Automated Driving: the Outside View
Also, the 2024 Progress Conference
Two quick programming notes before we begin:
Firstly, I'm introducing a new feature at the end of the newsletter; please let me know what you think! If you are reading this issue on e-mail, your e-mail client may truncate the post. You can see the whole thing by clicking on ‘view entire message’.
Secondly, welcome to new readers from Astral Codex Ten! You may enjoy reviewing the back catalogue, which features pieces on Tesla, Hyperloop, transit policy, the philosophy of progress, and other matters. This piece on A.I. risk is perhaps the most ACX-like thing I've written; Scott's authorial voice was a strong influence on it.
In every subject there are insiders and outsiders. They share an interest, but not a point of view.
Sometimes insiders understand the subject better. If I had questions about whether a particular drug was safe to take, I'd trust an pharmacologist over a doctor, and the doctor over a musician; especially so if the doctor and the pharmacologist agreed and the musician didn't.
Sometimes outsiders understand the subject better. Take fandom; ask a Batman insider—a fan who has read all the comics—why Batman doesn't kill the Joker, and you'll get tortured explanations resting on Bruce Wayne's ethics or psychology, which don't stand up to scrutiny. Ask a Batman outsider who hasn't read all the comics, and they'll have a more persuasive answer: if Batman killed the Joker, there wouldn't be any more Joker stories, and that would be bad for business.1
More high-mindedly, anthropology and sociology are places where the outside view has strong advantages. Alexis de Tocqueville wrote Democracy in America because as much as he admired Americans, he thought that he, as a Frenchman, could spot features of the American character that Americans were too close to see.
Against that example, James Mill (father of John Stuart Mill) wrote A History of British India despite never having been to India nor speaking any Indian languages. He argued that this ignorance made him a better historian of his subject: "A duly qualified man can obtain more knowledge of India in one year in his closet in England than he could obtain during the course of the longest life, by the use of his eyes and ears in India". No one takes this claim seriously today, and as one might have expected, Mill's lack of an inside view led him into a nest of errors.
The lesson is, if we want to really understand a subject, we need to take both the inside and the outside view.
Last week I returned from a trip to the Bay Area, where I visited the robotaxi firm Zoox, toured their workshop, and had a test ride. Then I attended the 2024 Progress Conference at Lighthaven in Berkeley. This was a bracing opportunity to discuss robotaxis with an audience of insiders and outsiders: the outlook for robotaxis, their weak points, and their implications for the future of cities.
Today I want to talk about the outside view.2 What do intelligent and well-disposed outsiders think about robotaxis? What questions would they like answers to? The Progress Conference was a wonderful opportunity for conversations along these lines. Before we get into the substance of those conversations, a word about the Conference itself.
In 2019 Tyler Cowen and Patrick Collison co-wrote an article, "We Need a New Science of Progress" that argued that the West was in a period of slower economic growth, intellectual dynamism, and cultural self-confidence, and had been since the 1970s; and called for a new field, 'progress studies', to understand how and why these things start and stop. This was to be an applied science, one that could be used to shake the West out of its torpor and make people excited about the future, and eager to build it. Five years in, the conference aimed to be a check-in on where things were at, and where they were going.
There are already several accounts of the conference online: to cite only three, there are Ben Parry's, Dean Ball's, and Kevin Kohler's. I agree with all of them. I found Dean and Kevin's reflections especially useful. The three of us were at the conference as Fellows of the Roots of Progress Institute (RPI), which not only put on the conference, but helped me to launch this newsletter. Likely for that reason, our perspectives are aligned. I'll confine myself to two observations.
Firstly, like Kevin, I found that 'progress studies' has become the 'progress movement'. There is still a wing of economic and institutional historians within it, but there are as many 'doers' as 'talkers' in the company. I'll quote Kevin and then paraphrase him loosely: "There is no party line and there is significant intellectual diversity. Some would describe themselves as supply-side progressives, others as state-capacity libertarians. Some are hard tech founders, some work in media, some work in policy... [but] common beliefs across most participants" include belief that technological progress has been a tremendous force for good; dynamism is better than stasis; human agency, freedom, and longevity are all goods; and the goal is not merely to describe problems, but solve them, with full awareness of and attempts to mitigate externalities and downsides.
Put another way, to quote someone whose name I can't quite recall: "the philosophers have only interpreted the world... the point is to change it".
