WEBVTT
00:00:10.240 --> 00:00:12.640
Hey, welcome back to the Lars, Larson Joe.
00:00:12.800 --> 00:00:16.320
It's a pleasure to be with you and always glad to get to your phone calls and your emails.
00:00:16.480 --> 00:00:22.559
I wanted to talk to Bryant Walker Smith, who's a University of South Carolina law professor and an engineer.
00:00:22.719 --> 00:00:26.000
So imagine that, getting both a JD and an engineering degree.
00:00:26.640 --> 00:00:28.160
Professor, welcome back to the program.
00:00:28.239 --> 00:00:28.719
How are you?
00:00:29.199 --> 00:00:29.920
Thanks so much.
00:00:30.239 --> 00:00:31.280
So tell me about this.
00:00:31.440 --> 00:00:34.000
Tesla has this self-driving technology.
00:00:34.079 --> 00:00:42.079
I'm not the biggest fan of Tesla or battery-powered cars, but uh I know that uh, you know, the the self-driving technology is probably coming.
00:00:42.159 --> 00:00:43.840
There's no way to unring that bell.
00:00:44.000 --> 00:00:48.320
But they just had a record month with 207 reported crashes.
00:00:48.479 --> 00:00:51.039
When the car is driving, let's start with a simple one.
00:00:51.200 --> 00:00:52.799
Who should take the blame?
00:00:53.600 --> 00:01:01.359
Yeah, so what you're referring to is the so-called full self-driving feature that you still have to drive.
00:01:01.520 --> 00:01:01.759
Yep.
00:01:01.920 --> 00:01:13.439
Uh Tesla has a so-called FSD that you have to drive, and they have another so-called FSD that in theory you don't if you happen to be really lucky in Austin.
00:01:13.920 --> 00:01:26.640
Um and so the statistics you're referring to come from reports of crashes involving the driver assistance system that Tesla very misleadingly calls full self-driving or full self-driving supervised.
00:01:26.879 --> 00:01:33.280
And the problem with those numbers is we just don't have context because we know the number of crashes that Tesla was reported.
00:01:33.359 --> 00:01:37.680
We don't know if the reporting is consistent in other automakers that also have driver assistance features.
00:01:37.760 --> 00:01:39.519
We don't know how many miles are driven.
00:01:39.680 --> 00:01:42.959
Um, part of that is on the data collection system, which should be better.
00:01:43.120 --> 00:01:52.799
Part of that is on Tesla, which should be much more forthcoming with uh a lot of its data, including some that would give us a lot more context to understand its performance.
00:01:53.040 --> 00:01:57.519
Uh you asked what happens when there's a crash, who's who gets blamed?
00:01:57.680 --> 00:01:57.760
Yes.
00:01:58.000 --> 00:02:06.959
And this comes back to is it the full self-driving in quotations that you still drive, or is it the quote unquote full self-driving that you don't?
00:02:07.200 --> 00:02:14.800
Um and very superficially, right, if you are a driver using a driver assistance feature, you are responsible for driving safely.
00:02:15.039 --> 00:02:28.159
If you are using a vehicle with an automated driving system, like a Waymo car, for example, a Waymo Robo Taxi, um, then you are not the driver and you have the same obligations as the passenger.
00:02:28.400 --> 00:02:29.840
But we can complicate that.
00:02:30.240 --> 00:02:37.599
So if you're driving a car and suddenly it veers to the right and you can't stop it, well, you're the driver, but that's not your fault.
00:02:38.000 --> 00:02:39.680
And you can't override your right.
00:02:40.960 --> 00:02:47.360
Uh well, in Tesla's driver systems feature, um, you're supposed to be able to override.
00:02:47.439 --> 00:02:58.159
Um, but there have long been concerns that because of confusion or design, drivers aren't able to do the thing they want to do at the moment that they're panicking.
00:02:58.479 --> 00:03:13.520
Um and if you're a pedestrian who's struck or uh you know another driver who's struck because a driver is misusing a system called full self-driving, well, then you could fairly say the driver is to blame.
