WEBVTT - The Great Google Car Crash of 2016 0:00:04.519 --> 0:00:14.680 Technology. Hey there, and welcome to Tech Stuff. I am 0:00:14.880 --> 0:00:23.040 Jonathan Strickland and dig today together again at last, Scott Benjamin. Hey, 0:00:23.040 --> 0:00:25.880 I'm back joining the show and and I actually have 0:00:25.920 --> 0:00:29.000 you here talking about cars as usual, but thank you. 0:00:29.000 --> 0:00:31.200 You know what, I always appreciate the invite. And uh 0:00:31.240 --> 0:00:33.559 and this one is again right in my wheelhouse here, 0:00:33.600 --> 0:00:37.280 so yeah, I'm excited about It's where I keep my 0:00:37.280 --> 0:00:41.200 wheels in the house. Yeah. So we we're talking today 0:00:41.240 --> 0:00:45.640 about a peculiar event, something that happened in February, and 0:00:45.880 --> 0:00:47.960 it's just taken me this long to finally get around 0:00:47.960 --> 0:00:51.320 and addressing it. You might have heard in previous episodes 0:00:51.360 --> 0:00:55.640 of Tech Stuff about how I would champion the fact 0:00:55.640 --> 0:01:00.760 that Google with their self driving cars had had enormous 0:01:00.800 --> 0:01:05.240 success flawless. You might argue success something like one point 0:01:05.280 --> 0:01:09.760 four five million miles driven without a single accident caused 0:01:09.760 --> 0:01:12.880 by the autonomous system. That there had been about fourteen 0:01:13.080 --> 0:01:16.440 or so accidents, but all of those were either the 0:01:16.520 --> 0:01:20.200 fault of a person manually driving the car in manual 0:01:20.319 --> 0:01:24.959 mode or another driver colliding with the autonomous car, but 0:01:25.080 --> 0:01:28.360 never the fall of the autonomous car itself. It was 0:01:28.920 --> 0:01:33.759 a perfect system until February, and that is the day 0:01:33.800 --> 0:01:35.959 of what what I'll cahol And I really don't mean 0:01:36.000 --> 0:01:39.000 to over dramaticize this at all, So maybe I'm titling 0:01:39.000 --> 0:01:42.680 this episode. I think maybe we should call this the St. 0:01:42.760 --> 0:01:46.399 Valentine's Day Google self driving Car Massacre. That is an 0:01:46.440 --> 0:01:50.480 excellent title and definitely not overly dramatic in any way. Well, 0:01:50.520 --> 0:01:53.520 I mean it's it's also funny because we're recording this, uh, 0:01:53.640 --> 0:01:57.080 the week after I've gotten back from south By Southwest, 0:01:57.160 --> 0:02:00.560 and this was a topic that was discussed heavily at 0:02:00.600 --> 0:02:06.560 south By Southwest because until this incident, it was a 0:02:06.680 --> 0:02:10.560 very easy sell to say autonomous cars are the way 0:02:10.600 --> 0:02:13.840 to go. And then this little accident happened, and and 0:02:13.880 --> 0:02:16.679 it it wasn't terrible. We'll get into the details of 0:02:16.720 --> 0:02:20.680 the accident. But this little accident happened suddenly it sounded 0:02:20.800 --> 0:02:25.639 like Google's autonomous car had caused an enormous pile up 0:02:25.680 --> 0:02:28.560 on the highway. Everyone was much more cautious. So you're 0:02:28.680 --> 0:02:30.760 maybe not buying my alternate title, then, is that what 0:02:30.760 --> 0:02:34.919 you're saying? What's your alternate No, No, that I'm totally no. 0:02:35.040 --> 0:02:36.440 I think it is a massacre, but I think it's 0:02:36.440 --> 0:02:39.400 a massacre in the sense of the public perception of 0:02:39.400 --> 0:02:44.280 autonomous cars. I see, okay, yeah, I'm thinking of from 0:02:44.280 --> 0:02:47.119 a pr stuff. I could I could take this opportunity 0:02:47.160 --> 0:02:51.080 to gloat and say, ah, they're not as infallible as 0:02:51.080 --> 0:02:53.840 you thought they're They're not perfect. But but I'm going 0:02:53.880 --> 0:02:56.679 to take a different stance here in this in this podcast, 0:02:56.720 --> 0:02:58.760 And and I think that as we as we talk 0:02:58.880 --> 0:03:02.360 through this, h we're gonna realize that they've been held 0:03:02.360 --> 0:03:05.080 to a much higher standard than they probably need to 0:03:05.120 --> 0:03:08.600 be right. And I know that's that's tough to to take, 0:03:08.800 --> 0:03:10.280 you know, when you when you just hear it that way, 0:03:10.320 --> 0:03:13.040 But listen to our argument back and forth about this, 0:03:13.200 --> 0:03:16.880 and and understand that they're being held to perfection when 0:03:17.120 --> 0:03:20.960 they probably shouldn't be when humans, I mean, we're not perfect. 