Secondly, I learned that if you want a conference to have a specific result, you should design for it. This conference was planned to produce stimulating conversations between strangers. That design was built into both the conference's 'software' and its 'hardware'. In terms of the former, RPI founder Jason Crawford said explicitly that though the team had gone to great lengths to assemble a full slate of interesting speakers and panels, if you went to all of the formal sessions, you were doing it wrong. Instead, conference-goers were to make it their business to meet other people, share notes, and build connections. Dean notes that "A one-on-one discussion I had with someone turned into a five-person conversation, and then, eventually, into a surprise debate... with what felt like 40 people watching. Not a second of it was planned." This sort of output was exactly what the conference aimed to produce.
That output depended on the conference's 'hardware', namely its venue, Lighthaven. Ben wrote that, in his estimation, it consisted of "three lots that have been joined together then laid out with one large interconnected open space, three larger structures, and a few smaller buildings... with larger lecture halls, seminar rooms, dormitories for visitors, a gym, kitchens, and endless conversation nooks."
If you want impromptu conversations, you need to give space where they can happen. Lighthaven has this in abundance.
One such conversation I had was with Scott Alexander. One might describe him as a 'celebrated blogger'. I would describe him as 'one mightier than I, the latchet of whose shoes I am not worthy to unloose'. We happened to sit next to each other at a session, and after it was over, Scott invited me outside to talk over self-driving cars, where they're at, and where they're going. Scott gave his side of this conversation in his own write-up of the conference. Reading it over, after the fact, it seemed to me that his account was worth unpacking. Scott is an exemplar of the outside view: intelligent, enthusiastic, but someone for whom mobility and transport are not principal concerns. Answering his questions here, in somewhat more detail, might be useful for others who share his qualities, people who approve of better transport without being dedicated to it.
So what sort of questions did Scott ask?3
Are robotaxis really happening?
Yes, robotaxis are not conceptual, they're actually active, and scaling up.
It's a fair question to ask because only a few people have seen them in action. I'm one of them: I foolishly took the wrong bus out of SFO, and to get to my destination by transit I had to disembark in San Francisco proper and walk several blocks to a Caltrain station.4 Along the way, I had the thrill of seeing a Waymo in full driverless mode traveling public streets, quietly and without any drama.
I was lucky enough to see it because right now, Waymo is only operating in San Francisco (and Phoenix and Austin). But scale is happening:
Waymo is currently serving 100,000 trips per week; will offer revenue service in Los Angeles soon; and is continuing to lobby SFO officials to be able to serve the airport
Zoox is currently operating in pilot mode in Las Vegas and Foster City, California, with plans to begin revenue service soon
Cruise, after a severe setback (about which more below), is operating test vehicles in Arizona and Texas, and will return to San Francisco later this year, albeit with safety operators onboard
And that's just in the United States. China has eight major robotaxi firms which have provided millions of paid rides, and are expanding service, to the consternation of domestic ridehail drivers.
Why is it taking so long?
It is taking longer than people anticipated. To cite only one example, in 2018—arguably the peak of robotaxi hype—one prominent 'futurist' predicted that by 2030, 95% of all trips would be accommodated by robotaxi. Today, in late 2024, the percentage of all trips taken by robotaxi in the United States is somewhere below 0.1%. Why has progress been so slow?
Firstly, it was always absurd to think the technology could emerge quickly. As Jason Crawford observed recently, big tech transitions are slow. Jason describes the centuries it took for the engine and the motor to replace other forms of power generation. Automobiles took less time than that, but it took more than two decades for the number of automobiles in the USA to exceed the number of horses, and more decades after that for the horse's role in personal transport to be eliminated.
The wave of pre-pandemic hype miscalibrated everyone's expectations about how fast this technology could emerge at scale.
Secondly, there are technical issues that make achieving scale difficult. It's easier to drive a car in clear, warm, dry, bright environments: this is as true for automated-driving systems (ADS) as it is for humans. For this reason, the large robotaxi companies have naturally chosen to begin offering service in places like this, like California and the American southwest. The hardware and software are constantly improving, but it will take longer for cities with snowy and icy winters to see robotaxis at scale.
Thirdly, and most importantly, the companies that intend to scale automated driving, both the robotaxis and the goods-movement firms, are deliberately moving slowly. They are well aware, individually and collectively, that most people who've never had an automated ride find the idea of non-human driving to be creepy. Part of this is decades of 'killer robots' in science fiction, but more of it is simple fear of surrendering control; for whatever reason, fear and mistrust are widespread. Earlier this month, J.D. Power noted that only 20% of the population has confidence that robotaxis can operate safely, meaning that a vast majority of residents of any city will be skeptical of robotaxis on their streets. All it might take is one incident that frightens the public; that might be enough to result in bans that would set the whole sector back years. Accordingly, the industry as a whole has adopted a thin-end-of-the-wedge approach: start small, have a light touch, go slow, build and maintain trust, avoid attracting negative attention. Start with zones in a city and gradually extend them; move city by city, rather than region by region.