00:03:13.680 --> 00:03:20.560
And also a company that allowed a product to be so marketed and to be so used is also at fault.
00:03:21.439 --> 00:03:39.039
Now, does that mean that if it's the kind that and you dis distinguish between the two, if it's the kind where it's full self-driving, but you still have to drive the car, but it leads you to believe that much of the driving will be done by the machine, that sounds like it's creating a moral hazard.
00:03:39.199 --> 00:03:51.919
I think what a lawyer would call a moral hazard, where they'd say, the driver, the uh the car is going to persuade you, I'll take on most of the driving most of the time, except when we crash into something, in which case it's your fault.
00:03:52.639 --> 00:03:54.960
It even before that, it just makes me dizzy.
00:03:55.039 --> 00:03:55.919
Does it make you dizzy?
00:03:56.159 --> 00:03:58.400
A full self-driving car that you have to drive?
00:03:58.560 --> 00:04:00.400
I'm a driver using this full self-driving.
00:04:00.639 --> 00:04:01.680
Yeah, yeah.
00:04:02.000 --> 00:04:11.199
Um there are ways to design a system to make everyone safer, and there are ways to design and market a system to make people less safe.
00:04:11.360 --> 00:04:17.439
And I worry very much about the marketing design of this particular driver assistance system.
00:04:18.000 --> 00:04:22.000
Okay, so can I make it one more step complicated, Professor Smith?
00:04:22.319 --> 00:04:23.040
Let's do it.
00:04:23.360 --> 00:04:23.600
Okay.
00:04:24.240 --> 00:04:25.600
The trolley problem.
00:04:27.519 --> 00:04:28.879
Do you know about the trolley problem?
00:04:29.920 --> 00:04:30.399
I do.
00:04:31.279 --> 00:04:32.800
Because that's in law school, isn't it?
00:04:34.720 --> 00:04:42.240
Where I'm standing there and the trolley is coming down the tracks, and if I don't do anything, the trolley is going to run right through five people.
00:04:42.480 --> 00:04:45.199
Uh, we won't decide on their age or anything else like that.
00:04:45.360 --> 00:04:50.720
Or I can pull a lever and the trolley will go off on a siding and only kill one person.
00:04:50.879 --> 00:04:52.720
But somebody is going to die.
00:04:52.959 --> 00:05:25.120
And the question becomes if you're and when it comes to automatic driving cars or full full uh driving cars, do you engineer the car to put the uh uh I guess the life of the occupant as the as the greatest value, or do you put or or if the car has the choice of saving the one occupant in the self-driving car or the five children who are in the crosswalk, does the car veer off the road, kill the occupant, and save the children, or does it go through the children and save the life of the occupant?
00:05:26.240 --> 00:05:27.759
Sociologically fascinating.
00:05:27.920 --> 00:05:32.560
Now, the narrow version of that, um, how many times have you confronted that driving to work, right?
00:05:32.639 --> 00:05:35.120
How many times in your life have you said, gee, who should I kill?
00:05:35.519 --> 00:05:36.000
Zero.
00:05:36.240 --> 00:05:42.160
So the narrow, the narrow instance is is virtually a non a non-issue, right?
00:05:42.399 --> 00:05:54.000
Uh and even these vehicles are not going to have the omniscience or the omnipotence to know exactly what will happen and to do everything perfectly to release this, reach this ideal situation.
00:05:54.160 --> 00:05:59.120
But the much broader version of that question is a really important one because we make it all the time.
00:05:59.680 --> 00:06:14.879
Do I get in a really heavy vehicle and go really fast, knowing that I'm gonna be very well protected by seatbelts and airbags and cushions and steel, but the person in the smaller car or the bicycle that I hit is gonna die?
00:06:15.040 --> 00:06:18.560
Like that's a that's an issue, the choice that we do make every day.
00:06:18.720 --> 00:06:24.240
And so those kinds of risk assessments aren't just a future hypothetical.
