The a16z Show - a16z Podcast: Exploding the Map

Starting point is 00:00:00 The content here is for informational purposes only, should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. For more details, please see A16Z.com slash disclosures. Hi and welcome to the A16Z podcast. I'm Hannah and we're here today talking about the evolution of cartography, how mapmaking is fundamentally changing in the age of autonomous vehicles. with Way Low, COO and Head of Product of DeepMap, which creates HD Maps for Autonomous Vehicles, and David Rumsey, map collector of one of the largest private paper map collections in the world, now at Stanford, and of the largest online digital map collection. We talk about how the tools we use have changed from sextants to measure the stars to computer vision and LIDAR, as well as how we think about what maps actually do and are used for. Maps have really evolved from incredibly primitive technology, right? Like sextants and stars, really, to now, I guess, satellite and camera and laser and LiDAR and beyond. So has the nature of mapping itself fundamentally changed or do we use maps differently now or is it the same just more than it has been in the past?

Starting point is 00:01:20 Well, I see, I never referred to sextants as primitive. They were, you know, the LIDAR of the day. I mean, Sextons opened up the whole world and created the ability to make scale maps and to map essentially beyond where you were. When would you say the beginning of sort of the golden era of mapping was? Well, it was really around the time of Columbus because the whole discovery of North and South America produced a phenomenal amount of desire to map it, to possess it. One of our favorite globes is the, it's called the Earth Apple, which was by Martin Behame, the year Columbus discovered America, 1492, and there's no North and South America on the globe. Amazing.

Starting point is 00:02:06 Of course, there's a huge Japan and China, beckoning. And completely not to scale either. Even though we know Columbus sailed west in order to go to China, seeing it on the globe without North and South America makes it so real. So that's when maps get really powerful. Well, and it's interesting because it's capturing an entire time too, right? It's a representation of all our knowledge at the time. Well, that's the history in the maps. That globe has information that you cannot get any other way. So if the bones of the maps were once, you know, longitude and latitude and medians, what are the bones of maps now today?

Starting point is 00:02:45 Where are we at in the state of mapping? Mapping has always been almost the frontier of technology. People who can make really good maps for that time tend to be the most technically advanced and mathematically advanced. Well, because it's always about processing information, right? It's always about getting the right information. It's about getting the right information and then do it in a very incredibly, the most precise way at the given time. 300 years ago, even 500 years ago. Their understanding of the environment is very limited, but somehow the needs. to predict into areas that they cannot visually see.

Starting point is 00:03:25 And today, back to your question, you know, what are the backbones of today's digital maps? Well, first of all, we have a lot of sensors that can enable us to see far, right? You know, there are satellites up in the air, there are, you know, airplanes up in the air. Even the cars on the road, a lot of them have a lot of sensors as well. We have radars, we have cameras and then so on. So a lot of these digital information is already getting collected and we can use that information to help us create digital maps. And they're being created all the time. The maps are constantly being redrawn. That's actually a very interesting question because as of today, that process is actually not happening.

Starting point is 00:04:08 We are collecting a lot of data, but the information that is actually being shared for map creation purpose is very small. So if you think about the maps that you actually use today, right? There are maps that you have on your phone. There are maps that are probably in your car navigation system as well. Those maps were not created from the GPS unit that you have on your car or the camera system that you have in your car and whatnot. They were created by special what we call mapping survey fleets. So companies like, you know, for instance, Google will build these special fleet that has a lot of

Starting point is 00:04:45 special sensors mounted on these special cars. They would hire human drivers to drive these road, collect a lot of data, and then they do a lot of offline processing. And at the end of day, some digital map of cities would be produced. And they get inserted into the cars. They get stored in the cloud and then send to your phone. But even though it's using new tools, it's sort of an old-fashioned model of thinking about map making, right? Like you send one pioneer out to map everything, and it's a static thing. Absolutely. We, in addition to collecting maps, I do collect the tools of map making. Oh, do you? And for mapping the West, there would be, you know, typically officers in the Army 1870 with a, we have a big wooden board that they would

Starting point is 00:05:37 keep on their arm as they drew the map from their horse, sort of like, not a self-driving car, but like, Like a drawing board? Like a portable drawing board. They would be moving through the environment. That sounds like the survey car or survey horse. Yeah, exactly. It was a survey horse. Totally.

