WEBVTT

00:00:01.960 --> 00:00:05.540
You're listening to There Auto Be A Law, the

00:00:05.540 --> 00:00:08.259
Center for Auto Safety podcast with Executive

00:00:08.259 --> 00:00:11.720
Director Michael Brooks, Chief Engineer Fred

00:00:11.720 --> 00:00:15.640
Perkins, and hosted by me, Anthony Cimino. For

00:00:15.640 --> 00:00:18.300
over 50 years, the Center for Auto Safety has

00:00:18.300 --> 00:00:25.480
worked to make cars safer. Hey, folks. Thanks

00:00:25.480 --> 00:00:28.980
for coming back. Today is Wednesday, July 15th.

00:00:29.359 --> 00:00:32.500
I hope everyone's having a wonderful day. Fred

00:00:32.500 --> 00:00:35.280
is out. He's being a test driver for Waymo this

00:00:35.280 --> 00:00:38.719
week. No, I'm kidding. That'd be amazing, though.

00:00:39.179 --> 00:00:42.600
And in his place, we have a frequent guest of

00:00:42.600 --> 00:00:45.359
the show. Phil Copeman's back. Hi, Phil. Hey,

00:00:45.479 --> 00:00:47.920
great to be back. We're glad you're here because

00:00:47.920 --> 00:00:50.820
we've got questions for you. We've, you know,

00:00:50.840 --> 00:00:53.119
behind the scenes, we have very specific questions

00:00:53.119 --> 00:00:55.420
for you and we're going to get to them. But first,

00:00:55.479 --> 00:00:58.219
let's start off with some news out of NHTSA.

00:00:58.670 --> 00:01:00.770
So NHTSA, as everybody knows, the National Highway

00:01:00.770 --> 00:01:03.890
Traffic Safety Administration, and their current

00:01:03.890 --> 00:01:06.670
head is a guy who came out of Apple. He's a lawyer,

00:01:06.769 --> 00:01:09.390
and they've always been very kind of hands -off

00:01:09.390 --> 00:01:12.409
around AV companies, and the administration's

00:01:12.409 --> 00:01:15.769
view is very laissez -faire. Let's go ahead,

00:01:15.909 --> 00:01:19.549
and this last week, he put out a letter, and

00:01:19.549 --> 00:01:21.310
I'm going to quote from some of it. It says,

00:01:21.900 --> 00:01:23.900
Every second matters when law enforcement officers,

00:01:24.099 --> 00:01:26.000
firefighters, or paramedics are answering a call

00:01:26.000 --> 00:01:28.180
because lives are on the line. That is why human

00:01:28.180 --> 00:01:30.239
drivers who impede these operations are subject

00:01:30.239 --> 00:01:33.260
to fines and even jail time. NHTSA is today issuing

00:01:33.260 --> 00:01:36.140
a call to action for AV developers and operators

00:01:36.140 --> 00:01:38.700
to immediately focus their resources on fixing

00:01:38.700 --> 00:01:42.659
this issue. We expect you, AV operators, to prioritize

00:01:42.659 --> 00:01:45.280
first responder interactions and will schedule

00:01:45.280 --> 00:01:47.439
meetings with driverless automated driving system

00:01:47.439 --> 00:01:49.939
developers by month's end to hear your solutions.

00:01:50.510 --> 00:01:52.829
NHTSA has already been active in this space and

00:01:52.829 --> 00:01:55.049
will continue to exercise our enforcement authority

00:01:55.049 --> 00:01:57.689
for developers that do not address significant

00:01:57.689 --> 00:02:00.870
safety concerns. Public trust on our roads is

00:02:00.870 --> 00:02:05.150
earned, not given. I mean, that sounds like Jonathan

00:02:05.150 --> 00:02:07.849
Morrison is a big fan of the show. That's my

00:02:07.849 --> 00:02:13.000
takeaway. What do you guys think? This is the

00:02:13.000 --> 00:02:15.400
most we've seen NHTSA do in a while on autonomous

00:02:15.400 --> 00:02:17.479
vehicles, seriously. I mean, they've done some

00:02:17.479 --> 00:02:21.080
stuff, but this is huge. I've heard people saying

00:02:21.080 --> 00:02:24.139
that they've not seen anything like this. I haven't

00:02:24.139 --> 00:02:26.020
been paying that close attention to NHTSA for

00:02:26.020 --> 00:02:27.639
30 years, like Michael has, so we'll see what

00:02:27.639 --> 00:02:31.939
he says. Yeah, I mean, there's kind of a, I guess

00:02:31.939 --> 00:02:33.460
they're kind of speaking out of one side of their

00:02:33.460 --> 00:02:36.629
mouth here. I mean, NHTSA is saying. We will

00:02:36.629 --> 00:02:39.409
continue to exercise our enforcement authority

00:02:39.409 --> 00:02:41.930
in these situations. You know, what they're not

00:02:41.930 --> 00:02:44.289
saying is what they're saying there is we're

00:02:44.289 --> 00:02:46.189
going to continue to look at these things through

00:02:46.189 --> 00:02:48.330
recalls as they already have. What they're not

00:02:48.330 --> 00:02:51.789
saying is we're going to put into play a full

00:02:51.789 --> 00:02:56.629
swath of safety metrics in regulation that you

00:02:56.629 --> 00:02:59.919
guys have to certify to. What they're saying

00:02:59.919 --> 00:03:01.560
here is, you know, we're going to continue to

00:03:01.560 --> 00:03:04.280
go down this path of plotting towards a recall

00:03:04.280 --> 00:03:08.120
that may or may not work, as we've seen. And

00:03:08.120 --> 00:03:11.639
at the same time, like what why the focus here

00:03:11.639 --> 00:03:14.879
on police and emergency responders? Why didn't

00:03:14.879 --> 00:03:17.580
Waymo get a letter like this when they were passing

00:03:17.580 --> 00:03:20.780
school buses or driving into floods? Why? You

00:03:20.780 --> 00:03:23.120
know, we've seen this administration continually

00:03:23.120 --> 00:03:27.080
try to align itself with law enforcement for

00:03:27.080 --> 00:03:29.439
political reasons. And they're kind of doing

00:03:29.439 --> 00:03:31.780
the same thing here. And my question is, you

00:03:31.780 --> 00:03:34.560
know, why aren't you addressing all of these?

00:03:35.209 --> 00:03:37.750
other circumstances that are happening with AVs

00:03:37.750 --> 00:03:40.449
that happen to normal people on the roads, right?

00:03:40.969 --> 00:03:43.090
Well, that's fair. Let me try and argue it the

00:03:43.090 --> 00:03:44.830
other way, because there's some good news here,

00:03:44.930 --> 00:03:46.750
but everything you said is fair, Michael. But

00:03:46.750 --> 00:03:49.210
let me try and bring the other side. The other

00:03:49.210 --> 00:03:52.830
side is, if you assume, and this is what you're

00:03:52.830 --> 00:03:54.569
having trouble with, if you assume that the only

00:03:54.569 --> 00:03:56.590
thing they're allowed to do is regulatory recalls,

00:03:56.590 --> 00:04:00.360
right? That from the rulemaking... It's it's

00:04:00.360 --> 00:04:03.259
we want heaven forbid they should stifle anyone's

00:04:03.259 --> 00:04:05.219
innovation. Right. So they're getting rid of

00:04:05.219 --> 00:04:07.099
the stifling innovation and there are, you know,

00:04:07.099 --> 00:04:09.979
the rules are it's, you know, open season. Put

00:04:09.979 --> 00:04:12.939
those suckers on the road. OK. And so they've

00:04:12.939 --> 00:04:14.939
decided the administration's decided that's the

00:04:14.939 --> 00:04:16.839
path they want to take. But given that's the

00:04:16.839 --> 00:04:19.040
path you want to take, the only stick they have

00:04:19.040 --> 00:04:20.920
left, you know, that's carrot. The only stick

00:04:20.920 --> 00:04:26.699
they have left is recalls. OK. And I agree with

00:04:26.699 --> 00:04:29.779
you. I'd like to see. more than that one stick.

00:04:29.939 --> 00:04:32.360
But given that's where we are, you know, let's

00:04:32.360 --> 00:04:34.360
say the administrator decides that's sort of

00:04:34.360 --> 00:04:36.779
the playing field, the hand he's been dealt,

00:04:36.800 --> 00:04:38.920
he has to play that hand, right? What he's doing

00:04:38.920 --> 00:04:42.220
is he's, I mean, why aren't school buses in there?

00:04:42.300 --> 00:04:43.839
Because there's already a recall on school buses,

00:04:43.959 --> 00:04:46.100
right? There's no point putting anyone on notice

00:04:46.100 --> 00:04:47.600
because there's already a recall on school buses.

00:04:47.680 --> 00:04:49.779
There's already a recall on construction zones.

00:04:49.980 --> 00:04:53.339
There's already a recall on some other things,

00:04:53.439 --> 00:04:57.459
right? So we've seen those recalls already. Now,

00:04:57.480 --> 00:04:59.639
there's a separate problem, which is, by the

00:04:59.639 --> 00:05:02.259
way, the recalls don't work. I'm going to come

00:05:02.259 --> 00:05:06.220
back to that at my gas life. But set that aside.

00:05:06.500 --> 00:05:09.600
You know, if the thing you can do is recalls,

00:05:09.600 --> 00:05:11.980
then he's put everyone to notice that, by the

00:05:11.980 --> 00:05:15.360
way, we consider messing with emergency responders

00:05:15.360 --> 00:05:19.319
a risk to safety we're not going to tolerate.

00:05:19.720 --> 00:05:22.019
And so he's setting policy, just like we know

00:05:22.019 --> 00:05:25.959
that stalling on a highway is a recall. You can

00:05:25.959 --> 00:05:28.300
argue that coming to a stop in a travel lane

00:05:28.300 --> 00:05:30.560
is or is not dangerous. You know, people argue

00:05:30.560 --> 00:05:32.199
this all the time. Robot taxis, what do they

00:05:32.199 --> 00:05:33.600
do? They come to stop in a travel lane, right?

00:05:33.740 --> 00:05:37.100
But NHTSA has a longstanding policy that if you're

00:05:37.100 --> 00:05:39.480
going to stop in a travel lane, that's a recallable

00:05:39.480 --> 00:05:42.660
offense. And it's one that we had to fight to

00:05:42.660 --> 00:05:45.680
get put into place. But you got it. You got it.

00:05:45.680 --> 00:05:48.600
It's there, right? And so this is, I see this

00:05:48.600 --> 00:05:51.240
as a policy statement saying, okay, just like

00:05:51.240 --> 00:05:54.899
stopping on a highway. is unsafe no matter what

00:05:54.899 --> 00:05:57.560
you argue, messing with emergency responders

00:05:57.560 --> 00:06:01.560
now is in that bin. So expect there to be, we

00:06:01.560 --> 00:06:04.240
sort of let you slide by before, but enough is

00:06:04.240 --> 00:06:06.579
enough. We're going to recall for messing with

00:06:06.579 --> 00:06:09.480
emergency responders. So this is sort of declaring

00:06:09.480 --> 00:06:12.899
some new turf for recall. So that's great. It

00:06:12.899 --> 00:06:15.779
goes, there's a limited amount that accomplishes,

00:06:15.779 --> 00:06:18.959
but it's a positive development. That's the way

00:06:18.959 --> 00:06:22.160
I look at it. I mean, I think it's positive in

00:06:22.160 --> 00:06:23.720
the sense that the administration is willing

00:06:23.720 --> 00:06:27.000
to speak out against the industry. I just find

00:06:27.000 --> 00:06:30.120
it very peculiar that that only happens when

00:06:30.120 --> 00:06:32.379
you when when the vehicles are interfering with

00:06:32.379 --> 00:06:35.899
their kind of pet partners in law enforcement.

00:06:36.060 --> 00:06:39.040
Well, yeah, all these other situations we see

00:06:39.040 --> 00:06:41.459
where they've been silent. Well, the question

00:06:41.459 --> 00:06:44.500
is, is this maybe this was the easiest step and

00:06:44.500 --> 00:06:46.120
there's another step coming. The question is,

00:06:46.120 --> 00:06:48.500
is this the only step or is it not? Of course,

00:06:48.519 --> 00:06:49.829
you're going to. you're going to pick the low

00:06:49.829 --> 00:06:51.889
-hanging fruit, right? And it's hard to argue

00:06:51.889 --> 00:06:55.050
that blocking ambulances is okay. This is sort

00:06:55.050 --> 00:06:57.569
of the easiest piece they could bite off, so

00:06:57.569 --> 00:07:00.810
they did it. And we'll see. Now, the other observation

00:07:00.810 --> 00:07:02.990
is this has been going on for literally years,

00:07:03.230 --> 00:07:06.350
right? Parking in firehouse driveways. So what

00:07:06.350 --> 00:07:08.370
changed? Well, the other thing I want to point

00:07:08.370 --> 00:07:11.490
out is that how many years has it been since

00:07:11.490 --> 00:07:14.029
NHTSA had an administrator who was there other

00:07:14.029 --> 00:07:17.540
than to do a quick EV deal? Three administrative,

00:07:17.699 --> 00:07:19.519
like what, 10 years, more than 10 years, right?

