0:04

Welcome to tech Stuff, a production of

0:06

I Heart Radios How Stuff Works. Hey

0:12

there, and welcome to tech Stuff. I'm your host,

0:14

Jonathan Strickland. I'm an executive producer with I

0:16

Heart Radio and I love all things tech. And

0:18

if you haven't noticed already, yeah, my voice

0:20

is all sorts of jacked up because I'm getting

0:23

over a cold. I apologize for that, but

0:26

the tech must go

0:28

on now. Just a couple

0:30

of years ago, the tech world

0:32

in general was pretty optimistic

0:35

about autonomous cars, and I include

0:37

myself in that group. I remember seeing

0:40

the remarkable progress that had

0:42

come out from the first DARPA

0:44

Grand Challenge up to about I

0:46

don't know ten

0:49

or so, and and it seemed like we were just on the

0:51

verge of having fleets of

0:53

robo taxis at our back and call. But

0:56

now we've gone on for several more

0:58

years and we're still at a point are only a

1:01

handful of companies are conducting limited

1:03

tests. Plus there

1:05

have been some high profile cases

1:07

of accidents involving vehicles operating

1:09

under autonomous or semi autonomous

1:12

modes that ended in tragedy.

1:14

So in this episode, we're going to take a look

1:17

at autonomous cars and where we stand

1:19

today. Now, let's start with I

1:21

think it helps if we run through the levels

1:24

of autonomy, and not everyone uses

1:27

these levels to talk about autonomy, and

1:29

to be honest, the barriers between levels

1:31

are a bit fuzzy, and sometimes we're not

1:33

really able to say where

1:36

we are at as far as

1:38

levels of autonomy. We can look back

1:40

at previous developments

1:42

and say, all right, well, judging on where we are

1:45

now, we'd say that this falls into level

1:47

two or level three, But it can be a

1:49

little difficult to see what level

1:51

we are currently in without

1:54

you know, truly remarkable

1:56

evidence. But in general, this

1:58

is a useful way to talk about how far along

2:01

we are uh as

2:03

far as getting fully autonomous cars.

2:06

So technically the levels

2:08

range from zero to five, so

2:10

that means there are really six levels.

2:13

However, level zero really means

2:15

there is no autonomy at all. So

2:17

with that type of vehicle, the human driver

2:19

is responsible for all operations

2:22

of the vehicle. Every driving

2:24

task is handled by the driver

2:26

alone. So some folks will say

2:29

that they're really just five levels of autonomy.

2:31

Zero would refer to vehicles that really,

2:34

honestly, they don't exist that

2:36

much anymore. Now you might be thinking

2:39

but hey, Mr smarty Pants

2:41

podcast person. I drive a car and

2:43

it doesn't have any autonomous vehicle features,

2:46

but depending upon whom you ask, features

2:48

like power steering or anti lock

2:50

breaks or cruise control and

2:53

other pretty common features

2:55

fall into the low level autonomous

2:58

range. It doesn't mean your car is at on this, but

3:00

it has some of the components

3:03

that are identified with this

3:06

concept of autonomy. So most

3:08

cars today are actually above

3:11

level zero if we go by that definition

3:13

there level one are higher. So

3:16

level one autonomy would apply to cars

3:19

where the driver still controls the vehicle.

3:21

The vehicle is still under driver control, but the

3:23

car has some driver assistance

3:25

features like power steering or antilock

3:28

brakes. The car might have what is called

3:30

an advanced driver Assistance

3:32

system or a d a S, and

3:35

the word advanced makes it sound a bit

3:37

fancier than it is at this particular

3:40

level of autonomy. The car

3:42

might have systems that help people steer,

3:44

or it might have systems that help accelerate

3:47

and or break, but the steering

3:50

and accelerating or steering and breaking can't

3:52

happen simultaneously. Either

3:54

one or the other can be taken

3:56

over by these systems, but not both.

3:58

At the same time, not with level one.

4:01

If we get up to level to autonomy, then

4:03

we're talking about partial automation.

4:06

The A d A s on these cars can do

4:08

stuff like control steering and breaking,

4:10

or steering and accelerating at the same

4:12

time, at least under certain circumstances.

4:15

But even in those cases, the car's driver

4:17

still remains primarily in control

4:20

of the vehicle. With this level of autonomy,

4:22

a driver would still not remove their hands from

4:24

the wheel, as the car would need the

4:27

humans participation to you know, works

4:29

safely. So with level to autonomy, you

4:31

still have to have your attention on the road,

4:33

you still have to have your hands on the wheel. It's just that

4:35

the car can occasionally kick in

4:38

and assist in various scenarios,

4:40

typically in very restricted cases.

4:43

Now at level three autonomy,

4:45

we're getting up to conditional automation.

4:48

These cars would still require a human driver,

4:51

but there can be times when the car systems

4:53

can operate the vehicle on its own and

4:55

the driver is essentially a passenger.

4:58

During those moments. The driver is still

5:00

supposed to monitor the environment. They're still supposed

5:02

to be prepared to take over the car should the vehicle

5:05

indicate it needs to hand over control

5:07

to the person behind the wheel. So ideally

5:09

there would be a system where the car would

5:12

identify a situation in which the driver

5:14

needs to take over, and then well

5:16

in advance of that situation becoming

5:18

imminent, it would alert the driver

5:21

to take over control of the car. Uh.

5:23

This is trickier, right. This is

5:25

harder to do than it than to

5:28

say, because the car would have to know far

5:30

enough in advance to be able to send

5:32

that alert to the driver, and the driver would

5:34

have to be able to respond to that. And while

5:37

we feel like our response time is

5:39

really fast, in computational

5:42

terms, we are snails.

5:45

We move super slow. So

5:47

this is actually pretty tricky, especially if you're talking

5:49

about a dynamic situation

5:52

where things are changing very rapidly. At

5:54

level three, autonomous cars are supposed to uh

5:57

do this seamlessly, and as I said, that

6:00

is a pretty tricky thing to do technically.

6:03

Most vehicle systems were looking

6:05

at now, especially the ones like Tesla's autopilot,

6:08

falls somewhere in level three. Level

6:11

four autonomy is at a point where

6:13

a vehicle can automatically operate itself

6:15

at least under certain conditions, but not

6:17

necessarily all driving conditions,

6:20

the vehicle would likely include

6:22

the option for a human driver to

6:24

take over operations, but under

6:26

normal, you know, conditions

6:28

that the car would pretty much drive itself. So

6:31

with level for autonomy, you would have self

6:33

driving cars that could

6:35

act as a self driving car for most

6:38

of the time, but also allow

6:40

a human driver to take over if the human

6:42

driver wanted to. UM.

6:44

Level five autonomy is a fully

6:46

autonomous car. The car can operate itself

6:49

under all driving conditions,

6:51

So any condition where a human would be driving a

6:53

car, a level five autonomous

6:55

car should be able to operate in that same

6:57

situation. There may not be any

7:00

steering wheel or any controls

7:02

at all in a vehicle, meaning there's

7:04

no option for a human driver to take

7:06

over. Now that's not our prerequisite.