Starting point is 00:05:55 But you can imagine the length of time for that drawing to get from that person onto the map. On the Pony Express. Yeah, years. Which they thought was revolutionary, you know, that it was very fast. So the speed has really increased, right? The mapping speed is what's increased. But the nature of it hasn't really changed so much. Or would you say no?

Starting point is 00:06:19 One thing that has changed drastically from paper maps is that maps are dynamic now. Maps are changing. You know, with paper maps, we have a sense of scale. The scale is printed onto the map and it's fixed. But with digital maps, the scale is totally fluid and variable. And the accuracy just depends on the size of the resolution. Mapping as a field has evolved or reinvented itself many times throughout history, and now we're reinventing it again.

Starting point is 00:06:52 What are those moments? Methods of printing, particularly, shape, mapping. So the first maps were done with woodblock. You could only get, you know, 100 good impressions. So maps were held by, you know, people who were rulers, who were wealthy and so on. Then it changed to copper engraving. You could make 500 impressions. copper engraving to lithography in the 19th century, thousands of maps and now chromolithography,

Starting point is 00:07:20 and then eventually digital distribution over the web. So maps have become ubiquitous. It's the access to the map. The access has been the most profound change. The actual depiction of space, I think, or the goal to depict space, has been remarkably consistent. It's democratizing the map. You know, not only the printing or the access of the map has. changed over time. But if you look at the general trend, maps tend to get more and more accurate

Starting point is 00:07:48 over time, just in general. Right. We're already doing that. A lot of people thought that, you know, mapping has been solved with today's commonly available navigation maps because they seem very accurate for human. Because we all have access. We already have access. They feel like they're updating. That's right. You get real time traffic on your Google Maps app and then so on and so forth. But if you actually dive deeper, maps today, including the digital maps, are actually built for human consumption only. So when or not, you are looking at, say, Google Maps or Apple Maps, ways and then so on, these are maps purposefully built to be, first of all, very easy to interpret by human beings and easy to use for navigation purpose. The navigation systems telling the humans

Starting point is 00:08:35 how to navigate that car according to some very simple instructions. It's also assuming that there is a human with human capabilities making those interpretations. That's absolutely correct. And what we are, the new period we're going to enter into is, or what's needed to be built for self-driving space is actually maps, built, purposefully built for robotics systems. So now we have to democratize beyond the human. That's right. That's right. So now we have a totally different set of demands that we need out of our maps. When we start looking at autonomous vehicles, we don't have enough right now to make sure that an autonomous car can predict exactly where something is and be safely in real time. Right. So what has to change about mapping to serve that? We need maps that are easy to interpret by robots. DeepMap is focusing on self-driving cars, but you can't generalize it to any robot that needs to roam around the physical world.

Starting point is 00:09:35 Yeah. Even though robots can all perform humans in certain aspects, in other aspects, they are intelligence-wise. Humans are actually much smarter. Things people tend or humans tend to take for granted, such as stopping at the right place, at an intersection, watching for the right traffic signal. Or making a split last-minute decision to avoid a raccoon in the road. That's right. Yeah. These decisions are very hard for robots to make. And as part of the decision-making process, the mapping becomes a very critical component of helping the robots to make the right decisions. Because they're essentially reading the whole world around them through the map. That's right.

