Given there’s not any momentum that I’m aware of to convert light rail in any American city to automated (and increase security with labor savings) despite this being a solved problem for decades, I’m skeptical we’re anything less than 20 years away from automated busses replacing bus drivers in Americas blue cities.
I mean, they really are. Look at America's seaports. They are much less efficient than Asian and European seaports because of the lack of automation and not being operational 24/7.
I find this fascinating because I am guessing the Euro seaports also employ Unions? Has anyone tried to take a "look how much better the German union is doing compared to yours?" approach when negotiating?
Unions are standing athwart the path of an oncoming train yelling "help us!" Why is it that (mostly non-union) Americans expressing outrage at poorly-executed trade policy that created economic devastation in specific industries is considered authentic and legitimate political expression, while unions trying to prevent economic devastation for their members in other economic sectors is considered greedy, grasping and counter-productive? All the embarrassment belongs to -- and all of the anger and frustration should be directed toward -- the public policy industrial complex (of which this blog and its ilk are a part), which has yet to devise or implement any credible or reliable solutions to these increasingly frequent failures to reconcile technology-driven efficiency and net economic benefit with stakeholder-specific harms. Typing shruggies and "that's sad"s while worshipping at Schumpeter's tomb doesn't really cut it. Not intellectually and certainly not politically.
It is possible to conclude in a given case both that a) the interests of the Union do not align with the interests of society as a whole and b) the Union is a perfectly reasonable organization acting in good faith.
The union doesn't have to be "greedy" or "grasping" to be counterproductive.
Also, as an aside, I don't think the tone you are using is very conducive to persuading others to adopt your position.
As to both logic and tone, please read the comments and many like them in this space. There is a LOT of reflexive, gratuitous union-bashing that does not confine itself to cases in which union interests are carefully and clearly distinguished from broader societal interests. Nor is it the case that -- whatever pretenses we all share -- the interests of "society as a whole" are somehow better known to or more accurately represented by commenters here than by unions making demands for their members. If this is meant to be a haven for consequentialist / aggregationist analysis, then the first principle is that all interests of all members of society are adverse to some degree all the time. Unions trying to capture more surplus for their members are competing with shareholders trying to capture more for their portfolios, are competing with executives trying to capture more for their salaries, are competing with consumers trying to demand lower prices. Reduced mortality from better AI driving is "paid for" with job losses and increased private costs for transportation and investment capital diverted from other productive uses.
Ultimately, nobody is against generating net positive sums from technological progress, but everybody disagrees about how best to allocate the net positive sums. Most people claiming to represent the interests of "society as a whole" are really just arguing for the net positive sum, which, as I say, is the easy part (and what, I believe, you are doing when you say union demands are counterproductive). My only "position" is that union demands should be seen as operating at the distributional level, on par with other distributional claims exercised through tax rates, spending cuts, minimum wages, executive compensation, etc. Calling unions "a few inefficient rent extractors" as one commenter here does (is that acceptable "tone"?) deliberately obscures the fundamental equivalence of all these claims and claimants, but heeding his call to develop "a coherent set of programs for wealth transfer" would make that equivalence clear in a hurry.
Any instance of a union is a good example of where unions are clearly not our friend. If you think poor people should get more stuff, you should support a coherent set of programs for wealth transfer, not a few inefficient rent extractors setting things up nicely for themselves.
That's fine for the 9 actual consistent neoliberals out there, but in the real world, the vast majority of anti-union people are also anti-government transfers.
I also, think as much as hurts neoliberals to hurt, people would rather be unionized and make more wages than be cut a pity check from the government because they're unskilled workers.
To your first point, if that is, in fact, a popular opinion, then the job is to convince folks otherwise. If it is an unpopular opinion, it would be nice to force them to own it.
To your second, "people" is way too broad. Once you account for union fees and the like, there are plenty of professions where non-union workers would probably make more. In general, Unions trade wages for job security, benefits, work/life balance, etc. Also, they solve for the lowest common denominators, meaning in any given union, the worst workers are overpaid and the best workers are underpaid. "People" is the whole bell curve, so I would disagree.
As soon as I'm convinced any sort of job has actual good ways to determine whose the "best workers" or "worst workers" outside of obvious outliers, maybe I'll start thinking about having so little solidarity with fellow workers I'd screw them over for a few bucks more a year.
If your only goal in life is to make as much as possible and screw everybody else on the way, then yes, I'm not shocked you're anti-union.
As far as convincing people, that's on the McKinsey grads and other neoliberals who stay awake at night in horror that a non-college educated auto worker, grocery cashier, etc. might be earning a little more than their supposed worth is, all as CEO's making combined, billions than their supposed worth to convince conservatives not to be ghouls.
I mean... literally every job I have worked at, there were a people that everyone could identify on both ends of the spectrum. If you are in a profession where this is hard, though, I can see how the union could benefit both employees and employers.
"If your only goal in life is to make as much as possible and screw everybody else on the way, then yes, I'm not shocked you're anti-union."