0:03:21.000 --> 0:03:26.679 Of course, there's a vast um a chasm between what 0:03:26.960 --> 0:03:30.040 the standards that they're held to versus the standard that 0:03:30.120 --> 0:03:32.240 human test drivers are held to. Sure, yeah, if you 0:03:32.280 --> 0:03:34.920 look at if you look at the standard driving test 0:03:35.360 --> 0:03:38.600 that you have to pass before you get a license, uh, 0:03:38.640 --> 0:03:42.920 I would argue most autonomous cars could likely pass such 0:03:42.920 --> 0:03:46.600 a test as close to flawlessly as you can get, 0:03:47.520 --> 0:03:50.760 but you don't have to be flawless when you take 0:03:50.760 --> 0:03:54.920 a driver's test. There's room for you to not completely 0:03:54.960 --> 0:03:58.600 do something perfectly, like if your parallel parking isn't exactly right, 0:03:58.760 --> 0:04:01.160 you're gonna get points to die did from your total, 0:04:01.360 --> 0:04:04.520 but you may still be you know, high enough, score 0:04:04.560 --> 0:04:07.200 high enough so that you could pass the full driver's test. Exactly. 0:04:07.320 --> 0:04:09.280 Knock over cone. It's not really a big deal, but 0:04:09.400 --> 0:04:12.560 an autonomous car knocks over a cone, everybody points at 0:04:12.560 --> 0:04:15.920 it and says, look at that thing, it's it's pilot junk. Yeah, exactly. 0:04:16.800 --> 0:04:19.600 It's interesting that you point that out too, because that 0:04:19.640 --> 0:04:22.040 ties into a different discussion. I saw it south by 0:04:22.080 --> 0:04:25.560 Southwest that wasn't specifically about autonomous cars. It was about robots. 0:04:25.760 --> 0:04:27.279 So this is a little bit of a tangent, but 0:04:27.360 --> 0:04:30.160 it it goes to illustrate the point you just made. 0:04:31.040 --> 0:04:34.640 So this this UH panel about robots. There was a woman, 0:04:34.920 --> 0:04:37.640 Layla taka Yama, who used to work for Google X 0:04:37.880 --> 0:04:41.520 but not in the autonomous car division. She talked about 0:04:41.560 --> 0:04:46.440 how she ran a a an experiment. She got a 0:04:46.480 --> 0:04:49.719 guy from Pixar to do a series of very simple 0:04:49.800 --> 0:04:54.320 animations to show people, the interactions between a robot and 0:04:54.400 --> 0:04:59.080 a person, and then to judge which robot is considered 0:04:59.400 --> 0:05:02.840 UH in elligent versus not intelligent. And the whole point 0:05:02.839 --> 0:05:06.200 of this was to show the differences between succeeding and failing, 0:05:06.360 --> 0:05:11.000 but giving no indication that the robot understands it succeeded 0:05:11.080 --> 0:05:15.159 or failed, or building in expressions for the robot to 0:05:15.560 --> 0:05:19.920 follow a success or failure UH to indicate it quote 0:05:20.000 --> 0:05:24.680 unquote understands what happened. And it was fascinating because they 0:05:24.720 --> 0:05:28.920 showed a very simple UH experiment with a robot trying 0:05:28.960 --> 0:05:31.000 to open the door. And again this is an animation, 0:05:31.880 --> 0:05:35.160 so there was a different scenarios. There's one where the 0:05:35.240 --> 0:05:38.840 robot opens the door and the door opens and then 0:05:38.839 --> 0:05:40.880 the robot just sits there and it's it's done because 0:05:40.880 --> 0:05:43.200 it's done what it was supposed to do. There's one 0:05:43.240 --> 0:05:45.919 where the robot opens the door and then kind of 0:05:45.920 --> 0:05:49.920 perks up like, oh, I did what I wanted to do. 0:05:50.320 --> 0:05:52.719 There was one where the robot fails to open the 0:05:52.760 --> 0:05:56.000 door and then does nothing UH. And then there was 0:05:56.040 --> 0:05:58.200 one where the robot fails open the door and then 0:05:58.240 --> 0:06:00.080 slumps down a little bit as if to say, oh, 0:06:00.120 --> 0:06:03.440 I'm disappointed I didn't succeed. They then asked people to 0:06:03.520 --> 0:06:05.880 judge which robots they thought were the most intelligent, and 0:06:06.000 --> 0:06:10.520 everyone said the robot that failed but showed disappointment was 0:06:10.600 --> 0:06:13.400 more intelligent than the robot that succeeded but didn't show 0:06:13.440 --> 0:06:16.960 any expression at all. And when you think about that again, 0:06:17.120 --> 0:06:21.400 it's holding robots to a standard that doesn't necessarily apply 0:06:21.480 --> 0:06:25.320 to them because of the human element this human robot interaction. 