7:09

You can have a level five

7:11

fully autonomous car that would still

7:13

have controls and still allow humans

7:15

to take over manually if

7:18

they if they chose to do so.

7:20

It's just that it's an option, it's

7:22

not. It's no longer a mandatory thing to

7:24

have those human based controls

7:27

with a level five autonomous car. UM

7:29

So we don't have any of these

7:31

yet, So really talking about this

7:34

is purely in the hypothetical. Arguably,

7:36

we have some that are in the level

7:39

four range, but there will

7:41

get to that they're under very strict

7:43

parameters, all right, So most

7:45

experts agree that the versions

7:47

of autonomous cars we've seen so far are

7:50

mainly in the level three and level four categories,

7:53

uh, creeping more toward a

7:55

firm level four. We're kind of in the early

7:57

stages of that, and there's several tests

8:00

programs that are operating almost

8:02

as if we're at level five. But

8:04

there's disagreement about whether or not technology

8:07

is really sophisticate enough to warrant us

8:09

calling any existing vehicle a

8:11

level four or level five autonomous car.

8:14

And so, while we have

8:16

some examples of cars and I'll talk about a

8:19

couple of them that lack control

8:21

systems for human drivers, they

8:23

are almost all prototypes and concept

8:25

vehicles or in very limited testing

8:27

situations. Uh. And so

8:30

therefore they don't really rank as level five

8:32

autonomous cars because while they lack the

8:34

controls, they cannot operate in

8:36

every situation an environment

8:38

that humans drive in. So

8:41

it's it's too early for us to talk

8:43

about deploying cars that have no way to hand

8:45

over control to human driver in

8:47

all regions and in all

8:49

you know, driving situations. Okay, so

8:52

let's do a very quick rundown on

8:54

the history of autonomous cars up to say, like

8:57

or so, and to see why sub folks like like

9:00

yours truly we're so bullish on

9:02

the future of autonomous cars.

9:04

So the history stretches back a good

9:06

long ways, particularly if we're looking at

9:08

stuff like power steering. But that's

9:11

getting way too granular. I'm not going to do

9:13

that. And the history is also really

9:15

complex, and that involves lots of different

9:18

disciplines converging into

9:20

the autonomous car form factor.

9:23

You have stuff like robotics, you know, sensored

9:25

development, artificial intelligence,

9:27

computational processing, power, range

9:30

finding technology, lots

9:32

of things that all have to come together. And

9:35

to really dive into the complex

9:37

history of all the technologies that are

9:39

coming together to make autonomous cars possible

9:42

would require a whole mini series of

9:44

episodes. So we're not going to jump

9:46

into all of that in this one. Instead,

9:48

I want to focus on things like the

9:50

DARPA challenges that were created

9:52

in the mid two thousand's. The first one

9:54

was in two thousand four and DARPA,

9:57

as you'll recall, is the research and Development

9:59

Arm of the depart I'm in the defense in the United States,

10:01

so it's technically an organization

10:04

that funds various other

10:07

groups to do R and D in

10:09

technologies that ultimately stand

10:12

to benefit the defense of the United

10:14

States. So while there are

10:16

other uses for those technologies

10:19

that don't directly relate to defense

10:22

or military systems, that's

10:24

the primary purpose for DARBA. So in

10:27

two thousand four they created this

10:29

this challenge. They called for teams

10:31

to build or convert vehicles

10:33

into autonomous cars

10:36

that were capable of navigating a long

10:38

distance desert course is

10:40

more than a hundred miles long, and there

10:42

it needs to be no human operators, so it

10:44

could not be remotely controlled, nor would

10:46

there be a driver in the vehicle. And the

10:49

idea was design a car that would

10:51

be capable of traveling a

10:53

predesignated route from

10:55

beginning to end. For that two thousand

10:57

four challenge, no team was able to complete

11:00

eat the challenge. Uh cars

11:02

failed. Some of them went off road

11:04

and got stuck, some of them just got

11:06

confused and stopped. So

11:09

no one completed it within the

11:11

time frame that DARPA had set.

11:13

But it's set the stage for subsequent

11:15

competitions. In two thousand five, DARPA

11:18

held another Grand Challenge again with

11:20

the desert course. This one was a hundred thirty

11:22

two miles long and this time

11:24

five teams were able to complete the route

11:27

and the winning team was from Stanford

11:29

Race. Stanford University was the Stanford

11:31

Racing team. They clocked the shortest

11:33

time on the course. By shortest

11:35

time, I'm still talking about a long

11:38

time. It took them six

11:40

hours fifty three minutes to

11:42

make the one thirty two mile

11:45

journey. That would mean that the average speed

11:47

for the vehicle when taken across the whole

11:49

course was somewhere around eighteen miles

11:51

per hour or approximately twenty nine kilometers

11:54

per hour, which isn't exactly tearing up

11:56

the track, but it was still a very impressive

11:58

achievement. I don't want to take away from what

12:00

they achieved. It was incredible,

12:03

especially for the time, but it's

12:06

not the sort of speed you would look at and

12:08

think, oh, well, this is the replacement for the modern

12:10

car. The next challenge

12:13

would happen in two thousand seven, and it switched

12:15

things up by requiring teams to design a

12:17

car capable of navigating through a simulated

12:20

urban environment complete with

12:22

traffic and traffic laws like

12:24

you know, traffic lights and stop signs and

12:27

simulated pedestrians. It

12:29

wouldn't be enough to design a car that

12:31

could detect a road and follow it,

12:33

or even a car capable of managing stuff

12:35

like how to how

12:38

to send torque two different

12:40

wheels. In order to get out of a tricky situation,

12:42

the cars would need advanced collision

12:45

detection and decision making capabilities.

12:47

They have to obey traffic laws, they'd

12:50

have to be able to adapt to potentially changing

12:52

situations, the kind of stuff you might find if

12:55

you're driving around a city.

12:57

So in that case, six

12:59

teams were able to finish

13:01

the course, Stanford Racing would actually

13:04

take second place that time. They

13:06

clocked in at just under four and a

13:08

half hours. First place went

13:10

to a group called Tartan Racing from

13:12

Carnegie Melon University and

13:14

they finished in four hours ten

13:16

minutes. Now, the purpose of these competitions

13:19

wasn't just to find out which groups

13:21

of smarty pants engineers were

13:23

able to build the best car. It

13:25

was an attempt to kick start serious development

13:28

in the various fields related to

13:31

making autonomous cars a possibility.

13:34

Engineers worked on all sorts of different

13:36

designs. Some incorporated lots

13:38

of optical cameras. Some used

13:41

lie dar, which is a type of laser based

13:43

range finding technology similar to radar.

13:46

So it works by zapping out a laser

13:48

and then detecting any reflections coming

13:50

back from that laser light. It uses

13:53

an array of sensors looking

13:55

for any evidence of that laser light

13:57

coming back to the sensor, and then

14:00

measures the time difference between when the laser

14:02

went out and when it picked

14:05

up the reflection, and

14:07

then, working with some math, it can

14:09

figure out how far away an

14:11

obstacle is from the vehicle. Not only

14:13

that, it can also figure out whether or not that obstacle

14:16

is moving, or if it's stationary, or if it's

14:18

moving away from or toward the vehicle.

