell when an enemy is surrendering. Second,
47:15
they say, deciding whether civilian
47:18
deaths are disproportionate inherently
47:20
requires human judgment. For
47:23
these reasons and others, the campaign says,
47:26
fully autonomous lethal weapons
47:28
should be banned. Getting
47:30
an international treaty to
47:32
ban fully autonomous lethal weapons
47:35
might seem like a total pipe dream, except
47:38
for one thing. Mary warm
47:40
In her colleagues already pulled
47:42
it off for another class of weapons,
47:45
land mines. The signing of this historic
47:47
treaty at the very end of the
47:50
century is this generation's
47:52
pledge to the future. The
47:54
International Campaign to Ban Landmines
47:57
and its founder, Jody Williams, received
47:59
the Nobel Peace Prize in nineteen ninety
48:01
seven for their work leading to the Ottawa
48:03
Convention, which banned the use, production,
48:06
sale, and stockpiling of an anti
48:08
personnel mines. While
48:10
one hundred and sixty four nations joined
48:13
the treaty, some of the world's major
48:15
military powers never signed
48:17
it, including the United States, China,
48:20
and Russia. Still,
48:22
the treaty has worked and even
48:24
influence the holdouts. So the United
48:26
States did not join, but it went
48:28
on to I think prioritize
48:30
clearance of anti personnel land mines and
48:33
remains the biggest donor to
48:35
clearing landlines an unexploded
48:37
ordinance around the world. And then
48:39
under the Obama administration, the
48:42
US committed not to use anti personnel
48:44
land mines anywhere in the world other
48:47
than to keep the option open for the
48:49
Korean peninsula. So slowly,
48:52
over time countries do I
48:54
think come in line. One major
48:56
difference between banning land mines and
48:59
banning fully autonomous lethal weapons
49:01
is, well, it's pretty
49:03
clear what a land mine is, but a fully
49:05
autonomous lethal weapon that's
49:08
not quite as obvious. Six
49:10
years of discussion at the United Nations
49:13
have yet to produce a crisp definition.
49:16
Trying to define autonomy is also
49:18
a very challenging task, and this is
49:20
why we focus on the need for
49:22
meaningful human control. So
49:25
what exactly is meaningful
49:27
human control? The ability for
49:29
the human operator and the weapon system to communicate
49:32
the ability for the human to intervene
49:35
in the detection, selection and engagement of targets
49:37
if necessary to cancel the operation.
49:40
Not surprisingly, international talks
49:42
about the proposed ban are complicated.
49:45
I will say that a majority of the countries
49:48
who have been talking about killer robots
49:50
have called for illegally binding
49:52
instruments and international treaty. You've
49:54
got the countries who want to be helpful, like
49:57
France who was proposing working groups,
50:00
Germany who's proposed political
50:02
declarations on the importance of human
50:04
control. There's a lot of proposals,
50:07
I think from Australia about legal
50:09
reviews of weapons. Those
50:11
efforts are being rebuffed by a smaller
50:14
handful of what we call militarily powerful
50:16
countries who don't want to see new
50:19
international law. The United States
50:21
and Russia have probably been amongst the most problematic
50:24
on dismissing the calls for any
50:26
form of regulation. As with
50:28
the landlines, Mary Wareham sees a path
50:30
forward even if the major military powers
50:33
don't join at first. We cannot stop
50:35
every potential use. What we want
50:37
to do, though, is stigmatized, so that
50:39
everybody understands that even
50:41
if you could do it, it's not right and
50:44
you shouldn't. Part of
50:46
the campaign strategy is to get other groups
50:48
on board, and they're making some progress.
