Smart Driving Cars episode 399: Waymo crashes examined by Understanding AI's Kai Williams

It's the Smart Driving Cars podcast. Glad you can spend some time with us. I'm Fred Fishkin,

along with the Faculty Chair of Autonomous Vehicle Engineering at Princeton University.

Alan Kornhouser. Hi, Alan. Hey, good morning, Fred. Good morning. And we have a guest joining

us for this edition from Understanding AI, a new writer there, Kai Williams. Hi, Kai.

Hi. Good morning. It's nice to be talking with you all today.

Well, you have a great background, Kai, and are supported by a Tarbell Fellowship at Understanding

AI. That's from the Tarbell Center for AI Journalism. This past week, Waymo released new safety

statistics, and the piece you have written for Understanding AI is titled, Very Few of Waymo's

Most Serious Crashes Were Waymo's Fault. Give us the overview, I think, to start with.

You looked at, I think, something like 45 major Waymo crashes?

Yeah, so Waymo releases this new safety data, and we want to check. There's a lot of ways you

could check whether they're being realistic and saying that their cars have been safer.

But one way to check is, by law, they're required to report all of their accidents

to the National Highway Transportation Safety Administration. So every time, even if it's like

someone nudges them and barely makes contact, any crash, they're required to report.

And there's a lot of really small crashes. So we wanted to focus on the most serious.

Those that either someone reported an injury or an airbag went off. And I was looking at the

last, basically, six months from February to mid-August. And there were 45 crashes with injuries

or airbags like that in the past six months. And so I went through all of them. And I want

to caveat, I think it's important to note that we only, for most of these, have Waymo's

side of the story, because they're the ones reporting to the NISTA. But even so, there

are some crashes that clearly aren't Waymo's fault. And that's a big category. So for instance,

a lot of times, a Waymo is stopped at an intersection. And then a car from behind

rear-ends the Waymo. And even though we're only hearing Waymo's side of the story,

it's pretty clear that it was not Waymo's fault because they were stopped in an

appropriate place. And so going through most of these crashes, almost all of them had a

pretty clear story of, oh, that's not Waymo's fault. But there were a couple that are less clear.

I mean, the one you lead off with in the piece here was Waymo's fault, but you point out

had nothing to do with the technology. Yeah. So the clearest example where Waymo was at fault

was they were driving in Austin, Texas, and a wheel came off. And no one else was involved

in the crash, luckily. But the passenger reported an injury. And that is clearly

Waymo's fault because no one else could have knocked the wheel off. But that's also less

interesting because it's not really Waymo's self-driving software that it was at fault.

It was some mechanical issue. Well, Kai, what I really loved about your article and what

you wrote is that you very, very, in some sense, methodically, and I guess with complete openness,

went through and discussed the crashes. I like to call them crashes or not accidents. They're

crashes that Waymo has been involved with and tried to deliver driverless mobility.

And of course, safety is an enormous issue in this whole process. And of course, we want to be

perfectly safe. I want to be perfectly safe. I mean, my goodness, what else would we want?

But really to be perfectly safe, I guess we, as I like to say, we'd stay home all the time.

And then the house would burn down and I'd burn down and I'd die and, you know, whatever.

Geez, I shouldn't have stayed home. Shouldn't have taken that risk. We should have taken

another risk. I also like to tell my students that I guess what makes, in some sense,

life worth living is that we don't know the future and there are some risks otherwise.

Be pretty boring to just not have risks all day. We should be so lucky.

God, why didn't she make us this way? I mean, come on, you failed us. I guess it was the app,

whatever. Oh, that was another thing, the glass in trouble. Never mind. We'd digress.

But safety is a really critical issue in all of this. And there's a lot that's written and there's

a lot that's discussed and the sky's falling. Who knows what or, you know, man, there's no problem.

And what I really appreciated about what you wrote is you just said, okay, here it is.

Let me with open eyes look at it right about it. Here it is. And I certainly hope

everyone reads what you wrote. And I think I'm trying to say it in the smart driving car e-letter.

And that's what I really like about it. And that's why I'm happy that you're here with us.

And so the issue is, is that as you identified, you know, okay, there are some situations

in which it's with, you know, oh, my goodness, you've got to report every crash.

I mean, even when every crash is reported by one entity, what do you compare it to?

And if you go through the whatever government statistics or whoever

accumulates the information about what you're going to compare it to, because I think you

have to make a comparison. You can't look at it just as one particular entity.

You have to look at it safety and comparison to how we currently get around, unless you don't

want us to get around anymore, which might be a certain thing. So therefore, you know,

it really is. And unfortunately with when one entity reports everything, and you're comparing

it to other things that don't report everything for good and bad reasons, then you can't even,

you can't do the comparison. And then of course, when you do a comparison, you really want to

compare it under some of the same situation. But of course, no situation is identically the same.

