Reverse engineering the incinerated Waymo
I suspect that everyone who has ever written a dissertation has felt a deep discombobulation at the seeming irrelevance of their research relative to the state of things in the world, even—honestly, especially—if it does have some topical urgency. Because what exactly does a long-ass document likely to be only half-read by five academics (or five academics and someday Chris Rufo if you’re unlucky) do for this world’s urgent crises? Even if you’re the kind of professional sociopath who’s able to land a tenure track job that requires you to turn the dissertation into a book, by that point your topical contributions are at a minimum three years too late and you’ll be lucky if even the tenure committee reads it? What does any of this do for people on the ground in real-world emergencies?
Right now, I am struggling to take my dissertation seriously because of the massive mobilization of federal military resources and ICE agents—and massive mobilization of everyday people fighting ICE’s disappearing of their loved ones and neighbors—in Los Angeles County. And it is maybe a bit of a reach, but there is a weird intersection of geospatial technology history with this situation: the Waymos. The (Brian Eno singing “Baby’s On Fire” voice) Waymos on fire.
Tactically, it’s pretty ingenious to call a 2,000 pound vehicle to show up to one’s location as a sort of “barricade as a service”, and given all of the cameras required for the vehicle to effectively navigate space it’s wise to neutralize its surveillant capabilities. The street really does find its own use for things! Alphabet’s self-driving taxi subsidiary is the culmination of years of technical developments in hardware design and geospatial computing, as well as years of data collection and (probably, this is more my personal theory) guys sitting in traffic on the way to Mountain View angry about all the productive hours they’ve lost sitting behind the wheel.
So, inspired by the cyberpunk image of a Waymo in flames, here is a little riffing on the some of the historical conditions that made that image possible.
Disclaimer: I honestly really didn’t want to get into self-driving cars in my dissertation because I think they’re fucking stupid, but they have come up quite a bit as a facet of the broader “let’s invest a lot of money into geo, including and especially open source geo” moment of the 2010s because they provided investors with a shiny goal to throw their money at. I’m not at all an expert, I’ve barely glossed the academic literature. This is a blog, which means it’s for me to force myself to think in public, and there’s probably a lot of shit I’m missing in here.
Waymo exists (partly) because Google wanted to save money on geospatial data
Geospatial data lies at the foundation of the incinerated Waymo: lidar, GPS, and meticulously precise road network information which ensures a self-driving taxi obeys all traffic laws and drives the right way down one-way streets on its way to a fiery death. Google’s collection and organization of this kind of data began with Project Ground Truth, an internal project to build a global base map from scratch using StreetView cars, computer vision, government road datasets, satellite imagery, and thousands of contract workers doing quality assurance and manual edits.
I mentioned this briefly in my post about Never Lost Again, but the section of Bill Kilday’s memoir where he talks about the origins of Google’s Project Ground Truth is, to me, unintentionally hilarious. By 2008, the emergence of Android smart phones and increasing iPhone use dramatically increased usage of Google Maps and Earth. Tele Atlas and Navteq licensed location data—painstakingly collected and quality-checked by individual mappers—to Google on a per-mapview basis, meaning the number of times a user viewed a map screen with Tele Atlas or Navteq data. (Honestly, the fact they didn’t negotiate this when they moved to the tile architecture now common to mapping seems like the real fuckup here?) Kilday describes the negotiations leading up to the decision to break with these third-party data vendors:
In the face of rapidly ballooning costs in 2008, Daniel [Lederman] and John [Hanke] went back to Navteq and asked instead for a flat-fee deal. Within forty-eight hours, Navteq came up with their exorbitant price—without spoken turn-by-turn directions.
I realize nothing about this passage is overtly comical, but the italicized without is very funny to me. Can you believe that a company whose bread and butter is selling navigation data to Garmin and car companies wanted to be paid a high sum and protect their business interests while negotiating with…post IPO, circa 2008 Google? The audacity!
Come on, man. Talking about how Tele Atlas and Navteq “wielded immense power over John and all of Google’s mapping innovative ideas” (side note to Kilday’s copy editor, I think maybe it should it be “innovative mapping ideas”?) as though Google is a helpless little baby startup unable to take its work to the next level because of meanie incumbents is silly. Kilday doesn’t provide numbers, so we just have to take him at his word that whatever these vendors asked for really was so much more expensive than the labor of taking everything in-house. (An anonymous comment on a 2007 article speculating over whether Google might buy then on-the-market Tele Atlas claims that Google paid $30 million a year for data from Tele Atlas, but I haven’t found any corroborating material on this point.)
That being said, I do believe that Project Ground Truth could have been long-run cheaper than data licensing for a couple of reasons. One, some of that QA labor was outsourced to India (Anthony Levandowski mentions setting up the India office for Ground Truth and scaling it to 2,000 workers in 3 months on his personal website—side note, Anthony Levandowski has a personal website with a “projects” section like he’s just some guy???). The cost of labor was probably cheaper than whatever they would have paid to buy Tele Atlas or keep paying a licensing fee. Another reason was the addition of Google Map Maker, which made it easy for users to contribute map data—a tactic that was understandably criticized by open data and OpenStreetMap nerds because Google treated unpaid user contributions as proprietary data. Map Maker was shut down in 2016 but user contributions to Google Maps apparently remain pretty big. A 2020 blog post from Google Maps product manager Kevin Reece noted that more than 20 million user contributions were added to Google Maps daily. Admittedly, in the post context it sounds like that statistic was for point of interest contributions; road edits are mentioned but numbers aren’t given for how many of those edits are regularly made. (Obviously, acquiring Waze in 2013 further augmented Google’s road data.)
