 My name is Mike Ostrowski, so I'm a professor at Stanford GSB. I do teach a first year course, so Jordan, if you have to take a course and you're interested in these topics, you can take mine. It's one of the many of the required ones. And it's unfortunately close to everyone else, it's only for MBAs. But I also teach some PhD electives, so if you're interested in these types of style of research, which is probably quite different from what you've seen. You know, reach out and there are lots of resources at Stanford for these type of stuff. And this paper is joined with Michael Schwartz, who is a Stanford alum who is now a chief economist at Microsoft. All right, so the topic of the paper is carpooling and the economics of self-driving cars. And it is going to be very much an economics talk. And the connection to energy is actually pretty straightforward. I'm going to be talking about ways to try to minimize energy consumption. I'm sure you've seen a lot of talks about improving energy production. I'm going to take the flip side. If we can save a chunk of energy from being consumed, it's just as good, maybe even better than having the same chunk of energy being produced. So that's kind of the spirit of the talking at the starting point. And I'll just start with this picture, which could have been taken anywhere in the world, right? What got me started on this whole project is the observation that traffic injection is one of the current problem in human condition everywhere in the world from the richest places like New York, Los Angeles, San Francisco to some of the poorest developing countries. As soon as countries get rich enough for people to have cars, you very quickly run into road congestion. And a lot of attempts have been made in the past to try to relieve it. Public transportation, encouraging people to carpool, building more roads, and mostly they routinely fail. And we're hopeful that this may change in the future, but a lot of work is needed. And our vision that I'm going to talk about in this paper is that there are three emerging technologies that will transform transportation and hopefully will relieve congestion and all the corresponding ills, which I'm not going to go through if you want to think about all the problems with congestion, you can read up on that. So the three technologies are one is self-driving cars. And I'm not going to talk about that. It's in the news. You've seen all about that. And in fact, as I'm going to mention a bit later, it's actually not essential. The other two, for these purposes, are going to be potentially sufficient. So the other two technologies are less futuristic, but potentially at least as impactful, maybe more. One is frictionless, time dependent tolls, and the other is convenient, efficient carpooling. I'm going to go through them and explain why I'm talking about them in future tense and the question we're trying to address here. What will a market powered by these technologies look like? How should it be designed and organized to get efficient optimal transportation? All right, so let's talk about why I'm talking about these other two technologies, tolls and carpooling in future tense, right? Because your reaction may be, well, we have tolls, right? And we have carpooling, so why am I talking about them in the future? And here is why. So let's talk about tolls first. So this is what tolls look right now. This is Bay Bridge, this is closed by. And if you want to charge tolls, the current technology involves constructing these gates and putting devices in cars. And then as the car passes through the gate, it gets charged a certain amount of money. Now, if you want to use that to raise revenue, this works just fine. But if you want to use it to guide people's commuting decisions, and try pushing people to drive later or earlier, this is problematic because each of those devices is very expensive, right? Building each one of those things, first of all, it's just cost a lot of money to build, then you also need to widen the road. Lots of resources go into this. And just to kind of illustrate that, even on Bay Bridge, you only have it in one direction. So even if you said, okay, let's use this particular toll system to regulate transportation and have time-dependent tolls. Well, you will only be doing it in one direction, to San Francisco. And then from San Francisco, you won't be able to control it. And putting one of those in every single intersection is just in practice. And it's just incredibly expensive, right? No one is going to do that. So for practical purposes, with some exceptions, we don't have tolls, right, with time-dependent, road-specific. So what's changing? And it's changing very soon. 2021 is not some kind of pine-the-sky dream. This is actually happening. And if you're interested, I can tell you more about what's going into this. Is the new technology where tolls are going to be collected based on GPS locators in the cars, right? At first, it seemed like just a technological difference. But the difference is crucial, because once you charge tolls based on GPS locators in cars, or equivalently, you can do it based on cameras. But take pictures of license plates and intersections, or some combination of those. Suddenly, it's expensive to build this infrastructure. But once you build it, you can now charge time-dependent tolls for every single intersection, every single road, right? And these can vary in time. And now you really can have actual road pricing that's used to optimize the flow of traffic. And so this picture is from Singapore. I