14:20

It can figure out all that. Uh, and

14:23

I've talked about that in past episode, so I

14:25

won't get into the whull technical details here,

14:27

but it was one of those key components

14:30

that's used in some, but not all

14:33

vehicles that are following under this

14:35

autonomous card development. It's interesting

14:37

because there are lots of different companies that are

14:40

working on autonomous cars. They are

14:42

not all relying on exactly the

14:44

same technologies to achieve

14:46

that goal. Some of them are much more heavily

14:48

focused on optical cameras. Some

14:51

of them are more focused on things like lidar

14:53

and other sensors. Some of them

14:55

involve a whole, you know, slew

14:58

of different technologies that are meant

15:00

to be both uh, you know, primary

15:02

systems and redundant systems. So it's really

15:04

interesting and it was. It was

15:06

really impressive to see these teams complete

15:08

the Urban Challenge, but again, it didn't immediately

15:11

make everyone think driverless cars would be available

15:13

right away. The challenges,

15:16

while impressive, didn't

15:18

compare to what the average human driver

15:21

deals with on a on a regular day. The

15:23

competition times were pretty long. The

15:25

average speeds were all below fifteen

15:28

miles per hour, so they're all below twenty

15:30

four kilometers per hour. At that speed,

15:32

it was clear that these vehicles were just the earliest

15:35

incarnations of technologies

15:37

that would power autonomous cars in

15:39

the future. So they were airing on the

15:41

side of caution, which frankly, you

15:44

want in the first place. You don't want to see

15:46

a lot of people say let's take some

15:48

chances, when when you're talking about

15:50

vehicles, I mean, they're human lives at stake.

15:53

Meanwhile, another narrative

15:55

drives home pun intended why

15:58

a lot of folks got really hyped up about

16:00

autonomous cars. It's also a

16:02

sobering line of thought.

16:05

And I'm talking, of course, about the

16:07

frequency of fatal car accidents

16:10

and how many of them can be traced

16:12

back to human error. Now,

16:14

getting global statistics is pretty

16:17

tough on this, so I'm going to focus on the United

16:19

States because we have a lot of organizations

16:22

in the US that track these kinds

16:24

of numbers, and you can kind of get

16:26

an idea of how big the problem

16:28

is. So in tween, the

16:30

National Safety Council released a report

16:33

that stated and estimated forty

16:35

thousand people had died in

16:37

car accidents in the United States. Uh,

16:40

that actually amounted to a decline of

16:42

one percent from two thousand seventeen.

16:45

That was when forty one

16:48

people died. Another four and a

16:50

half million people on top of that had

16:52

become seriously injured in car

16:54

crashes in Meanwhile,

16:58

the National Highway Traffics say D administration

17:01

in the United States said that of

17:04

serious car crashes result

17:07

due to human error or dangerous

17:09

choices. So, in other words, mechanical

17:12

failures only contribute a

17:14

very small percentage to the overall numbers.

17:16

When it comes to serious car accidents,

17:19

most serious car accidents aren't caused

17:21

by a tire blowout or

17:24

you know, a car failing in

17:27

some way. They're caused by humans

17:29

doing something wrong, whether

17:32

it's totally by accident or

17:34

someone just makes a really bad decision, like

17:36

they think, oh, sure there's no there's

17:38

no dashed line here, but I'm gonna go ahead

17:40

and try and pass this person on this windy

17:43

rural road because I bet nobody's

17:46

coming the other way. That's what we would call

17:48

a bad decision, So says

17:51

the tech optimist. If you

17:53

could create autonomous cars

17:55

that operates safely, you could eliminate

17:58

the vast majority of car crashes

18:01

and thus fatalities on the

18:03

road. You just remove the human error

18:05

element, and suddenly you're talking

18:07

about a staggering result,

18:10

and that is an incredibly powerful

18:13

motivator. Tens

18:15

of thousands of people wouldn't

18:18

die each year from these car

18:20

accidents. Millions more would

18:22

never be injured or affected by

18:24

the tragic loss of a loved one from

18:27

an accident. Then you start

18:29

moving outward, you go out another

18:31

circle. You think of this as a ripple effect and

18:33

you think, imagine all the contributions

18:36

those people might make in the future

18:38

that they'll get a chance to make because

18:41

they wouldn't have had this terrible car

18:43

crash. These are things we

18:45

never would see come to fruition

18:47

if they were to get in a fatal car crash,

18:50

and it becomes this butterfly effect issue.

18:52

And of course, we want to make

18:55

the roads safer for everyone. Now,

18:58

I'm sure all of you have already hit upon

19:00

the major issue here. The whole

19:03

concept of people being safer in

19:05

autonomous cars is contingent

19:07

upon those autonomous cars performing

19:10

better than humans already do and

19:13

in every type of situation in which

19:15

humans find themselves driving in. If

19:17

we can't get that right, then

19:20

we haven't made things safer at

19:22

all. All we would have done is

19:24

shifted the cause of the accidents

19:27

from human error to machine

19:29

error or computer error. So

19:31

we must be absolutely certain that the

19:34

vehicles we make meet a very

19:36

high standard if our goal

19:39

is to reduce car accidents.

19:41

So we have to prove that

19:44

these machines operate better than people

19:46

do in all the different situations

19:48

people find themselves driving in before

19:50

we can make any sort of declarative statement

19:53

of this is the best way forward.

19:55

Now, when we come back, I'll talk about why

19:57

this gets super tricky, and

20:00

talk about thought experiments

20:02

and things, and and also some

20:04

real world scenarios that kind of illustrate

20:07

why this is harder than what it

20:09

sounds. But first, let's take a

20:11

quick break. Before

20:20

the break, I positive that a future

20:22

with autonomous cars that all

20:24

but eliminate fatal car crashes hinge

20:27

upon building driverless

20:29

vehicles that are much better at driving cars

20:31

than humans are in all situations.

20:33

Now, we could get a bit more lucy

20:36

goosey here, but doing so brings

20:38

up some tough ethical issues. So, for

20:41

example, what if we knew that

20:44

machines were better. Right, autonomous cars

20:46

are better than human drivers, but they

20:49

are by no means perfect. So

20:51

what if we could be certain that autonomous

20:54

cars, if widely adopted, would

20:56

reduce those fatalities by half,

20:58

for example, but they would still be

21:00

at fault in the case of the other

21:03

half of fatal car accidents.

21:05

So let's say it's you know, I don't know, and

21:09

we have level four

21:11

autonomous cars that are pretty

21:14

reliably level four and

21:17

they are better as a whole than human

21:19

drivers are. So we've seen a vast

21:21

reduction in vehicles operated

21:23

by humans. And let's even assume that most

21:25

cars are now controlled by computers,

21:28

but let's also assume they're not perfect

21:30

now. Using the numbers from

21:33

if humans were still in control, we would

21:35

expect to see another forty thousand fatalities

21:38

due to human error. And I'm just using

21:40

that number as an example. I realized that in reality

21:43

we'd be talking about nine of

21:45

forty thousand. But now that cars

21:48

are in control, it means half

21:50

of those accidents are totally prevented.