Starting point is 00:10:20 What does that mean for the kind of map we're now going to need in the future of autonomous vehicles? The maps that are purposely filled for self-driving purpose are usually called high-definition maps or HD map for sure. They specifically refer to the maps that have extremely high precision. And we're talking about centimeter level accuracy or precision because the robots need very precise instructions on how to maneuver themselves and how to navigate themselves around the 3D space. Right. A few centimeters makes it really make difference when there's like maybe a curb there. Oh, absolutely. I mean, people tend to think, why do I need to say 5 centimeter or 10 centimeter actually? accuracy when I'm driving down the road. In most cases, you know, the tolerance for error

Starting point is 00:11:11 might be higher than that, but then there are going to be cases where if, you know, you're driving on, you know, I don't know, the road to Tahoe, there's literally cliffs on one side and there's really no room for error or any error. So the map needs to be extremely precise and it needs to contain a lot of information that, again, humans may take for granted. So not only we need to know where the lanes are, where the road boundaries are, we also want to know where the curbs are, how high the curbs are. If it's five centimeters in HD mapping, that's approaching something that's been sort of a holy grail for mappers forever, which is what we call the one-to-one map. It's the map of the world as big as the world. Wow.

Starting point is 00:11:58 Borges writes a short story about the cartographer that made a map that was one-to-one. and the problem, he couldn't unroll it. It made me realize that HD mapping is a one-to-one mapping. And yet we don't unroll it. We roll through it. It's so interesting because I think of like maps on some basic level as being condensing a huge amount of information and like taking one element of that information and showing it. Simplifying one down into one picture of like, you know, borders or topography or land versus C.

Starting point is 00:12:31 But now you're talking about actually way more than one element. It's exploding the simple map to as full, as comprehensive as possible. The map needs to describe every little thing on the road. And it needs to describe a lot of hidden things that you don't typically see on the map. Like what? Like for instance, the speed limit on the road that you see. You want to code into the map whether or not this lane is allowed to go straight. intersection or is required to make a left turn or right turn. David, am I right that it was only one,

Starting point is 00:13:07 sort of one, boiling things down to one dimension? Or is it actually much more multi-dimensional than I'm thinking? I think a little more multidimensional because it isn't just about scale. It's also about comprehensiveness. So the first Atlas was published in 1570 by Abraham Ortelius, and it was actually the best-selling book in Europe for years and years and years. And the reason it was important to people was there were maps of the entire world in the book. Oh, so it's always been important to capture as much. As much as possible. Now, these were not, obviously not anything like HD Maps or even digital maps, but they

Starting point is 00:13:47 were covering the whole earth. People love on old maps the notion of Terra Incognita, which is put over the center of Australia or the American West. The literally unmapped space that we call to. So I don't think it is gone, really. If we think about the oceans, I know Ways not talking about submarines being autopilot, not yet. But the whole bathymetry mapping is a complete, fascinating frontier. Basemetry mapping being underwater.

Starting point is 00:14:19 Being mapping the undersea areas. Oh, that's fascinating. But it actually sounds like we are at a whole new frontier, right, of trying to process all that information into a real map. Now you are, you're really just exploding what the map is. The map becomes reality. Well, and it's also, you make, you're immersive. It's an immersive map. So what kind of tools allow you to do that?

Starting point is 00:14:44 How are we actually starting to explode this map to map reality? On the very high level, obviously there's the hardware component as well, the software component. The hardware components are more visible because if you look at, you know, even a picture of a self-driving car, you will quickly recognize it's a self-driving car. mostly because it has a lot of sensors typically around its rooftop. And all these sensors are useful for map creation and map update purpose. We use a combination of different type of sensors that includes cameras, LiDAR, GPS, IMU, which is a unit that tracks the movement of the car and the radars as well. Are you using everything you possibly can?

Starting point is 00:15:29 Or have you made strategic choices about what the best combination and why those tools are? We try to actually make good use of all the sensors that are going to be mounted on a, let's say, a typical self-driving car. The reason is that self-driving car number one requirement is needs to be very safe. Yeah. We want to make sure we can take advantage of all the sensors in case when the car is running and one type of a sensor may fail or may get blocked. we can actually switch between different type of sensors in those. Oh, so it's like a backup plan as well. But why isn't just visual?

Starting point is 00:16:07 Well, visual's really important, but I mean, for the same reason, Way is saying today, in the past, they would use other methods triangulation. I mean, sex. It's about space, really. And taking mathematical calculations that they were doing essentially by hand, which are now done just instant. You know, either you measure with a ruler, or with, I don't know, horses.