0:06:26.160 --> 0:06:31.200 We're holding autonomous cars to a similar standard that perhaps 0:06:31.360 --> 0:06:34.560 is not fair. We're holding robots to a standard that's 0:06:34.600 --> 0:06:37.279 not fair, But that means that people who are designing 0:06:37.279 --> 0:06:40.359 autonomous cars and people who are designing robots have to 0:06:40.440 --> 0:06:44.480 take that into consideration because that's the way humans are. Well, 0:06:44.520 --> 0:06:48.640 this is interesting because you're you're mentioning specifically imitating human behavior. 0:06:49.080 --> 0:06:52.040 And this comes up in an article that I read 0:06:52.160 --> 0:06:55.080 in I let't see it was in the Verge, and 0:06:55.200 --> 0:06:57.880 the Verge you might have read the same thing article. 0:06:58.120 --> 0:07:00.120 It really is and a person by the name of 0:07:00.360 --> 0:07:03.640 Jennifer Haroun. She's the head of business operations for Google 0:07:03.680 --> 0:07:06.080 self driving project. And by the way, let's come back 0:07:06.120 --> 0:07:07.720 to the details of the accident and just a moment, 0:07:07.920 --> 0:07:10.520 we'll describe what happened. But she says that, well, you 0:07:10.560 --> 0:07:12.560 know what, maybe I need to back this up just 0:07:12.600 --> 0:07:15.160 a second here. How about this, Let's describe the accident 0:07:15.160 --> 0:07:17.320 and then we'll talk about what she said, because it 0:07:17.640 --> 0:07:21.040 plays perfectly into it, so and helping understand what happened. 0:07:21.080 --> 0:07:23.040 So here's here's how you can imagine it. All right, 0:07:23.080 --> 0:07:27.000 You've got an intersection, uh in Mountain View, California, which 0:07:27.040 --> 0:07:30.200 is where Google's headquarters is located, and you've got the 0:07:30.880 --> 0:07:32.960 Google self driving car. Correct me if I get any 0:07:33.000 --> 0:07:35.440 of this wrong, Scott, I'm going from memory. I'm doing 0:07:35.560 --> 0:07:37.960 mostly the same. It's so the Google self driving cars 0:07:37.960 --> 0:07:39.680 in the right lane and at once it's playing on 0:07:39.800 --> 0:07:43.080 making a right turn at this intersection. Now, at the 0:07:43.280 --> 0:07:45.880 corner of the intersection, there were some sandbags that were 0:07:46.040 --> 0:07:48.960 a partial obstruction of the lane. Then I think you're 0:07:48.960 --> 0:07:52.160 blocking a sewer entry a great maybe or something. Right, 0:07:52.240 --> 0:07:56.840 So the Google car detected that there was an obstruction, 0:07:57.320 --> 0:07:59.680 and so it had to plan an alternate way to 0:07:59.760 --> 0:08:02.040 make its right turn. It still wanted to follow the 0:08:02.120 --> 0:08:06.040 route that it had planned, so the change would have 0:08:06.040 --> 0:08:08.400 been for it to kind of edge into the next 0:08:08.480 --> 0:08:10.800 lane over, the next lane to the left before making 0:08:10.800 --> 0:08:16.280 a right turn. Behind the Google car, approaching at a 0:08:16.400 --> 0:08:19.960 blistering speed of fifteen miles per hour, was a bus. 0:08:20.400 --> 0:08:24.000 And so the Google car recognized there was a bus coming. 