Notably, J.D. Power also found that consumers who have actually ridden in a fully-automated vehicle have 76% confidence that the technology is safe, more than 50% better than among those who have never ridden in one. Familiarity breeds acceptance. This means that, given time, the robotaxis win. The firms' strategy aims to give their product that time.
So it's not the case that hostile regulation is throttling the industry?
It's reasonable to hypothesize that regulators are overindexing on the status-quo bias of the public and holding back promising developments. That is the pattern we have seen with housing construction, nuclear energy, supersonic flight, and so many other things that progress-movement people want. But while that applies in some cases of automated driving, in general it isn't so.
There have been NIMBY-style efforts to block automated driving at the local level in California, but the state has asserted its right to block the blockers, making it clear that regulation is a prerogative of state agencies, not cities. Further, those agencies—the California Public Utilities Commission and the Department of Motor Vehicles—are open to automated driving and are helping it to spread. This openness to innovation is a good thing for the industry, especially when compared to the reflexive hostility the pro-housing and pro-nuclear interests face in most places.
Against this, the fact that regulation is happening at the state level raises the stakes. Infamously, Cruise was keeping pace with Waymo as a leading-edge robotaxi operator, but in October 2023, after only being in revenue service in San Francisco for two months, one of its vehicles injured a pedestrian. The company took an adversarial approach to regulatory oversight, leading to stern consequences: the DMV revoked its right to operate. The company has still not returned to revenue service and has not announced when it might do so. Cruise seems to have recovered from that setback, but the lesson has been learned: a mistake in one market can lead to lockouts in many. This incident validates the go-slow, go-cautious approach the firms are taking.
Taking a broader view, California's regulators have been tested, and deserve praise, but other subnational jurisdictions do not. At the behest of the Teamsters union, the governor of Kentucky vetoed a bill that would permit automated trucking in the state, which the Teamsters described as "a threat to middle-class jobs"; thankfully the legislature overrode the veto. And British Columbia banned automated driving from the province, despite the fact that no vehicles covered by the ban are available for sale in Canada. So the possibility of some, or even many, overzealous regulators prioritizing the status quo and choking growth and progress remains real. In the coming years, as Waymo, Cruise, and Zoox, and automated-trucking companies like Motional and Aurora, attempt to spread their service footprints, there will be more occasion for regulators to introduce veto points.
So how should progress-movement people help driving automation to spread?
First, do no harm.
Automated-driving firms are moving at their own pace and we should let them, rather than pushing them to grow fast. The firms want to scale up, and seem to have landed on a collective strategy of moving as fast as they can without doing anything reckless that might turn populations, and so governments and regulators, against them. I've written before that Tesla's pivot to robotaxi poses risk to the entire industry, and that's still true. The WALTER theory remains valid, and all the companies know this. One firm representative told me, not for attribution (and, to be clear, it was not someone at Zoox) that after the Cruise incident there was a private meeting among sector firms where the Cruise rep was excoriated by everyone because their mishandling of the incident threatened the whole industry. The firms know the game they are playing, and we should let them play it and not kibitz.
The way to help is to address advocacy to governments: certainly to policymakers like those in Kentucky or British Columbia that try to default to 'no', and then to policymakers elsewhere that are merely indifferent. Such policymakers need to be reminded that the world needs progress and abundance, and the proper role of government is to facilitate these good things, just as it should facilitate housing construction, a twenty-first-century energy grid, and more. In that vein: I am working with the Canadian Automated Vehicle Initiative to help produce a national strategy for vehicle automation. It is under development, but I expect we will stress that national productivity and wealth depends on embracing new technology and not squelching it. This is a message that policymakers everywhere need to hear constantly.
What is so important about shared rides?
Scott wrote that "[t]he most interesting claim I heard was that self-driving cars could help the environment by encouraging carpools".
If by environment we mean the natural world, then this claim is shaky. If two people share a vehicle rather than taking separate vehicles, less fuel will be consumed. Does that help the environment? It would, if the trips in question were in gasoline-powered vehicles, but to the best of my knowledge, all planned robotaxis are and will be electric. To the extent that the electric grid runs on fossil fuels, sharing will help, but there are encouraging signs that even the mid-term future will be one of sustainable energy abundance; this was widely discussed at the conference, and Scott's post sums up those discussions well. In such conditions, the matter of sharing vs. exclusive trips makes little difference to the state of the natural world.