I mean, mathematically, boom, I mean, even the real line, you can't ever land on the same point.

I mean, if you get really crazy about this stuff. So, you know, the comparisons end up

being really tough to make. And really everybody would like to think if we're going to discuss this,

we have to discuss it in the same context. And I think you've done a good job with that. And you

especially with what you consider to be the crash that was most, for a probability very close to

one, that it was way most fault, okay, that, you know, wheel coming off. And then all of a sudden,

you say, Oh my goodness, why is this? And I would say, Oh my goodness, why is this in our

comparison with driverless technology? Because this is not, this is not have it have nothing

to do with the driver. It's a mechanical part. And of course, you know, where I end up in my own,

like I say, you know, learned opinion is that is that if one looks at that crashes in the automobile

mobility system, an overwhelming percentage of them involve a human in the loop. And if you look at

the humans involvement in the loop, in an overwhelming percentage of those things, some of us, not

necessarily everybody, everybody doesn't agree on anything anyway, but at least most of us see

that it is the safety problem is really a human misbehavior problem. So when you look at

drinking, when you look at base in the phoning, or whatever you want to call it,

when you look at excessive speeding, when you look at road rage, when you look at,

I was looking over here and didn't see it. It's, it's the human in the loop. And I like to call

misbehavior. It's misbehavior. We're not supposed to behave that way. You go to any,

any driving school, any, anybody gives you a license to drive. They say you shouldn't be

doing that. They shouldn't be running stop signs. Shouldn't even be running yield signs.

So then the, then, then the issue when you, when you get there is that really if, if NITSA or some,

you know, public agency who's trying to really do a good job of this really should be looking at

doing the comparison as to, you know, if, if what's certainly trying to deal with safety,

who, man, how do we deal with a human in the loop? And, and, and you can deal with a human

in the loop business trying to hate behave cornhouser. I mean, I'm the worst misbehavior,

but you know, I'm sorry, you know, it's behave, behave and behave. And, you know,

I've been told to behave for, you know, I won't tell you 39 years, I'm 39 years old,

whatever. And, and so out of that process, you know, I still haven't paid attention.

Okay. Unfortunately, and I don't know what NITSA should do to get me to behave. I mean,

I wish they would, but, but I mean, if you really want to solve this safety problem,

it's the behavior. And of course, some of us think that the other way to solve the safety

problem, Waymo, including a say, take the driver out of the loop. Okay. And of course,

hopefully out of that process, make it so that the driver doesn't even want to be in the loop,

is happy to not be in the loop. So then we can start taking a look at what the heck it is.

And I think it should be a comparison of it in the safety business versus driver out of the loop

versus driver in the loop. So if you do that, then of course, what you've, what you,

your decision, not to say how you reach your decision, but this first one who

drivers not in the loop here. I mean, it's the machine, you know, something fault

mechanical thing of the machine that's not driving a loop and driver out of the loop

shouldn't be part of the comparison. Shouldn't, shouldn't even, shouldn't even be mentioned.

I think it's at least that's what I interpreted you said. And I think I thought you said,

what I interpreted exceedingly well, and other people should realize it. And we should not go to,

hey, what is driver in a loop, driver out of a loop comparison. And let's go to those. So that's

a long lead into the rest of your article, which I think is fabulous. But again, it's

my personal view as I see it. Alan, he did ask the question in the piece, can we trust these

statistics from Waymo that claim its vehicles get into injury causing crashes, 60% less often

than human drivers, which gets to what you're, you're pointing out. How do you compare? And Kai,

tell us about that. So one, I hope in the future to write a more comprehensive piece

looking at Waymo statistics. So I don't want to say too much here, because I put the work in to

kind of carefully read through their methodology. When Tim at Understanding AI has, has talked with

kind of automated safety experts in the past, they've generally been positive about Waymo's

safety research, that it's serious and in good faith. And some of it's been peer reviewed.

And, and of course, you, of course, you always still want to have that grain of salt,

but it is a really difficult question to benchmark. And I think Waymo's done a,

one of the upsides of this whole autonomous driving thing is it's telling us more about

all the small things that happen that don't get report all the small crashes that happen,

that don't get reported usually, because they're not serious enough to

they're not serious enough to go to the insurance or the police.