Ah, I see we've moved on the part where you've already made up your mind about everything, and I must be the bad guy because I didn't should "You go, Jesse!" Carry on, then.
I think the wealth transfer is only one part of it, though—there are all sorts of employer abuses that wouldn't be corrected with just more money. But I'm a union member so I'm definitely biased here!
I think there's a range of anti-poverty programs that could prompt employees to say "take this job and shove it" to abusive workplaces, but that range probably overlaps considerably with one where labor force participation drops off a lot.
Well I'm a Southerner who is reflexively somewhat anti-union plus I had some really bad experiences with the UAW that cemented that view, so you are preaching to the choir.
But I don't want to come across as a crank on the topic, so limited my comment to a specific example where unions are very clearly harmful. =)
To be fair, pro-union people want all or most jobs to be unionized, or at least governed by a collecting bargaining agreement, as is the case in some European countries. So in theory everyone would benefit, not just a small number of unionized workers.
For sure I'm not sold on whether this would be a good thing, I wish Matt would do a piece on it actually.
That would address the issue of it being a small number getting the advantage of a union, but there's other issues to consider, too. For any given union, there are rents extracted to the benefit of the union members and the detriment of everyone else. If that were the only effect, then maybe economy-wide unionization would more or less even out. But those rents are also *inefficiently* extracted, making everyone poorer. Those countries in Europe with economy-wide unionization are much poorer than the United States.
I think it's clear that Matt is pro-union for political reasons, regardless of the economics.
When it comes to policy, Matt isn't a political hack. He holds positions because he believes the empirical evidence backs them up. He would never say "unions are good because lefties think so" - he certainly isn't afraid to critique lefties when he thinks they're wrong.
I'm skeptical of widespread unionization for the reasons you say. But yes, Matt does seem to be pro-union, so I wish he would do a piece making the case for it, beyond simply saying "the current US collective bargaining model is bad" - literally everyone already agrees on that, the only reason it persists is that the disagreements on an alternative model are too wide.
Matt has changed my mind on other issues, maybe he could do the same with unions. I would also love for someone to take a dive into why sectoral bargaining seems to work so well in Germany but so poorly for US automakers.
I think the reason most people in the comments are not as pro-union as Democratic politicians is they're mostly college-educated well-off in demand knowledge/etc. workers who have never needed the specific backing of a union to help them or protect them.
I don't think Matt is a political hack, but I also think he is somewhat of a political realist. He recongizes that unions are a huge boost to the Democratic party, so he supports them for that reason, even if the economic case doesn't make much sense. Not because he's a party hack, but because the Democrats tend to advance causes he believes in.
Hasn't he already declared for strong unions with sectoral bargaining (Nordic Model)? Think it was a couple years ago so maybe worth a revisit in this era of higher inflation and interest rates.
Management and business owners are also trying to extract rents. It is neither unique to unions nor are they necessarily the most effective rent extractors at the expense of the larger population.
"Management and business owners are also trying to extract rents."
Yep, they sure are! Which is why we have anti-trust laws that explicitly make that rent-seeking behavior illegal. But we have laws that explicitly make union rent-seeking behavior protected. I think that's bad.
I don’t think we have the demonstrated technology for driverless light rail. We do have demonstrated technology for driverless fully grade separated rail, like the Vancouver skytrain and the new Honolulu system. But you have to design the system with full separation of the relevant sorts.
It would be interesting to know which, if any, cities in Japan or Korea or China or Taiwan have made any move to convert existing rail to driverless, to know how feasible this is. (I’m not aware of any such cities - I think Copenhagen and Paris have a few driverless lines, that were built driverless from the start.)
In the US, BART trains are fully automated for travel, with drivers only operating the doors at stations I believe that the main blocker to automating this (besides drivers' unions, ofc) is the cost of installing platform screen doors at all stations.
Paris has done a couple of conversions. But having worked on feasibility studies of similar projects in my city, the issue is quite simple - you'd need billions to retrofit (most) existing lines to make them suitable for automation, and if you had billions available for capital investment (you probably don't), the RoI would be *much* higher on other projects (including entirely new automated lines!), so no one wants to do it.
Given the ubiquity of car use in America, there will be a larger public call for increasing ease of use to car commuters with this technology. So I would be optimistic; the pressure for regulatory changes could very likely be faster and stronger. And then once people use AVs to get to work from wider sprawl (and thus more people), they'll want more reliable transit for the final commute in.
I think the idea of self-driving vehicles is very exciting. But I wonder about the use case being developed versus what I would find useful enough to pay for. Robotaxis don't seem to offer that much savings versus an Uber or Taxi itself -- the driver just isn't that much of the cost when compared to the capital costs of a Waymo.
On the other hand, it would be very, very helpful to be able to have a few drinks, hop in the car and let it drive me home rather than risk a DUI. Or plop the family in the car in DC at 11:00 PM and wake up at Grandma's house in Jacksonville at 7:00 AM, fully rested and ready to go.
It seems the technology and regulatory path to the use cases I describe above will be measured in decades rather than years, though.