50:51
I think a big move in our favor
50:53
came in November when the United Nations
50:56
Secretary General Antonio Guterres,
50:59
he made a speech in which he called for them
51:01
to be banned under international law. Machine
51:05
the power and the
51:07
dispression to take human
51:10
lives are politically and
51:12
acceptable, are morally
51:14
impartment and should be banned
51:17
by international law. Artificial
51:23
intelligence researchers have also been
51:25
expressing concern. Since twenty
51:27
fifteen, more than forty five
51:30
hundred AI and robotics researchers
51:32
have signed an open letter calling
51:35
for a ban on offensive
51:37
autonomous weapons beyond meaningful
51:40
human control. The signers
51:42
included Elon Musk, Stephen
51:44
Hawking, and Demis Assabas,
51:46
the CEO of Google's Deep Mind. An
51:49
excerpt from the letter quote, if
51:52
any major military power pushes
51:54
ahead with AI weapon development, a
51:57
global arms race is virtually
51:59
inevitable, and the endpoint
52:02
of this technological trajectory
52:04
is obvious. Autonomous weapons
52:07
will become the Khalishnikov's Tomorrow
52:12
Chapter ten. To ban or
52:15
not to ban? Not
52:17
Everyone, however, favors the idea of an
52:19
international treaty banning all lethal
52:22
autonomous weapons. In fact,
52:24
everyone else I spoke to for this episode,
52:27
Ron Arkin, Missy Cummings, Paul Shari,
52:30
and Ash Carter oppose it, interestingly,
52:33
though each had a different reason and
52:35
a different alternative solution. Robo
52:38
ethicist Ron Arkin thinks
52:41
we'd be missing a chance to make wars safer.
52:44
Technology can, must, and should
52:46
be used to reduce noncombatant
52:48
casualties. And if it's not going to be
52:50
this, you tell me what
52:53
you are going to do to address that
52:55
horrible problem that exists in the world right
52:57
now, with all these innocence being slaughtered in the battlespace.
53:00
Something needs to be done, and
53:02
to me, this is one possible way.
53:04
Paul Shari thinks a comprehensive
53:07
ban is just not practical. Instead,
53:10
he thinks we should focus on banning lethal autonomous
53:12
weapons that specifically target
53:15
people. That is, anti
53:17
personnel weapons. In fact,
53:19
the Landmine Treaty bans anti personnel
53:22
land mines, but not say, anti
53:24
tank land mines. One
53:27
of the challenging things about anti personnel weapons
53:29
is that you can't stop being a person
53:32
if you wan't avoid being targeted. So if
53:34
you have a weapon that's targeting tanks, you can
53:37
come out of a tank and run away. I
53:39
mean, that's a good way to effectively
53:42
surrender and render yourself life of combat. If
53:44
it's even targeting, say handheld
53:46
weapons. You could set down your weapon and
53:48
run away from it. So do you think that'd be practical
53:51
to actually get either a
53:53
treaty or at least an understanding
53:56
that countries should forswear anti
53:58
personnel lethal
54:00
autonomous weapons. I think it's easier
54:03
for me to envision how you might get to actual
54:05
restraint. You need to make sure
54:07
that the weapon that countries are giving up it's
54:09
not so valuable that they can't still defeat
54:12
those you might be willing to cheat.
54:14
And I think it's really an open question how valuable
54:17
autonomous weapons are. But my
54:19
suspicion is that they are not as valuable
54:22
or necessary in an anti personnel
54:24
context. Former fighter
54:26
pilot and Duke professor Missy Cummings
54:28
thinks it's just not feasible to ban
54:31
lethal autonomous weapons. Look,
54:34
you can't ban people
54:36
developing computer code. It's
54:39
not a productive conversation to
54:41
start asking for bands on technology
54:44
that are almost as common as the air
54:46
we breathe. Right, So we
54:49
are not in the world of banning nuclear
54:51
technologies. And because it's a different world, we
54:54
need to come up with new ideas. What
54:56
we really need is that
54:58
we make sure that we certify these technologies
55:01
in advance. How do you actually
55:03
do the test certify that the weapon
55:06
does at least as well as a human. That's actually
55:08
a big problem because no one on the
55:10
planet, not the Department of Defense,
55:13
not Google, not Uber,
55:15
not any driverless car company understands
55:17
how to certify autonomous technologies.