21:52

But we still see fatal accidents that

21:55

claim twenty thousand lives.

21:57

On the one hand, we could look at

21:59

that scenari are you and say, based upon

22:01

what we know from past experience,

22:04

we would have seen many more people die in

22:06

accidents if humans were actually still operating

22:09

cars. But on the other hand, that's

22:11

all hypothetical, right, I mean, we can

22:13

only know anything based on

22:15

what actually happened, not on

22:17

what might have happened if things

22:19

had gone a different way. We can't be

22:22

sure. But more than that, though,

22:24

we're still talking about twenty thousand

22:26

people losing their lives and all

22:28

the ripple effects that that makes throughout

22:31

society, and moreover, we

22:33

have machines that are at the

22:35

fault for those twenty thousand lives

22:37

being lost, and the idea that people have

22:40

built machines that, through a failure

22:42

of some sort or another, resulted

22:44

in deaths is a very difficult

22:47

proposition to accept. Also,

22:49

it's just a key to think of in

22:52

terms like that. I mean, clearly, one

22:54

death is too many. We don't

22:56

want to see anyone die in a car

22:59

accident. Having a discussion

23:01

in which you compare a fewer number

23:03

of deaths and referring to it as quote

23:05

unquote better is something that's

23:07

pretty hard for us to process. It's

23:10

easier to do it the other way, right, I mean, it's

23:12

obvious that forty thousand

23:14

people dying is worse than twenty

23:17

thou people dying, But it's hard

23:19

to view it the other way because

23:21

anyone dying at all is awful.

23:24

Now. Part of this also really boils down

23:27

to a fear of handing over

23:29

control to a machine. I know a

23:31

lot of people bulk at that idea.

23:34

They don't like the idea of not being the

23:36

actual entity making decisions

23:38

behind the wheel. Confronting them

23:41

with statistics showing how human error

23:43

leads to catastrophe, doesn't

23:45

tend to sway them. I mean a

23:47

lot of people think, well, yeah, that's

23:49

other people. I am

23:52

not that person. Also, to be fair,

23:55

we don't have the evidence to show that computers

23:57

would necessarily be better, so

23:59

they're something to that right now.

24:02

Okay, let's let's get back to where we were in our

24:04

history. The Grand Challenges helped set

24:06

the stage for the next phase of development,

24:09

which was mostly the realm of startups

24:11

and some big companies, namely Google,

24:14

would hire participants from the Grand

24:17

and Urban Challenges to come and work

24:19

in new divisions dedicated to creating driverless

24:22

cars. The early pioneering work

24:24

was now shifting pun intended

24:26

into a phase of rapid iteration,

24:29

as engineers and computer scientists

24:31

and mechanics began to refine technologies

24:34

to help make them better. So going

24:37

from the first sort of proof of concept

24:39

approach to how do we make

24:42

this a better design so it does

24:44

the thing it does but more effectively.

24:47

Google's program began in earnest around

24:49

two thousand nine, not long after

24:51

the Urban Challenge. In twenty ten,

24:54

publications began to report on

24:56

the project. So it's been secret for about

24:58

a year, maybe almost two years. Google

25:01

had been testing vehicles in and around

25:03

the Mountain View, California, headquarters

25:05

for the company. And while the vehicles still had

25:07

manual controls and they still had a driver

25:10

behind the wheel, there were at least some

25:12

segments of some of these test drives

25:14

that felt totally under the control of the vehicle

25:17

itself. It was ranking up miles

25:19

of autonomous driving experience. It was gathering

25:21

data, and those people who are working in the

25:23

division use that data to further

25:25

refine their approach. By the

25:28

company had logged more than one hundred forty

25:31

thousand miles driven by autonomous

25:33

vehicles, which equals out to around two

25:36

five thousand kilometers. And that's

25:39

pretty, you know, respectable

25:41

distance. But let's compare that against

25:43

the miles that were driven by human drivers

25:46

in the United States. So in US

25:49

drivers accumulated nearly

25:51

three trillion miles

25:55

trillion. So that means

25:57

if you were to do a percentage and you were to say

26:00

how many how much percentage of miles did

26:02

Google cars drive compared to human

26:05

drivers in the US, in the

26:08

Google cars would account for about point

26:10

zero zero zero

26:13

zero zero four seven

26:15

percent of all miles traveled

26:17

vehicle miles. So not

26:20

you you can call it a fraction of a percent, But

26:23

even that is being generous. It's a fraction of a

26:25

fraction of a fraction of a percent. Now,

26:28

if you're familiar with the

26:30

idea of things like conducting surveys, you

26:32

know that sample size is really important.

26:35

Right, If you ask five people

26:37

a question, extrapolating those

26:39

five answers to try and apply it to

26:41

the population at large is a bad

26:43

idea. It's not a good sample

26:45

size. You don't have enough data to draw any

26:47

conclusions. It's definitely

26:49

bad science. So it makes little

26:52

sense to compare the results of autonomous

26:54

vehicles that haven't even come close

26:57

to accumulating a percentage

26:59

of the my was driven by the population at

27:01

large. You cannot compare the two because

27:05

the experience is so monumentally

27:07

different. Now, for several years,

27:10

Google's cars operated without

27:12

any accidents, at least not any

27:14

that were the fault of the driverless car

27:16

itself. There were a few

27:18

incidents, but they either happened

27:21

when the safety driver was operating

27:23

the car, so a human driver was driving

27:25

the Google car not autonomous vehicle

27:28

mode, or there

27:30

there was the fault of some other driver. Right,

27:32

someone in a totally different car got

27:35

into an accident with a Google car, and it wasn't the

27:37

fault of the autonomous system, but rather the

27:39

other driver. Those were really the only

27:41

two kind of categories of incidents that

27:43

happened in the early days of Google's

27:45

testing. So at first glance, it

27:48

looked like the driverless cars were truly

27:50

safer than a human operated vehicle. Right,

27:52

They had a much better record than human drivers

27:54

did, and it may very

27:57

well be the case that they were in

27:59

fact much much safer than human

28:01

drivers. But we have to go back to

28:03

the sense of scale here. So

28:06

in the United States, drivers travel

28:08

more than three trillion miles by vehicle

28:11

per year. I think the most recent one

28:13

was almost three point three trillion. We're

28:15

getting ridiculously high

28:17

in numbers, and there are around

28:20

forty thousand fatalities per year.

28:22

And for the sake of this example, will assume

28:25

all of those fatalities were caused by human error

28:27

or bad decisions, just to simplify

28:29

things. So if we do some rough math, we'll

28:32

see that that amounts to one death per

28:34

seventy five million miles driven. Now

28:37

that's my estimate just based on back

28:39

of the napkin. The actual estimates even

28:41

more generous than that. The National Safety Council

28:44

estimates that there's one point to five

28:47

deaths per one hundred million

28:49

vehicle miles driven. So

28:51

what does that mean for autonomous vehicles,

28:53

Well, they haven't driven close to a

28:55

hundred million vehicle miles. It

28:58

means those early days when we first learned

29:00

that Google had launched its project, there were so

29:02

few miles accumulated that you

29:04

can't draw any meaningful conclusions.