00:07:21.899 --> 00:07:24.220
When you say quick EV deal. No, he's the first

00:07:24.220 --> 00:07:26.740
confirmed guy for a long time, right? Yeah, we

00:07:26.740 --> 00:07:28.319
had a confirmed guy who was there for three or

00:07:28.319 --> 00:07:30.079
four months and he did some EV stuff and then

00:07:30.079 --> 00:07:32.660
he left, right? Yeah, they had a lot of, yeah,

00:07:32.720 --> 00:07:35.420
there was, that was Steve Cliff came from me.

00:07:35.439 --> 00:07:38.800
They were bringing people from CARB over to do

00:07:38.800 --> 00:07:40.920
the fuel economy. He's a CARB guy and he was

00:07:40.920 --> 00:07:43.100
there and he did his thing and that's cool. But

00:07:43.100 --> 00:07:45.620
that had nothing to do with. With like AVs, right?

00:07:45.680 --> 00:07:48.560
So set aside. Right. How many years without an

00:07:48.560 --> 00:07:52.439
administrator? Like more than 10, right? That's

00:07:52.439 --> 00:07:54.139
been a long time since they've had confirmed

00:07:54.139 --> 00:07:56.300
administrators. Not acting administrators. Because

00:07:56.300 --> 00:07:59.160
Ann Carlson after Cliff was not confirmed, right?

00:07:59.259 --> 00:08:02.759
And then Sophie after Ann Carlson was not confirmed.

00:08:02.939 --> 00:08:08.430
And before Cliff. Right. How do you confirm the

00:08:08.430 --> 00:08:11.110
right? So you had all these administrators and

00:08:11.110 --> 00:08:13.389
and I don't know how it works at NHTSA, but in

00:08:13.389 --> 00:08:15.089
other places when you're acting, you're told,

00:08:15.189 --> 00:08:18.069
all right, you're in charge. Don't break anything

00:08:18.069 --> 00:08:19.990
till we get a real administrator. Right. And

00:08:19.990 --> 00:08:23.410
so so now we have a real administrator. And so

00:08:23.410 --> 00:08:26.810
he it one explanation for why it took so long

00:08:26.810 --> 00:08:28.589
to do something about emergency vehicles is we

00:08:28.589 --> 00:08:30.009
didn't really have a confirmed administrator

00:08:30.009 --> 00:08:32.190
and they didn't maybe maybe they didn't feel

00:08:32.190 --> 00:08:34.090
they had the power to really do that. Don't know.

00:08:34.429 --> 00:08:36.129
Good question for Anne Carlson, if you can get

00:08:36.129 --> 00:08:39.350
her on the show. But I'm really glad to see that

00:08:39.350 --> 00:08:41.950
with a confirmed administrator, they're starting

00:08:41.950 --> 00:08:45.129
to take those bites, right? Now, the question

00:08:45.129 --> 00:08:47.669
is, what happens next? And the question I parked

00:08:47.669 --> 00:08:50.850
to the side is, by the way, if the recall doesn't

00:08:50.850 --> 00:08:53.470
work, what happens then? But again, that's my

00:08:53.470 --> 00:08:55.169
gaslight. We're going to get there. Yeah, that's

00:08:55.169 --> 00:08:56.990
kind of what I want to know is because the, you

00:08:56.990 --> 00:08:59.289
know, the administrator, you know, Morrison says,

00:08:59.409 --> 00:09:01.870
that is why human drivers who repeat these operations

00:09:01.870 --> 00:09:04.899
are subject to fine and even jail times. NHTSA

00:09:04.899 --> 00:09:07.299
doesn't have that kind of enforcement capability,

00:09:07.600 --> 00:09:09.820
right? Well, not for human drivers either. No,

00:09:09.820 --> 00:09:12.919
no, that's the thing. Yeah, that's the thing.

00:09:13.059 --> 00:09:15.120
He's putting this out there as kind of, hey,

00:09:15.179 --> 00:09:17.879
this is the way the world works and this is what

00:09:17.879 --> 00:09:20.340
we're going to do, but we don't have the ability

00:09:20.340 --> 00:09:24.519
to do that. They can just do recall, right? And

00:09:24.519 --> 00:09:26.299
meanwhile, manufacturers are lobbying behind

00:09:26.299 --> 00:09:31.500
the scenes to, you know, essentially... prevent

00:09:31.500 --> 00:09:34.039
states from enforcing those types of laws against

00:09:34.039 --> 00:09:36.720
them as well. Right. So that's an area where

00:09:36.720 --> 00:09:38.879
we need to watch out for. Right. Yeah. Yeah.

00:09:38.940 --> 00:09:41.480
State state preemption by federal laws, which

00:09:41.480 --> 00:09:43.360
don't cover driving behavior, are going to be

00:09:43.360 --> 00:09:45.639
a problem. Right. Right. Absolutely. And it's

00:09:45.639 --> 00:09:48.960
a it's a doesn't have the resources to be in

00:09:48.960 --> 00:09:51.179
the business of issuing traffic tickets for going

00:09:51.179 --> 00:09:54.460
through red lights. Right. It just can't do that.

00:09:54.500 --> 00:09:57.139
No. No, and they shouldn't be. No, that's not

00:09:57.139 --> 00:09:58.740
their role. It never has been. And we need to

00:09:58.740 --> 00:10:01.200
preserve state rights in those areas because

00:10:01.200 --> 00:10:03.460
it's traditionally what the state's traditionally

00:10:03.460 --> 00:10:06.740
regulated driver performance, right? Well, it's

00:10:06.740 --> 00:10:09.299
simple that NHTSA regulates equipment. States

00:10:09.299 --> 00:10:11.840
regulate what I'm going to call driver behavior.

00:10:12.960 --> 00:10:15.419
And there's a muddy middle in the middle when

00:10:15.419 --> 00:10:18.240
the driver's a piece of equipment. And the way

00:10:18.240 --> 00:10:21.279
I look at it is NHTSA should be able to regulate

00:10:21.279 --> 00:10:24.710
the ability of the... computer driver to behave

00:10:24.710 --> 00:10:28.529
properly, but the states should be able to retain

00:10:28.529 --> 00:10:33.230
the right to apply corrective force when they

00:10:33.230 --> 00:10:37.789
don't behave properly. If NHTSA says, well, this

00:10:37.789 --> 00:10:40.110
thing can't stop at red light, doesn't have the

00:10:40.110 --> 00:10:41.629
ability to stop at red lights because it doesn't

00:10:41.629 --> 00:10:43.350
even see them, okay, sure, that's a NHTSA thing.

00:10:43.809 --> 00:10:46.870
But if it passes a test and it stops at 10 red

00:10:46.870 --> 00:10:48.769
lights in a row, what's NHTSA supposed to do?

00:10:49.309 --> 00:10:51.230
And if it misses one red light in a thousand,

00:10:51.429 --> 00:10:53.389
you have a computer that almost right, you know,

00:10:53.389 --> 00:10:55.350
misses one red light in a thousand. Well, the

00:10:55.350 --> 00:10:56.970
states ought to be able to issue traffic tickets

00:10:56.970 --> 00:11:00.889
for that one in a thousand. Right. I mean, I'm

00:11:00.889 --> 00:11:03.490
excited by what happens next because I'm very

00:11:03.490 --> 00:11:06.769
naive and hopeful that he'll bring in these manufacturers

00:11:06.769 --> 00:11:09.549
and they'll say, yes, we're going to fix everything.

00:11:10.480 --> 00:11:13.779
Are you really? I am. I like to think very positive

00:11:13.779 --> 00:11:14.960
in the world. No, I think they're going to show

00:11:14.960 --> 00:11:17.259
up with a suitcase full of money. Like, you know,

00:11:17.279 --> 00:11:19.659
they're talking to FIFA referees like this. There

00:11:19.659 --> 00:11:23.440
you go. Infantino gate. You're younger than I

00:11:23.440 --> 00:11:25.519
am. And you're pretty cynical, at least as cynical

00:11:25.519 --> 00:11:30.860
as I am. So I don't know. Let me say a word about

00:11:30.860 --> 00:11:34.080
how this really works, though. So the thing,

00:11:34.120 --> 00:11:37.730
the most important part of that letter was. basically

00:11:37.730 --> 00:11:40.129
all the car companies being called to the, the

00:11:40.129 --> 00:11:41.789
AV companies being called to the principal's

00:11:41.789 --> 00:11:44.070
office, right? It's like, you're going to meet

00:11:44.070 --> 00:11:46.690
with us this month. Now they, these guys meet

00:11:46.690 --> 00:11:48.710
with NHTSA on a regular basis. They just wrote,

00:11:48.750 --> 00:11:51.029
but they got called to the principal's office.

00:11:51.149 --> 00:11:55.269
Okay. And the way this works is now I've never

00:11:55.269 --> 00:11:57.110
actually worked at NHTSA. Again, if you get an

00:11:57.110 --> 00:11:59.009
ex -NHTSA person, they can provide all the fine

00:11:59.009 --> 00:12:01.789
shading here. But my understanding is these guys

00:12:01.789 --> 00:12:03.789
meet all the time and that's good. That's appropriate.

00:12:04.600 --> 00:12:07.240
Why use a hammer when a quiet conversation will

00:12:07.240 --> 00:12:09.899
solve the problem? Makes sense, right? But these

00:12:09.899 --> 00:12:12.340
guys are being put on notice and not just with

00:12:12.340 --> 00:12:14.659
a letter, but in person. Hey, guys, you can't

00:12:14.659 --> 00:12:17.240
mess this up. And it's basically a soft power

00:12:17.240 --> 00:12:21.480
play. A lot of what NHTSA does isn't by... hammering

00:12:21.480 --> 00:12:23.480
on companies. It's having conversations saying,

00:12:23.500 --> 00:12:25.940
you've got to get on top of this or else we're

00:12:25.940 --> 00:12:28.279
going to have to hammer you later. And so this

00:12:28.279 --> 00:12:31.000
is the first step in a long -term escalation

00:12:31.000 --> 00:12:33.860
on this issue that ultimately ends up with investigations

00:12:33.860 --> 00:12:37.299
and recalls. But I hope, and I'm sure NHTSA hopes,

00:12:37.419 --> 00:12:40.340
and one assumes the manufacturers hope, it doesn't

00:12:40.340 --> 00:12:42.100
have to get there. It's like, fix the problem.

00:12:42.360 --> 00:12:45.179
Just, hey guys, here's your chance. Fix the problem

00:12:45.179 --> 00:12:47.289
so we don't have to pull out our hammer. So that's

00:12:47.289 --> 00:12:48.970
what's really going on. The letter doesn't sound

00:12:48.970 --> 00:12:50.629
like it does anything, but what's really going

00:12:50.629 --> 00:12:55.889
on is it's starting up the cycle of escalation.

00:12:56.830 --> 00:12:59.269
Yeah, I think it's good in that obviously I think

00:12:59.269 --> 00:13:02.090
this has happened because it's become a lot more

00:13:02.090 --> 00:13:04.909
prevalent in the news of, hey, this is what's

00:13:04.909 --> 00:13:08.610
happening with Waymo and specifically Waymo because

00:13:08.610 --> 00:13:09.710
they're the biggest player. Of course it's more

00:13:09.710 --> 00:13:13.029
prevalent. They're scaling up. Yeah. Once in

00:13:13.029 --> 00:13:15.590
a thousand. happens occasionally to 10 cars.

00:13:15.730 --> 00:13:17.610
It happens every day to a thousand cars. It is

00:13:17.610 --> 00:13:21.450
very simple. Or the example I like to use is

00:13:21.450 --> 00:13:24.330
if you have something that happens once a year,

00:13:24.470 --> 00:13:27.409
you can blow it off and say it's only once a

00:13:27.409 --> 00:13:29.250
year and maybe nobody even sees it, right? Right.

00:13:29.409 --> 00:13:31.710
Well, if that's once a year with a hundred cars,

00:13:31.830 --> 00:13:34.009
when you go to 10 ,000 cars, that's twice a week.

00:13:34.350 --> 00:13:37.429
All of a sudden people notice. Yep. And that's

00:13:37.429 --> 00:13:40.250
one of the things I constantly repeat around

00:13:40.250 --> 00:13:42.990
some of the exemptions that the industry is seeking,

00:13:43.070 --> 00:13:45.529
where they want to deploy 90 ,000 vehicles all

00:13:45.529 --> 00:13:47.990
at once on the road. I mean, that's where we're

00:13:47.990 --> 00:13:50.309
going to start seeing the biggest problems, I

00:13:50.309 --> 00:13:53.370
think, if that happens. Yeah, if you have an

00:13:53.370 --> 00:13:55.470
exemption and you put 100 cars in the road and

00:13:55.470 --> 00:13:59.350
you have a rare thing, and by the way, this only

00:13:59.350 --> 00:14:01.250
works if you plan on getting lucky most of the

00:14:01.250 --> 00:14:05.080
time. And, you know, let's say 1 % of the failures

00:14:05.080 --> 00:14:07.679
are dangerous and 99 % of the failures are benign.

00:14:08.059 --> 00:14:10.960
And you pay attention and you fix after the 99

00:14:10.960 --> 00:14:13.779
% of the errors, you probably get lucky, right?