00:06:24.319 --> 00:06:28.079
Um, they affect all kinds of design and policy questions today.
00:06:28.399 --> 00:06:34.879
Well, and and the other thing about it is that you and I might have to confront that on an occasion that we cannot foresee.
00:06:35.040 --> 00:06:40.319
But the programmer has to confront it by saying, Hey boss, how do you want me to program it?
00:06:40.480 --> 00:06:43.439
Do you want me to program it for minimal loss of life?
00:06:43.759 --> 00:06:53.439
You know, meaning if it's if it's got to be head on into the other car, which is gonna kill five people, or off the edge of the cliff, which is gonna kill the one person in the car.
00:06:54.160 --> 00:07:09.199
At some point, the programmer is gonna be told by by Tesla or somebody to program the programming to maximize uh what survivability of the person who owns the car or survivability of the other five people in the other car?
00:07:09.759 --> 00:07:16.480
That would be quite a memo to see, although it would be implicit in training data or implicit in in waiting decisions.
00:07:16.639 --> 00:07:23.199
The reality is Professor, as a lawyer, you would you would urge your client not to put that in writing, wouldn't you?
00:07:24.639 --> 00:07:28.959
I would urge a designer to take risks as they happen.
00:07:29.120 --> 00:07:39.839
You want to get energy out of the system, slow down as fast as you can, follow the road rules, and avoid harms as they arise.
00:07:40.240 --> 00:07:48.959
That's you know, that's the the best that we're gonna do in these very uncertain, unpredictable scenarios.
00:07:49.279 --> 00:07:52.800
Um Professor, I'm gonna look forward to talking to you the next time.
00:07:52.879 --> 00:07:57.360
That is Professor Brian Walker Smith, who is at the University of South Carolina.
00:07:57.519 --> 00:08:01.839
He is both an engineer and a lawyer, and we'll be back in just a moment.
00:08:01.920 --> 00:08:06.399
You're listening to the Lars Larson Show and the Radio Northwest Network.
00:00:10.240 --> 00:00:12.640
Hey, welcome back to the Lars, Larson Joe.
00:00:12.800 --> 00:00:16.320
It's a pleasure to be with you and always glad to get to your phone calls and your emails.
00:00:16.480 --> 00:00:22.559
I wanted to talk to Bryant Walker Smith, who's a University of South Carolina law professor and an engineer.
00:00:22.719 --> 00:00:26.000
So imagine that, getting both a JD and an engineering degree.
00:00:26.640 --> 00:00:28.160
Professor, welcome back to the program.
00:00:28.239 --> 00:00:28.719
How are you?
00:00:29.199 --> 00:00:29.920
Thanks so much.
00:00:30.239 --> 00:00:31.280
So tell me about this.
00:00:31.440 --> 00:00:34.000
Tesla has this self-driving technology.
00:00:34.079 --> 00:00:42.079
I'm not the biggest fan of Tesla or battery-powered cars, but uh I know that uh, you know, the the self-driving technology is probably coming.
00:00:42.159 --> 00:00:43.840
There's no way to unring that bell.
00:00:44.000 --> 00:00:48.320
But they just had a record month with 207 reported crashes.
00:00:48.479 --> 00:00:51.039
When the car is driving, let's start with a simple one.
00:00:51.200 --> 00:00:52.799
Who should take the blame?
00:00:53.600 --> 00:01:01.359
Yeah, so what you're referring to is the so-called full self-driving feature that you still have to drive.
00:01:01.520 --> 00:01:01.759
Yep.
00:01:01.920 --> 00:01:13.439
Uh Tesla has a so-called FSD that you have to drive, and they have another so-called FSD that in theory you don't if you happen to be really lucky in Austin.
00:01:13.920 --> 00:01:26.640
Um and so the statistics you're referring to come from reports of crashes involving the driver assistance system that Tesla very misleadingly calls full self-driving or full self-driving supervised.
