Tim: Hi Peter, how are you doing today?
Peter: Hey, great.
Tim: So your book has finally been published, and I’ve got a copy right here. So you’ve got, as we’ve discussed before in our previous conversations, your book is written in small segments, and one of the things that you talked about is safety. And I would like to talk a little bit about how things like autonomy in cars is going to affect road safety. So on that basic topic, what are some things that spring to mind about it?
Peter: Well, it is a big question because when you look at your computers, and computer scientists say the word ‘crash’ all the time, and they don’t mean anything bad about it. I think a lot of people are a little bit nervous that these same guys are going to be in charge of your car. But I think the good news is that the car industry has been using computers for a long time, and we don’t really have any stories about the computers causing trouble. And there is a lot of reason why that is: The computers in the car are updated differently. We don’t let people just download anything off the web into their car, and so they really are a lot more stable.
We have a long history of having computers in the car, so I am not as worried about that. You know, we really won’t know until we start rolling out the experiment for a while, but the good news is that the computers in the car, the robot drivers, don’t need to be perfect, they just need to better than humans. So they are not going to be texting at the same time, they are not going to be distracted because their boss yelled at them, they are not going to have kids in the backseat – there are so many things that the computers won’t have to worry about that I think it is possible that we are going to be happy with the computers even if they make an occasional mistake.
Tim: And sometimes there probably will be fatal mistakes, and that’s going to be a big issue the first time somebody is just run through by a car driven by a computer, because it will be an interesting question to know who gets the blame and how? What portion goes to the programmer who wrote an algorithm 20 years before, and what part goes to a manufacturer that sold it? What part is going to be assigned to the different actors involved? Because it certainly isn’t the driver and pedestrian. Now there are a lot of people involved.
Peter: You know, you are absolutely right. But the interesting thing is that I think that it is going to be easier than it is dealing with it when all the humans are involved. Because the way we do it currently with the human drivers is when we assume that the human has a lot of responsibility and then we get into these arguments – was it the driver? Was the driver 35 percent responsible and the pedestrian 65 percent responsible? And we get into that.
And what you notice is that a lot of states have said, we can’t assign blame, we are just going to have no fault insurance, and we are going to take care of everybody, we are going to have a big pool, and we are going to pay for it out of that, and that is going to be how we deal with the humans making a mistake. And I think you are right that there are going to be you can have these endless arguments, was it the library that the programmer used? Was it this program or was it that program? Or whether everybody in that scrum session who came up with the wrong idea?
But the interesting thing is that you are still going to have like one big corporation on the top of this. And we’ve dealt with this in the past, we deal with it in the pharmaceutical industry - the pharmaceutical industry has a pool of money that they set aside to care of mistakes. And as a society, we say okay, if this is good enough for a significant percentage and we can handle mistakes well enough, then it is okay. And I think that’s what is going to happen here. We are going to see a couple of corporations evolve, they are going to have those risk pools, and it is going to be simpler than the risk pools that we use with the current insurance system.
Tim: Do you think that the risk pools are going to be something that the car industry will come up with incrementally as we see more and more intelligence built into cars? Like right now we have, as you point out, a lot of software in cars, but it stops before the steering wheel, and that is a pretty big gap to go across.
Peter: You know you are right about the steering wheel, but the interesting thing is that a lot of cars have a computer in between the brake pedal and the brakes already - and that is the antilock brake system. And those systems are really simple, but they do turn the brakes on and off and on and off and on and off really quickly; so you also view it as a philosopher either way – is it the human that is pushing the brake pedal? Or is it the computer that is firing the brakes on and off rapidly? So we’ve tolerated a very simple system with the brakes and it is possible, we will end up tolerating the computers expanding their control.
Tim: There are a lot of things that the robots are probably going to do if not better like parking that we may see plenty of trade toward the positive, because I think things like passing at speed is something where a lot of accidents happen, and congested areas that a computer I should say won’t be distracted in the same way as a human.
Peter: You know you are right. We can end up with a system I think where society has a lot more central control – right now the roads are a chaos, and the nice thing is that since humans are relatively smart, and we have some basic rules of the road the chaos is tamed most of the time. But we can do an even better job of that. What happens is that if you look at a holiday weekend, everyone tries to funnel into the throughways and then you get deadlock.
Well, if you just have a simple reservation system up, well that can solve a lot of that problem. And you can have the roads flow. So we have a lot of opportunity for central control. And the other thing is that the cars can talk to each other, they can talk to each other nearby, and they will able to organize what they are doing.
Tim: One question that I have when I think of cars having any kind of AI is the fact that a driver can override right now; even if my car had a full intelligence system I’d really want a button that I could press to say, ‘no that bridge is out, no matter what your GPS internal map says’, and the ability to actually head off a bad decision. Where I was last weekend, there were some roads that were either unmarked or that Google didn’t know about. That’s one thing that I am curious how cars will adapt to situations where they have to use actual AI and not just follow a readymade routing system.
Peter: Yeah, I’ve had a lot of debates with some friends, computer scientists, artificial intelligence experts and stuff. And there are a lot of different ways that this can work out. One of the interesting questions is: Relying on the human to hit that override button at the right time, or giving the human just a little bit of responsibility may actually be worse than giving a human no responsibility whatsoever. Because when humans don’t have anything to do, they start doing their nails, they start talking on the phone or something, and so the human may end up being so distracted that they are worse than worse was, in an emergency situation.