0:08:24.880 --> 0:08:26.560 It was moving at a very slow speed at two 0:08:26.560 --> 0:08:31.040 miles per hour. The Google car said, well, based upon 0:08:31.120 --> 0:08:34.000 my programming, what I should expect happen is that the 0:08:34.000 --> 0:08:37.320 bus will slow down, allow me to move through. I'll 0:08:37.320 --> 0:08:40.920 clear the intersection, the bus will continue. What actually happened 0:08:41.040 --> 0:08:43.800 was the Google Car made the move into the lane, 0:08:44.200 --> 0:08:47.200 the bus continued, and there was a low speed collision 0:08:47.440 --> 0:08:49.560 and there were no injuries. No one was hurt. There 0:08:49.640 --> 0:08:52.040 was in fact a driver behind the wheel of the 0:08:52.080 --> 0:08:55.080 Google car. It's just the driver wasn't in control. The 0:08:55.120 --> 0:08:58.800 autonomous system was in control. And some people might say, well, 0:08:58.840 --> 0:09:01.200 you know, the bus driver just a let the car end. 0:09:01.559 --> 0:09:04.439 But Google actually said, and this is important. Google came 0:09:04.440 --> 0:09:08.160 out and said, we accept responsibility for this. This is 0:09:08.200 --> 0:09:12.000 something that uh, it's it's valuable that this information has 0:09:12.040 --> 0:09:13.560 come to light because it means that we need to 0:09:13.640 --> 0:09:18.680 revisit this particular part of the autonomous car programming. Now 0:09:19.480 --> 0:09:21.960 I thought that was really interesting. First of all, I 0:09:21.960 --> 0:09:25.120 have never heard of a company accepting responsibility for something 0:09:25.160 --> 0:09:28.440 so fast in my life. Well they did, and they didn't. 0:09:28.440 --> 0:09:30.240 I mean, there's a couple of versions of this. Now, 0:09:30.440 --> 0:09:32.600 you got the details of the accident, correct, Although I 0:09:32.600 --> 0:09:35.040 did hear, and this this is a bit confusing. I 0:09:35.080 --> 0:09:37.160 didn't hear that the lanes in this particular part of 0:09:37.200 --> 0:09:40.120 town are extremely wide, and so what happened was the 0:09:40.240 --> 0:09:43.319 car uh kind of edged itself over toward the curb. 0:09:43.840 --> 0:09:48.080 So it was I guess mimicking human behavior again. And 0:09:48.280 --> 0:09:49.880 I'll get to that in just a minute. But um, 0:09:49.920 --> 0:09:52.160 you know, once, once this accident happened and they said, 0:09:52.200 --> 0:09:54.840 you know, we do need to investigate this. They did 0:09:54.840 --> 0:09:57.840 that to the tune of about thirty five hundred new 0:09:57.920 --> 0:10:01.080 tests that they have now implemented since this accident, that said, 0:10:01.440 --> 0:10:03.040 we're gonna watch for this. You know that we need 0:10:03.120 --> 0:10:06.520 to understand a little more deeply that some of these 0:10:06.600 --> 0:10:09.400 larger vehicles may have a different and more difficult times 0:10:09.400 --> 0:10:12.760 stopping in traffic than a smaller vehicle will. And and 0:10:12.840 --> 0:10:14.920 that's the reason why some of these bigger vehicles like 0:10:15.000 --> 0:10:17.520 to continue on their path and think, well, maybe someone 0:10:17.559 --> 0:10:20.760 behind me will let them in. Google did say we 0:10:20.760 --> 0:10:24.000 were relying on an element of human kindness UM to 0:10:24.120 --> 0:10:26.360 let us into that lane. And that's normally what happens. 0:10:26.360 --> 0:10:28.079 It really does. Usually there's a back and forth or 0:10:28.120 --> 0:10:30.640 you know, maybe there's always gonna be that off handed 0:10:30.679 --> 0:10:33.000 time where you know someone does cut through and they're 0:10:33.040 --> 0:10:34.719 just like they're like, no, I need to get through 0:10:34.760 --> 0:10:36.920 that intersection in this light cycle, and no one is 0:10:36.920 --> 0:10:38.680 gonna stop me. Yeah. I mean I'm gonna be ten 0:10:38.679 --> 0:10:40.679 feet ahead of you when all this happens. And that's 0:10:40.720 --> 0:10:43.000 where I you know, that's that's my goal. But uh, 0:10:43.440 --> 0:10:46.000 usually what happens is it's an alternating pattern and they 0:10:46.000 --> 0:10:48.320 expected that to happen, and it didn't happen in this case. 0:10:48.360 --> 0:10:50.720 And this is where it plays right into what UM 0:10:51.120 --> 0:10:53.880 Jennifer had mentioned. Jennifer Harun, who is the head of 0:10:53.880 --> 0:10:58.200 business operations for Google self driving project, explained at and 0:10:58.200 --> 0:10:59.959 I think she was at the south By Southwest Kind 0:11:00.040 --> 0:11:03.319 Prince as well, and she said that the Lexus, it's 0:11:03.320 --> 0:11:05.679 a Lexus vehicle. It was outfitted with this gear so 0:11:05.720 --> 0:11:08.440 that it struck the bus in part because it was 0:11:08.480 --> 0:11:11.679 imitating human behavior. And that's I found that interesting. That 0:11:11.720 --> 0:11:15.400 she she kind of is deferring the fault here and 0:11:16.240 --> 0:11:18.720 to say, well, we were just imitating what we see 0:11:18.720 --> 0:11:20.720 on the streets and that's that's kind of what happened. 