I bet you scolds make it illegal to be intoxicated while occupying the “drivers seat” of an autonomous vehicle. This really is tragic, because reducing DUIs (and permitting freer and more joyous socializing in the suburbs) is a major part of the AV value proposition.
In fairness, this isn't just being a joyless scold; there is something to be said "no system is perfect, and if the car's control system goes on the fritz suddenly, a conscious human driver should be able to take over."
Of course, this also removes the possibility of sleeping in the car (red-eye driving), which would be another major use of autonomous vehicles, so... eh... I dunno? What is the cost-benefit ratio of "I want to get drunk and have my car drive me safely home" vs. "I may fall asleep in the car and never wake up if the driving system decides to break down at the wrong time"?
In safety engineering world, the man in the loop is often a key part of a redundant fault tolerant system.
Yet I wonder how alert we can really expect the 'driver' of an autonomous car to remain, such that they are able to quickly and correctly take over when the car goes on the fritz? Especially if it is presumably a very rare occurrence.
I would wager that even fairly diligent people will let their guard slip significantly.
I _might_ if the car's procedure was to say "I can't handle this", pull over safely to a stop, and _then_ ask me to take over. But never if I have to do it spontaneously.
And even the situation I described also assumes I've been through Driver's Ed.
"Yet I wonder how alert we can really expect the 'driver' of an autonomous car to remain, such that they are able to quickly and correctly take over when the car goes on the fritz? Especially if it is presumably a very rare occurrence."
Yes, this is a point that I've literally been bringing up for at least a decade in discussions about self-driving cars. Advocates for full self-drive frequently waive off questions about irreconcilable conflicts in safety commands and other issues by saying that the car will simply throw control back to the human driver, but, when I took driving classes many, many years ago, I remember being told that a driver has less than 2 seconds on average to react to an adverse event. Obviously, if you have a self-driving car that has to perform some sort of assessment of a situation and then return control to the driver, that's going to shorten the available time to react further (even if we're just talking milliseconds), so the system is (1) going to require a driver to be just as alert and aware as normal, while (2) simultaneously putting the driver into a state of tedium by relieving them of the hundreds or thousands of tiny adjustments a driver has to make under normal driving conditions.
Eh... most truly fault tolerant systems are built around the idea that the human is *guaranteed* to make a mistake, so the machine should work regardless of whether the human screws up.
Not entirely sure those are equivalent. I listened to Senator Chris Murphy speak the other week, and he talked a lot about the crisis of social isolation, etc. and he's a fairly normie Democrat.
Mask politics are tricky, and both sides got some stuff wrong in that debate. But I don't think that wholesale makes Democrats the ant-social policy. At least, I hope not...
I'm not denying communal drinking has benefits, I'm wondering where the tradeoff is between "I want to get drunk with my friends and then pass out in my car" vs. "I want to stay awake and aware so I could take over if Murphy's Law happens on my way home."
This can be answered almost a priori. Autonomous cars are safer than sober humans. Ergo, any policy regime should definitely try to substitute autonomous cars for drunk humans.
Why would you want to be in the drivers seat if you're drunk? Lie down in back. "Freer and more joyous" (by which I assume you mean "drunken") socializing is already a thing in the suburbs and can be accomplished in the front passenger or any of the rear seats.
Why do you say the driver just isn’t that much cost compared to the capital? If your driver needs to make $15 an hour, and the vehicle goes 10 miles per hour, for 100,000 miles, then the cost of the driver is $150,000 over the life of the vehicle. If the vehicle can get a longer life than 100,000 miles, or the driver gets more expensive, then the budget available for the capital expense to become a savings gets larger.
The challenge here is that present dollars are more expensive than future dollars, so 150k up front is going to cost you more than 150k over the lifetime of the vehicle (because of interest or equivalent financing cost paid or forgone.)
That said, labor is expensive enough and capital is cheap enough that your autonomous vehicle doesn't need to be *that* cheap up front for things to pencil out. And because robotaxis don't require as much downtime as human drivers, you can at least theoretically get good revenue-miles in fairly quickly.
For taxis / ubers / robotaxis you have to denominate and compare everything in costs per mile. Robotaxis can accumulate miles and recover capital costs much faster than either taxis or ubers, and it's not yet clear that their initial capital cost will be even 2x the capital cost of those human-driven vehicles. I don't expect the expensive, multi-sensor approach of Waymo to survive unless sensor costs drop dramatically.
AVs probably cost way more than $150,000 today. But in 10 years they'll probably be that cheap, and in 20 years the automation parts will probably be <$10,000
I wish a self-driving car was an option for my dad, who really shouldn't be driving anymore (at least, not without a backup that can automatically take over if he gets distracted or falls asleep) but doesn't want to give it up. I'm sure a lot of older folks are in the same boat, especially as more boomers reach their 70s and soon 80s.
But unfortunately it sounds like that type of usage is pretty far away as well.