mean, Singapore is a pioneer in road pricing. And this is from their land transport authority. They're launching this in Singapore in 2020. And next year, they're transitioning from this type of technology to a GPS-based one. And initially, they're not gonna play with pricing. Initially, they're just gonna charge the same exact prices for the same roads that they used to charge in the past. But then hopefully, they will transition to sort of more intelligent pricing. All right, so that's tolls. The second technology, that may sound even more surprising. Why am I talking about carpooling in future time? I mean, that technology is trivial, right? You put another person in the car, you're carpooling, you're done. What's the technology for? And to be more accurate, the technological improvement is gonna come from reducing the friction involved in finding good carpooling partners. So if you think about this picture again, which could be taken anywhere in the world, almost immediately you know there are lots of people going in the same direction from the same direction. This is like, you see this, like lots of these people are very good carpooling partners. You sort of really know that because they're stuck in the same part of the road. But they're not carpooling right now, it's just very inconvenient, right? It's a mess, you make plans with someone, you lose flexibility, then their plans change, their kids fall sick, they need to find a backup. It's just a mess. And empirically, people just don't carpool, very few people carpool. And of those who do, when you say, they're actually cars on carpool lanes, I see there are multiple people in the car. Well, almost all of them are either people traveling together for work, like construction workers going somewhere, or family members traveling together. The number of people who are like generous strangers who are carpooling just for the purpose of carpooling is just really trivial, really minimal. And so what's changing, there are now a bunch of startups. And this is very much an active area where lots of things are happening. So if you're interested in startups, that I think would be a very interesting one to participate in various startups trying to minimize frictions involved in finding carpool partners. So there is ways that's doing it based on maps. They already have a lot of people using their maps. So they're trying to leverage that into carpooling. There is SCOOP that's trying to do it based on people's co-worker networks. They try to match co-workers because then they trust each other. They're happy carpooling with each other, and they're going to the same place. But it's easier to find good matches. There is blah, blah car that's trying to more casual approach, starting from long distance carpools in Europe, and trying to move it to more regular commuting carpools. And I don't know which one of these is going to succeed, maybe multiple ones, but clearly a lot of people are working on this. So hopefully in the near future, frictions involved in finding good carpool matches is going to decrease substantially. So these are the technologies. And one of the points where real stress on the paper is that it's kind of intuitive that each one of those technologies is nice. And useful, but what's especially powerful is the interplay among them. So we have these green arrows that what they're meant to emphasize is that self-driving cars are going to make both tolls and carpooling much more attractive. And I'm going to walk you through that. And then tolls and carpooling dramatically reinforce each other. They're really highly complementary. And as I go through this, on the next few slides, I'm only going to be able to touch on these points. I only have 20 minutes and then time for Q&A. There is a paper on my website with the same title as the title of the presentation. So if you're interested, you can go and read them all of these kind of reasons and bullet points are elaborated on in the paper. All right, but I will give you a flavor of these arguments. So let's do this green kind of arrows one by one. So first, what's the connection? Why is there a green arrow? Why are self-driving cars making carpooling more attractive? And the basic logic, the basic intuition is that if you think about public transportation, if you think about just any kind of situation where one person is driving one or more people, more other people, the cost of the person driving others around is a very substantial part of that system, right? Drivers are expensive, not many people can afford to take Uber to and from work every day, right? You will do it kind of once or twice a week. But it gets really expensive because you have to pay for another human being to spend their time driving you around, okay? And for that reason, if you think about public transportation, most places in the world, the cost of that human being is split among many co-passengers, right? So you have large buses, you have large trains, and part of the reason is precisely to split that cost of that human across multiple co-passengers. Once you have self-driving technology that goes away, suddenly it becomes feasible to have smaller vehicles that are more convenient, that don't make as many stops. And basically, the way we think about this is once you have self-driving transportation, the line between kind of solo private driving and public transportation is gonna be completely blurred, right? Instead of being kind of a discrete difference, do I take my car to work or do I take a bus? Public transportation becomes a continuous decision, right? Do