29:06

Now. To be fair, I don't think many people

29:09

were trying to argue that autonomous

29:11

car technology as it was in two

29:13

ten, was already clearly superior to human

29:16

driving. This was still an early

29:18

testing phase. This was a point where it wasn't

29:20

about showing that the technology was already

29:22

better than humans. It was rather showing,

29:25

hey, we've created technology that will allow

29:27

this card to navigate and maneuver

29:30

through human environments without

29:33

making it a problem. So it wasn't even

29:35

that our our standard is higher

29:37

than human capability. It's more

29:39

like, can this machine operate

29:41

at the same level as humans within

29:44

certain parameters pretty

29:46

restrictive parameters. Skip ahead

29:48

a few years, several companies invested

29:51

in driverless car technologies that

29:53

included big car companies, you know

29:55

Toyota and Chrysler and others GM

29:57

they've all invested huge amounts of money

30:00

in autonomous car research and development.

30:02

UH. It also included startup companies independent

30:05

startups that either we're working on components

30:07

for autonomous cars like light our

30:10

systems, or they were attempting to

30:12

convert or build fully autonomous

30:14

vehicles themselves. And then there

30:16

were ride hailing companies, most notably

30:19

Uber, that we're also investing

30:22

billions of dollars in this technology

30:24

with an eye on replacing

30:26

the fleets of human operated vehicles

30:29

that were the bread and butter of their company

30:32

to uh turn them all

30:34

over to robotaxis. So instead

30:36

of having human drivers over at Uber, you

30:38

know, Uber at the highest level wants

30:41

to replace them with autonomous

30:43

vehicles for reasons that

30:45

are complex but mostly come down to money.

30:48

So meanwhile, consumer vehicles

30:50

were getting more and more sophisticated,

30:53

and higher end vehicles started sporting

30:55

some really nifty features that

30:58

relate to autonomous cars are

31:00

semi autonomous in themselves. Some

31:02

of them are more modest, like lane assist

31:05

or breaking assist safety features.

31:08

Some are a little more spectacular,

31:11

like the self parking capabilities that some

31:13

cars have where they can park

31:15

themselves and and pull out of parking

31:18

spaces all by themselves, like that's

31:20

pretty cool. They weren't intended

31:22

to make consumer cars autonomous,

31:25

but were rather positioned as sort of value

31:27

added options for cars, like this

31:29

is something nifty this car has, other

31:31

cars don't have it. Don't you want to buy this car?

31:34

And they give a hint of what might be

31:37

in days to come. In Elon

31:41

Musk started talking about an autopilot

31:43

like feature for cars, and sure

31:45

enough, the following year, Tesla

31:47

unveiled a driver assist suite

31:50

of features called autopilot.

31:52

Now, personally, and I've talked about this before,

31:55

I think naming it autopilot was the

31:57

wrong move. I feel like the word

31:59

auto pilot has a loaded

32:01

meaning to it. It conveys a sense that

32:03

the car will take care of everything

32:06

for you, and that's not necessarily

32:08

the case. In fact, that's not the case at all.

32:10

The company tried to walk that back

32:12

a bit, uh not by renaming

32:15

it, which I think they needed to do, but

32:17

they included messages, and this

32:20

they also need to do, But they included messages that said

32:22

drivers were not meant to remove

32:24

their hands from the wheel or to take

32:26

their attention away from the road, that

32:29

these systems can assist,

32:31

but they don't replace the need for a driver,

32:34

and you have to agree to that

32:36

before you can enable the

32:38

autopilot feature. So the

32:41

goal was saying, well, you have to acknowledge

32:43

the fact that no, this is not meant for it

32:45

to be an autonomous car, and not

32:47

to go off on too much of a tangent. But I

32:49

feel as though Elon Musk might be a little

32:52

too aggressive with his

32:54

projections about autonomous cars.

32:57

And I don't mean to suggest that Elon Musk and

32:59

Tesla are interchangeable. I

33:01

do see that happening a lot in techt circles,

33:03

where people will use one or the other

33:06

interchangeably, and they are two different

33:08

entities. But maybe Tesla

33:11

the company's bravado stems

33:13

from Elon Musk's own personality.

33:16

But whatever the case, autopilot has

33:18

proven to have its own limitations,

33:21

and we saw that manifest in some

33:23

rather high profile and tragic

33:25

accidents. Beginning in ten,

33:29

there have been several fatal accidents

33:31

involving Tesla vehicles operating

33:33

in autopilot mode. The first

33:35

one took place on January twenty, two

33:38

thousand sixteen, in China, and

33:40

the most recent examples I know about

33:42

took place on December two,

33:45

thousand, nineteen, and there are actually two

33:47

crashes with fatalities that day involving

33:50

Tesla vehicles reportedly engaged

33:52

in autopilot. I say reportedly, because

33:55

I don't have access to all the data, I

33:57

don't know if conclusively

34:00

they've discovered that both of these vehicles were

34:02

actually operating an autopilot mode. One

34:04

of these happening California and the other

34:06

happened in Indiana, both in the United

34:08

States on December nineteen.