My Dad is in his mid 70s and lives in the suburbs of orange county. He and my mom frequently drive to LA to visit my sister and her grandkids. He purchased a Tesla Model 3 a few years ago without the full self driving software. There was a recent update to the software that has been such a significant inlmprovement that they gave him a free 1 month trial of full self driving. He and my mom have been using it to drive door to door from OC to LA with no issues. The software still requires supervision but even at this point he trusts it more than my mom's driving and he chooses to use it because it puts him in the role of an observing passenger and reduces driver fatigue. He is now no longer worried about needing to move out when he is no longer able to drive reliably.
This is a great use case and good example of how the current, imperfect version of FSD can still provide benefits. I have been using the same trial for the last few weeks and while I agree that "supervising" is much less taxing than driving, I am still surprised by the number of mistakes the current (12.3) version makes. I get at least one significant mistake per trip (wrong turn despite correct mapping) and one scary error every 5 trips or so (veering off-map to follow train tracks intersecting a street at grade at a 45 degree angle). It's impossible for me to know how close this version is to "real" FSD that does not require supervision at the technical level, but it does not feel "decades" away by any means.
This is a real problem that forces many older people to put themselves and others at risk. It's shocking how difficult it can be for an elderly person to get by without a car, even in a relatively compact and service-rich independent living community. We are currently going through this with my wife's 92 year old mother in SoCal. We were able to convince her to give up driving (with some help from the DMV) and have had pretty good luck with a service called "gogograndparents", which offers uber and lyft services with drivers trained to expect and cater to elderly clients, as well as a phone number to book rides for non-app savvy seniors. Not a perfect solution by any means, but might be good enough. Good luck!
I think the capital-intensive Waymo model is going to lose. The cheap Tesla model will win. Then the capital cost is a drop in the bucket compared to paying the driver. And then the transition can happen very, very quickly.
it certainly seems that lots of people (even very smart folks in the space) have just completely written them off for quite arbitrary reasons without even considering the quite real advantages the business model has in terms of cost structure / scalability...
I definitely agree that if Tesla solves self-driving in a fully generalized way, it’ll be a huge win for the company and its business model.
But the bear case is worth discussing here. This is a problem with very high execution difficulty— significantly harder than working out a patchwork of effective local solutions. And in its own way, it’s actually also going to be very capital-intensive to solve— as the evolution of language models etc has shown, incremental improvements in neural network performance tend to require increasing compute and training data inputs by orders of magnitude (eg: GPT-4 has about ten times as many parameters as GPT-3), and it’s far from clear that Tesla will get to acceptable fully-generalized performance before these scaling costs get prohibitive. Management’s record of over-promising and under-delivering in this domain many, many times also makes it a lot harder for investors to trust that they actually have a good handle on the problem and a good path to solving it. I don’t think it’s impossible that they pull it off, but there actually are good reasons to be skeptical.
The advantage of the solve-more-limited-cases Waymo approach is that we know it actually works with feasible levels of compute and training data.
Tesla actually has an order of magnitude more data to work with than anyone else (over 1.3 billiion miles of driving data, versus Waymo's 40 million).
Additionally, from a capital stand point...Tesla has been lucky to (somehow?) subsidize this by getting people to pay thousands of dollars a head to beta test the software for them...rather than rely on low wage test drivers to drive in circles on a map.
Per Elon, they are "no longer compute limited" on the training side and the most recent cadence of FSD revisions has proven that (there is a new one going out every 2 weeks, it seems), so it seems like most of that capital has already been spent.
I'm guessing the take up will probably follow a similar path to cars themselves. I get the sense in popular imagination, cars as a mass-produced product widely available to the masses was a quicker process than it really was. Henry Ford and the Model T is held up historically for a variety of reasons, but I think one misconception that's been created about the Model T is that it allowed us to go from a world where everything was horse and carriage to suddenly everyone had a car. But by the end of the 20s, still only 25% of Americans families had a car. https://www.digitalhistory.uh.edu/disp_textbook.cfm?smtid=2&psid=3396#:~:text=In%201929%2C%20a%20quarter%20of,the%20American%20way%20of%20life.
I bring this all up because I think Matt makes a pretty good case that we're in the 1918-1920 stage with driverless cars. Once they allowed on highways, it really will be the case that there is a real use case to take a driverless car over an Uber. Even if we are faraway from everyone owning a driverless car outright, I think we are on the cusp sooner than you think of driverless cars fundamentally changing the nature of urban and suburban living just like Model-T cars did in the 20s.
Until there's competition between multiple robotaxi companies in the same market, there's no real reason for a robotaxi company to charge much less than what a human-driven taxi would cost.
I remember seeing a study a few years ago where some researchers hired chauffeured cars for a group of people to simulate how people would use privately owned AVs. The big takeaways were that total VMT increased ~50%, trip volume increased 25%, and 80% of the number of new trips were deadheading (driving the car with no passengers in it). Transit usage cratered among people with access to a private AV. I’d be nervous of broad based adoption of private AVs without a major congestion tax to manage demand on our roads - the increase in VMT and deadheading would cause huge snarls of congestion without a way to manage it.