I take my car to the small car? Or am I willing to take a bigger vehicle and take more stops? And in the paper, we go through a lot of different bullet points emphasizing what are the convenient features that make this combination particularly valuable, okay? And I'm gonna skip them in the interest of time. All right, the second green arrow that I wanna talk about is why does autonomous transportation make road pricing more attractive? First, there are logistical reasons. Basic one being if you have autonomous self-driving cars, by definition, they have to know where they are and they have to track it. So attaching road pricing on top of that is just very easy. Just logistically, just much easier. The same thing for passengers kind of telling the car where to go, just displaying the pricing information. So that's kind of more mechanical. But there is also a very important economic reason, this kind of typical reasoning in economics. But if you haven't taken a lot of economics, this kind of thinking may be new, is the equilibrium effect of self-driving cars during peak hours. So here's what I mean. So when you think about self-driving cars, you're like, this is gonna be wonderful. Right now, when I'm stuck in traffic, I have to sit there and wait my time, maybe I'll listen to an audio book or to the radio. But it's still nowhere near as productive as actually like sleeping or doing my work or doing something meaningful, right? And when I get a self-driving car, it's amazing like the same half an hour, I will actually be able to do productive things, right? So that's wonderful. How can that possible be bad? Well, the way it's gonna be bad is that now everyone else is thinking the same way. And so now, people who use to say, I'll leave my home at 4.30 in the morning or 5.00 in the morning, in order to beat traffic, they say, I'll still, I'm gonna leave at 8.30, because guess what? Even though I'm gonna be stuck in traffic, it doesn't matter because I'll just work, I'll make my conference calls, I'll sleep, I'll have breakfast. It really makes no difference. There is absolutely no reason for me to try to beat traffic. So suddenly, precisely because the disutility of being stuck in traffic drops, the number of people who are willing to encourage the disutility goes up, and the whole thing can equilibrate to the point where all the benefits get undone. So Jesus, in the initial state of the world, you're stuck in traffic for half an hour, and you're wasted half an hour, and now you're stuck in traffic in a self-driving car, you're having breakfast and you're sleeping, but now it takes you two hours. Because a lot more people are now driving during that time. So that's terrible, right? It's a terrible outcome, and again, if you're thinking about not just kind of human well-being, but also like economic costs, resource costs, and all this other thing, I mean, it's just terrible, right? Like the amount of energy that's wasted is just enormous. And we're not the only ones worried about that. You can find out the cause. Here's one from Elon Musk, and the reason why I put this one here is because of all people, you would expect him to be the last one to mention this because he's working on self-driving cars, right? So Tesla is trying to push them, but he's warning that a lot of people think that once you make cars autonomous, that they'll be able to go faster, and that will alleviate congestion, and to some degree that will be true, but the amount of driving that will occur will be much greater with shared autonomy, and actually traffic will get far worse. So in a world like that, you really do need something like road pricing in order to give people incentives to still continue to try to beat traffic, or maybe take public transportation, or find other ways not to generate more congestion. So that's the second green arrow. And now let me talk about the third one. Actually, there is a pair at the bottom, these two. You can see the tolls on carpooling that reinforce each other, and I think this is the most important one, in part because, who knows, maybe self-driving cars won't materialize in our lifetimes. It's still kind of science fiction, and I hope they will, but we don't know. And even if they don't materialize, a lot of the things I'm saying are still going to be relevant, and in particular, this interaction at the bottom. So let me tell you why that is. And it's a relatively simple observation, but we think it's really, really important. So why is there a complementarity between carpooling and road pricing? What does complementarity mean in economic terms? It means that each of those by itself is only moderately effective, but together they're incredibly valuable. So it's the combination of the two that's really going to work, whereas each one separately may not be that powerful. So let's do a very simple example. It's based on a paper by William Vickrey from 1969, and I don't expect you to know this paper. It doesn't really matter. You should be able to get the intuition for what I'm describing, even if you don't know it. So consider a congested road. Just think about a morning commute. Here is point A, point B, maybe there's a bridge into New York City, or Bay Bridge, whatever. For simplicity, just one segment of the road, and we're only thinking about commute in one direction. Suppose it has a certain capacity of what the throughput of the road is. And suppose twice as many people want to take this road during the available time. So obviously there's congestion. Everything is bad. There is not enough space for