0:11:20.760 --> 0:11:23.839 So it's a it's a double deferral in a way, right, 0:11:23.880 --> 0:11:26.080 because first they say we were counting on an element 0:11:26.080 --> 0:11:28.080 of human kindness, which is already kind of a deferral 0:11:28.120 --> 0:11:31.720 in itself. You're you're essentially saying, well, we were concert 0:11:31.720 --> 0:11:33.640 we were thinking that the bus driver would be a 0:11:33.679 --> 0:11:37.280 decent human being and not a Lexus hating scumbag. That's 0:11:37.440 --> 0:11:40.920 I'm paraphrasing what they said. Obviously, I'm taking a little liberty. 0:11:40.960 --> 0:11:44.040 And then in this one you're saying she's also saying, well, 0:11:44.280 --> 0:11:46.680 we designed the car to behave the way we see 0:11:46.920 --> 0:11:50.679 actual cars behaving on the road. So in both cases 0:11:50.720 --> 0:11:53.440 you're almost it's a little bit of backing away from 0:11:53.520 --> 0:11:56.920 taking full responsibility exactly. And this imitating human behavior that 0:11:56.960 --> 0:11:59.760 she's talking about was that they had recently taught the 0:12:00.000 --> 0:12:02.319 eals to hug the right hand side of that lane 0:12:02.720 --> 0:12:05.320 when they're making that right hand turn, and that that's 0:12:05.320 --> 0:12:08.880 when it's encountered the sandbags that were unexpected, and so 0:12:09.080 --> 0:12:10.760 this is what I find interesting is if it's a 0:12:10.760 --> 0:12:13.320 wide lane and she's saying that it was hugging the 0:12:13.400 --> 0:12:15.840 right hand side of that lane, trying to make the 0:12:15.880 --> 0:12:18.560 turn as most humans do. If it was in the 0:12:18.600 --> 0:12:20.880 center of the lane, she's saying, if it had just 0:12:20.920 --> 0:12:22.959 behaved as they normally would do it, you know that 0:12:23.000 --> 0:12:24.880 it would be in the exact dead center of that lane. 0:12:25.240 --> 0:12:27.520 The bus wouldn't have had the gap, I guess, and 0:12:27.520 --> 0:12:29.520 try to try to make it through that gap, so 0:12:29.840 --> 0:12:32.080 it would have just been behind the car. Never would 0:12:32.080 --> 0:12:34.840 have happened. So she's saying, in effect, because we're trying 0:12:34.880 --> 0:12:37.040 to make it mimic human behavior, and we were hugging 0:12:37.040 --> 0:12:40.240 that right side, that's why this accident happened, maybe we 0:12:40.280 --> 0:12:42.439 should have done that. But then again they come back 0:12:42.480 --> 0:12:46.319 and say that that's absolutely necessary for them to mimic 0:12:46.400 --> 0:12:48.959 human behavior, because if they don't, that causes trouble as well. 0:12:48.960 --> 0:12:51.040 There's proather issues that are right, like if a if 0:12:51.040 --> 0:12:55.120 a vehicle, let's say that it's an autonomous car heading 0:12:55.160 --> 0:12:58.560 toward an intersection, the light goes from green to amber, 0:12:59.440 --> 0:13:04.760 and there's technically enough space for the car to break 0:13:05.320 --> 0:13:08.679 safely and come to a complete stop as the light 0:13:08.760 --> 0:13:13.640 turns red, knowing that most humans would just gun it, 0:13:14.559 --> 0:13:16.360 or at least just continue at the same speed to 0:13:16.360 --> 0:13:20.120 go through the intersection while it's still amber. You might 0:13:20.240 --> 0:13:22.920 want to think about that when you're designing your autonomous 0:13:22.920 --> 0:13:25.480 car so that you don't cause a pile up behind 0:13:25.520 --> 0:13:29.160 you like you don't. If the person directly behind the 0:13:29.200 --> 0:13:31.520 autonomous car expects the car in front of them to 0:13:31.559 --> 0:13:34.839 continue through the intersection, you could potentially get rear ended. 