Tim: You don’t want to wake up in the middle of a crash.
Peter: Exactly. And so one of the things that some people have talked about was like the AI all of a sudden when they get worried and they get screwed up, and they don’t know where they are, they are able to call like a central dispatch. And there will be somebody there, a human, who will be able to kind of look immediately, pick up the camera feeds and figure out what’s going on and then take over.
Tim: Ground traffic control, basically?
Peter: Yeah, some kind of ground traffic control. And then if you have that, that might be better than having the people in the car given a choice and the people in the car to have an override. And I think that they will always be happy if they have that override switch, and we will probably have it for a while, and then the cars will just kind of quietly pull over to the side of the road and stop. But I don’t think you can rely on humans to do the right thing in an emergency.
Tim: That is an interesting thought. Usually I am more worried about machines making mistakes and that may be an irrational fear but we’ve heard stories of GPSes going wrong, where it will take you to the nonexistent bridge or the road that is closed in winter. I would really like a secret button myself but like you said, that may be a worse solution. It is a little bit scary. I see it as making a lot more sense now in cities, does that ring true at all?
Peter: Yeah, I think you are right. It would make people happy, and there are going to be times when humans do the right thing. And humans are really good at jumping out of a model and they are able to see that the model is not working right - and computers aren’t really good at that at all, they just keep going, and they are stuck in their own little world and they stay on the hamster wheel. And humans are better at jumping out of that.
Tim: Not many weeks ago, there was the attack in the marathon in Boston, and the city was suddenly a different place – vehicle traffic was stopped, there were ambulances, of course, and those ambulances probably had to make a lot of last minute split second decisions about what roads to go down, and how to avoid crowds – is that a place you would see either by ground control, or with human drivers or sort of an override switch?
Peter: Yeah, I think humans are still going to be there, and at least for the foreseeable future there are going to be human driven cars that can handle these situations. There may always be human controlled vehicles and the robots will just take over 90 to 95 to 99 percent of the traffic. Because you know you are right, when the whole system goes completely, humans are much better at those kinds of situations than computers.
Tim: I bet cops will prefer to have cars that they can drive as well.
Peter: Oh sure.
Tim: The way that things like an artificially driven car, not artificially, but autonomously driven car is going to change traffic patterns - that gets shunted off to the side a little bit if there are still humans, because we got to deal with the human common denominator, right?
Peter: Well sure, and I don’t know how much longer humans will be on the road, and I don’t know how long they will work, but the neat thing about the cars is that you can really add a lot of flexibility to things; you can fit more lanes in, you can pack the cars in more tightly, and you can do things where you can make a road going downtown at rush hour be a 100 percent one-way and then once rush hour is over it is 60 percent one direction and 40 percent in the other. And it can shift maybe five or seven times a day. And you point out about humans, the neat thing is that humans do a pretty good job of it. There is one road in Washington DC-Connecticut Avenue-which switches from four in one direction and two in the other to three and three and it switches a couple of times during the day. A friend of mine lived on Connecticut Avenue and he said it is really surprising but it works. And it rarely fails. Which is odd. And you think that it wouldn’t be the case but if that’s what humans can do, we can probably do much better with the computers.
Tim: You know hurricane and tsunami egress lanes have come into fashion in the last decade or so too. And I haven’t personally seen any problem with them, I haven’t had to use them in an emergency but I have certainly seen those lanes and they seem to be well marked and accessible. So I certainly don’t think those are an imminent danger and that is another case of how we can adopt current road widths and available concrete and put cars in a different pattern on top of it.
Peter: That would be ideal. In a situation like that, the centrally controlled situation can easily choose and make 100 percent of the lanes out of town he heading out of the town, and no one will come in, and that is something that we can fix pretty quickly.
Tim: Now Peter you talked as we discussed about a lot of small aspects in this book. How did this book get written? It seems to be in a series of small anecdotes or vignettes more than your typical technically oriented book is. How did this model come to your mind?
Peter: Well, there are a couple of things. I mean I started making a list of all the different things that would change and it was clear that were lots and lots of little things, it really was kind of and it was hard to fit it in to the standard narrative where you look where someone will try to come up with a book and they’ll have one narrative and an arc, and then they’ll have something heroic or wonderful to happen at the end that ties it all together. So there were two different things you had to deal with – one, you had a 100 or so different threads; in this case, I just concentrated on 80 of them.
And there is not an ending. Because this is just kind of a guess about where we are going, and it is going to be obsolete soon, and we are going to be moving in all these diverse directions. And it is just kind of a catalyst - to get people thinking. So then I started looking around, and the interesting thing you start seeing on the internet is that there are just a lot more people who are writing lists. And the list is becoming the dominant narrative form, and the arc, and the way that people try to fit things to make it work like fiction isn’t working as well on the internet.
So I figured out, well, why don’t we try to do that on a book level, book length work. It has a lot of advantages, you can snack, you can go from chapter from chapter, you don’t have to read the things linearly, and that’s great. But again it doesn’t have that really nice satisfying ending at the end but I don’t think there is one of those yet, because this is just the beginning.
Tim: You think the robot drivers are going to be cute like this?
Peter: I wish, I wish. Then you could choose one with the right level of cuteness and your particular taste, but no, I don’t think we are going to see them at all, they are going to disappear, they are going to be buried, the same way that the Mac buried all the computer on the screen, it is going to be hidden in one little corner of the car, we are not even going to know where they are.