everyone. And now someone, one of those apps, comes out and there is an amazing carpooling app. You're still better off driving on your own. You still prefer a lot of people like driving on their own. But the disutility from driving solo is small. It's just some delta. Well, guess what? This doesn't do anyone any good. Because I would like other people to carpool, and it doesn't cause them very much. But I'm still individually better off by delta driving solo rather than carpooling with someone. So you have this amazing carpooling technology. Didn't improve anything. So now let's think about the other one, the classic economic solution that I'm sure you've heard of, road pricing. And this is what actually Dicker's paper was about. Let's compute optimal tolls based on people's willingness to pay their value of time. All of that machinery is actually not that complicated. But I'm not going to go into this. So you determine these optimal tolls. But suppose there is no carpooling. Everyone is driving solo. So what happens in this world? So it is true that you fix carpooling. Oh, sorry, carpooling. You fix congestion. So the optimal tolls have the feature that road is going to flow freely. That's wonderful. But there is still twice as many people who want to go during this period, which means a bunch of them are departing too early. A bunch of them are late for work. And your tolls have to be high enough to encourage them to do that. Or put a different discourage them from driving during peak times. And if you do the math, it turns out that these tolls are so high that for the drivers, they undo all of the benefits of free-flowing traffic. So a mathematical statement is that each individual driver after these tolls is going to be just as well off as they were before you introduced the tolls. What's different is that what they used to pay in terms of their time of being stuck in traffic, now they pay in dollars. So society overall is better because instead of wasting resources, the government now collects that revenue. But now you're relying on the government to do something intelligent with it. There is a lot of trust, political economy issues. And the drivers themselves, they don't feel any benefit. They're like, look, I'm now being gauged. And if they have budget constraints, which most people in the world do, suddenly it's like, wow, I need to make this money somewhere. It's just really not a pleasant outcome. And every time toll roads are proposed, a lot of people get upset. And it's bad even if people have kind of similar willingness to pay. If you have richer people and poorer people, there is a whole new dimension that rich people are able to go when they want to go, and then poor people have to stay home. So that's, again, politically really unattractive. And the last point is that even though you're solving the problem that roads are no longer congested, you still have a bunch of people arriving at work at suboptimal times. A lot of them arrive too early. A lot of them arrive too late. Technologically, you have not increased the throughput of the road. You just kind of rearrange people in time. So each one of those, they have somebody but not much. So now suppose, again, in the same very simple example, you have both. You have convenient carpooling, and you set tolls optimally. So now the magic thing happens, because now all you have to do is to set the tolls a bit above delta, or more accurately, a bit above delta over 2. That gives people incentive to carpool, because you're now setting them just above their disutility from carpooling. And now each car carries two people and your relief congestion. And notice that even if the government just steals all their revenue, or destroys it, or burns it, or loses it, or whatever, all the drivers are still substantially better off. And everyone gets to work on time. It's really this combination that's really powerful. All right, so I have five minutes left. So I'm not going to take you through the model. So I'll just kind of advertise what we have in the paper, and then I'll make one last observation. So in the paper, we set up a mathematical model of such a market. There are lots of models of transportation of this, so there are lots of papers on road pricing, starting with Vickra. What's special about our paper, and why we need a new framework, is that we also introduce this coalition formation component, where we also need to figure out who is driving with whom, and how much they need to pay, and how those payments are determined. So in the model, I'm not going to take you through it. We have a set of riders. We have a set of road segments, time stamps. We have some capacities for those road segments, some trips that people may want to take. And then we look at various assignments. And then what's important is there are going to be two sets of prices in the system. One set of prices is the optimal tolls set by the government, but the other set of prices is going to be how much the riders actually pay. And very importantly, if the toll from A to B is $7, and seven people are sharing a vehicle, then each of them only pays a part of that. It could be each one pays a dollar. It could be some other sharing, but the tolls are segment-specific and not person-specific. That's what makes it powerful. And then our main result of the paper is that in this model, if you set up these prices correctly, you get the optimal outcome, where the total welfare net of costs is optimized. And obviously, I don't have time to take you through the math, but as I said, it's all in the paper. All