And they, it seems like they do pretty careful things to balance out the fact that the types

of miles that Waymo drives are not on freeways, and are constrained to relatively small

geographic areas in major American cities. So from what I've seen, they seem thoughtful, but

I guess my best answer is come back to me in a couple months when I've like read things more

carefully and talk to a couple of outside people. And I don't want to comment too heavily right

now, because, as I said, I haven't gone through the work to have a well informed enough opinion on

that. Yeah. Okay. Congratulations on taking that approach. Yeah, I think that's great. You've taken

that approach, but I think you're cutting yourself short. Okay. And so I'll extend what I've read

and what you said. And in some sense, and I think, I think, I think I've always, I've

always marveled at how careful Waymo has approached this whole thing and how seriously

they've taken the safety thing and realize actually from the very beginning that, you know,

if they're not safe, they're not in business. Okay. I mean, if they can't solve the safety,

if they can't be reasonably safe and being reasonably safe means, you know, well, whatever

the heck that ends up meaning, because society is going to have to decide what the heck that

reasonable is. They're not in business. Now, we're now arguing about, you know, what is reasonable.

And one can have a very, you know, analytical, whatever approach to it. Although, unfortunately,

that very analytical approach is fraught with all the caveats that you just said.

And so still somebody's going to say, well, you know, I mean, I didn't, I mean,

actually probably free razor easier be better than we are in freeways and it isn't local. But

that's my opinion, you know, I mean, I forget. So in some sense, you know, somebody throws

that stone, I mean, really, you're throwing the stone the wrong direction. It is really

tougher to do it, my opinion, in places where they're doing it on San Francisco streets,

with all the chaos that actually happens there with, as we all know, what happened to

GM Cruz, I mean, it's the misbehavior that goes on around you is that you've got to sit there

and dodge the misbehavior. And then you're being called to blame. Okay, I mean, if everybody else

behaved, there wouldn't be a problem. You know, the rear end stuff, oh my goodness, it must be

Waymo's fault. They stopped at a stop sign. I don't. Boom, I hit them. Okay, why didn't they go

through? I mean, are people really, I mean, really, you want to blame Waymo for

for behaving when you didn't behave? I mean, you know, which dog ate your homework? I mean,

come on. And this is it's almost ridiculous of stuff that comes out out of some of this. I mean,

what we're supposed to have it, I'm sure, I'm sure, I'm sure I shouldn't speak for Waymo.

They can program those things to go through stop signs, just like I go through stop signs.

Therefore, we won't have any more rear enders in there. Okay, great. Okay, really, that's what you

want out of out of putting that stuff out there. Cut it out. Okay, so I get very, I get maybe too

animated when when when some when some of these comparisons are put out there. I think the

what what what you or what I think you pointed out and why I want I think, you know, you're

independent looking at it, you don't have a you really don't have a horse in this race,

you know, and and and you pointed out to my goodness, where they've been good,

where they've been good. And and and oh my goodness, we don't know if they're lying,

cheating and stealing here. Okay, we don't. But I don't know. I think you're right.

Like in comparison to other, especially other robot taxi companies who are already a lot of

them are are decent. Like, I think that Cruz was a was a decent robot taxi company,

where their technology was again, pretty safe. And then they had this, this kind of

terrible accident that kind of ended the business. So all of the terrible crash, it

wasn't an accident, it was a crash and it was a crash that was induced by somebody

hitting running somebody in the crosswalk. Okay, and flinging him. Okay, that's what it did. And

that guy never got a guy or gal never the police. I don't think even look for I shouldn't

blame, you know, San Francisco police for not looking for him. Did they ever catch that guy?

Are you kidding? And the company ends up having to be closed down and everybody and you

know, who knows how many billions of investment down the tubes? Okay, I mean, it's and saying,

oh my goodness, they were maybe they weren't as safe as they could. I mean, if a person gets flung

in front of you, you know, f equals ma doesn't help you. And then instead of being in the crash

avoidance business, which is what Waymo is in, which is crash avoidance business, you're

in the crash mitigation business. But guess who's in the crash mitigation business? Nitsa.

Okay, the things that Nitsa does is done and has done in safety is seatbelts,

airbags, crush zones. What are those things? They're crash mitigation. Okay,

yes. Finally, they're saying maybe automated emergency braking systems have to be in all

cars so that you don't go boom. Okay. All right. And so and so you know, there's a real

difference here. But go ahead. But what I want to say is that I think Waymo has clearly done

a lot of work. And I appreciate that when when going through this data, they,

they don't redact their narratives. They don't redact their, and then they provide more details

voluntarily after the fact with a couple months of the time, like this specific location,

which can sometimes be helpful. And the specific date which Nitsa redacts. And

and also I think whether or not the cruise crash was no matter the kind of specifics of that crash,

I think what that shows is all of Waymo's incentives are against serious crashes,

like very, very heavily in that also like at least two companies cruise and Uber self driving

program have basically been ended by bad crashes. And so it's clear that Waymo is trying really,

really hard to avoid that potentially ironically at the cost of slowing down adoption of a safer

technology. I don't blame Waymo because I think it's clear where their incentives are to be very

cautious. And so for instance, with freeway driving, I'm guessing that they could they probably

could expand to freeways at this point, like they're having they're comfortable enough with

their employees taking Waymo's to work going on a freeway, but they're still in testing.