34:11

Now, Tesla states that autopilot is

34:13

meant as a driver assist feature

34:16

and it's only semi autonomous. But at

34:18

the same time, Elon Musk has said repeatedly

34:20

that his goal was to get a fully autonomous

34:22

vehicle on the road by the end of

34:24

twenty nineteen, which now has been pushed

34:26

back to sometime in the first quarter of so

34:30

there are some conflicting messages coming

34:32

out. Since a fully autonomous car

34:34

and I'm talking about something that we would at least classify

34:37

as level four, if not level five, is

34:39

well beyond just a driver assist

34:42

mode. And I should also add that Tesla

34:45

drivers have a responsibility to use

34:47

these features safely and as

34:50

intended. If someone is

34:52

taking their attention off the road, or

34:54

they're sitting back from their steering wheel,

34:57

or they're taking a nap, or they're watching Netflix

34:59

or whatever, that's dangerously

35:02

irresponsible behavior, and they

35:04

are accountable for it. I don't

35:06

want to give the implication to you guys

35:08

that I think Tesla the company is fully

35:11

to blame in this case. I actually think it's

35:13

a shared responsibility, and that

35:15

you've got some drivers who are eager to

35:17

test out admittedly really cool

35:19

and technologically advanced features,

35:22

and you have a company that might message

35:24

out these features in a way that isn't

35:27

perhaps the most realistic or responsible

35:29

method. It's a really bad

35:32

combination, right. You've got people

35:34

who are tech heads who are eager

35:36

to play with the newest stuff. You've got a

35:38

company that's Bill's reputation on creating

35:41

super cool new stuff. It's

35:44

only natural that you

35:46

get when you combine those two, you can

35:48

get some bad situations if they haven't been

35:50

messaged properly. And I really feel

35:52

that Tesla bungledness that

35:55

the rollout needed to be done in such a way

35:58

where there was never the implicate atian

36:01

that this was an autonomous

36:03

mode. Uh. Saying hey it's

36:05

not autonomous after you've already called it autopilot

36:07

and put the idea in the into people's heads is

36:10

a little late in the game. So I think

36:12

that that all parties here

36:14

share accountability. It's not just

36:17

Tesla the company's fault, and it's not

36:19

entirely the driver's faults, although I

36:21

think it's more their fault than the company's. Honestly,

36:24

I mean, we're all adults, right, you should

36:26

be if you're driving a car, and if you're an adult,

36:28

you should be able to make the determination of hey,

36:30

this is a bad idea. I should

36:32

also add that Tesla is

36:34

not the only company that has had autonomous

36:37

or semi autonomous vehicles involved in

36:39

fatal accidents. There was a case in

36:41

Tempe, Arizona, involving Evolvo

36:44

that had been converted into a semi autonomous

36:47

vehicle that was being

36:49

operated under Uber and

36:51

that car hit a pedestrian while an autonomous

36:53

mode, and the pedestrian died

36:55

as a result of that accident. So

36:58

Tesla is not the only company

37:00

that has had tragedy befall

37:03

it due to you know, failures

37:05

in autonomous systems. Getting

37:07

back to the scale argument

37:10

for a second, when we're talking about autonomous

37:12

systems allegedly at fault for accidents

37:15

that lead to fewer than

37:17

a dozen deaths, you could say, like,

37:19

well, it's all tragic. You

37:21

never want to see anyone die.

37:23

One death is really too many, but still

37:26

twelve less than twelve, that's that's so much

37:28

fewer than you know, forty thousand. And

37:30

you might be tempted to say these are tragic accidents,

37:33

but if you look at how many are caused by humans, there's really

37:35

no comparison. But once again, you

37:37

have to remember that humans account for way more

37:39

vehicle miles traveled by several

37:42

orders of magnitude. So

37:44

really the only way you could compare the two is

37:47

if you had autonomous systems driving

37:49

as many miles as humans are

37:51

driving, and then you'd have to see if

37:53

they still stacked up favorably, if those numbers

37:56

were still matching up are still mismatched,

37:59

like if if a ton of this car is still accounted for, you

38:02

know, uh, significantly fewer

38:04

accidents. But we can't say that because

38:07

the autonomous cars are driving far fewer

38:09

miles than humans are. So

38:12

it is true that most accidents involving

38:14

autonomous vehicles seemed to be the fault

38:16

of human drivers. You know, it's not like

38:18

most of the accidents we hear about were caused

38:21

by the autonomous vehicles themselves. It

38:23

tends to be that someone else,

38:25

some other human, caused the accident.

38:29

But the case of these

38:31

fatalities, it does look like it was the

38:33

autonomous system at fault, and that's truly

38:36

truly concerning um. And

38:39

also, you know, when when it's when

38:41

it's a person who's at fault.

38:44

We understand that people make mistakes, and

38:47

we can feel, at least in some

38:49

cases, we can feel some sympathy for a person

38:52

where perhaps the situation was truly

38:55

out of their control, that

38:57

that situation was was partcularly

39:00

extreme or unusual,

39:03

and so we can feel so some sympathy for the

39:05

person. But when it's a machine, then

39:07

we've already surrendered control up to it,

39:10

and that's where it

39:12

gets particularly scary. You

39:14

know, we have to trust in the machine, and

39:16

when the machine betrays that trust

39:19

by failing, that's a big problem. So

39:21

what happens when there are no controls

39:23

at all? The humans can access more

39:25

on that in just a moment but first, let's take another

39:28

quick break. One

39:36

of the challenges autonomous car companies

39:38

and engineers have faced is how do you

39:40

balance between computer and manual

39:43

control of a car? You know, how

39:45

should control switch from one

39:48

to the other. When should an automated

39:50

system take over to avoid an

39:52

accident like a collision

39:54

prevention system, or when should

39:56

a driver be able to override autonomous

39:58

commands and bring the vehicle under manual

40:01

control. Doing this is not as

40:03

straightforward as you might think, and

40:05

and doing it in a way that's safe

40:08

and has a seamless transition of control

40:10

is really hard. But what

40:13

if there's no question about

40:16

it at all? Because there are no controls

40:18

to take See back in twenty four Google

40:21

showed off a driverless car prototype

40:24

that had no steering wheel, had no

40:26

accelerator, no brake pedal, so

40:28

there were no controls for a human to take

40:30

over. The car would only operate

40:32

autonomously because there were no

40:34

other options. The prototype

40:36

worked with a smartphone app and acted as

40:39

sort of a ride hailing or robo

40:41

taxi service. Users could

40:43

summon a car using the app and

40:45

they would indicate where they were wanted to go within

40:48

a very restricted range

40:50

of operation. Like it was geo

40:52

fenced, so you couldn't go beyond

40:55

a certain border that

40:57

was pretty limited, and that

41:00

meant that the vehicle had

41:02

a lot of variables reduced, right it

41:05

It cut back on the types of conditions

41:07

and routes and situations the car

41:09

might encounter, and thus

41:12

made the problems of

41:14

having an autonomous car slightly

41:17

less complicated. There's still opportunities

41:20

for complications, but you've drastically

41:23

reduced them because you've reduced the variables.

41:25

Well. The vehicle used an electric

41:27

motor that was good for about one miles

41:30

of driving per charge, and it boasted

41:33

a top speed of twenty five miles per hour. So

41:35

this little car would only really be suitable for transportation

41:38

and restricted situations such as the

41:40

campus of a big company like I don't

41:42

know, Google. It wasn't intended

41:45

as a practical vehicle for widespread adoption,

41:47

but rather another iterative step

41:50

towards fully autonomous cars. The

41:52

robotaxi vision is one that tends

41:54

to be the most common across the autonomous

41:56

car space. That's largely because the technology

41:59

used to of cars autonomy, you know, the

42:01

the sensors, computers,

42:04

robotic systems, that kind of stuff they

42:06

don't come cheap, and a vehicle would

42:08

cost significantly more than a

42:10

manually operated vehicle a traditional

42:13

car, So most experts agree

42:15

that the future of autonomous cars, at least

42:18

in the near term, will

42:20

be in fleets that are operated by companies

42:22

like Uber or Lift. They will

42:24

be ride healing vehicles or robo

42:26

taxis, and they will take passengers to

42:28

their destinations, and then those

42:31

cars will then move on to pick up their next

42:33

fair, or they'll return to some sort

42:35

of h Q for recharging or

42:37

maintenance or whatever. It's unlikely

42:39

that we're gonna see autonomous vehicles offered up

42:42

for private ownership right away for the

42:44

most part, due to the prohibitive

42:46

expense of this additional technology.

42:49

The Google's experiment pointed out both

42:51

the advances of the tech and the

42:54

limitations of autonomous car technology.

42:57

Yeah, the car had no controls, which

42:59

is what you would expect only if you had

43:01

a level five autonomous car. But

43:04

it also had very strict geo

43:06

fencing restrictions and operational

43:09

restrictions, so it couldn't go very fast,

43:11

it couldn't venture very far, it wouldn't

43:13

likely encounter unusual situations.