I too was amazed that Matt didn't consider the zombie miles problem, and that, as Nathan just said, didn't consider that cars not in parking spaces still have to be physically present somewhere. And yes, I'm guessing that Matt would just counter with congestion pricing, and that's fine, but there's a limit to how effective cars, automated or not, can travel with all the merging and intersections that occur. At some point, there's no way getting around the need for mass transit.
It seems plausible AVs empower much more aggressive sprawl outside a city with commuting to a mass transit stop (bus or light rail) for the final 30 minutes. In fact, a few zombie miles to a charging location could help solve the parking problem that would otherwise hinder this mass adoption for dropping off at a bus or train station to work.
I don't see why that would be the case. When people originally got cars and spread out into the suburbs, that did not spur the development of train stations with big parking lots so they could commute into the city... it spurred the development of suburban office parks (where most American jobs now are), and the replacement of many buildings in America's downtown cores with parking, for the jobs that remained. The jobs sprawled too.
In the UK cities have been introducing low and ultra low emission zones. You're vehicle must be more efficient than some standard or you need to pay a (hefty) fee to drive into the zone. I think it was started in London and has since been rolled out to more cities so it can't be completely politically toxic.
I could see these type of schemes ratcheting up until driving to mass transit station on the outskirts will be cheaper and more convenient than driving all the way to the centre.
I think there is enough in the centre (as opposed to in the suburban office parks) to make that worthwhile
But when the suburban office park took off, there wasn't much remote office work possible to do and the car needed to be driven by the occupant. If both of those variables change for a fraction of workers, you end up with a different calculation.
If anything the move to remote work has intensified the desire for short, simple commutes when people do go in. The pattern we've seen since COVID is that office occupancy has been hit harder in city centers than suburbs. Few people are going to want to mess around with transferring to a train to a dense job center when they could just commute to an office in or near their suburb - whether the first part of the trip is in a self driving car or not.
It seems like the zombie miles problem is a problem of sole ownership vs highly available on demand per-trip renting?
If I own an AV, it's going to do zombie miles for me, and also sit around and do nothing a lot too.
I thought the dream is more like yellow cabs in Manhattan - they're just everywhere, they're hopefully mostly doing useful work, and they're so plentiful the service seems on demand.
Right, as soon as I read the words "you don't need a parking space" I knew there was trouble, because that means that the cars are taking up space on the roads instead. There aren't secret out-of-everybody's-way parking alcoves that AVs would disappear to during downtime; they'd just keep roving around, and we have solid research on TNCs (Uber and Lyft and so on) that that's exactly what happens.
They can of course double, triple, quadruple... park since they can just wake up and shuffle themselves around when one needs to enter service. So while "no parking space" may be an exaggeration, it could still be transformational to land use.
We would end up building those out of the way parking spots, but it would take time. Eventually having a big garage every few blocks would be Waymo efficient than the current system of street parking.
If these are zero emission vehicles (electric yes, but tires also produce emissions, that's still a problem that needs solving) and they're cruising slowly and safely, I'm not sure deadheading is a big problem.
For robo taxis, I wouldn’t think so. If demand is high, ie a time of high traffic, there won’t be much deadheading. Pickups and drop offs would be close. If demand is low there would be some deadheading, but then it wouldn’t matter as much. Deadheading is definitely an unintended consequence of this tech that needs to be monitored.
In the robotaxi vision, existing roads become the new parking lots. During peak, the robos are all driving on the road, during lull, those robos that are still working can easily and safely navigate around the "waiting" robos that are parked in the same roads. But this vision is not fully compatible with a "mixed" system of robos and human drivers, so you will need some version of private vehicle / human driver bans to make it work. This is hugely inefficient land use when compared to transit's footprint, but as a society we have consistently chosen the convenience (and personal efficiency) of door-to-door transport over land use efficiency, so I think that will probably continue.
"Liked" in part for use of "deadheading" -- a term I unselfconsciously use on occasion at work (a law firm) and then have to deal with blank looks from everyone else.
One of the biggest business model challenges for autonomous vehicle operators will probably be figuring out how to maximize utilization/avoid deadheading-type situations. I think that in practice, robotaxis will be most economical if they also serve as delivery or small-load short-range freight vehicles some of the time, and I also suspect that if they become common enough, some businesses will adopt models and locations to take advantage of the cheaper leg of the trip. (Analogous to people containerizing recycling and soybeans to ship from the US to China.)
I like AV grocery deliveries coupled with food waste / compost pickups. Laundry and dry-cleaning pickup / drop-off is obvious. Amazon should be picking up all its packing waste after each drop off (if it isn't already), offering some kind of discount/credit for packing returned in re-usable condition. Some percentage of robo-SUVs that take commuters into city centers could be redeployed as package delivery vehicles in the city during working hours, then re-assigned for the commute out, then used as suburban meal delivery vehicles and DUI-avoidance taxis in the evening.