But if by environment we mean the quality of urban life, then this claim is on firm ground. City streets are highly constrained spaces. The more vehicles are on them, the more congested the roads are, reducing everyone's travel time. If multiple people share the same vehicle, each rider sacrifices a little convenience, as their personal trip takes longer as the vehicle diverges from the straightest path to accommodate others; but congestion is slightly less than it otherwise would have been, making all road users somewhat better off. Sharing is better for ridehail operators too, as they can charge multiple riders for (at least part of) the same trip, making them more money.
Put this way, we can see sharing is a collective-action problem. If everyone rode in shared cabs congestion would be reduced and everyone's trips would be better; time savings from the lack of congestion would more than offset the time penalties of slightly-inefficient routing. But there is a strong incentive to defect and ride alone (that is, in one's own car, or as the only passenger in a ridehail vehicle). Defectors get the time savings of reduced congestion, and also the savings of taking a direct route. Game theory tells us what the end state is: the rational move is never to cooperate by taking a shared ride, because that approach incurs costs without benefit.
Ridehail operators, seeing this, took steps to normalize shared rides, and provide riders incentives to choose it: by opting for a Lyftline or UberPool trip, users were charged less. Indeed, in an effort to overcome rider reluctance, they were charged significantly less, with some Uber insiders claiming the discounts cost the company hundreds of millions of dollars.
The discounts were so high because fundamentally, people didn't like to share rides. Travel time was one issue, but there were (and are) more important obstacles. As Sartre observed, hell is other people: they can be noisy, smelly, or ill-behaved. That bad behaviour can take many forms, ranging from 'mildly irritating' to 'actively menacing'. And of course other people can spread disease; during the Covid pandemic, willingness to share rides plunged, and both Uber and Lyft retired their shared-ride offerings. They have since returned as UberX Share and Lyft Shared, but these services have yet to reach the scale of their pre-pandemic levels, which were anemic to begin with.
The need for shared rides isn't going away. Is it possible to overcome users' stubborn reluctance to embrace it? I think so.
Ridehail vehicles today are optimized for use by family and friends, and put everyone in close proximity. Sitting with strangers in vehicles like this will always feel unsettling to many. But with the advent of ADS, robotaxi companies can experiment with other vehicle forms.
This Zoox vehicle illustrates one way such experimentation can go: four seats in a pod, facing each other. It's a promising start, and one that opens the door to future experimentation. Imagine a vehicle the size of Tesla's proposed Robovan but with different doors, each opening into a separate compartment. Such a vehicle could offer both shared rides and privacy at once, which, to my mind, is the only way to square this circle.
To be clear: no firm is proposing such a vehicle, or has even indicated they are planning one. But I am confident that, eventually, some firm will. The logic of sharing is inexorable: the social goal of reducing congestion, and the capitalist goal of charging multiple users for the same ride, mean that ultimately someone will find a way to make private spaces in a public vehicle feasible. My confidence rests on the same foundation that the overall progress movement does: the application of appropriate technology, market incentives, and regulation can, if done thoughtfully, help to make the world a better place, delivering benefits to everyone.
Taking the outside view of robotaxis teaches us something about technological progress more broadly: the path from invention to widespread adoption requires not just solving technical challenges, but addressing human concerns and social friction points. The robotaxi firms' measured approach shows this understanding in action—moving deliberately and building trust may be slow, but in fields that involve the use of public assets and public space, it's the only way to make lasting progress. Automated driving at scale will, when it arrives, be obviously better than what it replaced: movement of people and goods will be cheaper, safer, and easier. To secure that future, it needs to emerge carefully, avoiding unnecessary blocks and vetoes by arriving carelessly. Patience here will be a virtue; but so will reminding policymakers and regular people of the advantages this technology will unleash. That is what the progress movement can do for robotaxis; and by being an inescapable advertisement for the benefits of new technology, robotaxis will be good for the progress movement.
Off-Ramps
This is a new occasional feature I’ll be adding to the end of some posts where I provide a few links to the most interesting things I’ve read on Substack lately.
Thanks for reading Changing Lanes! Please let us know how we’re doing by answering the poll below. And if you’d like to respond to this post, please leave a comment.
If you end up having this argument, just tell the insider that Batman doesn't do it because he knows it's futile: Gotham City executed the Joker once and it didn't take.
The inside view is interesting too, particularly of Zoox, but that will be the subject of a future piece.
I emphasize "sort of". These are not verbatim, but paraphrases.
Actually, I didn’t take the wrong bus; I took the right bus but going in the wrong direction. The fact that the airport is not a terminal destination but a midpoint on a line is wild service design.
Andrew, well done, this essay. I took some notes for our End of Driving textbook, 2nd edition. John