55:20
So four driverless cars can come
55:22
to an intersection and
55:25
they will never prosecute that intersection
55:27
the same way a sun angle
55:30
can change the way that these things think. We
55:32
need to come up with some out of the
55:34
box thinking about how to test these
55:36
systems to make sure that they're seeing the world.
55:39
And I'm doing that in air quotes in
55:41
a way that we are expecting
55:43
them to see the world. And this is why
55:45
we need a national agenda to understand
55:48
how to do testing to
55:51
get to a place that we feel comfortable
55:53
with the results those you are successful
55:55
and you get the Pentagon and
55:58
the driverless car folks to actually
56:00
do real world testing, what
56:02
about rest to world? What's going to happen?
56:06
So one of the problems that we see in
56:09
all technology development is
56:11
that the rest of the world doesn't
56:13
agree with our standards.
56:17
It is going to be a problem going
56:19
forward, So we certainly
56:22
should not circumvent testing because
56:24
other countries are circumventing
56:26
testing. Finally, there's
56:29
former Secretary of Defense Ash Carter Back
56:31
in twenty twelve, ashe was one
56:33
of the few people who were thinking about the
56:36
consequences of autonomous technology.
56:38
At the time, he was the third ranking official
56:40
in the Pentagon in charge of weapons
56:42
and technology. He decided to
56:44
draft a policy, which the Department
56:46
of Defense adopted. It
56:49
was issued as Directive three thousand
56:51
point zero nine Autonomy
56:53
in Weapons Systems. So
56:55
I wrote this directive that
56:58
said, in essence, there will always
57:00
be a human being involved
57:03
in the decision making when it comes
57:05
to lethal force in the
57:07
military of the United States of America.
57:10
I'm not going to accept autonomous weapons
57:12
in a literal sense because
57:14
I'm the guy who has to go out the next morning after
57:17
some women and children have been accidentally killed
57:19
and explain it to a press conference
57:22
or a foreign government or a widow.
57:25
And as suppose I go out there, Eric
57:27
and I say, oh, I
57:29
don't know how it happened, the machine, did it? Are
57:32
you going to allow your Secretary of Defense
57:35
to walk out and give that
57:38
kind of excuse. No way
57:41
I would be crucified. I should be crucified
57:43
for giving a press conference like that, And I
57:45
didn't think any future Secretary Defense should
57:48
ever be in that position, or
57:51
allow him or herself to be in that position. That's
57:53
why I wrote the directive, Because, ashe
57:55
wrote the directive that currently prevents
57:57
US forces from deploying fully autonomous
58:00
lethal weapons, I was curious to know
58:02
what he thought about an international
58:04
ban. I think it's reasonable
58:07
to think about a national ban, and we have and
58:09
we have one. Do I think it's reasonable
58:12
that I get everybody else to sign up to that.
58:14
I don't because I think that
58:16
people will say they'll sign up and then not do
58:19
it. In general, I don't
58:21
like fakeery in serious
58:25
matters, and that's too easy
58:27
to fake. That is the fake
58:29
meaning to fake that they
58:31
have forsworn those weapons,
58:34
and then we find out that they
58:36
haven't, and so it
58:38
turns out they're doing it, and they're lying about
58:40
doing it or hiding that they're doing it. We've
58:42
run into that all the time. I
58:45
remember the Soviet Union said it signed
58:47
the Biological Weapons Convention. They
58:49
ran a very large biological
58:51
warfare bird. They just said they didn't all
58:54
right, but take the situation. Now, what
58:56
would be the harm of the
58:59
US signing up to such a thing, at
59:01
least building the moral approbrium
59:04
around lethal autonomous weapons,
59:06
because you're building something else at the same time,
59:08
which it's an illusion of safety for
59:11
other people. You're conspiring
59:14
in a circumstance in which they are lied to
59:16
about their own safety, and I
59:20
feel very uncomfortable doing that. Paul
59:23
Shari sums up the challenge as well.
59:25
Countries are widely divergent interviews
59:27
on things like a treaty, but there's
59:29
also been some early agreement that at
59:32
some level we need
59:34
humans involved in these kinds of decisions.