right, last point I want to make. And it's actually very important. It's not directly related to what we have in the paper, but although we use some of this intuition, you would often hear in the newspapers the following complaint observation statement. So this is from the Wall Street Journal from three and a half years ago, but it's like super typical observation, where people say, look, the absurdity of our century-old ad hoc approach to mobility is captured in one statistic. The utilization rate of automobiles in the US is about 5%. For the remaining 95% of the time, 23 hours, our cars just sit there, a slow, awful cash burn like condos at the beach. I'm sure you've seen the statistics in many times, many places. And our view, our claim of what we explained in the paper, like this is a total red herring. And just to give you an analogy, I have a roll of paper towels in my kitchen, and it sits unused probably 99.9% of the time. And no one think of that as a slow, awful cash burn. Why? Well, because I'm kind of using it gradually. Once I use it up, I get a new roll. And if I was sharing it with my neighbors, well, it would get used up twice as fast. There would not be that much benefit in doing that. The same is true for cars. So we do a simple numerical example in the paper. If you have an empty car, it's really not that expensive for that car to sit and wait. Now, if you have parking issues, that's a whole different consideration. But in the world where parking is available, like in the suburbs, it's really not a problem. So we do a typical numerical example where cars die after 200,000 miles or 15 years. New car is $30,000. Real interest rate is low. And then we just do the basic math. What is the cost per mile of using this car for different levels of car utilization? And if you use your car rarely, like if your car's annual mileage is 5,000 miles per year, if you triple that, you really do have substantial savings. Because in that world, when you don't drive the car too much, it does die of old age. It's like if my roll of paper towels is sitting kind of unused for 10 years, eventually the ants will eat it or they will turn yellow or something bad will happen. So there is an aging component. But you have to use it very, very slowly for that to kick in. Once you get to some reasonable levels, even 15,000 miles per year, which is not crazy. Most people who commute to work regularly in the car hit that number or something close. From that point on, even if you multiply your car utilization by a factor of 5, which is crazy, crazy high, very hard, you're only saving 2 cents out of 48 cents per mile. So again, when you see this type of lament, that's really not a problem, at least in the current world. Now compare that to car utilization as in how many people are in the car. So that's the second bullet point. And then I'll get to the first one. So if you think about car utilization as what fraction of time the car is on the road, that's not relevant. That's just not a relevant number. If you think that the way to save energy and save the planet is just to make sure that cars drive stay idle more, it's not going to work. It helps you a little bit. However, if you think about car utilization in terms of how many people are in the car at the same time, when you double car utilization in that sense, you are increasing energy efficiency and everything else by a factor of 2. If five people in the car, you are doing better by a factor of 5, again in contrast to this picture where when you go from 15,000 to 75,000, so you utilize car five times and intensely in the time dimension, that really doesn't do you any good. And I'm going to skip the first bullet point, so I'll just stop here and take questions. Yes. Hi. So with respect to this last point of filling up cars with more people, that should work, yeah. But another option could be to just make cars a lot smaller. And I know that a lot of companies have been doing concepts in that area for many decades now. And the stats supported, I think, that a given car or at a given moment, 80-something percent of the cars are occupied by one person or so. So I wonder why carpooling is such a popular suggestion for a solution. And personal mobility doesn't seem to be. So, OK, a couple of answers to this. I don't know, your guess is probably as good as mine, but I'll throw out a couple of conjectures. First of all, for you individually, a bigger car is going to be safer. If you take your smart car and I hit it with my Hummer, I don't have a Hummer, but suppose I did, I know who is going to be worse off. So on the individual level, each person feels like, oh, I want the safest car available. So it naturally tends you towards heavier and bigger cars. And trying to regulate that is just incredibly hard. I can imagine you're telling your mom with kids, no, you cannot have the safe car because it's bad for the other drivers. You have to drive the small, dinky cars. It's politically very hard. But I think from the point of view of congestion, I actually don't know. I don't think that having one smart car, even if it's a single passenger, but it still kind of takes a reasonable amount of space, more than a motorcycle. It's an inefficient design. But again, if you think from the point of view of congestion, if you replace a typical five-person sedan with a smart car, I don't think it creates that much less congestion. Because there is still a lot of congestion coming from breaking before, breaking after. So all of that space around the cars, I think if you can put two of them, if you can make