00:14:14.519 --> 00:14:18.480
But when you put 100 ,000 cars on the road, you're

00:14:18.480 --> 00:14:21.700
going to get such a flood of problems that that

00:14:21.700 --> 00:14:23.679
1 % is still going to happen all the time and

00:14:23.679 --> 00:14:25.879
be dangerous, right? It's a numbers game. Flood,

00:14:25.879 --> 00:14:29.820
he says. Yeah, let's jump into that. So, Phil,

00:14:29.879 --> 00:14:32.710
you wrote a good sub stack piece. expanding on

00:14:32.710 --> 00:14:35.389
this specific letter more. And I'm going to quote

00:14:35.389 --> 00:14:38.509
from yours. Uh, no, no, this is, this is my,

00:14:38.590 --> 00:14:40.950
this is one of my favorite things. And I'm going

00:14:40.950 --> 00:14:44.049
to cross examination. No, no, no, no. The cross

00:14:44.049 --> 00:14:45.809
examination is coming up and it's not really

00:14:45.809 --> 00:14:48.669
that it's, we have some confusion. So, uh, quoting

00:14:48.669 --> 00:14:51.090
from you, the endlessly repeated AV industry

00:14:51.090 --> 00:14:53.230
mantra that computers don't drive drunk most

00:14:53.230 --> 00:14:55.549
necessarily extend to. They also don't get a

00:14:55.549 --> 00:14:57.750
free pass just because metrics might show that

00:14:57.750 --> 00:14:59.809
drunk or otherwise impaired humans struggle with

00:14:59.809 --> 00:15:02.629
emergency response situations. I love that, and

00:15:02.629 --> 00:15:06.669
I hope that comes – that's what we were just

00:15:06.669 --> 00:15:08.350
talking about, that this comes forward and real

00:15:08.350 --> 00:15:11.590
progress happens there. But my question is, and

00:15:11.590 --> 00:15:13.929
I know Michael has the same thing, so there's

00:15:13.929 --> 00:15:16.730
the emergency response situations, which we'd

00:15:16.730 --> 00:15:21.559
call an edge case, right? Incorrectly call it

00:15:21.559 --> 00:15:24.279
edge. Well, incorrectly depending who you talk

00:15:24.279 --> 00:15:25.600
to. And that's I think that's what we're getting.

00:15:25.659 --> 00:15:27.919
Yeah, that's what we get into because we go back

00:15:27.919 --> 00:15:30.000
to the NHTSA letter and read the part Jonathan

00:15:30.000 --> 00:15:33.059
had wrote about edge cases. Yeah, I think, you

00:15:33.059 --> 00:15:35.019
know, that kind of underlies some of this confusion

00:15:35.019 --> 00:15:37.580
that I think everyone's experiencing to some

00:15:37.580 --> 00:15:40.559
extent. Right. So around edge cases. So my hope

00:15:40.559 --> 00:15:43.000
is that you can expand on what. What do you see

00:15:43.000 --> 00:15:44.960
is what is, what exactly is an edge case? I think

00:15:44.960 --> 00:15:47.940
the, the typical nomenclature is that, Hey, it's

00:15:47.940 --> 00:15:50.240
something that's very rare and unexpected and

00:15:50.240 --> 00:15:53.159
no one could possibly train for. Right. But that's

00:15:53.159 --> 00:15:55.580
not how we're seeing it used from AV companies

00:15:55.580 --> 00:15:58.720
in particular. Well, that's what AV wants you

00:15:58.720 --> 00:16:02.320
to believe. Yeah. Right. So please. All right.

00:16:02.539 --> 00:16:04.919
All right. So this, this ends up being a towel,

00:16:05.080 --> 00:16:12.039
right? Sorry, Fred. Hope you enjoy it. I'm pinch

00:16:12.039 --> 00:16:15.980
-hitting for Fred. Okay. All right. So let's

00:16:15.980 --> 00:16:17.820
do this slightly long version because hopefully

00:16:17.820 --> 00:16:20.779
this sort of puts people's minds in the right

00:16:20.779 --> 00:16:24.500
place. And I did an article on this, and you

00:16:24.500 --> 00:16:26.320
can put a point in the show notes to this, but

00:16:26.320 --> 00:16:27.840
let me sort of give the quick version, right?

00:16:28.659 --> 00:16:30.919
Back at the dawn of time when there was analog

00:16:30.919 --> 00:16:34.279
electronics, but computers weren't a thing, which

00:16:34.279 --> 00:16:38.559
is before I was born. Barely. Okay. What you

00:16:38.559 --> 00:16:41.279
had was you had components that would work properly

00:16:41.279 --> 00:16:44.340
in a range of conditions, but would have trouble

00:16:44.340 --> 00:16:47.320
beyond that range. Okay. And so you might have

00:16:47.320 --> 00:16:49.320
a circuit that works between three volts and

00:16:49.320 --> 00:16:51.740
five volts, but doesn't work below three volts

00:16:51.740 --> 00:16:55.419
and burns up above five volts. Okay. And you

00:16:55.419 --> 00:16:59.669
might have something that works at... above freezing,

00:16:59.830 --> 00:17:01.970
but below boiling or something like that for

00:17:01.970 --> 00:17:03.610
temperature. So you'd have temperature ranges,

00:17:03.669 --> 00:17:05.990
you have voltage ranges. And what you would do

00:17:05.990 --> 00:17:08.109
is say, well, this thing's supposed to work between

00:17:08.109 --> 00:17:11.450
4 .5 and 5 .5 volts. That was a really common

00:17:11.450 --> 00:17:13.430
back in the day, four and a half to five and

00:17:13.430 --> 00:17:15.490
a half volts is supposed to work. And you would

00:17:15.490 --> 00:17:19.470
test it at 4 .5 volts or 4 .49 or whatever. You

00:17:19.470 --> 00:17:22.329
test it at 4 .5 and 5 .5. And if both of them

00:17:22.329 --> 00:17:24.390
worked, you'd say, okay, it probably works in

00:17:24.390 --> 00:17:28.079
between. And so. The low voltage cutoff was an

00:17:28.079 --> 00:17:30.000
edge and the high voltage cutoff was the edge.

00:17:30.079 --> 00:17:31.559
And they didn't have to work beyond that, but

00:17:31.559 --> 00:17:32.819
you want to make sure they work right on the

00:17:32.819 --> 00:17:35.559
boundary. Right. Okay. And for temperature, you

00:17:35.559 --> 00:17:38.819
tested it freezing. You tested it at whatever

00:17:38.819 --> 00:17:40.680
the maximum temperature, probably not boiling,

00:17:40.799 --> 00:17:42.779
probably like 120 Fahrenheit or something like

00:17:42.779 --> 00:17:45.019
that. And there's also vibration. They had these

00:17:45.019 --> 00:17:46.680
chambers in automotive called shake and bake

00:17:46.680 --> 00:17:49.980
chambers where you'd rattle this electronics

00:17:49.980 --> 00:17:52.900
around. Okay. And you take the temperature down

00:17:52.900 --> 00:17:54.420
to the minimum and you take the temperature up

00:17:54.420 --> 00:17:56.180
to the maximum and you'd see if they could work

00:17:56.180 --> 00:17:58.400
for 10 ,000 hours or whatever it was. You know,

00:17:58.400 --> 00:18:02.779
it's fun. Okay. So edge cases were a minimum

00:18:02.779 --> 00:18:06.240
or maximum specification. And a corner case was

00:18:06.240 --> 00:18:09.359
a pair, lowest voltage and highest temperature,

00:18:09.640 --> 00:18:12.200
stuff like that. So combinations of edge cases.

00:18:12.559 --> 00:18:14.759
All right. And so that's sort of the history.

00:18:15.140 --> 00:18:18.359
Now, computers come along and people in software

00:18:18.359 --> 00:18:20.400
talk about edge cases, but it's a little weird.

00:18:20.910 --> 00:18:22.730
Because if you have a number, if you have an

00:18:22.730 --> 00:18:25.130
integer, there's a minimum that can be represented

00:18:25.130 --> 00:18:27.650
and a maximum that can be represented. So you

00:18:27.650 --> 00:18:29.609
test the minimum and the maximum in software.

00:18:29.990 --> 00:18:32.710
But also zero is a special number because division

00:18:32.710 --> 00:18:35.990
by zero really ruins your day. Okay. And that's

00:18:35.990 --> 00:18:38.349
not an edge at all. It's this discontinuity in

00:18:38.349 --> 00:18:40.450
the middle. So you can call it an edge case,

00:18:40.589 --> 00:18:42.589
but it's not really an edge anymore. It's just

00:18:42.589 --> 00:18:46.069
a special value. Division by zero is bad. One

00:18:46.069 --> 00:18:48.589
is fine. Minus one is okay. Zero blows up. All

00:18:48.589 --> 00:18:52.450
right. They can still call them edge cases, but

00:18:52.450 --> 00:18:55.450
they aren't edges in the analog electronic sense.

00:18:55.630 --> 00:18:59.309
They're a boundary between behaviors, but not

00:18:59.309 --> 00:19:02.670
necessarily a minimum or maximum. So now you

00:19:02.670 --> 00:19:05.069
get self -driving cars and they say, well, there's

00:19:05.069 --> 00:19:09.589
edge cases, okay? And then, so software in general

00:19:09.589 --> 00:19:12.569
mutated edge case from a value -based thing to

00:19:12.569 --> 00:19:14.710
being, well, it's a rare combination of conditions.

00:19:16.009 --> 00:19:18.390
And the self -driving car guys wholeheartedly

00:19:18.390 --> 00:19:20.410
embrace this. They say edge case is a rare combination

00:19:20.410 --> 00:19:24.250
of conditions. Now, there's an intuition here

00:19:24.250 --> 00:19:29.569
that breaks, okay? If you believe that the testing

00:19:29.569 --> 00:19:32.769
conditions represent the environment precisely,

00:19:33.130 --> 00:19:37.809
then you can say that you should be able to handle

00:19:37.809 --> 00:19:40.289
everything that's common correctly and the only

00:19:40.289 --> 00:19:43.609
thing left are the things that are rare. Right.

00:19:43.809 --> 00:19:46.170
Right? And the things that are left that are

00:19:46.170 --> 00:19:48.089
rare are probably not simple things. They're

00:19:48.089 --> 00:19:51.250
probably combinations of things. All right. So

00:19:51.250 --> 00:19:54.650
you might see a stop sign on a pole. And you

00:19:54.650 --> 00:19:57.269
might see a school bus. You might not have seen

00:19:57.269 --> 00:19:59.450
a stop sign on a school bus, which means something

00:19:59.450 --> 00:20:01.750
completely different than a stop sign on a pole.

00:20:02.450 --> 00:20:10.089
Yeah. Right? Okay. And so you can call those

00:20:10.089 --> 00:20:12.490
edge cases if you want, but it's no longer the

00:20:12.490 --> 00:20:16.259
original meaning. It's rather unique combinations

00:20:16.259 --> 00:20:18.880
that you sort of haven't dealt with. Now, if

00:20:18.880 --> 00:20:21.200
you believe that you're diligent and always fix

00:20:21.200 --> 00:20:23.059
everything you see the first time you see it,

00:20:23.079 --> 00:20:26.539
that there's a failure, then what you expect

00:20:26.539 --> 00:20:30.339
is edge cases that you fail on to be rare. But

00:20:30.339 --> 00:20:32.519
that's not what we're seeing. What we're seeing

00:20:32.519 --> 00:20:35.240
is failures of things that happen all the time,

00:20:35.299 --> 00:20:38.079
right? And there's some intuition. If it's rare,

00:20:38.319 --> 00:20:40.180
then people wouldn't have thought of it. That

00:20:40.180 --> 00:20:42.259
intuition is broken in the notion you fix them

00:20:42.259 --> 00:20:44.849
all is broken. So we're seeing things like floods

00:20:44.849 --> 00:20:49.329
and construction zones that robotaxis, autonomous

00:20:49.329 --> 00:20:51.930
vehicles struggle with, even though they've seen

00:20:51.930 --> 00:20:53.609
them a bunch of times. They just haven't really

00:20:53.609 --> 00:20:55.910
fixed it. And by the way, it's very challenging

00:20:55.910 --> 00:20:58.829
to fix. But saying, well, it's an edge case,

00:20:58.849 --> 00:21:00.950
so we're going to ignore it. Well, we're back

00:21:00.950 --> 00:21:03.710
to scale. If you have 10 robotaxis, in fact,

00:21:03.730 --> 00:21:05.390
there's a story from Waymo's early days back

00:21:05.390 --> 00:21:06.990
when they're Google. They didn't worry about

00:21:06.990 --> 00:21:10.450
construction zones because they're so rare. Are

00:21:10.450 --> 00:21:14.839
you kidding me? You know, I'm in the Northeast

00:21:14.839 --> 00:21:17.660
where there's two seasons. There's winter and

00:21:17.660 --> 00:21:20.859
road construction. Exactly. I'd like to live

00:21:20.859 --> 00:21:22.559
in that part of the country. But they're in California

00:21:22.559 --> 00:21:24.819
where they don't get to get winter road damage

00:21:24.819 --> 00:21:26.339
and a bunch of other things. And also with 10

00:21:26.339 --> 00:21:28.940
cars, maybe road construction is relatively rare.

00:21:29.799 --> 00:21:32.140
OK, so this is the other thing people say, well,

00:21:32.160 --> 00:21:33.700
it can't be an edge case because I've seen it.

00:21:33.759 --> 00:21:36.500
Well, if you have 10 cars and you haven't seen

00:21:36.500 --> 00:21:39.640
it from the point of view of a computer, it's

00:21:39.640 --> 00:21:41.859
an edge case because it's never seen it. Right.