43:16

So because of that, it wouldn't be

43:18

Level five anyway, because you've you've

43:20

limited the scenarios

43:23

where it would be operating in the first place, it

43:26

would not be driving into all the different situations

43:28

that a human driver would encounter. A

43:31

truly autonomous vehicle would need to be

43:33

able to handle everything, all

43:35

sorts of unpredictable situations. The average

43:37

person isn't likely to encounter a truly

43:40

unusual experience on any

43:42

given drive, right, It's not

43:44

like if you drive down the road you're going to

43:46

see every single outlier.

43:49

That's very unlikely. However,

43:52

when you have a collective three trillion

43:54

vehicle miles traveled per year, you're

43:57

bound to get some pretty extreme

44:00

situations somewhere in those

44:02

three trillion miles. So you might have a

44:05

person who has to drive through a dangerous environment,

44:07

like maybe mud slides

44:09

are coming across a road, or when

44:11

people were evacuating parts of California

44:13

that were affected by wildfires, or

44:15

there might be you know, animals in

44:17

the road. There could be people in

44:19

the road. Weather effects

44:22

can be unpredictable, and they can change driving

44:24

conditions rapidly. There are all

44:26

sorts of things that humans encounter every

44:29

year, with varying degrees of

44:31

success and maneuvering around or

44:33

through them. And if we actually

44:35

do see autonomous cars take up more

44:37

of the car landscape, those autonomous

44:40

cars are also going to encounter those situations

44:42

too. It's just a matter of the odds, you

44:44

know. And there are a lot of unanswered

44:47

questions about how these cars are going

44:49

to deal with those situations when they arise,

44:52

and that includes the famous trolley problem

44:54

dilemma. Now, in the classic trolley problem,

44:57

you're presented with a hypothetical situation

45:00

in which a trolley is out of control. It's

45:02

moving down the tracks, uh and it cannot

45:04

stop. So if you do nothing,

45:07

if you do not act, the trolley

45:09

will continue down the track and it's

45:11

going to hit a group of five people. It's gonna

45:14

there's no doubt it will kill those five

45:16

people. However, there next to a lever,

45:18

and if you pull that lever, you will send

45:21

the trolley down a side track, so

45:23

it will miss the five people, but it will definitely

45:25

hit and kill one person. So

45:28

if you do nothing, five people die, But

45:30

if you act, one person dies.

45:32

So does making the choice to pull the lever

45:35

amount to murdering that one

45:37

person? Did you just choose to kill

45:39

that person. Does doing nothing mean

45:42

that you've murdered five people or does

45:44

it just mean that you allowed five people to die?

45:46

Is there any meaningful difference between those two

45:48

things. Well, these are all questions

45:50

and ethics, but with autonomous cars it

45:53

gets into less hypothetical territory. You

45:55

have to actually start to answer these questions.

45:57

Cars may very well encounter, since

45:59

you, ations in which there is no

46:02

way to avoid injuring or

46:04

killing someone. So in those

46:06

cases, what do the cars do you

46:08

know who? How do the cars choose

46:11

which person is to be put at

46:13

risk? How do they decide what

46:15

action to take? Do they try to protect the people

46:18

who are inside the car at all costs,

46:20

so in other words, yeah, we're gonna make this decision

46:23

which will protect the people who are inside the car.

46:25

By anyone else there they are

46:27

fair game. Or do they try to protect people

46:30

who are outside the car who maybe don't

46:32

have the benefit of the car's other safety

46:34

features. Maybe you build it into

46:36

an autonomous car that the people

46:39

inside the car are allowed

46:41

to encounter a bit more risk because

46:43

your thought is, well, the inside the car

46:45

is very safe, so we want to make sure we

46:47

protect say a pedestrian or bicyclist.

46:50

We don't want the car to hit them because

46:52

they will suffer way more damage than

46:55

the people inside would. So we're going

46:57

to make that decision. That's a that's a possible choice

46:59

too, But these are not necessarily answered

47:01

questions. There are questions that are being answered

47:04

as people are designing these vehicles. One

47:06

benefit that autonomous cars might

47:09

have is that organizations overseeing

47:11

them could, at least in theory, use

47:14

the collective information across an

47:16

entire fleet of autonomous cars

47:18

to improve performance of each vehicle

47:20

within that fleet. So if

47:23

one car were to encounter a really

47:25

unusual experience, engineers

47:27

could take the data from that experience

47:30

and tweak the behavior of all the cars

47:32

across the fleet. So when

47:34

one individual encounters something

47:37

new, everyone learns

47:39

from that experience. So it's sort of like

47:41

the borg in Star Trek. It's

47:44

a collective and that's a big advantage

47:46

over human beings, right because when it comes to

47:48

humans, the person who experienced

47:51

something, they might learn from

47:53

that experience, but that that

47:55

learning, that knowledge doesn't automatically

47:58

spread across the population and general

48:01

So in that way, autonomous

48:03

cars can have a big advantage over human drivers.

48:05

If that is used properly

48:08

on the flip side, when it comes to something

48:10

as potentially deadly as a vehicle,

48:13

it's pretty cavalier to say, well, the

48:15

cars will learn as they go, and we'll apply

48:17

that knowledge to all the vehicles. They'll get better

48:19

the longer they drive, because

48:22

if learning also includes accidents

48:25

that could potentially result

48:27

in injuries or fatalities, that's

48:29

a really steep price to pay for

48:32

knowledge. And we're seeing more companies

48:34

developed vehicles that have no manual

48:37

control systems. You know, Google came out with

48:39

There's in but that's

48:41

not the only case of it. In January,

48:44

g MS Autonomous Car division, which is called

48:46

Cruise. Originally it was an independent

48:48

startup, but GM gobbled them up. A couple

48:50

of years ago, they unveiled a driverless

48:52

car called Origin. And the Origin,

48:55

like Google's prototype, has no steering

48:57

wheel, has no accelerator, no brake pedal.

49:00

It has seats that all face inward.

49:02

They're kind of like, you know, think imagine two benches

49:05

with with backs, but the two benches

49:08

are facing each other, so the people

49:10

sitting in what would consider to be the front

49:12

of the vehicle would have their backs to the

49:14

windshield and they'd be looking back

49:16

at the people sitting in the back seats, who'll

49:18

be looking forward. Uh. Now,

49:21

it's about the size of a crossover suv,

49:23

and that means there's a pretty good amount of space inside

49:26

the vehicle. So while you are facing

49:28

the other folks, like if if you're in the front seat,

49:31

you're facing the folks in the back and they're facing you. Because

49:33

there's so much space, you're not likely to accidentally

49:36

kick each other or anything. It looks pretty roomy.

49:38

On top of that, the car has a cool

49:41

little keypad on the doors. And

49:43

the idea is that a production model

49:45

of this car would be used like a robotaxi.

49:47

So you would hail a ride on your smartphone

49:50

and this little robo car would come driving

49:52

up to you, and then it would give

49:55

you a multi digit pass code.

49:57

You would get one on your app and you would

49:59

look at that ask God, and you would type the numbers into

50:01

the keypad and that would open the doors.

50:04

So that way, you know, some unauthorized

50:06

person wouldn't just jump into your car and

50:08

then go gallivanting off without you.