0:13:35.400 --> 0:13:37.680 That happens a thousand times around. I mean more than that, 0:13:37.760 --> 0:13:41.199 but it happens all over the world, especially in Atlanta, 0:13:41.320 --> 0:13:43.520 where the rule is that the light turns red, three 0:13:43.559 --> 0:13:46.679 cars get to go through. It is so true, isn't it. 0:13:46.960 --> 0:13:49.240 It seems like once one goes through, two more follow you. 0:13:49.840 --> 0:13:53.000 That's crazy. Yeah, I've seen it happen in multiple places 0:13:53.040 --> 0:13:56.120 around the city. There's some neighborhoods where it's worse. I'm 0:13:56.120 --> 0:13:58.800 not gonna name any names, Buckhead, but I'm just saying, 0:13:59.240 --> 0:14:02.400 never accelerate immediately on a green light anywhere in this area, 0:14:02.440 --> 0:14:05.440 because you can expect there's gonna be that that odd 0:14:05.440 --> 0:14:07.240 ball car that comes through it, right, and there's gonna 0:14:07.240 --> 0:14:09.400 be like a car that's been waiting to turn left 0:14:09.440 --> 0:14:11.240 the whole time, and they say, I'm not waiting another 0:14:11.320 --> 0:14:13.520 light cycle, I'm just going now, like even if they're 0:14:13.520 --> 0:14:15.800 behind the stop line. Yeah, I'm sure. I'm sure it 0:14:15.800 --> 0:14:19.960 happens everywhere. It's it's particularly bad in these congested areas. Yes, 0:14:20.360 --> 0:14:24.160 So to your point, it is important to take those 0:14:24.160 --> 0:14:26.880 things into consideration when designing the autonomous car. You don't 0:14:26.920 --> 0:14:29.720 want an autonomous car to drive like an inconsiderate jerk 0:14:29.720 --> 0:14:31.880 of a driver. But at the same time, you can't 0:14:31.960 --> 0:14:36.080 have it be so clinically precise that it is standing 0:14:36.120 --> 0:14:38.760 out from all the other drivers. The only way that 0:14:38.800 --> 0:14:40.840 works is if you get to a point where you're 0:14:40.880 --> 0:14:43.760 at a saturation point with autonomous cars on the road, 0:14:44.240 --> 0:14:46.960 where then you can affect the behavior on a mass 0:14:47.000 --> 0:14:50.800 scale across a fleet of cars and not have that 0:14:50.960 --> 0:14:56.240 issue of human drivers having awful interactions with robotic drivers exactly. 0:14:56.320 --> 0:14:59.120 Here's here's the way they state at these These spokesman 0:14:59.160 --> 0:15:01.520 stated and say it's vital for us to develop advanced 0:15:01.520 --> 0:15:04.320 skills that respect not just the letter of the traffic code, 0:15:04.640 --> 0:15:07.160 but the spirit of the road. I think that's a 0:15:07.160 --> 0:15:08.600 good way to put it. The spirit of the road. 0:15:08.640 --> 0:15:11.160 I understand that. I completely get that when I read 0:15:11.200 --> 0:15:14.240 it is that, Yeah, there's little rules here and there 0:15:14.280 --> 0:15:16.880 that we'd bend, but everybody bends them, and you you 0:15:17.040 --> 0:15:19.800 expect you you understand how other drivers are gonna behave 0:15:19.800 --> 0:15:22.560 in the same situation, and you expect that to happen, 0:15:23.080 --> 0:15:25.240 and you behave in that way, and it all works. 0:15:25.480 --> 0:15:28.160 But when something comes in, a spoiler comes in, um 0:15:28.240 --> 0:15:30.400 and it follows exactly to the letter of the log 0:15:30.480 --> 0:15:33.479 the way that's supposed to happen, that person maybe the 0:15:33.560 --> 0:15:36.760 you know, the standout right. Another great example of that 0:15:36.800 --> 0:15:39.400 in Atlanta would be, uh. We have a couple of 0:15:39.400 --> 0:15:41.840 different highways that run through the city and one that 0:15:41.960 --> 0:15:46.960 runs around the city and is often thought of as 0:15:47.040 --> 0:15:48.920 the type of highway that if you get on it, 0:15:48.960 --> 0:15:51.480 you have to speed. You cannot go the speed limit 0:15:51.480 --> 0:15:54.080 on two five. It's just too dangerous because everyone else 0:15:54.160 --> 0:15:57.240 is going