00:21:42.480 --> 00:21:45.200
But people have a lot more awareness and also

00:21:45.200 --> 00:21:47.519
people learn by hearing stories and reading news

00:21:47.519 --> 00:21:50.579
articles, which cars learn by they if they've

00:21:50.579 --> 00:21:54.059
taken on the road. So there's there's two fallacies

00:21:54.059 --> 00:21:55.420
that arise here. Well, it's an edge case, so

00:21:55.420 --> 00:21:57.619
it's rare. So it doesn't matter. I'm sorry. Now

00:21:57.619 --> 00:22:01.119
there's three. Oh, I'm on a Python skit. One

00:22:01.119 --> 00:22:05.579
is just because it rare is rare doesn't mean

00:22:05.579 --> 00:22:09.670
it's not severe. Power outages, earthquakes,

00:22:10.210 --> 00:22:13.390
sure, they're rare, but that doesn't mean you

00:22:13.390 --> 00:22:14.849
can ignore them because they're rare because

00:22:14.849 --> 00:22:18.069
it isn't about frequency. It's about risk. And

00:22:18.069 --> 00:22:20.089
risk is a combination of severity and frequency

00:22:20.089 --> 00:22:23.710
in its basic form. OK, the next one is just because

00:22:23.710 --> 00:22:25.849
you haven't seen it doesn't mean you shouldn't

00:22:25.849 --> 00:22:28.990
expect it. I have no idea how many I'm going

00:22:28.990 --> 00:22:30.309
to come up with. I'm just going to keep spitting

00:22:30.309 --> 00:22:31.730
them out, okay? I'm loving this. Just because

00:22:31.730 --> 00:22:33.150
you haven't seen it doesn't mean you can't expect

00:22:33.150 --> 00:22:36.269
it. There's this standard, UL4600, that lists

00:22:36.269 --> 00:22:38.150
all this. Did you think of this? And part of

00:22:38.150 --> 00:22:40.210
the reason is so you can fix it before you actually

00:22:40.210 --> 00:22:43.829
mess up in the real world, okay? Another one

00:22:43.829 --> 00:22:46.170
is even if you have training data, if you don't

00:22:46.170 --> 00:22:49.750
have enough training data, it may sort of dissolve

00:22:49.750 --> 00:22:51.630
into the statistical noise. You may not learn

00:22:51.630 --> 00:22:54.500
the less you need to learn. Okay. So, you know,

00:22:54.579 --> 00:22:58.380
if you have one bad day and a bunch of good days

00:22:58.380 --> 00:23:01.640
and you're 99 .9 % accurate, well, yeah, that's

00:23:01.640 --> 00:23:03.839
because you ignored the bad day. That's a problem.

00:23:03.920 --> 00:23:05.799
See the previous comment about severity, right?

00:23:06.740 --> 00:23:09.779
And in other words, so there's a bunch of issues,

00:23:09.819 --> 00:23:11.839
but when people say it can't be an edge case,

00:23:11.920 --> 00:23:13.819
they're thinking if they've seen it, it's not

00:23:13.819 --> 00:23:17.960
an edge case. And I understand that's intuitive,

00:23:18.140 --> 00:23:21.000
but it's... They're thinking they haven't seen

00:23:21.000 --> 00:23:22.700
it's an iEdge case. And the car companies are

00:23:22.700 --> 00:23:24.359
saying it's rare, so it doesn't matter. They're

00:23:24.359 --> 00:23:27.480
ignoring severity. But what you have to look

00:23:27.480 --> 00:23:30.099
at it is from the point of view of the car. The

00:23:30.099 --> 00:23:32.980
computer in the car, either it knows how to do

00:23:32.980 --> 00:23:36.819
something or it doesn't. Right? It's either trained

00:23:36.819 --> 00:23:39.319
or it's not. And so if it gets something wrong,

00:23:39.539 --> 00:23:42.339
that means from the car's point of view, it's

00:23:42.339 --> 00:23:44.339
an iEdge case. It's something it hasn't seen

00:23:44.339 --> 00:23:46.980
enough to learn the lesson. And it could be a

00:23:46.980 --> 00:23:49.420
very common scenario where we're all like. Duh,

00:23:49.420 --> 00:23:51.740
we know how to do that. In the early days, we

00:23:51.740 --> 00:23:53.700
found that people wearing yellow were an edge

00:23:53.700 --> 00:23:56.299
case. Why is that? Because it hadn't trained

00:23:56.299 --> 00:23:57.660
in rain and it hadn't trained in construction

00:23:57.660 --> 00:24:00.740
zones. And that's where people wear yellow. Other

00:24:00.740 --> 00:24:02.619
than yellow raincoats and high -vis vests, you

00:24:02.619 --> 00:24:05.740
don't see a lot of yellow on people, right? Yellow

00:24:05.740 --> 00:24:08.420
must mean road sign because people don't wear

00:24:08.420 --> 00:24:10.039
yellow. Why? Because I've never driven in the

00:24:10.039 --> 00:24:11.619
rain and I don't drive in construction zones.

00:24:11.859 --> 00:24:16.000
If I wear yellow, I look like the sun. So I now

00:24:16.000 --> 00:24:18.789
wear bright blue instead of yellow. It's just

00:24:18.789 --> 00:24:21.509
one of my personal quirks. Oh, no. The bright

00:24:21.509 --> 00:24:23.529
blue raincoat, not the yellow one. There's not

00:24:23.529 --> 00:24:25.849
a lot of bright blue in nature, right? Yeah,

00:24:25.849 --> 00:24:28.450
there's not, but at least it's blue. You know,

00:24:28.490 --> 00:24:30.529
at least they don't look like a road sign. Right.

00:24:30.650 --> 00:24:34.309
But this is a problem that I run into with kind

00:24:34.309 --> 00:24:35.849
of the Silicon Valley mindset where it's this

00:24:35.849 --> 00:24:38.970
very myopic view, whereas if I haven't seen it,

00:24:39.009 --> 00:24:42.450
it doesn't exist. And they're the ones training

00:24:42.450 --> 00:24:46.130
these babies. This is almost a religion in the

00:24:46.130 --> 00:24:49.150
data science. machine learning community and

00:24:49.150 --> 00:24:51.069
engineering and a bunch of other things. There's

00:24:51.069 --> 00:24:53.190
like in the agile, they say, well, you're never

00:24:53.190 --> 00:24:54.769
going to don't put in something you don't need

00:24:54.769 --> 00:24:56.390
because you're not going to need it later. Only

00:24:56.390 --> 00:24:59.150
do stuff you care about. Right. And so in data

00:24:59.150 --> 00:25:01.289
science, I've seen a number of folks say, well,

00:25:01.309 --> 00:25:06.130
what is it? The data was set you free. Only look

00:25:06.130 --> 00:25:08.250
at the data, ignore everything that isn't in

00:25:08.250 --> 00:25:09.690
the data because it's not in the data doesn't

00:25:09.690 --> 00:25:12.069
exist. And if what you're trying to do is build

00:25:12.069 --> 00:25:15.349
a web app and get it up and running. And, you

00:25:15.349 --> 00:25:18.150
know, and if 1 % of users aren't served the 99

00:25:18.150 --> 00:25:21.569
% payer bill, I get why you look at that. But

00:25:21.569 --> 00:25:26.170
again, that ignores severity, right? So, yeah,

00:25:26.289 --> 00:25:28.049
we finally come down to my definition of an edge

00:25:28.049 --> 00:25:31.829
case. It is no longer minimum or maximum or combinations,

00:25:32.170 --> 00:25:35.750
okay? An edge case, the only definition that

00:25:35.750 --> 00:25:38.170
works is an edge case is something the car has

00:25:38.170 --> 00:25:41.049
not been trained to do well enough. There's not

00:25:41.049 --> 00:25:43.049
enough training data. Now, you may say that's

00:25:43.049 --> 00:25:44.549
sort of a circular argument. It's like, well,

00:25:44.549 --> 00:25:46.289
that's where they forced us by using all these

00:25:46.289 --> 00:25:48.589
excuses, right? Sure. You know, an edge case

00:25:48.589 --> 00:25:50.890
is something that needs more training data to

00:25:50.890 --> 00:25:53.230
fix, assuming more training data will fix it,

00:25:53.289 --> 00:25:55.230
which is a whole different discussion, okay?

00:25:55.430 --> 00:25:57.289
Yeah. Now, if you look back at Administrator

00:25:57.289 --> 00:26:01.049
Morrison's letter, what's interesting is he's

00:26:01.049 --> 00:26:04.970
not saying, he's not actually saying edge cases

00:26:04.970 --> 00:26:07.630
are okay or edge cases are bad. What he's saying

00:26:07.630 --> 00:26:11.500
is... Stop using edge case as an excuse because

00:26:11.500 --> 00:26:14.339
we don't buy it. So he's not weighing on what

00:26:14.339 --> 00:26:16.240
an edge case might be or might not be. What he's

00:26:16.240 --> 00:26:18.759
really saying is calling it an edge case doesn't

00:26:18.759 --> 00:26:21.059
get you off the hook. And that's fair. I completely

00:26:21.059 --> 00:26:23.140
support that. Saying, well, it's an edge case.

00:26:23.220 --> 00:26:24.480
What are we supposed to do? It's like, well,

00:26:24.579 --> 00:26:27.359
if it's a high severity edge case, you're supposed

00:26:27.359 --> 00:26:30.160
to fix it before you go on the road or have a

00:26:30.160 --> 00:26:34.079
plan. Why don't they have a... Just a very simple

00:26:34.079 --> 00:26:35.920
heuristic built into these systems. Say, hey,

00:26:35.960 --> 00:26:37.700
if you come across something you don't understand,

00:26:37.960 --> 00:26:41.759
move to a minimal risk condition. Well, they

00:26:41.759 --> 00:26:45.519
kind of do, but it's not working. Well, with

00:26:45.519 --> 00:26:48.400
like road flooding, they claim that's an edge

00:26:48.400 --> 00:26:50.299
case and they're like, let's drive through it,

00:26:50.359 --> 00:26:52.519
which is. I mean, isn't it a how do you know

00:26:52.519 --> 00:26:54.779
what you don't know type of question there? Yes,

00:26:54.779 --> 00:26:57.910
exactly. So what they're doing, different rant,

00:26:58.029 --> 00:27:01.089
also an article that I did. I had five articles

00:27:01.089 --> 00:27:03.509
that really hit a lot of current topics with

00:27:03.509 --> 00:27:07.410
Junko Yoshida and I have this podcast, podcast

00:27:07.410 --> 00:27:10.529
sub stack. Yeah, we need to push your podcast.

00:27:10.670 --> 00:27:12.970
Yes, it's a great podcast. We'll have a link

00:27:12.970 --> 00:27:15.230
to it. So this is extremely relevant because

00:27:15.230 --> 00:27:17.769
one of my articles is edge cases. And what we

00:27:17.769 --> 00:27:20.809
do is we have a podcast where I write an article,

00:27:20.950 --> 00:27:24.390
we do a podcast, and then Junko writes an article.

00:27:24.940 --> 00:27:27.539
You know, as a set, we hit five different topics

00:27:27.539 --> 00:27:29.859
and one of them was edge cases, right? And we

00:27:29.859 --> 00:27:33.200
had a guest. We had Kristen Poland from NTSB

00:27:33.200 --> 00:27:35.960
on as a guest. That was amazing. We had historian

00:27:35.960 --> 00:27:38.279
Peter Norton on as a guest and that was season

00:27:38.279 --> 00:27:40.920
one. So yes, please put the pointer, plug it.

00:27:40.940 --> 00:27:43.079
They're all there. And season two starts this

00:27:43.079 --> 00:27:44.559
fall. We're going to have five more episodes

00:27:44.559 --> 00:27:46.759
and we need to get guests and all this stuff.

00:27:46.839 --> 00:27:49.720
So we're funded for another season for this fall

00:27:49.720 --> 00:27:52.710
and then we'll see what happens. But going back

00:27:52.710 --> 00:27:55.009
to edge cases, where we are is edge cases are

00:27:55.009 --> 00:27:56.829
things where they don't have enough training

00:27:56.829 --> 00:28:00.849
data to fix it, assuming more data can fix it.

00:28:01.369 --> 00:28:04.869
And if you say you don't have enough training

00:28:04.869 --> 00:28:07.690
data, well, then you need to get more training

00:28:07.690 --> 00:28:09.710
data if it's high severity. It's that simple.

00:28:11.049 --> 00:28:13.190
Claiming it's an edge case is a description of

00:28:13.190 --> 00:28:16.130
why it failed. It's not an excuse. Well, it's

00:28:16.130 --> 00:28:19.869
rare, so it doesn't matter. Well, you know, earthquakes

00:28:19.869 --> 00:28:22.500
are rare. It doesn't mean... They don't matter.

00:28:24.039 --> 00:28:27.099
Power outages are rare. Power outages are rare.

00:28:27.480 --> 00:28:30.640
Floods are rare. This is kind of where the whole

00:28:30.640 --> 00:28:33.500
thing gets discombobulated. We're talking about

00:28:33.500 --> 00:28:37.079
events that are rare when whether something's

00:28:37.079 --> 00:28:40.099
an edge case or not has almost zero to do with

00:28:40.099 --> 00:28:42.059
the rarity of the event. It has everything to

00:28:42.059 --> 00:28:44.759
do with whether the computer is capable of handling

00:28:44.759 --> 00:28:49.900
it. In a perfect world. The unhandled edge cases

00:28:49.900 --> 00:28:53.039
are so rare, you've never even seen them to train

00:28:53.039 --> 00:28:55.579
on them. If you take a pure machine learning

00:28:55.579 --> 00:28:57.940
approach, the way you look at it is you only

00:28:57.940 --> 00:29:00.339
train on stuff you see. You don't go looking

00:29:00.339 --> 00:29:03.480
for trouble, okay? And so in edge cases, the

00:29:03.480 --> 00:29:05.180
stuff that hasn't happened yet, so you didn't

00:29:05.180 --> 00:29:08.559
train on it, okay? And from a technical point

00:29:08.559 --> 00:29:11.240
of view, I'm okay with that definition, but there's

00:29:11.240 --> 00:29:13.700
two issues. One is there's stuff that has happened

00:29:13.700 --> 00:29:16.069
that they still didn't train enough on. Right.