I have heard the argument that we should ban self-driving cars because people should be taking public transit and self driving cars will displace public transit. I suspect that public transit could hold on to some riders if it is safe, reliable, fast, clean and pleasant, but that means improving the current transit system.
Correct - this is the real issue and congestion management taxation / TOU charges are the only viable solution. Transportation has a huge peak simultaneous use problem that AVs do not make better and will likely make worse, at least in the near term. In the early days of this discussion (MY's piece today is basically Tony Seba from 2020) a lot of emphasis was put on increasing throughput and reducing congestion by using AVs capable of moving very fast and very safely (so safe that they would not have to stop at lights, for example). Another strain of the discussion was a combination of systems to reduce traffic and deadheading with multi-rider vehicles, including AV carpools for school pickup / dropoff (a huge peaking problem in many suburbs) and smallish AV buses / vans for routes underserved by existing transit. TOU charges and VMT taxes obviously reinforce the latter.
But really, nothing works politically until there is widespread acceptance that scarcity economics apply to road-space in many places at some times of the day.
Seems like converting public transit to smallish AV buses / vans for routes underserved by existing transit is a big opportunity to increase bus frequency on a given route, and to have more routes
I live in an area in Asia with a relatively high number of people with private drivers.
When you drive your own car, you generally park relatively close to your destination. This pressure allows parking garages to exist.
When you have a driver (or an autonomous vehicle, presumably) the driver avoids paid parking and simply drives to some neighbourhood where they can park for free.
If you're commuting to work you could probably tell your autonomous car to find a free parking spot within 10-15 miles. You could even have it move every 2 hours or whatever to avoid those kinds of rules.
edit: I think the logic would be that if AVs become common, most people would not own one, because the economics would be insanity. It'd be like owning a plane. Why would an averagely wealthy person own a plane, with its enormous capital costs, instead of just buying plane trips when you need one?
And obviously the economics are different. If VMT and trip volume has a direct impact on marginal costs, then they probably don't go up (i.e. if you don't own the AV, there is no economy of scale).
I dunno. Can't deal with snow and ice and requires absolutely massive capital investment to expand aren't really convincing me that the technical problems are going to be imminently solved and it is underhyped.
The transition will happen slowly and then all at once. As soon as one company (probably Tesla) has cheap robot taxis, everyone else will copy them to get cheap robot taxis. Just how no one could do what ChatGPT could, and after ChatGPT was released, there were half a dozen clones within a six months.
I wouldn't be sure of that at all. It wouldn't surprise me if the artificial neural networks used for self-driving cars were trained on conditions largely without snow and ice, and bear in mind that artificial neural networks have only passing similarities to the natural ones and don't necessarily have the same capacity for generalizations as natural ones. There's also the question of what sensors are being used on the cars. If they are using ones more easily confounded by snow and ice, that could be an issue, too.
Humans use sensors that are confounded by snow and ice, but most of us have enough common sense to pull over and scrape the ice off the windshield if it gets to that point.
I'm thinking more of situations where the electronic sensors would be confounded in places where humans' sensors would not.
Also, you bring up an interesting point. Humans often can tell when their sensors are producing input that gives them a very uncertain view of their surroundings, and they'll respond by taking that uncertainty into account, say, by (as in your example) pulling over. Dealing with uncertainty is a very open question in machine learning, and it’s far too easy for an AI to deal with uncertain inputs by being confidently wrong, even in conditions where a human would say "I don't know".
I don’t think solving the problem of, for example, being the first vehicle driving on a road with six inches of newly fallen snow, is only training. But that certainly strikes me as a place where non-human sensors could be engineered to have an advantage.
I think you misunderstood. "Training" is a term of art in machine learning, and if one is talking about supervised machine learning in particular, it's basically nonlinear regression.
Think of it as a kind of curve fitting, where you have a bunch of sample points-- pairs of inputs and outputs -- generated from some process, a function with lots and lots of parameters, and you try to adjust those parameters so that (1) the function gets as close to those points as possible, and (2) the function can approximate the original process, that is, given an input, generate close to the same output(s) that the original process would generate. Doing that second part *reliably* is hard.
I don't know, human overconfidence is a huge problem in driving and almost every other activity. At least AI can be "tuned" - I get a very different ride when I set Tesla FSD to "Assertive" mode than when I set it to "Chill"...
I'd say that humans and AIs tend to be overconfident in different ways. A human wouldn't mistake a staticky picture of a panda for a gibbon, but an AI might (and one already had). It can be difficult to predict how an AI will go wrong, as it can do so in ways that seem arbitrary and don't fit with human intuition.
I think the bigger issue is weird reflections off every surface around you, particularly if you are sensing distances by emitting laser or radar or something else humans don’t usually use.
Humans who learned to drive in areas with significant snow and ice are very good at driving in it. Automated vehicles that learned to drive in warm climes (Waymo in San Francisco, Ford in Miami) probably could learn if they were instead tested in wintry places (Minneapolis or Waterville, Maine, say).
A large portion of both the US and global population lives where snow and ice just aren’t a problem. But importantly if there is a good toehold of commercial viability the remaining technical problems will start to be addressed much faster.