And I think part of that is if you mess up on a freeway, then the it is you have to be much more

confident in your technology to be able to deal with the consequences where it's off the freeway.

Like if you're on a city street, it's almost always safe to slam on the brakes if you're unsure

or if or if something weird comes up, but you can't do that will well going 70 on the freeway.

And so I think I want to be cautious and careful and not make too many claims about

the safety of research. And I think it's fair to say that Ray Waymo's Waymo has a has a well

deserved reputation of being careful and being cautious and being and making good safe technology.

Well, I'll I'll go even farther. They definitely are with with no remorse on any of it. I mean,

no, no caveats on any of it. There is no doubt that they have been absolutely focused on safety

and so on and so forth. And they do it and they whatever and it is a constant. And I don't question

it at all. I don't I don't also don't question GM cruises either. Okay, they got stuck and some

sense I didn't I don't think I really ever questioned Ubers in in in the in the Lane

Herzberg crash. You know, I just my goodness, you know, people people are people, you know,

whatever. And maybe they should have been more careful not gone 41 miles and I've been gone 39

because of course, the automated emergency braking system, we could point to who who wrote the

code and so on to do that. And then some and in some sense, a lot of this stuff ends up coming

down is that we don't know what we don't know. I mean, it's a situation that that existed with

respect to to to cruise. I mean, I don't know if any of us ever imagined a scenario in which

a hit and run driver would hit some a pedestrian in a crosswalk and and and fling that pedestrian

in front of our vehicle. And somehow, I don't know, didn't have the sensors to determine whether or not

she was still there. And and she's decided to get out of the way of causing any more problems,

which is sort of a standard practice on on roads and which you, well, as far as we know,

it's minor and and pull over. I mean, the fact that they misinterpreted their data set to not

realize she was there, I would and and didn't happen to have a camera in the bumper to look

for such a thing because nobody ever thought that there would be such a thing. Although,

guess what? Now people have cameras in the bumper. And I sure hope all these systems

sit there and look, hey, can we go over anything that's in front of us when we start? Why? Because

we learned something out of the the Waymo the the cruise thing. And as far as I'm concerned,

we learned it so well that it won't happen again. It won't happen with with Waymo. I'm

I would imagine their stack looks at before they start out every time.

Is there is there anything there? So in some sense, you know, the whole process of having these crashes,

the reason they're reported and real reason they're reported is to fix them so they don't happen

in the future. It's not that anybody Waymo certainly not Waymo out there like a bunch of

loose cannons on some deck just running havoc over safety. I mean, come on. I mean,

is how can anybody even suggest that? Okay, but I but I think what you're what I see you helping do

in doing this has taken a fresh look at some of this, putting it very well out there for people to

read. And I think we can move on from the safety issue to how in the heck do we now get some

value out of what we've created? Okay, besides safety or the key safety improvement that we've

done is every bunny that rides Waymo is not driving. Okay, and I'm sure that everybody

that rides Waymo is certainly not would not ever misbehave when they're driving. Oh my goodness.

That's not their customers that of course, they're the perfect people in the world.

But maybe there are a few that are not there that would have misbehaved. I wonder why

why aren't we capturing some that value? And in fact, you know, moving to a system. And so

when one compares safety, at least my problem, my view on this thing is that

oh, man, they are safe. And sometimes we aren't even counting the fact that they're taking they're

taking misbehaviors, people that misbehave may it might just be one, maybe just two, who knows

how many it is, whatever, okay, off the roads, so that they're not misbehaving and causing

so actually they're making a whole damn thing much safer. And then of course, what's not

pointed out in what NHTSA doesn't collect and is what about all the situations which they haven't

reported yet, in which case somebody was misbehaving for which their system cause was the difference

and not crashing with that misbehavior. I think Waymo's biggest challenge is not

their system crashing. It's the avoiding of misbe of situations that misbehavior by

the driving public would cause the crash. Okay. And as you pointed out in some, you know,

running into the back end. But I suspect they also look in the back and how many of these

situations where somebody would have and they eased out into the stop sign and kept the person

from running into them because they were going to run the stop sign. And of course, the second

article in the e-letter talks about, you know, with the Tesla system about this woman and says,

hey, the system kept her from hitting somebody that runs through the stop sign.