00:29:16.170 --> 00:29:18.950
And the other one is if you care about safety,

00:29:19.109 --> 00:29:21.369
you have to think about things that haven't happened

00:29:21.369 --> 00:29:23.369
yet because it's not OK to kill someone to learn

00:29:23.369 --> 00:29:25.910
the lesson if it was an obviously foreseeable

00:29:25.910 --> 00:29:28.309
lesson. Right. I mean, there's some things you

00:29:28.309 --> 00:29:31.130
just, wow, never thought of that one. The stuff

00:29:31.130 --> 00:29:33.089
I've seen is like, well, didn't occur to me.

00:29:33.250 --> 00:29:36.829
OK. And some of them are very subtle. So yesterday,

00:29:36.910 --> 00:29:39.490
yesterday I was driving by a fire truck that

00:29:39.490 --> 00:29:41.730
opened its doors and there were stop signs painted

00:29:41.730 --> 00:29:44.640
on the inside of the cab doors. Oh, that's a

00:29:44.640 --> 00:29:47.299
new one. Never saw that one before, but I get

00:29:47.299 --> 00:29:48.839
it. I get, you know, they don't want you driving

00:29:48.839 --> 00:29:50.819
past a fire truck while the crew's climbing.

00:29:50.880 --> 00:29:53.160
It's like, okay, that's interesting, right? That

00:29:53.160 --> 00:29:56.079
was an edge case for me personally. But, you

00:29:56.079 --> 00:29:58.279
know, if I were in the AV business and I were

00:29:58.279 --> 00:30:00.180
talking to firefighters, I hope I would have

00:30:00.180 --> 00:30:01.859
noticed that when I was taking a station tour

00:30:01.859 --> 00:30:04.259
and program my cars to stop for fire trucks and

00:30:04.259 --> 00:30:06.349
the doors are open. Right. Yeah. And that's another

00:30:06.349 --> 00:30:09.269
weird stop sign, right? It's not the same as

00:30:09.269 --> 00:30:12.589
a stop sign. Okay. Different rant. Stop signs

00:30:12.589 --> 00:30:15.769
have the following meanings. I can have more

00:30:15.769 --> 00:30:18.849
fun by doing Q &A, but I'm mindful of time. I

00:30:18.849 --> 00:30:20.410
have fun with journalists with this sometimes.

00:30:20.609 --> 00:30:22.730
Okay. So one meaning of a stop sign, stop sign

00:30:22.730 --> 00:30:24.869
on a pole in the ground means stop and then go,

00:30:24.950 --> 00:30:28.690
right? Stop sign on a pole that is being held

00:30:28.690 --> 00:30:31.609
by hand, but is not in the ground. as opposed

00:30:31.609 --> 00:30:34.049
to a stop sign in the ground that has someone's

00:30:34.049 --> 00:30:35.690
resting their hand on, completely different,

00:30:35.829 --> 00:30:38.630
right? But if the stop sign is not actually planted

00:30:38.630 --> 00:30:39.869
in the ground and just resting on the ground,

00:30:40.029 --> 00:30:42.309
that means stop and stay stopped because that's

00:30:42.309 --> 00:30:44.730
a construction worker directing traffic and you're

00:30:44.730 --> 00:30:46.509
waiting for it to spin around at the slow, okay?

00:30:47.210 --> 00:30:50.329
Another stop sign is a stop sign held by a person

00:30:50.329 --> 00:30:53.430
on a paddle and that depends on if they're serious

00:30:53.430 --> 00:30:56.829
about it or not. If they're serious about the

00:30:56.829 --> 00:30:58.670
stop, you stop and say stop because that's the

00:30:58.670 --> 00:31:00.789
school crossing guard. But if the stop sign is

00:31:00.789 --> 00:31:02.769
sort of dangling by their side, they're not serious

00:31:02.769 --> 00:31:07.710
and you ignore it. That's a tough one. But that

00:31:07.710 --> 00:31:10.789
is the semantics, right? It makes you wonder

00:31:10.789 --> 00:31:15.059
if we need to not reinvent. the vehicles to interpret

00:31:15.059 --> 00:31:17.539
the stops, but reinvent the stop signs to be

00:31:17.539 --> 00:31:20.859
more accurate, right? Like a stop go sign has

00:31:20.859 --> 00:31:23.660
maybe a green border and the stop stop sign,

00:31:23.880 --> 00:31:25.859
you know. But it's stop on both sides. So school

00:31:25.859 --> 00:31:27.579
crossing guards, it has to be stop on both sides,

00:31:27.640 --> 00:31:30.180
not slop slow, right? Okay. Now, you know, if

00:31:30.180 --> 00:31:32.039
it's upside down by their side, it's pretty obvious.

00:31:32.160 --> 00:31:34.079
People can figure this out. And if they're not

00:31:34.079 --> 00:31:36.160
sure, they slow down and the guard, you know,

00:31:36.180 --> 00:31:37.559
the guard's out in the middle of the road. There's

00:31:37.559 --> 00:31:40.380
a bunch of contextual clues. But we're not done

00:31:40.380 --> 00:31:42.160
yet. There's stop sign in a school bus, which.

00:31:42.720 --> 00:31:45.339
Which never means stop and go. It either means

00:31:45.339 --> 00:31:48.640
stop and stay stopped if it's rotated out. But

00:31:48.640 --> 00:31:50.359
if it's flattened against the school bus, it

00:31:50.359 --> 00:31:52.519
means disregard, even though you can see it,

00:31:52.539 --> 00:31:56.099
right? Yep. Okay. So I was saying earlier, edge

00:31:56.099 --> 00:31:59.940
cases are about contextual clues. I mean, you

00:31:59.940 --> 00:32:02.000
thought a stop sign was just a stop sign. No

00:32:02.000 --> 00:32:05.440
way, man. It's super complicated. Right. And

00:32:05.440 --> 00:32:09.299
that's your stop sign. Let's explain people how

00:32:09.299 --> 00:32:15.650
to yield. Okay. People wonder why you owe 4 ,600,

00:32:15.769 --> 00:32:17.789
300 pages long. It's like, you have no idea how

00:32:17.789 --> 00:32:19.450
much stuff you have to deal with for this stuff.

00:32:19.730 --> 00:32:24.269
Oh my word. That one isn't even in there. It

00:32:24.269 --> 00:32:26.309
just says stop signs. It's, but you're supposed

00:32:26.309 --> 00:32:27.670
to know all this. You're supposed to figure this

00:32:27.670 --> 00:32:32.490
out. I love that. So we know AVs are going to

00:32:32.490 --> 00:32:34.630
continue to have this problem with edge cases.

00:32:34.630 --> 00:32:37.369
It's not going away. It may be what they define

00:32:37.369 --> 00:32:40.690
as an edge case. Yeah, well, I mean, it may be

00:32:40.690 --> 00:32:42.769
the thing that prevents, you know, prevents the

00:32:42.769 --> 00:32:45.190
industry from, I think it definitely is the thing

00:32:45.190 --> 00:32:46.890
that's preventing the industry from moving as

00:32:46.890 --> 00:32:50.130
quickly as their investors want them to. Yeah,

00:32:50.170 --> 00:32:52.609
rare events, heavy -tailed rare events are the

00:32:52.609 --> 00:32:55.230
technical thing, the biggest technical challenge

00:32:55.230 --> 00:32:58.170
for the industry. And the biggest non -technical

00:32:58.170 --> 00:33:00.609
challenge is acceptance of people, you know,

00:33:00.609 --> 00:33:02.529
that they make mistakes and it gets complicated,

00:33:02.750 --> 00:33:06.529
right? Yeah, and it looks like, you know, we're

00:33:06.529 --> 00:33:08.920
not, I mean. judging by what we're continuing

00:33:08.920 --> 00:33:12.200
to see as, you know, Waymo tests and as other

00:33:12.200 --> 00:33:14.480
companies put their vehicles out on the road

00:33:14.480 --> 00:33:17.559
to test, we're not seeing these events go down.

00:33:17.700 --> 00:33:20.819
We're seeing them go up. Is there, you know,

00:33:20.839 --> 00:33:24.500
I don't know, is this, is it a fundamental problem

00:33:24.500 --> 00:33:28.940
to the future of autonomous vehicles here? Well,

00:33:28.940 --> 00:33:32.859
well, kind of, yeah. Let me do my gas light if

00:33:32.859 --> 00:33:34.440
I can. Yeah, good. We're going to go into that

00:33:34.440 --> 00:33:36.619
next. Perfect. Let's tease it up. All right.

00:33:36.640 --> 00:33:39.180
Let's do it. One of the things to be careful

00:33:39.180 --> 00:33:42.660
of is if you have a constant failure rate and

00:33:42.660 --> 00:33:44.099
you double the number of cars, you're going to

00:33:44.099 --> 00:33:46.960
see twice as many bad things happen, right? And

00:33:46.960 --> 00:33:49.240
so be careful saying we're seeing more events.

00:33:49.339 --> 00:33:51.559
Well, is that because there's more cars? More

00:33:51.559 --> 00:33:54.359
cars is different than the technology getting

00:33:54.359 --> 00:33:57.059
worse, right? But if you're seeing more events,

00:33:57.220 --> 00:33:59.920
it's clearly a failure of the technology to improve

00:33:59.920 --> 00:34:03.420
fast enough. to keep the events under control

00:34:03.420 --> 00:34:07.619
as they scale, right? Now, which brings me to

00:34:07.619 --> 00:34:10.420
my gaslight. The gaslight, I thought about 10

00:34:10.420 --> 00:34:11.880
ways to go at this, so I'm just going to take

00:34:11.880 --> 00:34:13.480
a shot. We'll see where we end up, all right?

00:34:13.920 --> 00:34:18.639
But the gaslight is the notion that geofencing

00:34:18.639 --> 00:34:23.829
is going to solve edge cases, right? Well, doesn't

00:34:23.829 --> 00:34:25.849
it just kind of ignore them? No, it just puts

00:34:25.849 --> 00:34:27.969
them aside. Well, yeah. So I told you there's

00:34:27.969 --> 00:34:30.050
a bunch of ways this goes, right? So this can

00:34:30.050 --> 00:34:32.610
be interactive, but I have a certain place I

00:34:32.610 --> 00:34:35.150
want to go, right? So one of the things people

00:34:35.150 --> 00:34:37.809
say is, have said since the beginning, is if

00:34:37.809 --> 00:34:40.090
there's something you can't do, geofence it out

00:34:40.090 --> 00:34:42.909
or exclude it from your ODD. So if you don't

00:34:42.909 --> 00:34:45.150
know how to do traffic lights, don't drive on

00:34:45.150 --> 00:34:47.510
roads with traffic lights. Well, that makes sense.

00:34:47.789 --> 00:34:52.900
Okay? Sure. I mean, interstate highways without

00:34:52.900 --> 00:34:55.960
traffic lights is a viable way to deploy, right?

00:34:56.340 --> 00:34:58.119
You don't have to do traffic lights on an interstate

00:34:58.119 --> 00:35:00.599
highway, except for a few places. There's an

00:35:00.599 --> 00:35:02.260
Indian reservation where there's a traffic light

00:35:02.260 --> 00:35:04.820
in the middle of an interstate highway. Geofence

00:35:04.820 --> 00:35:08.019
that guy out. Okay. But it's still viable because

00:35:08.019 --> 00:35:09.860
that's a pretty well -characterized situation.

00:35:10.400 --> 00:35:13.739
But then it gets trickier when you say we can't

00:35:13.739 --> 00:35:16.980
do floods, so we're going to geofence out places

00:35:16.980 --> 00:35:22.019
that are flooded. Or we can't do emergency response

00:35:22.019 --> 00:35:23.719
scenes. We're going to geofence out emergency

00:35:23.719 --> 00:35:27.079
response scenes. And the problem with that is

00:35:27.079 --> 00:35:31.860
it takes time to set up the geofence. And when

00:35:31.860 --> 00:35:34.820
you had 10 cars, it probably worked. When you

00:35:34.820 --> 00:35:37.440
have a city flooded with cars, see what I did

00:35:37.440 --> 00:35:39.500
there? You have a city flooded with cars, and

00:35:39.500 --> 00:35:42.980
now there's floodwater, all right? The chance

00:35:42.980 --> 00:35:46.219
of a car wandering into the area with the floodwater

00:35:46.219 --> 00:35:49.619
in the next five minutes is quite high. Yeah.