At the likely pace of driverless car development and implementation, by the time we should worry about their ability to deal with snow and ice we won't have to worry about their ability to deal with snow and ice.
Driving through flooded streets (hi, Ken from MIA!) and on melting asphalt -- that may be another thing.
This is exactly right, and Matt just breezing by it was silly. You literally cannot use AVs in the much of the eastern half of the country for most of the year! My city has over 180 days with precipitation, what are those vehicles supposed to do?
Matt, I generally like your articles, but sometimes you display an obliviousness to the obvious.
You talk about cab & truck drivers losing their jobs as "sad". It isn't "sad", it's a bloody crisis for the individuals involved, particularly if they didn't just lose their job, but that entire class of jobs has been automated away. I'm glad you'll benefit from saving a few cents on your Amazon deliveries (as will millions of others) but that doesn't obviate the harm to the people who's trade has vanished.
Perhaps you'll suggest they learn to code? Do I need to spell out the history here?
Much like free trade, the benefits are broadly distributed (and likely a net positive for the economies involved), but the pain is sharply concentrated. And one reason why we've become more reactionary is the complete ignorance of "knowledge workers" such as yourself have of the trades. If you don't have a realistic plan for how millions of suddenly-endangered taxi & truck drivers have to make a living, particularly older ones, don't be surprised by the viciousness of their reaction, particularly at the ballot box.
Just to quibble - being a truck driver is a career, and it's very tough to lose a career. Being an Uber driver is by design just a random job, and Uber drivers can pick up other random jobs. And unfortunately all the cab drivers already lost their careers to Uber drivers 10 years ago.
If you look at the mobile aloha project at Stanford, we have Rosie the Robot you can build today, for around 15K (project includes a BOM to build yourself).
It isn't just drivers. Hotel maids are now automatable. Robots are being trained right now that have massive increases in dexterity - partly from advances in reasoning capability of LLMs and especially as those get implemented in smaller, faster and less expensive to run models.
Of course, most email jobs that involve taking information from one thing and passing it back and forth are on the way out too. GPT4 class AI can do most of that, and do it just as poorly as a 40th percentile mediocre middle manager, which in many contexts, is probably good enough for government work. The automation will hit most levels.
I personally don't see this going down without a means to distribute the productivity gains in some manner from AI, but that is a political question, not an economic one.
Agreed. I think it's important to take a broader view of the poltical landscape and coalition politics.
America is currently a threat from a fascist movement. The conservative movement is inspired by Viktor Orban in Hungary, who managed to change election rules to lock himself into power and use the government to reward his allies an punish his opponents.
If the Dems are seen as being ok with throwing truck drivers, bus drivers, etc. to the wolves, that just reinforces the image as elitists who don't care about working people and will cause us them to lose more working class support. The loss to society from the end of democracy would greatly outweigh the efficiency gains from driverless vehicles, so we need to prioritize coalitional politics here.
I feel like the time horizon gap between economical autonomous trucking and "I don't need my car anymore because I can reliably schedule an autonomous taxi to and from work every day" is a really big one. Actually I'm kinda skeptical that second one ever pencils out.
Yes and no, right? Some places have enormous fleets of non-automated taxis available for rush hour, and it seems like only a matter of time before the self-driving versions are strictly superior to that.
“Some places have enormous fleets of non-automated taxis available for rush hour…”
And the residents in those places have already decided if the trade off of paying to commute by taxi is worth not having to own a car. I don’t see what changes when it comes to driverless taxis, unless those are dramatically cheaper to hire.
Right, it seems like the "self driving" part is a relatively small hurdle to that sort of paradigm. Seems you'd have a lot more challenges redefining the "car" part to even begin to make that plausible.
Uber is already that. Replacing existing Uber vehicles with FSD-enabled Teslas feels like the easy part of the problem. "Cheap" is relative. Robotaxis will under-price (non-robo) Ubers enough to take market share, then raise prices to meet the market demand.
The economically viable model for this does not involve these vehicles waiting around all day, which is what commuter vehicles do 90% of every day and what Ubers do most of the day. Robotaxis can be cheap when capital costs can be offset with additional economically productive activity. This would entail using them for many more jobs than just ferrying one human to work and sitting in a parking lot doing nothing for 8 hours, or worse, incurring costs for staying parked (typical non-AV commuter vehicle) or for just driving around (typical Uber). The additional jobs could be as simple as providing back-up electrical capacity while parked and plugged in (or off-taking mid-day solar curtailments in CA), or as complex as running around delivering packages. The efficiency comes from improved logistics and automated price signaling ($2.50 / hour for backing the grid right now v. $1.50 net for delivering that package v. $0.04 per kWh for offtaking 50kWhs over the next hour), which is software driven so ultimately cheap and scalable.
Maybe. But that’s ultimately only viable if the vehicles have a comparative advantage in those tasks. I don’t see why a machine capable of safely self-driving on crowded roads while keeping passengers comfortable and/or entertained would do those other tasks as cheaply as machines purpose-built as specialists. Sure, it’d be better to make some income from contributing in tiny ways to grid reliability and the like, but that doesn’t mean that income will offset the costs rathe than just make the losses smaller.