And they certainly did, but my question on that is Waymo doesn't even report that.

NHTSA hasn't even asked them to report that. Why shouldn't that be counted on their safety side,

you know, along with, you know, discussion of how many, you know, whatever they've had,

wheels falling off. I mean, it may be a computational thing where you probably would

want to do a lot of simulations. Like if you're Waymo and you're trying to make that claim,

like occasionally you can show a video, but solve it like kind of saying the counterfactual that

like human in the same position. They do that. They actually run simulations. Like I know they

do that. Okay. They do do that. Okay. They look, they know it, but NHTSA hasn't asked them to present

that as part of the safety case that they're out there. I mean, you know, if they go out there

and do that, I mean, people who knows what, I don't know what people's reaction would be,

whatever. Go ahead. Sorry. But I think it is suggestive in the data that a large number

of crashes aren't just not Waymo's fault, but seem difficult to be, to defensively drive about.

Like if I, even if I'm a great defensive driver, it's hard for me stopped at a light to do anything

about the car directly behind me. Whereas there are comparatively few crashes that have occurred

in intersections. Instead of the like 20 rear ends, there were I think

maybe five that occurred in intersections, which I don't know the safety data off the top of my

head, but I'm guessing for the average human driver, the ratio is a little bit like fewer

rear ends is the main main crash. And there's more kind of crashes that may not have been

a human's fault, but may have been avoidable with slightly quicker reflexes or something.

And you see at least two of the crashes that kind of occurred when a car straight into Waymo's

right of way, they crashed into something else and then just gotten like thrown into

like very quickly in a way that of course, I wasn't there, I don't know, but is plausible

that it was completely unavoidable from Waymo's perspective. And so I think you can still kind

of infer all the crashes that didn't happen from the data by comparing what crashes show up less

than you'd expect, given the total number of crashes that Waymo has reported.

And I'm guessing I think that Waymo probably has done some research on this angle, and it's

definitely something I'll be paying attention to when I sit down and read what they've written.

But even not having done that, I think you can see some evidence.

This is terrific work, Kai. And Alan, you've got a link to the to the full piece

in the newsletter. The site is understandingai.org. And elsewhere in the newsletter, and Kai,

we want you to participate as much as you want to, looking at some of the other headlines here.

Alan, from Smart Driving Cars, Adam Jonas, Morgan Stanley, AI is about to get physical,

and you had, I guess, an up close look at this.

Yes, I mean, it was such a treat to have Adam in my class on last Monday. And so

instead of making us look at the video, he actually did it for us and went through the slides. And

of course, I had some other comments to make along the way. And we had an absolutely wonderful

give and take question and answer period afterwards. But what I can share with everyone

is at least point them to the link of his newest sort of lecture on where AI is going,

and so on. And I also made sure that that Adam is certainly aware of

understanding AI and UK and Timothy, certainly, because I guess Adam, I guess Adam not only is

the head research analyst for automotive that he's been for, I don't know, 15 years or now, and so

on. But there's now encompassing AI, and I guess his views on where AI is going are just whatever.

I mean, they're Adam Jonas views. For those of you that don't know Adam, I think Adam made

the call on Tesla 15 years ago. And I think a lot of people who paid attention to him got very

rich. I buy high, sell low. I should have paid attention to Adam, then Adam's a good friend.

I'm going to pay attention to him now, though.

There's more on Waymo in the newsletter. From the Verge, this headline, Robo Taxis has public

transit. Waymo thinks so. And this is about the company's Robo Taxis joining the Flex

Micro Transit Service in Chandler, Arizona. Two bucks a ride, half that for seniors and wheelchair

users as well and free for middle and high school kids. I mean, I guess, I don't know,

I guess alphabet is contributing to public good. Fantastic. Unless they actually can put it out

there at the price is the cost or the price is the cost plus profit. Oh my goodness. I mean,

when did they become that good? Are they that good? I mean, they're great on safety. I didn't

realize it was that cheap. I didn't realize they could give a wheelchair ride. How can they

give a wheelchair rides for a buck? I think whatever. And then I have a little discussion

there about the difference between costs and price. But I don't know who's picking up the tab.

I guess it's really, maybe that's in your, maybe that's in your next headline. Waymo

gets the green light for airport service in San Francisco. They're going to pick up the

tab. I don't think we have enough rich people go on the airports to pick up the tab

in Chandler. But maybe I haven't, I haven't done the balance sheet and the arithmetic on that.

Of course, I don't have the source of the data. I think I do put some things

in the letter about the New Jersey transit struggles to do an equivalent thing in New Jersey.