00:01:26.879 --> 00:01:33.280
And the problem with those numbers is we just don't have context because we know the number of crashes that Tesla was reported.
00:01:33.359 --> 00:01:37.680
We don't know if the reporting is consistent in other automakers that also have driver assistance features.
00:01:37.760 --> 00:01:39.519
We don't know how many miles are driven.
00:01:39.680 --> 00:01:42.959
Um, part of that is on the data collection system, which should be better.
00:01:43.120 --> 00:01:52.799
Part of that is on Tesla, which should be much more forthcoming with uh a lot of its data, including some that would give us a lot more context to understand its performance.
00:01:53.040 --> 00:01:57.519
Uh you asked what happens when there's a crash, who's who gets blamed?
00:01:57.680 --> 00:01:57.760
Yes.
00:01:58.000 --> 00:02:06.959
And this comes back to is it the full self-driving in quotations that you still drive, or is it the quote unquote full self-driving that you don't?
00:02:07.200 --> 00:02:14.800
Um and very superficially, right, if you are a driver using a driver assistance feature, you are responsible for driving safely.
00:02:15.039 --> 00:02:28.159
If you are using a vehicle with an automated driving system, like a Waymo car, for example, a Waymo Robo Taxi, um, then you are not the driver and you have the same obligations as the passenger.
00:02:28.400 --> 00:02:29.840
But we can complicate that.
00:02:30.240 --> 00:02:37.599
So if you're driving a car and suddenly it veers to the right and you can't stop it, well, you're the driver, but that's not your fault.
00:02:38.000 --> 00:02:39.680
And you can't override your right.
00:02:40.960 --> 00:02:47.360
Uh well, in Tesla's driver systems feature, um, you're supposed to be able to override.
00:02:47.439 --> 00:02:58.159
Um, but there have long been concerns that because of confusion or design, drivers aren't able to do the thing they want to do at the moment that they're panicking.
00:02:58.479 --> 00:03:13.520
Um and if you're a pedestrian who's struck or uh you know another driver who's struck because a driver is misusing a system called full self-driving, well, then you could fairly say the driver is to blame.
00:03:13.680 --> 00:03:20.560
And also a company that allowed a product to be so marketed and to be so used is also at fault.
00:03:21.439 --> 00:03:39.039
Now, does that mean that if it's the kind that and you dis distinguish between the two, if it's the kind where it's full self-driving, but you still have to drive the car, but it leads you to believe that much of the driving will be done by the machine, that sounds like it's creating a moral hazard.
00:03:39.199 --> 00:03:51.919
I think what a lawyer would call a moral hazard, where they'd say, the driver, the uh the car is going to persuade you, I'll take on most of the driving most of the time, except when we crash into something, in which case it's your fault.
00:03:52.639 --> 00:03:54.960
It even before that, it just makes me dizzy.
00:03:55.039 --> 00:03:55.919
Does it make you dizzy?
00:03:56.159 --> 00:03:58.400
A full self-driving car that you have to drive?
00:03:58.560 --> 00:04:00.400
I'm a driver using this full self-driving.
00:04:00.639 --> 00:04:01.680
Yeah, yeah.
00:04:02.000 --> 00:04:11.199
Um there are ways to design a system to make everyone safer, and there are ways to design and market a system to make people less safe.
00:04:11.360 --> 00:04:17.439
And I worry very much about the marketing design of this particular driver assistance system.
00:04:18.000 --> 00:04:22.000
Okay, so can I make it one more step complicated, Professor Smith?
00:04:22.319 --> 00:04:23.040
Let's do it.
00:04:23.360 --> 00:04:23.600
Okay.
00:04:24.240 --> 00:04:25.600
The trolley problem.
00:04:27.519 --> 00:04:28.879
Do you know about the trolley problem?
00:04:29.920 --> 00:04:30.399
I do.
00:04:31.279 --> 00:04:32.800
Because that's in law school, isn't it?
00:04:34.720 --> 00:04:42.240
Where I'm standing there and the trolley is coming down the tracks, and if I don't do anything, the trolley is going to run right through five people.