above the speed limit. Massive truck traffic, yeah 0:15:57.280 --> 0:16:01.400 and yeah, and there are enormous, enormous semis uh rushing 0:16:01.440 --> 0:16:03.120 down there, and you don't want to get you don't 0:16:03.120 --> 0:16:05.720 want to be poking along when they come up behind you. 0:16:06.040 --> 0:16:09.440 So again, an autonomous car would need to have that 0:16:09.480 --> 0:16:12.640 information and take that into account unless you've got to 0:16:12.680 --> 0:16:14.960 a point where you had so many autonomous vehicles on 0:16:15.000 --> 0:16:18.160 the road that it was no longer a concern. Yeah, 0:16:18.200 --> 0:16:20.200 and this is where we we discussed this yesterday because 0:16:20.200 --> 0:16:22.000 we're talking off air about this just a little bit 0:16:22.000 --> 0:16:25.320 too prep for today. And uh, the idea would be 0:16:25.400 --> 0:16:28.240 that's kind of like schooling, almost like fish schooling, UM, 0:16:28.360 --> 0:16:30.680 and that the cars know where the other one is 0:16:30.720 --> 0:16:32.840 at all times and they can communicate between them. The 0:16:32.880 --> 0:16:35.360 problem is when you throw in the human driver element 0:16:35.400 --> 0:16:38.200 into that mix, or you know, if if if you 0:16:38.240 --> 0:16:40.960 have just one autonomous car among all humans. That's the 0:16:41.040 --> 0:16:43.240 other problem with the other issue, and right now, that's 0:16:43.280 --> 0:16:45.480 the battle that they're fighting right right. So once we 0:16:45.480 --> 0:16:47.840 get to a point where there's that that tipping point 0:16:48.320 --> 0:16:51.880 one way or the other, then things will be very different. 0:16:51.960 --> 0:16:55.000 But there's going to be some growing paints. And this 0:16:55.080 --> 0:16:57.800 also leads into something that I talked about earlier in 0:16:59.360 --> 0:17:01.680 UM when I went to ce S Toyota had their 0:17:01.680 --> 0:17:07.240 big AI discussion. You know, they're they're investing millions of 0:17:07.240 --> 0:17:12.120 dollars in AI research for autonomous cars and beyond. And 0:17:12.200 --> 0:17:15.280 one of the things they've talked about was how autonomous 0:17:15.320 --> 0:17:18.920 cars in general are really really good at handling all 0:17:18.960 --> 0:17:22.880 the mundane stuff that you would typically encounter on a 0:17:22.920 --> 0:17:26.239 normal day driving from point A to point B. What 0:17:26.320 --> 0:17:29.320 they are not good at is dealing with stuff that's 0:17:29.400 --> 0:17:33.000 outside of that norm, and the sandbags that we talked 0:17:33.000 --> 0:17:36.400 about earlier would be a great example of that. It's 0:17:36.440 --> 0:17:40.040 some form of obstruction that's partially blocking off part of 0:17:40.040 --> 0:17:46.119 the road and that ends up causing a different scenario. 0:17:46.640 --> 0:17:51.439 And sometimes the card behaves in a way that works 0:17:51.440 --> 0:17:54.440 out for everybody. In this case, it didn't, And it's 0:17:54.480 --> 0:17:57.680 not it's not that the car couldn't handle the situation. 0:17:57.720 --> 0:18:00.399 It's just that the method that the car used turned 0:18:00.400 --> 0:18:03.760 out to not be reliable. Yeah, this was an extremely 0:18:04.040 --> 0:18:06.800 slow speed crash, as we've said, and the bus was 0:18:06.840 --> 0:18:09.080 traveling fifteen miles per hour in the other lane trying 0:18:09.119 --> 0:18:11.439 to get through that gap. But the Google car was 0:18:11.480 --> 0:18:14.320 traveling i think they said to two miles per hour. Yeah, 0:18:14.480 --> 0:18:18.080 very very slow, very slow speed. So the thing is 0:18:18.160 --> 0:18:21.240 like with the with the the compensation for this, you know, 0:18:21.280 --> 0:18:23.800 the thirty five hundred tests that they're now going to 0:18:23.880 --> 0:18:27.400 run additional tests, uh to determine or to to find 0:18:27.440 --> 0:18:30.520 a way around that situation. Said, it's never gonna happen again. 0:18:31.000 --> 0:18:33.720 We're gonna do everything we can, but to to think 0:18:33.720 --> 0:18:36.360 about it that way, to say, the thirty five hundred 0:18:36.359 --> 0:18:39.120 tests that are gonna allow this vehicle to to think 0:18:39.160 --> 0:18:42.320 about that exact situation and never let it happen again. 