Starting point is 00:16:30 Right. Triangulation. You're trying to get some kind of distance matter, right? Through triangulation with camera data, you can do some triangulation and get distance estimate for sure. But when you're talking about extreme high precision measurement, cameras oftentimes is just not enough. And the lasers gives you a precise measurement of depths or distance in the 3D space.

Starting point is 00:16:56 If we're talking about sensors that are mounted on self-driving car, LiDAR and the cameras are always working together. Generally speaking, they're all running very fast multiple times per second. As the car is driving, for instance, at fairly high speed, I don't know, 70 miles per hour on the highway, the car is generating or collecting a lot of data as it's driving at that high speed. with both the cameras, the Liders, the radars, and all the other sensors. So you have a little like map-making bubble, basically that you're traveling around the world in, also reading the map. It's consuming the map and creating the map at the same time.

Starting point is 00:17:38 And it's not one bubble, it's lots of bubbles. Well, that's what's interesting, right? Because maybe, because we were always consuming and creating the map at the same time, would you say? Or is that different? Well, there often was a time lag, though, between, you know, the sources of information long ago would be all the way from sextants that we talked about, but also itineraries. You know, people would walk or go by horse and say how many hours it took from this town to the next and they could extrapolate what the distance was.

Starting point is 00:18:09 Oh, yeah. Yeah. So it was never as instant as it is today. What fascinates me about the 21st century mapping is it's so open. You know, way saying that the cars on the road will actually share information with each other. Sharing information in the history of mapping has been really important in moving accuracy ahead. And not every state in the world believed in sharing information. For instance, at one time in the period of explorations, the Spanish made extraordinary maps, but they did not share them.

Starting point is 00:18:49 they viewed the information as totally proprietary. And highly valuable. But their competitors, for instance, in the Pacific Northwest of the United States, the English, were releasing all of their map information. As soon as Vancouver explored the area around Seattle, all that information went to press in Britain and was widely used. So it's their names that lasted. And it was their settlers that came.

Starting point is 00:19:15 So the whole notion of open content goes back a long time as being. a very successful business model in the mapping world. That's fascinating. Generally, those who hold information really tight, it may have short-term value, but long-term, it does not have nearly as much value as sharing. So I think that's one very important theme in terms of the history of cartography. I love that. So we talked a little bit about the sensors.

Starting point is 00:19:41 Are there ways that you gather the information, other sources that are perhaps widely available that you guys are pulling in as part of this? The other part is obviously the software. And then the software piece is really interesting because, as you mentioned, each car is a little bubble collecting data. And obviously the software needs to be there to power the hardware to collect, record the data. And then this information somehow needs to be shared. Yeah. So where are you right now with your one-to-one map of reality?

Starting point is 00:20:12 I mean, is that, are you already? How long does it take to get that kind of depth of map? is does one car driving through one time do it? Is that enough? One soldier on a horse? I don't think one soldier on a horse is sufficient. A single car cannot map the entire world. The reason is that when we think about how the, we not only think about how the map gets created, we also think about how the map needs to be maintained and updated and updated to reflect the changes on the road. So instead of having one soldier with one horse, you know, I don't know, drive. down the road once, we designed for an aggregated effect, multiple cars driving down the same

Starting point is 00:20:55 road, and we aggregate all that data from multiple drives, multiple cars together. And the more cars you have on the road, or the more data that we collect from, the higher quality the map will be. Right. A constellation. A constellation. Yes, many bubbles. So you're dealing with enormous amounts of data that you have to instantly update, right? Receive and process and send, And what are some of the challenges or the ways that you think about dealing with that? Cost is definitely a big challenge because you don't want to send tons and tons of data over cellular network. So the way that we think about how to solve the problem is to almost categorize what information that we need to share in real time and what information we don't need to share in real time. There can be, to David's point, a slight lag.

Starting point is 00:21:43 That's right. Interesting. How do you qualify those different types of? Well, things that really impact driving behavior needs to be shared and then distributed to other cars that maybe be affected, such as, you know, accident on the road. Right. Or maybe, you know, in the last raining season that we had in the Bay Area, there were a lot of landslides, not trees falling down on the road. How about faster things like a deer runs across the road? You can, but I'm not sure if that information is very valuable to share with the car behind you.