00:35:49.679 --> 00:35:52.420
Right. And so what we saw with Waymo is they

00:35:52.420 --> 00:35:56.619
had a recall for floods and the recall didn't

00:35:56.619 --> 00:36:00.260
work. OK. Actually, they had a fix for floods

00:36:00.260 --> 00:36:02.619
and the recall didn't work and the fix didn't

00:36:02.619 --> 00:36:04.760
work. I don't remember the exact sequence. And

00:36:04.760 --> 00:36:07.280
the story they told in the I think was in the

00:36:07.280 --> 00:36:09.219
recall letter, which was really interesting,

00:36:09.420 --> 00:36:12.559
was, well, and I'm going to paraphrase and read

00:36:12.559 --> 00:36:14.800
into it what was probably going on. OK, because

00:36:14.800 --> 00:36:17.849
they. You know, they're always a little bashful

00:36:17.849 --> 00:36:20.030
in these reports. Do not admit ever doing anything

00:36:20.030 --> 00:36:23.409
bad. But apparently what happened was their plan

00:36:23.409 --> 00:36:25.769
for floods was the cars still don't know how

00:36:25.769 --> 00:36:28.250
to not drive into floodwaters. Their plan was

00:36:28.250 --> 00:36:30.469
when the National Weather Service issues a flood

00:36:30.469 --> 00:36:33.690
alert, they set up a geofence for low -lying

00:36:33.690 --> 00:36:35.989
areas that are prone to flooding and tell the

00:36:35.989 --> 00:36:39.349
cars to stay away. Well, that's not a bad thing.

00:36:39.489 --> 00:36:43.050
You know, that's like a 90 % plus solution, but

00:36:43.050 --> 00:36:46.349
it's not 100%. Yeah. I mean, it relies on accurately

00:36:46.349 --> 00:36:48.630
mapping all the areas where floods might be occurring,

00:36:48.889 --> 00:36:51.190
right? Which is tough. And their excuse for why

00:36:51.190 --> 00:36:53.090
they drove into floodwaters, the Weather Service

00:36:53.090 --> 00:36:55.449
didn't get the alert out until after the flooding

00:36:55.449 --> 00:36:58.429
had started. And one of their cars drove into

00:36:58.429 --> 00:37:00.289
it between the time the flooding started and

00:37:00.289 --> 00:37:04.300
the alert went out. They didn't quite do it,

00:37:04.320 --> 00:37:06.159
but, you know, blaming the National Weather Service

00:37:06.159 --> 00:37:08.679
is not an excuse for driving into floodwaters.

00:37:08.980 --> 00:37:13.219
No. They didn't actually say that, but they sort

00:37:13.219 --> 00:37:17.019
of led you up in that direction, right? And we

00:37:17.019 --> 00:37:20.119
also have seen stuff now, that one's documented.

00:37:20.860 --> 00:37:23.760
We've seen school bus incidents go down in Austin,

00:37:23.880 --> 00:37:25.820
but not eliminated. There was one that happened

00:37:25.820 --> 00:37:27.659
in March, and now we're into summer break, okay?

00:37:28.099 --> 00:37:31.829
And I speculate. because in fact, I said this

00:37:31.829 --> 00:37:34.949
was a possible way to help things out back when

00:37:34.949 --> 00:37:38.309
it was happening, was Austin's had an app, a

00:37:38.309 --> 00:37:40.650
phone app for like 10 years that tells you where

00:37:40.650 --> 00:37:43.130
the school buses are, which is great for parents.

00:37:43.639 --> 00:37:45.559
You know, where's my bus? Oh, my bus is 30 minutes

00:37:45.559 --> 00:37:48.320
late. I'm not going to go stand out in 120 degree

00:37:48.320 --> 00:37:50.400
heat waiting for my bus, you know, wait till

00:37:50.400 --> 00:37:52.500
it comes. It's a tracking. All the buses have

00:37:52.500 --> 00:37:54.460
trackers and they track an app. And if I were

00:37:54.460 --> 00:37:57.039
Waymo on day one, I would have said I'm having

00:37:57.039 --> 00:37:59.239
trouble with buses. I'm going to do a mobile

00:37:59.239 --> 00:38:01.699
keep out zone around every bus so I don't run

00:38:01.699 --> 00:38:04.219
into them. But the app's not going to be perfect.

00:38:04.519 --> 00:38:05.920
Transponders are going to die. You know, who

00:38:05.920 --> 00:38:09.980
knows? Stuff goes wrong. So these kind of...

00:38:10.280 --> 00:38:14.300
Keep -out zones for transient, mobile, things

00:38:14.300 --> 00:38:17.840
that aren't a stop sign set in cement, right?

00:38:18.039 --> 00:38:20.460
Even those change, but less frequently, right?

00:38:20.739 --> 00:38:24.860
Things that are very, very variable. Doing keep

00:38:24.860 --> 00:38:28.420
-out zones is like a B -plus kind of solution.

00:38:28.679 --> 00:38:32.260
Yeah, it's only as good as the technology underlying

00:38:32.260 --> 00:38:34.219
what's behind it, right? And the data, there's

00:38:34.219 --> 00:38:36.480
always a response time to, well, we'll set up

00:38:36.480 --> 00:38:39.579
a response for construction zones. Part one,

00:38:39.739 --> 00:38:42.519
it may take a while. There may be delay and some

00:38:42.519 --> 00:38:44.539
car is going to wander into it. OK. And part

00:38:44.539 --> 00:38:48.099
two, sometimes there's ad hoc stuff that isn't

00:38:48.099 --> 00:38:51.519
even reported to the city. A tree branch falls

00:38:51.519 --> 00:38:54.639
down and somebody sets up a cone. It's not going

00:38:54.639 --> 00:38:56.860
to be in the city database in 30 seconds. Right.

00:38:57.119 --> 00:38:59.360
Or there's a big truck trying to back out of

00:38:59.360 --> 00:39:01.420
a driveway and there's some folks in vests who

00:39:01.420 --> 00:39:03.400
are directing traffic. And that's not going to

00:39:03.400 --> 00:39:06.519
be in the city database. Right. And a water main

00:39:06.519 --> 00:39:09.139
breaks is never going to be covered by the NWS.

00:39:09.699 --> 00:39:13.420
Or a bridge, yeah, water main break or a place

00:39:13.420 --> 00:39:15.460
that usually doesn't flood floods today because

00:39:15.460 --> 00:39:17.440
reasons, right? You know, because it's winter

00:39:17.440 --> 00:39:19.579
and there was a bunch of snow that melted. Who

00:39:19.579 --> 00:39:23.900
knows, right? So it catches the majority of them,

00:39:24.019 --> 00:39:26.820
but it doesn't catch all of them. So the gaslight

00:39:26.820 --> 00:39:29.679
is anyone who says, well, we're going to have

00:39:29.679 --> 00:39:33.760
a database and or. vehicle to infrastructure

00:39:33.760 --> 00:39:37.099
or vehicle to vehicle warnings about weird stuff

00:39:37.099 --> 00:39:41.320
that happens. And it's like a 90 % plus solution,

00:39:41.400 --> 00:39:43.679
but it's not 100 % partly because it will never

00:39:43.679 --> 00:39:46.300
be perfectly accurate. And partly because the

00:39:46.300 --> 00:39:48.460
first car that gets there still has to deal with

00:39:48.460 --> 00:39:50.579
something that it hasn't been told about. And

00:39:50.579 --> 00:39:55.179
what if that first car is a river taxi? So if

00:39:55.179 --> 00:39:57.380
you have a system that's like within a factor

00:39:57.380 --> 00:39:59.559
of 10 of where you need it to be, it's really

00:39:59.559 --> 00:40:02.090
good, but occasionally has trouble. Then arguing,

00:40:02.289 --> 00:40:05.150
well, this central database gives us that last

00:40:05.150 --> 00:40:07.349
factor of 10 that we need to squeeze out. Okay,

00:40:07.469 --> 00:40:10.070
completely. It's going to reduce your exposure,

00:40:10.250 --> 00:40:13.789
right? And maybe what's happened with Waymo,

00:40:13.849 --> 00:40:15.710
again, speculating is maybe they're not running

00:40:15.710 --> 00:40:17.750
past school buses because their system of keeping

00:40:17.750 --> 00:40:19.809
away from them is pretty good, but it's not perfect.

00:40:20.150 --> 00:40:22.030
And sure enough, one of the times it didn't work,

00:40:22.090 --> 00:40:25.550
they ran past, one got caught. It tells me they

00:40:25.550 --> 00:40:27.130
have more work to do. On the other hand, I'm

00:40:27.130 --> 00:40:29.510
glad to see it wasn't dozens a month. That's

00:40:29.510 --> 00:40:32.079
improvement. This is a good thing. But the gaslight

00:40:32.079 --> 00:40:35.099
specifically is anyone who tells you that telling

00:40:35.099 --> 00:40:37.559
cars what to stay away from is 100 % solution,

00:40:37.840 --> 00:40:42.380
not buying it. I like that as your gaslight.

00:40:42.420 --> 00:40:44.480
I think it's a very bold one because you did

00:40:44.480 --> 00:40:46.860
come on this show months ago and we were talking

00:40:46.860 --> 00:40:48.199
about the school bus problem and we're like,

00:40:48.260 --> 00:40:51.360
yeah, we can just geofence school buses. So you've

00:40:51.360 --> 00:40:53.699
essentially called yourself out a little bit

00:40:53.699 --> 00:40:56.219
as a gaslight. I and everyone else have been

00:40:56.219 --> 00:40:58.619
saying geofencing works and we were wrong. Yeah,

00:40:58.619 --> 00:41:00.960
that's... I mean, I wouldn't want to say I was

00:41:00.960 --> 00:41:03.119
wrong until we saw it, but it turns out in the

00:41:03.119 --> 00:41:06.019
real world, it doesn't work. And we had a recall

00:41:06.019 --> 00:41:08.960
based on it not working. Well, so everyone thought,

00:41:09.079 --> 00:41:11.139
including me, that it was going to be good enough.

00:41:12.019 --> 00:41:14.719
We're all wrong. I mean, that's going to be a

00:41:14.719 --> 00:41:16.300
tough gaslight to beat. Michael, what do you

00:41:16.300 --> 00:41:21.639
got? Well, I'm going with Waymo since Fred's

00:41:21.639 --> 00:41:23.659
not here to do it. I'll take Waymo this week.

00:41:24.969 --> 00:41:27.889
And Mike, I'm basically responding to a blog

00:41:27.889 --> 00:41:31.250
post they put out. I believe it was last week

00:41:31.250 --> 00:41:34.670
where July 7th, they are they are proclaiming

00:41:34.670 --> 00:41:37.650
that not all miles are equal. Why time and location

00:41:37.650 --> 00:41:40.949
matter when benchmarking autonomous safety and

00:41:40.949 --> 00:41:43.889
touting a couple of, you know, once again. Hold

00:41:43.889 --> 00:41:46.389
on, Mike. I didn't know that not all miles were

00:41:46.389 --> 00:41:49.929
equal. Really? Really? Oh, come on. I mean, that's

00:41:49.929 --> 00:41:51.670
my whole point here, Phil. You know, you were

00:41:51.670 --> 00:41:56.380
going to win Gaslight, not anymore. This is nonsense.

00:41:56.980 --> 00:42:00.119
Like NHTSA and everyone in the traffic safety

00:42:00.119 --> 00:42:03.159
community have been looking at time and location

00:42:03.159 --> 00:42:06.940
when looking at vehicle safety for decades now.

00:42:07.420 --> 00:42:10.659
And Waymo is kind of coming out here and pretending

00:42:10.659 --> 00:42:12.960
they just discovered this problem and they're

00:42:12.960 --> 00:42:14.880
using it to benchmark their vehicles, despite

00:42:14.880 --> 00:42:18.199
the fact that virtually everyone who has criticized

00:42:18.199 --> 00:42:22.800
Waymo's safety data has pointed out. You know,

00:42:22.800 --> 00:42:26.260
where and when they're conducting these studies

00:42:26.260 --> 00:42:29.960
is a big part of the problem. So just for that,

00:42:30.039 --> 00:42:32.900
I'll end briefly on this or quickly on this.

00:42:33.019 --> 00:42:36.000
You know, this is nonsense. Waymo is pretending

00:42:36.000 --> 00:42:38.139
to have discovered something that we've all known

00:42:38.139 --> 00:42:40.940
about for decades and touting it as some type

00:42:40.940 --> 00:42:43.340
of, you know, something they're doing in their

00:42:43.340 --> 00:42:45.099
safety when, in fact, it's something they should

00:42:45.099 --> 00:42:46.820
have been considering the whole time. This is

00:42:46.820 --> 00:42:48.619
not some new discovery. They should have been

00:42:48.619 --> 00:42:51.170
dealing with this for 10 years or more now. Since

00:42:51.170 --> 00:42:54.349
Fred's not here, I have to ask, so Michael, does

00:42:54.349 --> 00:42:56.510
this mean that miles in a Piggly Wiggly parking

00:42:56.510 --> 00:42:58.670
lot are different than miles in other parking

00:42:58.670 --> 00:43:02.909
lots? You know, since we do not federally track

00:43:02.909 --> 00:43:06.110
at all miles in Piggly Wiggly parking lots, maybe.

00:43:06.190 --> 00:43:09.389
What time of day is it? What season? What's the

00:43:09.389 --> 00:43:11.530
weather like? Is the Piggly Wiggly open? Yeah,

00:43:11.650 --> 00:43:13.489
it's a good point. It has the Piggly Wiggly closed.

00:43:14.300 --> 00:43:16.800
I'm going to my gaslight is going to be just

00:43:16.800 --> 00:43:19.400
like Michael. I'm not going to change, but I'm

00:43:19.400 --> 00:43:21.119
going to slightly different perspective. And

00:43:21.119 --> 00:43:23.179
this is going to be, as I mentioned earlier,

00:43:23.380 --> 00:43:26.960
the myopic view of Silicon Valley software engineers.