0:18:42.400 --> 0:18:44.720 Where where just kind of noses out into uh, into 0:18:44.840 --> 0:18:48.040 a lane that appears open. That's remarkable. I mean it 0:18:48.119 --> 0:18:50.600 just lets you know that, uh, they're there are tens 0:18:50.640 --> 0:18:55.119 of thousands of of UM programs are or or thoughts, 0:18:55.200 --> 0:18:57.080 I don't know how to say it, that are they're 0:18:57.119 --> 0:18:59.600 going through this thing at all times. You know, um, 0:18:59.640 --> 0:19:03.159 all these calculations, calculations and parameters and just you know, 0:19:03.520 --> 0:19:05.800 if this then that you know, those scenarios are being 0:19:05.880 --> 0:19:08.359 run all the time. It's just incredible, mind boggling, it 0:19:08.400 --> 0:19:10.320 really is. And and I was looking into you said 0:19:10.359 --> 0:19:13.479 one one point four or five million miles have been driven, 0:19:14.080 --> 0:19:16.880 uh flawlessly, right, I mean they hadn't have any problems, 0:19:17.160 --> 0:19:19.840 you know at at fault. I guess the autonomous vehicle 0:19:20.440 --> 0:19:23.680 Do you know how much they test on a daily 0:19:23.680 --> 0:19:27.159 basis and on actually on a daily basis. Okay, well, 0:19:27.200 --> 0:19:28.520 let's see, I've got I got a note here I 0:19:28.520 --> 0:19:31.320 should have looked for that reading. Okay, here we go. 0:19:31.440 --> 0:19:33.280 All right, So, actually this is a per week and 0:19:33.320 --> 0:19:37.200 then a per day thing. They drive ten thousand miles 0:19:37.240 --> 0:19:39.399 per week and that's like, you know, somebody in a 0:19:39.480 --> 0:19:43.360 vehicle on the road, ten thousand miles per week. Per day, though, 0:19:43.400 --> 0:19:47.920 this number is incredible. Per day, they're driving three million 0:19:47.960 --> 0:19:52.440 miles of computer simulation miles, so three million the equivalent 0:19:52.480 --> 0:19:54.960 of three million miles. That's because they can quickly just 0:19:54.960 --> 0:19:58.040 go through that and have multiple systems running these things, 0:19:58.040 --> 0:20:00.800 you know. So so the amount of testing that they 0:20:00.840 --> 0:20:02.919 do in a year is just unbelievable. I don't have 0:20:03.040 --> 0:20:05.560 yearly stats or anything, but you can extrapolate those numbers 0:20:05.560 --> 0:20:08.560 to that. Well and and uh, the other point in 0:20:08.640 --> 0:20:11.679 the Toyota press conference that was interesting to me, and 0:20:11.720 --> 0:20:13.080 this goes back to what you were saying at the 0:20:13.119 --> 0:20:15.679 beginning of the show about holding autonomous cars at a 0:20:15.720 --> 0:20:19.760 different standard than we hold human drivers. Uh. They talked 0:20:19.760 --> 0:20:24.880 about how you offer a lot of the autonomous car 0:20:24.920 --> 0:20:30.000 industry talks about the hundred million mile um benchmarks, saying 0:20:30.040 --> 0:20:33.200 that you want you want a hundred million miles traveled 0:20:33.920 --> 0:20:37.560 of proven safety, and they said, you know that's not enough. 0:20:37.600 --> 0:20:40.119 You need to go much bigger than that, a hundred 0:20:40.160 --> 0:20:43.880 billion miles. And I thought, wow, that is I mean, 0:20:43.960 --> 0:20:47.199 I get it for you want that many miles so 0:20:47.240 --> 0:20:52.160 that you can encounter as many possible different situations as 0:20:52.160 --> 0:20:56.400 you might encounter on the road. Because obviously, if you 0:20:56.840 --> 0:20:59.399 if you plan a system and it's great for handling 0:21:00.040 --> 0:21:03.440 percent of the situations, that's fine until you run into 0:21:03.480 --> 0:21:06.120 that one percent. And when you do figure out how 0:21:06.160 --> 0:21:08.960 many cars are on the road in the United States 0:21:09.040 --> 0:21:14.240 alone traveling on any given day, you realize the odds 0:21:14.280 --> 0:21:18.440 are Eventually, I mean, statistics show statistics proved like odds are, 0:21:19.480 --> 0:21:22.160