59:37
What's not clear is at what level is that
59:39
the level of prisiople choosing every
59:41
single target, people deciding
59:44
at a higher level what kinds of targets are
59:46
to be attacked. How far are we comfortable
59:48
removing humans from these decisions. If
59:51
we had all the technology in the world, what
59:54
decisions would we want humans to make it more? And
59:56
why what decisions in the
59:58
world require uniquely human judgment and
1:00:01
why is that? And I
1:00:03
think if we can answer that question, will
1:00:05
be in a much better place to crapple with
1:00:08
the challenge of a hotness weapons going forward.
1:00:17
Conclusion, choose your planet,
1:00:20
So there you haven't fully autonomous
1:00:23
lethal weapons. They might
1:00:25
keep our soldiers safer, minimize
1:00:27
casualties, and protect civilians,
1:00:31
but delegating more decision making
1:00:33
to machines might have big risks
1:00:36
in unanticipated situations. They
1:00:39
might make bad decisions that could spiral
1:00:41
out of control with no Stanislav
1:00:44
Petrov in the loop. They might
1:00:46
even lead to flash wars. The
1:00:49
technology might also fall into
1:00:51
the hands of dictators and terrorists,
1:00:54
and it might change us as well
1:00:56
by increasing the moral buffer between
1:00:59
us and our actions. But
1:01:01
as war gets faster and more complex,
1:01:04
will it really be practical to keep
1:01:06
humans involved in decisions? Is
1:01:09
it time to draw a line? Should
1:01:11
we press for an international treaty to
1:01:14
completely ban what some call
1:01:16
killer robots? What about a
1:01:18
limited ban or just a national
1:01:21
ban in the US? Or
1:01:23
would all this be naive? Would
1:01:25
nations ever believe each other's promises.
1:01:28
It's hard to know, but the right
1:01:31
time to decide about fully autonomous lethal
1:01:33
weapons is probably now, before
1:01:35
we've gone too far down the path. The
1:01:39
question is what can you do
1:01:42
a lot? It turns out you
1:01:44
don't have to be an expert, and you don't
1:01:46
have to do it alone. When enough
1:01:48
people get engaged, we make wise
1:01:51
choices. Invite friends over
1:01:53
virtually for now in person
1:01:55
what it's safe for dinner and debate
1:01:58
about what we should do. Or
1:02:00
organize a conversation for a book
1:02:02
club or a faith group or a campus
1:02:04
event. Talk to people
1:02:06
with firsthand experience, those
1:02:08
who have served in the military or been
1:02:10
refugees from war. And don't
1:02:13
forget to email your elected representatives
1:02:15
to ask what they think. That's
1:02:17
how questions get on the national
1:02:20
radar. You can find lots
1:02:22
of resources and ideas at our website
1:02:24
Brave New Planet dot org. It's
1:02:27
time to choose our planet. The
1:02:30
future is up to us. ED
1:02:40
don't want a truly autonomous car.
1:02:42
I don't want to come to garage and the concess.
1:02:45
I've fallen in love with the motorcycle and I
1:02:47
won't drive you today because I'm autonomous.
1:02:55
Brave New Planet is a co production of the Broad
1:02:57
Institute of MT and Harvard Pushkin Industries
1:02:59
in the Boston Globe, with support
1:03:02
from the Alfred P. Sloan Foundation. Our
1:03:04
show is produced by Rebecca Lee Douglas
1:03:06
with Mary Doo theme song
1:03:09
composed by Ned Porter Mastering
1:03:11
and sound designed by James Garver, fact
1:03:14
checking by Joseph Fridman and a stitt
1:03:16
An enchant. Special
1:03:19
thanks to Christine Heenan and Rachel Roberts
1:03:21
at Clarendon Communications, to
1:03:23
Lee McGuire, Kristen Zarelli and Justine
1:03:25
Levin Allerhans at the Broad, to Milobelle
1:03:28
and Heather Faine at Pushkin, and
1:03:31
to Eliah Edie Brode who made
1:03:33
the Broad Institute possible. This
1:03:36
is brave new planet. I'm Eric
1:03:38
Lander.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More