Starting point is 00:22:17 That's true. Right. Yeah. There are certain. Only in as much as like there might be more. Right. That's right. Yeah.

Starting point is 00:22:24 That's possible. Or if you have, I'm just thinking out and out if you have a delivery van who is spilling out, I don't know. Yes. Oranges. Yeah. Right. That spill may last.

Starting point is 00:22:36 Longer than a fleeting second. Yeah. Yeah. But as a human driver, if you needed to respond to that scenario in a real time and the car behind you would respond. in a similar fashion, then that information should be shared as quickly as possible. If it's something that needs to be in the broader driving ecosystem instantly. Right.

Starting point is 00:22:57 Let's say, you know, there's a new road construction happening, which may take quite some while to actually complete. And as the cars are going over that stretch of road, you can keep collecting the data until the change actually gets into or once the change is actually in effect. and then you distribute that information more broadly to all the cars. It's so interesting that it's sort of exploding the time dimension of map too. You know, it's like becoming a living organism. It's a very much organic process, the way I think about it.

Starting point is 00:23:31 So now you have all this data that we're collecting in these deep maps. What happens to those? Where do they live and how long do they live for? You know, are we storing up this map of the world that will, you know, that has, Is recording it in a way we've sort of never seen before? There are a lot of data being collected now, and there are going to be a lot more data being collected in the future once you have more self-driving cars on the road.

Starting point is 00:23:59 This is where cloud infrastructure comes into place, of course, is the ultimate storage place. And it's also the place where 99% of the computation is done for map creation and map updates. But then each individual car will have its own, I don't know, memory cells as well as its own intelligence. Another term people tend to use is called edge computing. Essentially, each car will carry some storage and some computing power so that it can make its own decision independently.

Starting point is 00:24:31 Right. If, you know, it's completely offline, it should be still fully functional. So the maps are constant being called by the rest of self-driving software stack, such as the perception system, such as the planning and control system. again, many times per second, the interpretation. The interpretation of the map. And as you come across a change, a change detection module needs to kick in and say that looks different from what my map tells me.

Starting point is 00:25:00 Now, I need to decide if that's an important enough change to be distributed in real time or, you know, through the edge compute process, the computers on the cars need to decide whether or not I need to actually maybe. enter an active data collection mode and then share that data with the cloud. So the decision making is separate from the map. It's completely two separate functions. So on a very high level, the self-driving software stack consists of four components. The first component is called the perception system. You can think of it as the eyes of humans. It's trying to see what's around the car and figuring out, for instance, do I see a human crossing

Starting point is 00:25:44 the road or do I see the signal light being red or green? Another piece is called localization and localization module basically tells the car where you are in the 3D space and what's actually around you and says, okay, you are 153 centimeters from the next stop line and here's a crosswalk of this width. You need to do X, Y, and Z with the car. And that's what the planning and control module will kick in and say, okay. okay, I'm going to slow down and then make a full stop at the next intersection. And then the first component is the mapping component. And the mapping, you can think of it.

Starting point is 00:26:25 Mapping is having tentacles into all these three things that I just mentioned earlier. Because for instance, for localization, mapping and localization actually work very, very tightly together, constantly comparing. It's where you are in the map. Exactly. Basically. He needs to know what is supposed to be here. And then it can tell the difference, right? If I know there's supposed to be intersection and a crosswalk and then I see a moving object around like crossing it, that's probably a pedestrian. That seems like something really new to me that a map can not only have a have represent what's actually there, but what a sort of shadow map of what it thought was there. You know, what it was once there before or might be might have been there on Thursday. you know, but isn't there now? It's a living thing. I think it's a real revolution in mapping.

Starting point is 00:27:16 I've never seen anything like it. And, of course, I want to collect it. Well, let's see, what does it like, what does it look like to the car? Does the car see something? Does a passenger see anything? Has David seen what this map looks like that he wants to collect? Right. Oh, yes.