And, you know, when you base and not to jump to the bottom line, but when you when you go to it,

you know, the public subsidy for those rides is $88 each. So I guess as long as the public's

willing to pick up the tab on 88 and, you know, let's do it. But, but I don't know. I don't know

how you scale that. I mean, if you, yeah, I mean, yeah. The interesting, there's an interesting

comparison in that one of the recent new, like, entrance to the robotex, I mean, they've been

around for a while, but his main mobility, who isn't backed by like a really big tech company.

And so has been forced to like find profit centers earlier. And there, it seems like a lot of their

offerings are public transportation related, like in an arbor. So it's interesting that Waymo is

entering. Yeah. And that's, that's, that's very much to their credit that there are looking at

providing mobility to those who most need mobility and a better mobility than,

than without, without them society has put out there for these folks. And so I think it's what

I'm trying to do also, but whatever we'll see whether or not I'm any more successful than

somebody else on it, it's very tough. And, and of course, what made mobility,

you know, what's picking up the tab there. I don't, well, I don't know what's picking up,

but I might speculate. I like to spec. I mean, I just throw stuff out there. That's why people

don't like me. I just throw stuff out there. You know, there, there, the venture venture

industry has to be putting it out there, I guess, because that's typically what the venture

industry does with, with startups, they're a startup. Or it's, you know, it's some sort of

public entity that's as, as we taxpayers in New Jersey are picking up the tab on the $88

shortfall on each of those rides that, that New Jersey transits demand responsive system

accumulates. And so, you know, as long as we have that, that's fine. I mean, I think,

I think the real, the real inroads in, let's say, public transit or what is public transit

needs to be, my goodness, is it, is it possible or where could one provide this and not ask,

not require public subsidy and wait and see. Allen, in the newsletter, you have this.

Mackenzie had a report from a conference that held and it's headlined.

The inflection point has arrived. Waymo's vision for the future of mobility share.

And the link is to a video that's, that's been edited down some and you had some comments about that.

Yeah, I think they must have edited all the good stuff out of it. Again, that's my opinion.

I thought Mackenzie was better than that, but never mind.

From Business Insider, Uber's CEO says RoboTaxi's could displace drivers in 10 to 15 years and

create a big societal question. He's talking about many ride hailing drivers losing their jobs.

That's something I guess everybody talks about.

Yeah, I, I of course don't see it. I respectfully disagree. I mean, you know,

if you look at the market that ride hailing, it's not ride sharing. Some people call it ride sharing,

there's no ride sharing. You know, I use it. I use it. I guess I use it all the time.

I've claimed I will never again drive to Newark Airport because it's very simple. I mean,

Newark Airport charges so much to park that I can't afford it. And it's cheaper for me to

basically take Uber or Lyft to Newark Airport and I get chauffeured. It's like wonderful and

so on. And I don't have to worry about it. I can actually do work while I'm going to the airport.

Do I do work or am I playing games? Oh, of course I'm working. I mean, goodness.

I'm making use of my time. It's just like all the academic reports say that I'm going to do if

I get to be chauffeured, of course. And I think that's not going to go away. I just don't

think that's I think that. I mean, my goodness, when it when the when the when the car pulls up,

he, oh, Professor Kornhäuser, can I take your bag and put it in a trunk? I mean, Waymo can't do

that for me. There's nobody there to take my bag, hold my hand, open the door, be nice to me. Okay.

So in fact, the market for the for the gig worker to earn some amount of money to

begin to feed their families. I don't think it's going to go away.

Kai, do you have thoughts on that?

I disagree a little bit. I can imagine there being at the top level of the market, if you can

afford it for someone to grab your bag and and for it to be a luxury item. But I think

if you can manufacture enough of these cars, which should happen eventually, though it's it's been

a little slow thus far, then the comparison between the car driving and you have to pay the driver

versus the car driving period without any driver needing to be paid. I'm guessing that you can

eventually offer a lower price point to the where for most people, they're going to prefer

the cheaper, safer autonomous vehicle. And that's safer safer than the safer than an Uber driver.

No, no, no, well, I guess I want I want to I want to support Uber and Lyft and so on. And

there's a lot of stone thrown against their drivers and so on. And and and my experience with

their drivers has been nothing but superb and and and and and didn't misbehave. And even though

John Nash and his wife were essentially killed on the New Jersey turnpike by a taxi driver.

Okay, now that's a horrendous claim misprobably misbehaving. Okay, but or I shouldn't put probably

he might he might have been misbehaving. Okay, which is absolutely tragic, at least tragic,

you know, John Nash, I mean, whatever. John has Nobel Prize in economics for

Nash equilibrium in case you didn't know whatever. So can watch the movie made about him, whatever.