00:04:42.480 --> 00:04:45.199
Uh, we won't decide on their age or anything else like that.
00:04:45.360 --> 00:04:50.720
Or I can pull a lever and the trolley will go off on a siding and only kill one person.
00:04:50.879 --> 00:04:52.720
But somebody is going to die.
00:04:52.959 --> 00:05:25.120
And the question becomes if you're and when it comes to automatic driving cars or full full uh driving cars, do you engineer the car to put the uh uh I guess the life of the occupant as the as the greatest value, or do you put or or if the car has the choice of saving the one occupant in the self-driving car or the five children who are in the crosswalk, does the car veer off the road, kill the occupant, and save the children, or does it go through the children and save the life of the occupant?
00:05:26.240 --> 00:05:27.759
Sociologically fascinating.
00:05:27.920 --> 00:05:32.560
Now, the narrow version of that, um, how many times have you confronted that driving to work, right?
00:05:32.639 --> 00:05:35.120
How many times in your life have you said, gee, who should I kill?
00:05:35.519 --> 00:05:36.000
Zero.
00:05:36.240 --> 00:05:42.160
So the narrow, the narrow instance is is virtually a non a non-issue, right?
00:05:42.399 --> 00:05:54.000
Uh and even these vehicles are not going to have the omniscience or the omnipotence to know exactly what will happen and to do everything perfectly to release this, reach this ideal situation.
00:05:54.160 --> 00:05:59.120
But the much broader version of that question is a really important one because we make it all the time.
00:05:59.680 --> 00:06:14.879
Do I get in a really heavy vehicle and go really fast, knowing that I'm gonna be very well protected by seatbelts and airbags and cushions and steel, but the person in the smaller car or the bicycle that I hit is gonna die?
00:06:15.040 --> 00:06:18.560
Like that's a that's an issue, the choice that we do make every day.
00:06:18.720 --> 00:06:24.240
And so those kinds of risk assessments aren't just a future hypothetical.
00:06:24.319 --> 00:06:28.079
Um, they affect all kinds of design and policy questions today.
00:06:28.399 --> 00:06:34.879
Well, and and the other thing about it is that you and I might have to confront that on an occasion that we cannot foresee.
00:06:35.040 --> 00:06:40.319
But the programmer has to confront it by saying, Hey boss, how do you want me to program it?
00:06:40.480 --> 00:06:43.439
Do you want me to program it for minimal loss of life?
00:06:43.759 --> 00:06:53.439
You know, meaning if it's if it's got to be head on into the other car, which is gonna kill five people, or off the edge of the cliff, which is gonna kill the one person in the car.
00:06:54.160 --> 00:07:09.199
At some point, the programmer is gonna be told by by Tesla or somebody to program the programming to maximize uh what survivability of the person who owns the car or survivability of the other five people in the other car?
00:07:09.759 --> 00:07:16.480
That would be quite a memo to see, although it would be implicit in training data or implicit in in waiting decisions.
00:07:16.639 --> 00:07:23.199
The reality is Professor, as a lawyer, you would you would urge your client not to put that in writing, wouldn't you?
00:07:24.639 --> 00:07:28.959
I would urge a designer to take risks as they happen.
00:07:29.120 --> 00:07:39.839
You want to get energy out of the system, slow down as fast as you can, follow the road rules, and avoid harms as they arise.
00:07:40.240 --> 00:07:48.959
That's you know, that's the the best that we're gonna do in these very uncertain, unpredictable scenarios.
00:07:49.279 --> 00:07:52.800
Um Professor, I'm gonna look forward to talking to you the next time.
00:07:52.879 --> 00:07:57.360
That is Professor Brian Walker Smith, who is at the University of South Carolina.
00:07:57.519 --> 00:08:01.839
He is both an engineer and a lawyer, and we'll be back in just a moment.
00:08:01.920 --> 00:08:06.399
You're listening to the Lars Larson Show and the Radio Northwest Network.