00:43:27.260 --> 00:43:30.139
If I'm not aware of it and I just discovered

00:43:30.139 --> 00:43:33.360
it, I am Christopher Columbus. I've discovered

00:43:33.360 --> 00:43:35.219
something brand new and I've dealt with this

00:43:35.219 --> 00:43:38.519
for forever. I remember. 20 plus years ago, some

00:43:38.519 --> 00:43:41.760
software engineer who was really well known said,

00:43:41.860 --> 00:43:46.039
hey, I built this amazing web scraper using Ruby

00:43:46.039 --> 00:43:47.719
on Rails and we'll go out and we'll pull down

00:43:47.719 --> 00:43:49.559
all of your website content and do this thing.

00:43:49.739 --> 00:43:53.199
And and my response was W get, which was just

00:43:53.199 --> 00:43:56.719
like, yeah, this was built 20 or not that 10

00:43:56.719 --> 00:43:59.380
years earlier, like all of this stuff has been

00:43:59.380 --> 00:44:01.400
in there. And that's the view of unfortunately

00:44:01.400 --> 00:44:04.679
what I see a lot with with engineers out in California

00:44:04.679 --> 00:44:08.039
where they don't do. any sort of historical work

00:44:08.039 --> 00:44:11.139
to see what came before, what tools already exist,

00:44:11.320 --> 00:44:13.719
what information's there. So example - Is that

00:44:13.719 --> 00:44:15.760
because if you didn't invent it, you can't sell

00:44:15.760 --> 00:44:18.639
it? You don't even know if it's that cynical.

00:44:18.780 --> 00:44:22.340
I think it's just lazy. It's not reading the

00:44:22.340 --> 00:44:27.699
manual. It's a weird quirk of computer science.

00:44:28.360 --> 00:44:31.039
So for example, if I was a Waymo engineer, if

00:44:31.039 --> 00:44:34.139
I used a tool called Google, I could have put

00:44:34.139 --> 00:44:37.750
in - Traffic crashes by time of day and city

00:44:37.750 --> 00:44:40.849
study. And I discovered, hey, NHTSA has publicly

00:44:40.849 --> 00:44:43.449
made all of this data that Waymo just discovered

00:44:43.449 --> 00:44:49.309
online going back 15, 20 years. So that's my

00:44:49.309 --> 00:44:52.349
gaslight is I stick with Michael and Waymo and

00:44:52.349 --> 00:44:57.489
the lazy, lazy, laziness of people just to use

00:44:57.489 --> 00:45:01.289
their own tools. That's my takeaway. But I'm

00:45:01.289 --> 00:45:04.489
going to have to say that, Phil, you do win because

00:45:04.489 --> 00:45:08.250
you, called out yourself as a gaslighter. Can't

00:45:08.250 --> 00:45:12.369
beat that. Yeah, that's still pretty good. Yeah,

00:45:12.449 --> 00:45:16.030
that's our gaslight rants. I wanted to go back

00:45:16.030 --> 00:45:18.989
to one thing around edge cases before we leave

00:45:18.989 --> 00:45:22.889
that. Because looking at it, you know, we often

00:45:22.889 --> 00:45:25.269
talk about remote drivers here as well. Like,

00:45:25.269 --> 00:45:29.510
is a remote driver... A potential solution, a

00:45:29.510 --> 00:45:33.110
mitigation. What's the role of the remote driver

00:45:33.110 --> 00:45:37.269
and the actual humans here who can, you know,

00:45:37.269 --> 00:45:39.610
as a remote driver can go into a situation that

00:45:39.610 --> 00:45:42.190
the computer and the vehicle can't properly interpret?

00:45:42.409 --> 00:45:44.630
That may be a very common situation that a remote

00:45:44.630 --> 00:45:47.289
driver can and assist the vehicle. I mean, are

00:45:47.289 --> 00:45:50.030
remote drivers almost necessary at this point?

00:45:51.190 --> 00:45:54.170
Yes. That was another topic joke when I covered,

00:45:54.289 --> 00:46:00.869
but let me unpack that, okay? For years, I thought

00:46:00.869 --> 00:46:03.269
that edge cases were the limit to being able

00:46:03.269 --> 00:46:05.349
to deploy Robotaxis, that you had to solve edge

00:46:05.349 --> 00:46:07.750
cases to deploy Robotaxis, and I just didn't

00:46:07.750 --> 00:46:09.809
see how you could do it because of the statistical

00:46:09.809 --> 00:46:12.829
issues. There's just so many rare things with

00:46:12.829 --> 00:46:14.710
high severity. How do you catch them all? It's

00:46:14.710 --> 00:46:17.409
like you're hunting Pokemon. You're playing Pokemon

00:46:17.409 --> 00:46:19.409
Go. There's an infinite number of Pokemon. How

00:46:19.409 --> 00:46:21.639
in the world do you catch them all? And the industry

00:46:21.639 --> 00:46:23.280
was saying, well, we're going to catch enough

00:46:23.280 --> 00:46:25.280
that we're going to just not sweat. If you go

00:46:25.280 --> 00:46:28.980
from 40 ,000 people dead a year to 40 on U .S.

00:46:29.019 --> 00:46:31.440
highways, people are going to suck up the 40

00:46:31.440 --> 00:46:34.460
because it's a thousand times better. That was

00:46:34.460 --> 00:46:36.800
basically their argument. And I said, I don't

00:46:36.800 --> 00:46:38.719
see how you get there, but you're the guys spending

00:46:38.719 --> 00:46:40.500
tens of billions of dollars. Maybe you get there.

00:46:40.659 --> 00:46:43.579
They didn't get there. So they declared defeat

00:46:43.579 --> 00:46:46.519
on edge cases. Edge cases have kicked their butt.

00:46:46.619 --> 00:46:50.389
OK. They can't do it. So what was the plan instead?

00:46:50.550 --> 00:46:53.329
The plan is remote assistance. That every time

00:46:53.329 --> 00:46:54.829
there's an edge case, remote assistance comes

00:46:54.829 --> 00:46:57.570
in to unpack their edge case and fix it for them.

00:46:58.650 --> 00:47:01.110
That's the good news. And if they can make that

00:47:01.110 --> 00:47:04.190
work, that's the only way we get viable robotaxis

00:47:04.190 --> 00:47:06.570
anytime soon. Because edge cases, they just couldn't

00:47:06.570 --> 00:47:08.289
solve them. So they use remote assistance. Remote

00:47:08.289 --> 00:47:11.349
assistants are not a temporary one -year crutch.

00:47:11.469 --> 00:47:15.010
They're a forever thing. Okay. Until they develop

00:47:15.010 --> 00:47:17.309
artificial general intelligence, which doesn't

00:47:17.309 --> 00:47:20.119
exist yet and won't exist. Probably in my lifetime.

00:47:20.340 --> 00:47:23.159
Okay. We'll see. Not in my lifetime. I plan to

00:47:23.159 --> 00:47:26.039
live a long time, but I still think. I'm not

00:47:26.039 --> 00:47:29.099
that old. All right. So we're going to need remote

00:47:29.099 --> 00:47:31.539
assistance forever. And that's okay because it

00:47:31.539 --> 00:47:35.199
isn't about purity of autonomy. It's about making

00:47:35.199 --> 00:47:37.300
a business case that you can deploy the technology

00:47:37.300 --> 00:47:39.500
safely with few enough people you make a profit.

00:47:39.679 --> 00:47:41.719
And I'm okay with that. All right. But here's

00:47:41.719 --> 00:47:45.559
the catch. For this to work, I have an article

00:47:45.559 --> 00:47:47.980
that goes into more detail. At a high level,

00:47:48.019 --> 00:47:49.539
for this to work, three things have to happen.

00:47:50.340 --> 00:47:52.440
The first thing is the car has to know it's in

00:47:52.440 --> 00:47:56.539
trouble. All right. So you've reduced. Let me

00:47:56.539 --> 00:47:58.139
come back to that one. First is the car has to

00:47:58.139 --> 00:47:59.860
know it's in trouble. The second, it has to be

00:47:59.860 --> 00:48:02.260
able to ask for help. And the third is the remote

00:48:02.260 --> 00:48:04.920
person has to be able to help it. All three of

00:48:04.920 --> 00:48:07.119
these are challenges. And we've seen all three

00:48:07.119 --> 00:48:11.239
during the big blackout. We actually saw all

00:48:11.239 --> 00:48:14.480
three of these things be a problem. Okay. So

00:48:14.480 --> 00:48:16.320
the, and during 4th of July, actually during

00:48:16.320 --> 00:48:18.440
4th of July, we saw all three of these, right?

00:48:18.599 --> 00:48:20.639
So it has to know it's in trouble. Well, we saw

00:48:20.639 --> 00:48:22.719
Waymo drive over fireworks because it didn't

00:48:22.719 --> 00:48:25.380
know that it could catch it on fire. And it didn't

00:48:25.380 --> 00:48:28.699
ask, presumably didn't ask for help. Okay. At

00:48:28.699 --> 00:48:30.480
least some of these things, it's not asking for

00:48:30.480 --> 00:48:34.699
help. It's just making a mistake. And we've reduced

00:48:34.699 --> 00:48:37.179
the, it has to know how to drive safely everywhere,

00:48:37.360 --> 00:48:40.059
which is an impossible problem. to the merely

00:48:40.059 --> 00:48:42.340
extremely difficult problem if it has to know

00:48:42.340 --> 00:48:45.099
when it doesn't know what to do, which you still

00:48:45.099 --> 00:48:47.480
have to train it to know not what to do. But

00:48:47.480 --> 00:48:50.380
there's a reason for optimism that knowing that

00:48:50.380 --> 00:48:53.460
it's not in your statistical database is a different

00:48:53.460 --> 00:48:57.219
question than, have you been trained on? Being

00:48:57.219 --> 00:49:00.820
able to say, I'm not sure, is different than

00:49:00.820 --> 00:49:03.159
saying, oh yeah, I know that thing I've never

00:49:03.159 --> 00:49:05.780
seen and I'll get it right. So they have to...

00:49:06.139 --> 00:49:08.579
The emphasis is no longer on training on everything.

00:49:09.260 --> 00:49:12.980
The emphasis is on being able to, with extremely

00:49:12.980 --> 00:49:14.920
high accuracy, know that it hasn't been trained

00:49:14.920 --> 00:49:18.039
on it, ask for help, part one. Okay, this is

00:49:18.039 --> 00:49:20.119
the only way robotaxis and autonomous trucks

00:49:20.119 --> 00:49:22.559
are viable. The only way. Has to know when it's

00:49:22.559 --> 00:49:25.719
in trouble. And I'm going to say every embarrassing

00:49:25.719 --> 00:49:27.559
picture we've seen is a failure of one of these

00:49:27.559 --> 00:49:30.119
three things. It drove into floodwater without

00:49:30.119 --> 00:49:32.099
stopping and asking for help. Well, it didn't

00:49:32.099 --> 00:49:34.420
know it was in trouble, okay? Part two is it

00:49:34.420 --> 00:49:36.559
has to be able to get help. Well, if the communications

00:49:36.559 --> 00:49:39.420
network is overloaded because 4th of July or

00:49:39.420 --> 00:49:41.760
because earthquake or because blackout or because

00:49:41.760 --> 00:49:44.559
whatever, flooding, it's not going to be able

00:49:44.559 --> 00:49:49.659
to get a data link, right? And I'm told that

00:49:49.659 --> 00:49:52.260
you can reserve dedicated bandwidth on data links,

00:49:52.420 --> 00:49:55.380
but they don't do it because money. And by the

00:49:55.380 --> 00:49:57.719
way, you have to worst case that, which brings

00:49:57.719 --> 00:50:00.880
us to number three. which is even if it knows

00:50:00.880 --> 00:50:03.219
it's in trouble and gets itself to a safe waiting

00:50:03.219 --> 00:50:06.400
position before it gets in so deep it can't be

00:50:06.400 --> 00:50:10.280
rescued, right, easily. And it has a data link

00:50:10.280 --> 00:50:12.760
to a remote operator. The remote operator has

00:50:12.760 --> 00:50:15.820
to, first of all, be there. So they have, I think

00:50:15.820 --> 00:50:18.119
Waymo has a few dozen operators on at any time.

00:50:18.280 --> 00:50:20.940
So if 200 robo -taxis all get stuck at the same

00:50:20.940 --> 00:50:23.099
time because blackout, somebody's going to be

00:50:23.099 --> 00:50:26.369
last. One of the emergency responder complaints

00:50:26.369 --> 00:50:29.329
was they waited almost an hour for a remote assistant

00:50:29.329 --> 00:50:31.369
to tell them to connect with them because they

00:50:31.369 --> 00:50:33.769
were overloaded. Right. So so there has to be

00:50:33.769 --> 00:50:36.449
enough of them. And so you need is a call center

00:50:36.449 --> 00:50:39.289
problem. You know, your your crisis is important

00:50:39.289 --> 00:50:41.610
to us. Please wait for the next available. Right.