Okay, whatever. I knew him. Okay, whatever. But it the issue, my issue, okay, my issue with

respect to the airport thing is, when I go to the airport, most of the time, okay, I'm not paying.

I got an expense account that pays. I don't know what percentage of people that go to the airport

are going expense account, but most of the people go on an expense account, you know,

the go on expense account, you know, they tend to tip really well. Okay, Uber drivers and Lyft

drivers know that it is it is a it is the market for them to be able to somehow be able

to accumulate enough money to pay for the car, pay for the gas, pay for all the other stuff, pay for

do all that, spend all the time waiting for whatever between and so on, empty reposition

themselves to be able to scrape a living, which is what they're doing in the most part.

And I think and I certainly hope that that at least those of us that that do do that

and go to the airport that way, which is not every person that goes to and from the airport,

but so on, sort of continue to to frequent the Ubers and Lyfts and provide and provide

a work opportunity for those folks, which I don't know if you're I'm you know, if you really look

at them, you know, it's been a godsend to them to have Uber Lyft. Why? Because those 10 people who

don't I like to tell my students don't want to work for the man. Okay, they want the

independence, the opportunity to be able to choose their times, when they do it, when they do

that and the gig worker concept that Uber brought to the world with respect to this is, you know,

as valuable as the gig worker concept that Apple brought to the world with their app store,

you know, for coders who just want to write code and really don't want to or don't care about

marketing, about sales, about just sitting their basement of their parents home and

write code. All of a sudden, they got an outlet for their creativity because of the app store.

And in some sense, you know, that's really the value of what ride hailing brought to this world.

It gave people who needed rides instead of driving themselves be able to use them

with all kinds of value to them and provide them a reasonable work environment for the people

that do that. I hope and I, you know, in those of us that develop driverless cars, I would hope we

wouldn't go in there and ruin that market for them. Okay, leave that one to them. Okay.

All right, find there are so many other needs in our society for really efficient

affordable mobility that can change people's lives for the positive. Not in some sense make no

difference to me as to whether it's automated or not. So therefore, I don't provide me any value.

But all of a sudden, the tracks from the value, the value gained by the person that's doing the job.

I mean, we should be looking at this technology if it, in fact, is safe, which I think it is.

How do we use it to improve society? First, let's use it there for, we have so few of them.

There are 1.1 billion person trips a day in the USA, 1.1 billion.

Do we really have to go, go, go after the 0.1% going to the airport?

Is that really the place where we should be putting this thing? Is that really where,

where the $330 billion that have been invested in this technology? Is that really,

and it's finally safe? And we're not finally over, as you pointed out, over the safe hurdle?

We're not going to apply this to, to that solve that problem? I don't know.

Kai, if you didn't know, Alan is the co-author of a book called

The Real Case for Driverless Mobilities. So, you know, I mean, I've got, I mean,

people are sick of hearing me on this thing. And they are, but, but I didn't really,

really, there are, they're unfortunately, you know, this is a great country. And many of us

have cars, we have mobility that comes out of our ears. Okay. And safe mobility that comes out of

our ears. And here comes another one, another safe mobility that comes out of our ears.

But my goodness, this one is, this one is really important because if we, if it is safe,

which, which I think you've pointed out, I believe you, it's safe, then, then we should be looking

at where do we put this to use, where it really delivers the most value. In other words, for one

exists, pump it up there, pump the value proposition, not go someplace where the value, you know,

and you're barely, it's even questionable as to whether or not you're providing any value.

And, and if you're really worried about your bottom line, which you should be, this is a

capitalist society here, and an alphabet deserves to make a good return on this, on this moonshot,

because it is supposed to be a moonshot, whether it gets to be a hundred X or not.

Oh, it's got a long ways to go. I don't know, I think we should also be looking at

where can we, where, where can we, where can we apply this word? People really need a ride.

And, and, and whatever we've had up up to this point in time to give them a ride just

isn't working. Uber and Lyft are too expensive for them. Okay, so they're walking.

Alan, Alan, this might be a good time for you to give us an update on handy rides if we're going

to include one here. I guess, you know, we're still live, you know, we're given, we gave revenue

rides last week, you know, we're very happy where we are, we'll see where it gets to.

One more, one more headline from the newsletter from Inside EVs, and I think you alluded to this a

little bit ago, the car that stopped itself, Tesla collision avoidance assist lives up to

its name. This was about an incident posted on TikTok, of course, that went viral,

no FSD or autopilot involved here. A Tesla appeared to detect a rapidly approaching

vehicle that went through a red light and engaged emergency braking before the other

driver could react and before the driver could react and avoided a collision here.