Starting point is 00:27:31 He's appetite. It's been able. So, I guess data is data at the end of the day. How we visualize is up to, you know, how. how we want to consume it. Yeah, it's just like it was the knowledge. Everyone visualized the knowledge in a different way in the Renaissance. And in the digital map, you have even more options.

Starting point is 00:27:52 You can color code it. You can apply different lifestyle, so on and so forth. But generally, you have two audiences now, the humans and the robots. The robots don't really need to see anything. They need to consume the data. And they consume the data through APIs and so on. So what's the point of having a visual? representation.

Starting point is 00:28:12 And at all, then. Those are for humans. As part of the map creation and update process, we will have human moderators to check the computer output to see, you know, there are always ambiguities in the real world. There are certain intersections that are so complex that are hard even for humans to interpret the traffic rules. Yeah. I know those type of, I used to call what, dysfunction junction.

Starting point is 00:28:39 That's right. So we have visualization tools to help us as part of the map creation and maintenance process. But again, data is data and data can be useful in so many different ways. There are so many aesthetic choices that go into map making, which I'm sure part of why you love map so much. It's an art. So what kind of aesthetic choices did you guys make or how are you, you know, that when the human is looking at this deep map? that you guys have been thinking about? Map creation and maintenance, I think for us, at least,

Starting point is 00:29:16 functionality kind of trumps over the beauty aspect because we need to make sure the quality of the map is very, very high. Otherwise, our human operators won't be able to work as effectively. So I think productivity and efficiency is what we design for. But when we are displaying the maps for other purpose, you know, for simulation, for virtual reality, then, you know, depending on which use case we're looking at, we can make the right, you know, design choices. So it's not difficult to change.

Starting point is 00:29:50 Maps have always been also a representation right of law in some form, of sort of territory and rules and where people are allowed and where they are. How do you guys think about some of the legal issues? Maps have always been very sensitive in law. There are claims and counterclaims. The Sea of Japan is that, you know, others want to call it the Sea of Korea or maps have always had a function, an administrative function, you know, a taxing function, a governing function. Wars, awful as they are, produce an enormous amount of mapping. And I've collected widely.

Starting point is 00:30:26 And fought over mapping, too. And fought over mapping. I got a three-volume Atlas that was from 1786 that had been captured from a French vessel. And it was a fight between the British and the French right off of Brittany and France during the American Revolution. And the British captain, in his notes, felt that the three atlases he got were more valuable than the ship. Amazing. And, you know, it's just in those days, that kind of knowledge was very, very powerful. One of the problems today is, of course, it's so maliable.

Starting point is 00:31:01 Yeah. It can be counterfeited. It can be falsified. It isn't printed. So, you know, people will put out false claims on maps and they circulate and then you don't know what it's real. So how do you guys make sure that you protect that information, those three valuable atlases on the ship? It's going to be a big challenge for HD mapping because as of today, even government regulations are not clear, even though digital maps are widely available today. but there are countries who heavily, heavily regulate their geospatial data.

Starting point is 00:31:37 You know, China is one, South Korea is one. There's actually a whole bunch of other countries who... Just like Spain back in the day, right? Yeah, explicitly forbid exporting geospatial data. Wow. You cannot move data out of South Korea about the boundaries and the roads of South Korea streets. So that problem actually already exists today with just the normal navigation maps. When it comes to HD maps, you can imagine all sorts of sensitivity because you have these cars with very high-resolution, LIDAR and cameras constantly scanning the street and then possibly seeing into people's private driveways as well.

Starting point is 00:32:17 So being able to protect people's privacy and fulfill some government regulations. From security point of view, we can do a lot of encryption and so on to make sure the data don't accidentally leak out. Having said that, the future remains to be seen. U.S. government is fairly active, you know, trying to both protect and advance the self-driving technology and make sure privacy is, people's privacy is well protected. And it's playing out literally week by week. That's right. And the other governments are doing the same.