the lanes half as wide, maybe you can get them. But that's just such a massive, massive change. So I think it's the kind of thing if you're designing a city from scratch, that's something that I think might be viable. You just decide, or like a retirement community, and you just find a way to enforce that. Taking an existing city, especially like something like the Bay Area where we can't even build an extra mile of barge and trying to impose something like that, I just think politically, it's just much harder. Yes? So to build on what Tela said, does your road pricing model allow for taxing basically one person commuting via some kind of a huge, enormous gas tax? So what we show is you don't need to do that. So what we show is that it's sufficient to tax things purely based on the road segments. And then the way what you want is going to work is that if I'm driving with three other people, I'm paying one third of that toll. And then if you want to help poorer people, you can remove all the other taxes. So you can say, we're not going to have a gas tax. We're not going to have a sales tax on cars. But we will have pretty high tolls. So for people who are carpooling, then their net effect is going to be better. Because they're saving a lot of money on those taxes. Now they're paying something on taxes. They're not better off. And people who insist on driving solo, they will be hit the harder. But wouldn't gas tax, aside from having a similar effect to tolls, also reduce possibly the carbon? Yeah, so you can have both. So I mean, there are two separate policies. It's actually important to think about this. So in economics, there is just basically general basic lessons. You should tax the thing that you're worried about. So for the purposes of carbon emissions, like for example, if you want people to switch to electric cars, then gasoline tax is the right thing to do. Because it forces them to use a different type of fuel. But what it doesn't give you is kind of time pattern. If you want people to drive at 5 in the morning, or 10 AM, and avoid rush hour, gas tax doesn't give you that, whereas this thing does. And so if I could set an optimal policy, you would have both. One to encourage people to use the right type of fuel and energy, and the other one to encourage them to drive at the right time. Yes? Do you think people might modify their car to have a GPS, but then change, remove the part which sends out the signal for where they are for the tolls? I mean, we regulate a lot of things. And I think it depends on the country. It depends on the legal environment. But yeah, I mean, you can mandate it to be done at the level of the car manufacturer. Cars already have this OBD or DB computer, so you can just have that as part of that. But I take a question from that part of the room, because I know I naturally gravitate to, yes? Last question. All right, and then I'll be around. You can ask me for another 10 minutes. So my question, again, at piggybacks, a little bit about off of what was already asked, which is there's a lot of work being done on the technology behind self-driving cars. It seems somewhat inevitable at this point that it will come. My question is, I hope so, but my question is on the things that you're talking about, specifically the policy architecture, do you think there's enough work being done, and a little bit on the technology side as far as the ancillary technologies you would need to capture things like carpooling and tolling and things like that. Is there enough work being done to receive the self-driving cars once they're ready to go? Or are the state and local governments going to be lagging so far behind that? OK, that's an excellent question. And my answer is an emphatic no. And that's the reason why we wrote this paper. It's literally the way we started working on it. We'll then realize the paper is relevant even in the world without self-driving cars. But the way we started working on this is about five years ago, we said all this stuff is happening, how is the world going to be different? And at first, we were just talking. And then over time, we realized people are just not thinking about these types of stuff. And with road pricing and policy, one thing that over the years becomes clear is that status quo is very powerful. Like a bunch of this stuff, you have to get them right from the start. It's very hard to, if people are used to the idea that roads are free, there are lots of other government sources that are not free. My water in my house is not free. Why are the roads free and water is not? It's actually not clear, but it's just people are very used to it. It's very hard to change. So in particular, if I was advising policymaker, how do you get something like this in place, given that people are used to the idea that roads are free? And one way to do it, which I'm sure a lot of self-driving car people will hate, is you start out by just doing this for empty self-driving cars. You say, what we're going to do is we're just going to charge empty. When the self-driving car is empty, they have to pay these things. And then when people see that it's not the end of the world, you just have kind of reasonable instrument. Then you can start expanding this in all self-driving cars. So you kind of gradually build from them. But the transition quite, I mean, it's massive. Because once people are used to something, it's just incredibly hard to take it away from them. Even if on the overall grounds, it's the right thing to do. With that, we thank you, Michael. Thank you for being here. Thank you.