00:50:41.750 --> 00:50:45.130
It's where you end up. And so but but if people

00:50:45.130 --> 00:50:48.469
can die and in cars in bad situations, you can

00:50:48.469 --> 00:50:51.170
end up there. What do you need to do? Well, you

00:50:51.170 --> 00:50:54.590
need you need to have a surge plan. To have surge

00:50:54.590 --> 00:50:56.710
staffing for when the crisis hits and you can't

00:50:56.710 --> 00:50:58.630
predict it, it's just all of a sudden, I got

00:50:58.630 --> 00:51:01.110
200 cars, they're stuck. What's our plan? And

00:51:01.110 --> 00:51:03.550
even if you do have the operators, they have

00:51:03.550 --> 00:51:07.369
to have the ability to extricate the car. And

00:51:07.369 --> 00:51:09.309
a lot of times that involves driving the car

00:51:09.309 --> 00:51:12.969
out of the way, which is very tricky to do safely

00:51:12.969 --> 00:51:15.809
remotely. And Waymo says, Waymo most of the time

00:51:15.809 --> 00:51:17.489
says we don't do that. That's not true. They

00:51:17.489 --> 00:51:20.449
have some operators who can do it, but they're

00:51:20.449 --> 00:51:23.360
reluctant to use that for good reason. All right.

00:51:23.800 --> 00:51:26.800
Well, so every time you see a firefighter or

00:51:26.800 --> 00:51:29.780
police officer get in a Waymo and move it, that

00:51:29.780 --> 00:51:32.079
means the remote operator wasn't able to fix

00:51:32.079 --> 00:51:34.760
the problem. But if we if we want a million of

00:51:34.760 --> 00:51:36.300
these things instead of 100 of these things,

00:51:36.340 --> 00:51:37.900
they have to be good at knowing when they're

00:51:37.900 --> 00:51:40.380
getting into trouble. There has to be the ability

00:51:40.380 --> 00:51:42.219
to reach out to remote operators and the remote

00:51:42.219 --> 00:51:44.980
operators have to have the ability to get them

00:51:44.980 --> 00:51:47.119
out of the situation without waiting an hour

00:51:47.119 --> 00:51:50.199
for a tow truck. Now, now, if you had a million

00:51:50.199 --> 00:51:53.289
robotaxis. And one robot taxi a day somewhere

00:51:53.289 --> 00:51:57.750
in the U .S. needed a tow truck. Human cars need

00:51:57.750 --> 00:52:00.570
tow trucks, too. I'm OK with that. But that's

00:52:00.570 --> 00:52:02.230
not what we're seeing. What we're seeing is at

00:52:02.230 --> 00:52:04.610
a scale massive enough to really disrupt emergency

00:52:04.610 --> 00:52:07.309
operations. And this is why NHTSA put out that

00:52:07.309 --> 00:52:10.710
letter. Now, we've seen that with these. I think

00:52:10.710 --> 00:52:12.690
it was in Texas where there's the police officer

00:52:12.690 --> 00:52:15.670
trying to get into the Waymo and the Waymo operators

00:52:15.670 --> 00:52:18.130
like, what's your name? What's your badge number?

00:52:18.780 --> 00:52:20.880
Can you repeat that to me? And the cops like

00:52:20.880 --> 00:52:23.860
open the door. I need to move this immediately.

00:52:24.219 --> 00:52:26.880
What was your how do you spell your name? What

00:52:26.880 --> 00:52:28.980
kind of shoes are you wearing? No, it should

00:52:28.980 --> 00:52:31.039
be. Hold your badge up to the camera. OK, you're

00:52:31.039 --> 00:52:34.480
good to go. Right. I mean, it's yeah, it's mind

00:52:34.480 --> 00:52:37.659
blowing. So one more just update before we go

00:52:37.659 --> 00:52:39.960
into recalls. We have an update on one of my

00:52:39.960 --> 00:52:41.440
favorite topics. That's right. New York City

00:52:41.440 --> 00:52:45.190
congestion pricing. So, you know, the big doom

00:52:45.190 --> 00:52:47.110
and gloom New York City congestion pricing, it's

00:52:47.110 --> 00:52:49.590
going to cause the end of humanity. And instead,

00:52:49.710 --> 00:52:51.570
what they've shown is that traffic speeds have

00:52:51.570 --> 00:52:54.849
improved, which is great. Congestion has dropped.

00:52:55.030 --> 00:52:57.710
Traffic to restaurants and businesses has improved.

00:52:57.929 --> 00:53:00.030
Now there's a study. I'm going to quote from

00:53:00.030 --> 00:53:02.530
a Jalopnik article. A new academic paper published

00:53:02.530 --> 00:53:05.090
on Monday found that ambulance travel times within

00:53:05.090 --> 00:53:08.130
the congestion zone have improved by 63 to 70

00:53:08.130 --> 00:53:10.590
seconds. adding to the laundry list of benefits

00:53:10.590 --> 00:53:12.789
of congestion pricing. I hope it's not news to

00:53:12.789 --> 00:53:14.329
you that traffic moves faster when there are

00:53:14.329 --> 00:53:18.550
fewer cars on the road. And 60 to 70 seconds

00:53:18.550 --> 00:53:20.510
doesn't sound like a lot, but when you're trying

00:53:20.510 --> 00:53:23.789
to save somebody's life, that is massive. I've

00:53:23.789 --> 00:53:26.250
spoken to first responders and they've said at

00:53:26.250 --> 00:53:28.110
times the only time they've saved somebody's

00:53:28.110 --> 00:53:30.369
life is because we pick them up and the hospital

00:53:30.369 --> 00:53:32.429
was a block and a half away. They said if the

00:53:32.429 --> 00:53:34.809
hospital was two blocks away, that person didn't

00:53:34.809 --> 00:53:37.869
live. So another huge benefit to congestion pricing.

00:53:39.309 --> 00:53:42.750
Have there been any identifiable negatives to

00:53:42.750 --> 00:53:44.989
congestion? I mean, we've heard and talked about

00:53:44.989 --> 00:53:47.610
a lot of the positives, but have we seen any

00:53:47.610 --> 00:53:51.110
of the objections that were put in front of congestion

00:53:51.110 --> 00:53:53.050
pricing when they were putting it into play?

00:53:53.210 --> 00:53:55.210
I don't even hear. Have we seen any of that happen?

00:53:55.269 --> 00:53:57.170
I don't even hear people complaining about the

00:53:57.170 --> 00:54:00.489
fee. Like, that's the thing. And it's, you know,

00:54:00.510 --> 00:54:05.150
it's nine bucks or whatever. I mean. I, I, I,

00:54:05.250 --> 00:54:07.489
I personally don't hear any complaints about

00:54:07.489 --> 00:54:09.369
it, except when I look at my easy pass bill and

00:54:09.369 --> 00:54:10.670
wonder why my wife's driving into congestion

00:54:10.670 --> 00:54:13.570
zone. Now, sometimes she has to on Saturdays.

00:54:13.889 --> 00:54:17.269
So we heard a lot of objection to this idea when

00:54:17.269 --> 00:54:19.449
it was being propped up and put into play. But

00:54:19.449 --> 00:54:22.949
since then, it's really hard to object to. Yeah,

00:54:22.949 --> 00:54:25.070
it's it's been great. I mean, New Jersey was

00:54:25.070 --> 00:54:27.130
the biggest one who was objecting to it at first.

00:54:27.150 --> 00:54:31.349
And I think they've, you know, their their governor

00:54:31.349 --> 00:54:33.719
is kind of. Been quiet on it now because he's

00:54:33.719 --> 00:54:35.400
like, oh, people on New Jersey transit buses,

00:54:35.699 --> 00:54:38.340
their commutes cut, you know, 20 minutes. They're

00:54:38.340 --> 00:54:41.300
not really going to complain. And with that,

00:54:41.340 --> 00:54:43.199
let's jump into a couple of recalls. How's that

00:54:43.199 --> 00:54:45.739
sound, guys? That sounds great. Sounds great.

00:54:47.440 --> 00:54:52.739
First up, Kia, 462 ,869 vehicles, the 2020 to

00:54:52.739 --> 00:54:57.389
2024 Telluride. And this is an external impact

00:54:57.389 --> 00:55:00.010
with excessive force to the front power seat

00:55:00.010 --> 00:55:02.550
side cover and or seat slide knob may cause the

00:55:02.550 --> 00:55:04.690
power seat switch to become dislodged, internally

00:55:04.690 --> 00:55:07.309
misaligned or otherwise damaged. And this can

00:55:07.309 --> 00:55:10.110
lead to overheating of a seat motor, increasing

00:55:10.110 --> 00:55:12.809
the risk of fire while driving or while parked.

00:55:12.929 --> 00:55:16.289
Oh, no, that's not a good. Is this a park outside?

00:55:16.469 --> 00:55:21.840
No, it's not a park outside. It was a. Oh, it

00:55:21.840 --> 00:55:24.360
is a park outside. I think it is. I think we've

00:55:24.360 --> 00:55:26.460
got two park outsides this week. Hold on. Is

00:55:26.460 --> 00:55:28.719
this a pants on fire recall? Is that what I'm

00:55:28.719 --> 00:55:30.579
hearing? It looks like it is. Yeah, your pants

00:55:30.579 --> 00:55:33.920
will get on fire. Also, it looks like, you know,

00:55:33.920 --> 00:55:37.480
an external impact with excessive force to the

00:55:37.480 --> 00:55:39.840
power seat side cover and slots. I mean, yeah,

00:55:39.960 --> 00:55:42.059
you bumped it. What else? What are they talking

00:55:42.059 --> 00:55:43.980
about there? I assume if a car drove into you,

00:55:44.000 --> 00:55:45.960
but in that case, you're parking outside at a

00:55:45.960 --> 00:55:49.289
repair place. I think it's an edge case because

00:55:49.289 --> 00:55:51.409
you bumped it. We could not possibly have ever

00:55:51.409 --> 00:55:53.469
seen someone bumping it hard enough to do that.

00:55:53.489 --> 00:55:55.690
No one has ever bumped a button in their life.

00:55:56.269 --> 00:56:01.250
Next up, Honda 325 ,588 vehicles, the 2018 to

00:56:01.250 --> 00:56:04.710
2020 Honda Odyssey. Water can enter the rear

00:56:04.710 --> 00:56:07.849
view camera assembly. Yeah, I got you there.

00:56:07.889 --> 00:56:11.010
It's a different rear view camera than we typically

00:56:11.010 --> 00:56:13.429
seen, which is software. This one actually has

00:56:13.429 --> 00:56:15.710
a physical cause, which is water and corrosion.

00:56:16.190 --> 00:56:18.989
Again, edge case. No one water. You know, it

00:56:18.989 --> 00:56:21.570
doesn't go to the path of least resistance ever.

00:56:23.170 --> 00:56:27.329
OK, so this one looks like so owners owners are

00:56:27.329 --> 00:56:28.889
going to hear about this in late August. I think

00:56:28.889 --> 00:56:34.110
they've got to replace a part on the camera.

00:56:34.190 --> 00:56:36.329
So this one. I think they're going to have to

00:56:36.329 --> 00:56:39.230
get a difference. It looks like they are getting

00:56:39.230 --> 00:56:41.690
a different supplier, Sony, to do a different

00:56:41.690 --> 00:56:45.389
design for the component here. So that's actually,

00:56:45.409 --> 00:56:47.349
they're getting that done pretty quick for a

00:56:47.349 --> 00:56:50.909
new component in a little over a month. So compared

00:56:50.909 --> 00:56:53.110
to the four recalls we looked at last week, where

00:56:53.110 --> 00:56:55.929
owners may be waiting until mid -27 to get a

00:56:55.929 --> 00:56:59.610
fix, this is pretty good. Yeah, well, Ford's

00:56:59.610 --> 00:57:01.489
not having any recalls this week. And the last

00:57:01.489 --> 00:57:06.070
recall, Chrysler. 12 ,592 vehicles, the 2026

00:57:06.070 --> 00:57:11.949
Ram 1500. The premium headlamp assemblies may

00:57:11.949 --> 00:57:14.130
experience intermittent parking lamps and daytime

00:57:14.130 --> 00:57:17.570
running lights and illumination. What's happening

00:57:17.570 --> 00:57:19.969
here? Their lights are going on and off. Parking

00:57:19.969 --> 00:57:21.829
lamps and DRLs, which turn off or flicker when

00:57:21.829 --> 00:57:26.550
not intended, may reduce. Okay. Yeah. Oh, so

00:57:26.550 --> 00:57:30.610
it sounds like a short. This looks like another

00:57:30.610 --> 00:57:32.920
one where they're going to have this. Similar

00:57:32.920 --> 00:57:34.519
to last week where it's going to be a while.

00:57:34.699 --> 00:57:37.139
Owners are going to get an interim notification

00:57:37.139 --> 00:57:41.280
at the end of this month. But there is no final

00:57:41.280 --> 00:57:44.800
notification date and no indication of when the

00:57:44.800 --> 00:57:49.079
repair will be available. Oh, boy. And with that,

00:57:49.079 --> 00:57:53.679
thank you, Phil, for joining us. I suggest everyone

00:57:53.679 --> 00:57:57.079
go listen to the Phil and Junko on AV Safety

00:57:57.079 --> 00:58:02.150
podcast. One of our favorites. You know, go to

00:58:02.150 --> 00:58:05.429
autosafety .org and click on Donate. And next

00:58:05.429 --> 00:58:07.630
week, Fred should be back. Is Fred back from

00:58:07.630 --> 00:58:10.010
being a crash test dummy next week? Is he back?

00:58:10.889 --> 00:58:13.409
I assume it depends on how the crash test goes,

00:58:13.409 --> 00:58:15.929
right? Okay, fair enough. And with that, thanks,

00:58:16.030 --> 00:58:18.170
everybody. We'll be back next week. Bye -bye.

00:58:18.289 --> 00:58:23.170
Thank you. Bye. For more information, visit www

00:58:23.170 --> 00:58:25.210
.autosafety .org.