Yeah, I mean, you know, it's one thing to take the driver out of the loop and then make a driver

list. It's another thing to put this stuff into all our vehicles to again, do the defensive

driving that we somehow or seem to be have limited capabilities of doing to avoid the

misbehavior that's out there because because it's not no misbehavior out there. I mean, it's

everything's easy. Everything easy. All this stuff does it. I don't I can't see how anybody

can suggest it doesn't. Okay. But the misbehavior out there and then to deal with the misbehavior

finally, you know, last year, year and a half ago, finally, NHTSA said, finally,

everybody has to have an automated emergency braking system in there. Okay. And it should be

automated emergency braking system that really is triggered so as to avoid the accident rather

than being crash mitigation rather than just reducing the severity of the accident or accident

crash, excuse me for using the wrong name. I mean, that's where it should be. The challenge,

the real challenge with doing that is of course, the false positive. Okay. Since technology

can't be made perfect, there is a probability that there will be a false positive. And unfortunately,

when the false positive occurred and why did my car break? I don't know. Oh, I got to take it back.

This is 11 here. Give me my money back. And then, you know, no manufacturer can can can

put a product out there in which you're going to use it for a while and then you bring it

back to and say it's a lemon on my money back. And so of course, you know, the way we've been

getting around this before is a heck with it, you know, we won't even try to help you. Go

ahead and crash. It's your problem. It's your car. You're in control. Go do it. I mean,

as opposed to sitting there and saying, Oh, I was hitting the gas and the damn thing hit

the brake. Oh, my goodness, it went over my desires. Why did it do that? It must not be any good.

Or in fact, you don't even know that it saved your butt. Has happened here. And at least this

one part, unfortunately, it's one viral thing. How many times this happens, we don't even know.

No, it's as demanded that it be reported. They should demand that that be reported as much as

they demand that the crashes be reported from from from from Waymo and the others.

And I think we talked about it before. Hey, you know, so come on. That's a

correct. Sorry. I should say

understanding AI has nothing to do with my irrational discussion here. So they are

totally exempt from, you know, not to be, Oh my goodness. And maybe they'll choose not to appear

with me because I'm so irrational or just mine. Okay, come after me. I mean,

you know, there's very little left of me anyway. So, you know, come after me. I was going to point

out that Kai said before Kai was talking before that, you know, if they can, when we're talking

about replacing drivers, etc. that if Waymo can manufacture or others can manufacture enough

vehicles, we left out the Tesla aspect of all this, which may have, you know,

all these vehicles on the road already with lots of capabilities. So yeah, well,

it's not only it's not only Tesla, but I think also, you know, one has to give credit to

to Toyota certainly has been out there. I think if you look at people we've we've

discussed it before the on the IIHS testing and videos that for years have been showing us

so which one of these automated emergency system system work and don't work. I think, you know,

we should point out Subaru with their eyesight has been absolutely, you know,

leading to make sure that these things, you know, do set off early enough to avoid the crash,

as opposed to just mitigating it, you know, I've commented about Mercedes, unfortunately,

whatever, I think, you know, they they wait to 1.6 seconds, the collision before they

they need these things, I think it's waiting a little too long. But, you know, we can disagree

or whatever. I think the whole industry needs to go to a point in which, yes, absolutely,

we need to put the we need these things need to be mandatory in all the cars and they need to be

actuated in time to avoid the collision as opposed to just mitigating it and mitigated if

that's all you can do. Absolutely. I'm not saying don't do any mitigation, but dammit, don't

don't wait around, you know, just to cover your butt, because you're not sure,

which is really what happened in the Uber situation. But never mind, we won't go back

through and rehash that whole thing. Well, sorry, hey, we dragged you into a lot of stuff. Really,

okay, okay. It's kind of that. No, seriously, I mean, I wrote to Tim, Timothy, you know,

immediately after I read read what you wrote. And then I just said it's

it's a it's a really good, good piece. And thank you so much for for taking the time with

this guy. Really, well done. And again, it's understanding ai.org. Thanks, Kai. And thanks

to Kai and thanks to Timothy for putting a whole going thing together. I think it's

anybody that's dealing with any things in AI should be

subscribing and reading that. You guys are doing a great job. Thank you. And thank you all for the

invitation. Okay. And you can find us at smart driving car dot com. My tech reports are at

textination.com. Thank you for watching or listening. Have a great weekend.

Smart Driving Cars episode 399: Waymo crashes examined by Understanding AI's Kai Williams

About this episode

Request an Explanation

Want to learn more?

More from Smart Driving Cars Podcast