Cost-cutting rationale aside, it’s hard to believe that Sebastian Thrun (one of the people who Kilday claims championed the project and later co-founder of Waymo) didn’t also have some vision for the applications of this tech to autonomous vehicles, or that Larry Page approved Ground Truth solely on the basis of saving money. Thrun and Levandowski’s whole thing was stupid robot car shit when they came to Google via some Stanford DARPA challenge. They had to have seen the dual application here.
Economics of self-driving cars: seems bad, and yet
It’s very weird how many articles about Waymo’s pilot programs emphasize how great and successful they are and then note several paragraphs in that the company is still totally losing money. In a New York Times article last year a former Waymo executive now at Rivian called the autonomous driving sector “an industry of one: Waymo”, given how many auto companies and tech firms have exited the market. The reasons for those exits are manifold but bottom out in the fact that making a safe, efficient autonomous vehicle for consumer travel is just super fucking expensive. Alphabet can afford to stay in the game because they hold a lot of the relevant means of automation (the robust data, the internally developed hardware) and can afford to subsidize an unprofitable sector for far longer than a VC-backed firm or a car company that doesn’t have the in-house competencies.
There’s a way of reading this unprofitable setup as Alphabet wisely and patiently playing a long game.I think this framing maybe gives them too much credit? Anecdotally, every former or current Googler I know has described the company a bunch of little fiefdoms holding their particular little territory, not a place of Socratic debate and long-term thinking. The egos and interpersonal dynamics of VPs plays a much bigger role in determining what products or orgs live or die. It seems more likely that some small number of people with power and influence within Alphabet are keeping Waymo alive for some mix of true believer zealotry, ego, and sunk cost reasons.
Anyway, every time someone complains about vandalizing the Waymos in Los Angeles as some kind of violence against Alphabet’s balance sheet, just remember that Waymo isn’t profitable and never has been! They already lose money, the amount lost by destroying the car is pocket lint in the Alphabet balance sheet realm.
Why were/are people so into the dream of the driverless car
Answering the “how we get to the incinerated Waymo” question requires answering another: why did so many people put so much effort into making people think that self-driving cars were necessary, cool, and inevitable? And why did people buy into it? Why do they continue to do so?
There’s a “trying to sound like a guy who cares about people” answer, which is safety—humans get into car accidents, autonomous vehicles…only sometimes get into accidents and somehow it’s always the human’s fault when that happens. For the geospatial tech companies that went all-in on them, it’s a way to keep getting investment money: our routing engine or our basemap or whatever will be the backbone of an entire industry so we will make so, so much money. There’s an underlying fantastical “look we live in the shiny future I was promised by popular culture” aspiration angle, which overlaps with the “I’m just so fascinated by a technical puzzle and/or the audacity of the premise” angle.
Media complicity—unintentional or not—also seems important to understanding the atmosphere of inevitability that developed around self-driving cars. Re-reading articles from the 2012-2017 window about Google’s geospatial endeavors is sort of depressing because there’s so much implicit trust in the company that what it’s working on won’t possibly have negative ramifications. There’s so much admiration for the ambition of Project Ground Truth and the self-driving car as a project, and so, so many Borges references made without acknowledgment that Borges was talking about hubris in that story. Of course there were already critics posing pointed questions—I knew most of them—but it’s easy to forget how long it actually took for critical perspectives on tech to gain cultural cachet.
One aspect of Waymo vehicles that’s come up a lot amidst the Los Angeles protests is their surveillance capacity—404 Media reported in April on police using footage from a Waymo in a (human-driver) hit-and-run investigation. I don’t really think that in the process of building out this tech anyone supporting it was also thinking “ah yes, thank goodness another way to put networked, always-on cameras into public space such that state surveillance becomes even more powerful”—maybe some small sliver of intelligence community types or politicians, but not as some organized endeavor. (Organized endeavors by shitty people in power do exist, but recent events suggests they tend to look like Heritage Foundation projects and WhatsApp groups of billionaires mad about Columbia students in tents.)
I doubt that the burning Waymos are going to end driverless cars as a project. It might make it a slightly more expensive project because Waymo might introduce fortifications to protect their cars (adding to the speculation that their future is as a luxury product/service for the ultra-rich) but again, I don’t think that economic viability is 100% what’s driving (ha ha) Alphabet here.
Did writing this help me feel less fucked up
I mean, kind of? Most of it probably doesn’t fit into the dissertation or becomes sort of a thing-for-the-introduction. I’d gone into this project assuming that I’d only really be able to research Google obliquely without primary sources—I don’t have the right inroads to get sources there and the company can actually make good on pursuing NDA violations in a way that, for example, a company like Mapbox probably won’t (also, given I’m talking to a lot of people from their open source era very little actually was under NDA). I said up top that I didn’t want to get into driverless car stuff with the dissertation because I think they’re dumb; they’re also just part of a much more proprietary business culture that I don’t expect to get access to. (Though if you are a reader who would like to tell me everything I got wrong in this post and violate some NDAs in the process, I’m all ears! Email me!!)
In a way, maybe the most surprising thing about the burning Waymos of Los Angeles is that this didn’t happen sooner. Apparently the company is now limiting pickup sites to be outside of expected protest locations in other cities. As the current federal regmine’s campaign of terror escalates and public resistance grows in kind, I wonder how much smaller Waymo’s terrain will become.