50:11

You would be able to unlock the

50:13

car yourself because you had a one time use

50:16

key code. That's a decent concept

50:18

for a working robotaxi, but the fact remains

50:21

that we haven't hit level five autonomy

50:23

yet. At best, we have limited level

50:26

four. Most of the vehicles we've seen in testing

50:28

can perform autonomously, but only

50:30

with pretty tight restrictions like that along

50:33

specific predefined routes or

50:36

within very strict geo fencing,

50:38

or at particular times of the year or

50:40

even particular times of the day.

50:43

Again, that helps reduce the variables

50:45

that the car mine encounter on any given

50:47

day, and it gives it the best chance to operate

50:49

safely, but that really limits

50:51

how useful the cars are in practical

50:54

applications. For autonomous cars

50:56

to work as an alternative to manually controlled

50:58

vehicles, they need to brand and pretty much all

51:01

the same conditions that regular cars

51:03

do without restrictions, and we just

51:05

aren't there yet, and we might not

51:07

be for several more decades.

51:10

The Prognos Research Institute

51:12

actually identified four factors that are

51:15

in the way of autonomous vehicles. They

51:17

include technological maturity, which

51:19

is what I was just talking about. Infrastructure

51:21

development, so having you

51:24

know, cities that are designed

51:26

in such a way that they can allow for

51:28

autonomous cars the inertia

51:31

of the fleet. This means that you

51:33

know, there's a ton of manually controlled

51:35

vehicles out in the world already, right The

51:37

vast majority of cars that are out there

51:40

are manual control vehicles.

51:42

They might have some limited autonomy, but for the most

51:44

part, they're controlled by humans. It would take a

51:46

very long time before autonomous

51:48

vehicles represent a significant percentage

51:51

of the overall vehicles on the road, let

51:54

alone a majority. So it

51:56

will take many, many, many years to

51:59

wean off of a human controlled

52:01

cars and go to autonomous cars, barring

52:04

any legislation that outlaws

52:06

vehicles um or human

52:08

controlled vehicles, I guess I should say. And then finally

52:10

we have legal hurdles

52:12

to overcome the regulations that

52:15

are going to be coming out around driverless

52:17

cars. We're seeing a lot of money

52:20

poured into research and development to push

52:22

the technological limits further

52:24

and to establish the foundation for truly

52:27

autonomous vehicles. But I wonder

52:29

if these various companies and their

52:31

investors are really in it for the

52:33

long haul, so to speak, because

52:35

I suspect it's going to take a pretty long

52:38

time to get to a point where we feel there's

52:40

really reliable safe level

52:42

five autonomous vehicles in the world, let

52:45

alone a world in which governments have also

52:47

agreed and have caught up

52:49

and have defined the legal parameters

52:52

for the operation of these vehicles. Because

52:54

you know, it's one thing to prove the technology works.

52:56

That doesn't necessarily mean that technology will

52:58

be legal to to operate, right

53:01

Like, governments tend to move a lot more slowly

53:03

than technology does. So if investors

53:06

are willing to play the long game, then

53:09

I think their investments will ultimately pay

53:11

off. But it's going to take

53:13

a long time, which means

53:16

lots of repeated investments are going to be

53:18

required to keep these companies going, to keep

53:20

them innovating and improving

53:22

technologies. And meanwhile, there's

53:24

not going to be an actual market for them

53:26

to capitalize on outside

53:28

a few, you know, test programs that

53:30

don't really count, because there's

53:33

no way that the revenue they're generating

53:35

is actually eclipsing

53:37

the cost of operation. It's

53:39

got to be a money losing proposition right

53:41

now in all the different test cases at

53:44

scale, with fully

53:47

legal vehicles that are embraced

53:50

by the general public shore it could

53:52

work from a financial standpoint

53:54

right now, though it's all just proof

53:56

of concept that hasn't uh

54:00

seem full fruition. Now. I still

54:02

believe in autonomous cars. I

54:04

still believe they will ultimately make the roads

54:07

safer and reduce the number of deaths

54:09

and injuries from car accidents. I

54:11

just think it's going to take a lot longer that

54:14

I had previously imagined.

54:16

And that's not necessarily a bad

54:18

thing. This isn't important enough issue

54:21

that we have to make sure we get it right

54:23

that we can deploy vehicles in ways that

54:26

makes sense, that are truly safe,

54:28

that are ethical, that there

54:30

are in as an ideal implementation

54:33

as we can manage UH.

54:35

And we have to make sure

54:37

that it makes financial sense

54:39

too, right. We need to have h make sure

54:41

that it truly represents an affordable way

54:44

to get around that eliminates the need for stuff

54:46

like garages and parking

54:48

lots and dense urban centers. Those

54:50

areas could be reclaimed and used for

54:52

other stuff, and that stuff

54:55

might be far more productive than just being a

54:57

storage place for a car when it's not being

54:59

in use. Personal ownership could

55:01

really be on a serious decline in that kind

55:04

of future, replaced with on demand

55:06

car service, and the cars

55:08

that are in service would be used much more frequently

55:11

rather than just sitting idle and taking up space

55:13

for the vast majority of their existence. If you think about

55:15

your average car um

55:17

it's the amount of time you're

55:19

actually using it versus the amount of time

55:21

it's just sitting there doing nothing is

55:23

staggering, right. So if you're able to

55:25

make more use of the vehicle,

55:29

uh, then it's a more efficient

55:31

use of the technology. It's a it's a better investment

55:35

for all the the materials

55:38

that went into making that vehicle. So

55:40

you could argue, well, this makes more sense from

55:43

multiple perspectives if we're

55:45

able to make better use of this technology

55:47

and not just have it sitting someplace taking up

55:49

room. But it's

55:52

it's a lot a lot of things have to fall into place

55:54

for that future to come true. I

55:58

think it's a future that it

56:00

makes sense, but only if we can get the

56:02

tech just right and before then,

56:05

what we're really risking is making

56:08

bad decisions that just make

56:10

it harder to get to the

56:12

right future. So we have to be

56:14

careful in how we're testing these things.

56:16

We have to minimize risk while

56:19

maximizing our our ability

56:22

to learn things, which is a very

56:24

tricky thing to do because Ultimately, you do

56:26

have to start deploying autonomous cars

56:28

into populated centers or

56:31

else. All you've done is created something that

56:33

works really well in the lab, but

56:35

not well in the real world, and

56:37

that would be useless to us because most

56:40

of us don't live in a lab. I know

56:42

I don't, not since

56:44

two thousand fourteen, but

56:46

that's another story. Guys. If

56:49

you have any suggestions for future topics

56:51

for tech Stuff, reach out to me.

56:54

You can find me on social media.

56:56

I'm on Facebook and on Twitter with

56:58

the handle tech stuff h s

57:01

W and I'll talk to you again

57:04

really soon. Hext

57:09

Stuff is a production of I Heart Radio's How Stuff

57:11

Works. For more podcasts from I heart

57:13

Radio, visit the I heart Radio app,

57:15

Apple Podcasts, or wherever you listen

57:18

to your favorite shows.