Starting point is 00:32:51 We will, I think, in a few years of time, a lot of dust will settle and we'll see what is the, you know, both from the technology side, what we can do as well as from. regulation side what the government is waiting to do. For instance, who should own the data? As the driver of the car, do you own the data? Or as the private homeowner, do I own the data around my private home? Or does the car maker? Or does the map maker? Or should the government own the data? What's the classic thinking about that in terms of maps? Was it the person who drew the map that owned the data? Or was the person who was hired to draw the map or what? The same confusion that we have today. I think all these issues just,

Starting point is 00:33:31 speak to the power of maps, and they're just getting more powerful all the time, because they embody all of these kinds of sensitivities. They tell history, and they record our lives, and particularly these, now these living dynamic maps, which we've never had before. I love this that you talk about the maps as capturing a moment in history, too, because if you're capturing this kind of, this depth of map and this representation of a kind of immersive, of worlds that we've never had before. It is capturing an incredible moment of time. Do you think about the historic value of capturing the world like that? I was saying to way earlier as we were coming over here, please hold on to the data. Because it's like capturing, it's like a model of our entire world.

Starting point is 00:34:22 There are no out-of-date map. Before we had our conversation, I, to be honest, I didn't. Because, you know, when it comes to digital mapping. Because the value is to have it in the present. It's always about freshest data. And that's the way it should be. That's exactly what you should be doing. But at the same time, you should preserve the data. Yeah. No, that's fascinating. I mean, I used to be an art book publisher and one of my favorite books on the list at Princeton was the Barrington Atlas, which is just a wonderful representation of what the world used to be. And you can think about people doing these kinds of looking backwards in a totally different way. Last thing, you know, when we think about terra incognita and then it becomes terra cognita, right?

Starting point is 00:35:04 It opens up all sorts of new population shifts and trade routes and products and economies. We've talked a lot about the autonomous vehicle. What do you see some of the other sort of unexpected consequences of mapping out the world like this possibly being? Essentially, when we are creating HD maps, we're creating a digital infrastructure of a city's road network. And they do try to create that digital infrastructure today because almost every city in the U.S. has a GIS department. They send out a surveyors. They map the road. The public works department oftentimes do the same.

Starting point is 00:35:41 This process is static because it's very costly to do. It's very labor intensive and can't repeat these surveys all the time. On the other hand, when we have cars with the right sensors running on the road, they're constantly. constantly updating the road. So the process of actually having a central government, having to maintain that digital infrastructure, can suddenly be solved with people with cars or just cars. And this information is tremendously useful beyond just self-driving. Having seen maps evolve, what strikes you as some of the possible consequences of thinking of these exploded maps? Well, everybody becomes a mapper, which is very powerful. Everybody takes own

Starting point is 00:36:27 ownership and because they contribute as they're driving to making the great map. I think it will have an effect on people to be part of the knowledge creation system. I'm not sure exactly how, but I think it's part of empowering people. And historians. And then for historians to have access to, you know, what people were doing 50 years earlier. I mean, you can imagine like a Barrington's Atlas that you could literally be submerged into. It's almost too much to see patterns in this vast amount of data. Oh, that's a fascinating idea. To also read it just for human patterns.

Starting point is 00:37:06 It's always been something we've wanted so badly was to be able to do, you know, optical character recognition in old maps. But it's impossible to do because the text is going up and down and their different fonts and... They're artistic. That's why we love them. But with the deep maps, really, they are data to begin with. So it completely gets over that particular issue. And I see deep maps being made in rivers and seas. We're opening up a whole new frontier of mapping that is mapping at a very, very high

Starting point is 00:37:44 definitional level, which creates a whole new kind of map. When you map at that level of accuracy, you are making a living map. a map that is full of so much information that we've never had. You're really mapping the world one to one. And that's sort of the Holy Grail. So there's still tons of Terra Incognita that needs to be mapped. And it's all in front of us. I think it's a big, big paradigm shift. When we all become mapmakers. We're all mapmakers, yeah. What a wonderful note to end on. Thank you both so much for joining us.

Starting point is 00:38:18 Yeah, thank you too. Thank you very much.

The a16z Show - a16z Podcast: Exploding the Map

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.