I'm not sure what paradigm you are using to compare the alternatives. If the question is "how much would I pay to commute in a robotaxi rather than buying a commuting car, paying for parking, etc.", then it's pretty easy to calculate the breakeven point. Net capital costs for purpose-built robotaxis are going to shake out in the 3-4 cents per mile range, "fuel" cost in the 5-7 cents per mile range and maintenance costs (almost entirely tire replacement) in the 4-5 cent/mile range, for a total of 15-16 cents per mile or $15 per 100 miles traveled. That is a very low cost base from which to deploy a wide range of services from 15 cent/mile delivery to 5 cent per kWh battery based services. These "cars" are gig economy robots entering an existing market for some of the services they can provide and an emerging market (especially on the energy side) for other services they can provide. Every element of their cost base (AI and logistics software, high capacity batteries, electricity) is currently on or will soon be on a declining cost curve (except -- again and weirdly, tires). There will be a market for these services at the margin and beyond.
I don’t have the subscription for the first article. The second article doesn’t show anything about trends, just current occupancy rates. I suppose for peak congestion purposes, it doesn’t matter whether a higher fraction of people are doing office commutes, or if total white collar employment in these cities has gone up - what matters for number of cars on the road at peak is the number of people going to offices at peak.
But for the “rush hour” issue for self-driving cars, what really matters is the *ratio* of peak to off-peak travel, to figure out what fraction of vehicles sit around unused and depreciating the rest of the day.
I have a (pretty expensive) CoStar subscription - their news feed has a monthly story about office occupancy levels from data aggregated by Kastle, a company that sells building access control systems. Those reports have shown very definite trends of more people going to the office. The trends are not uniform: The 2021 increases were much larger than subsequent ones, big drop in December 2021 from the Delta wave, large dips around holidays, etc. But the general trend was large increases Q2 - Q4 2021 and then a leveling off with much smaller gains overall through today. I’ll see if I can find a story later and grab a fair-use screenshot.
None of this is to say remote work won’t increase over the long term, only that we haven’t seen that yet.
I will note that the title of the chart is referring to this detail:
"Office attendance averaged 51.2% of pre-pandemic levels in 10 large cities tracked by Kastle Systems for the week ended March 27, down slightly from the prior week’s 51.3% and remaining below the peak 53% reached in the final week of January."
I don't read too much into that. As you can see, it's a pretty messy chart looking at it week to week, but the general trend appears to be a levelling off after climbing out of the Delta wave.
If you're looking for more data on the topic, CoStar, Kastle, and Placer.ai are the best sources I know of. None are cheap.
This article is *aggressively* incorrect in being completely blind to the fact that the first 90% of the problem takes the first 90% of the development time, and the next 10% of the problem takes 90% of the development time, and the last 0.1% takes the most development time of all.
What is the technical issue that makes operating in Houston, Dallas, San Antonio, Orlando, Tucson, San Diego, Sacramento, Las Vegas so different from operating in Austin, San Francisco, and Los Angeles? Just expanding to the full city of all of these cities would be pretty transformative.
I will say that if you get 99.9% of the stuff solved, they should already be far superior to human drivers at that point.
The 2 most likely failure modes would be:
1) Sometimes you have to call in and have someone remote drive the car around the place where it just gave up
2) Some areas will not be suitable for these cars - but almost nobody needs to go those areas, and those that do will either have a car they drive themselves or rent one.
This is true, but already baked into the tech timelines. The 99.9% was solved years ago, all progress and projections on AV are about solving for the 0.1%.
What do you mean here? I know Sharty was being vague, but you seem to have something more specific in mind. What is the 99.9.% that's solved in your mind?
In my mind, the jump from L2 autonomy (driver assist functions) to L3 autonomy (driving is *mostly* automated, but requires full human attention still) was the 90% mark, but we're just *barely* there now. L4 is still years away.
The 90% etc. framework that Sharty used, and I continued is indeed vague (90% of trips? Miles? Driving hours? Regions?) We could spend a lot of time trying to define it better and then debating progress and projections. But Waymo is currently running L4 in areas of AZ and CA - the tech is here, at least in certain climates. I'd say they've solved for at least 99.9% of the technical needs in those regions, and the proof is they're doing it! (Cruise might've solved for 99.5%, and that wasn't enough...again this is really vague and debatable language.)
If Waymo was told they had an a) unlimited capital expenditure budget, and b) that regulators and public perception would only hold them accountable to being at par with human-level safety metrics, we'd be at L4 in much of the country right now. That's what Matt means by expansion being primarily blocked by economic and political hurdle right now.
So I disagree with Sharty's contention - that this article is "aggressively incorrect" in not treating the timelines right. It's correct that the last 0.1% is much harder than the first 90%, but incorrect in stating that this fact isn't already baked into the assumptions of this article and of every company and regulator that's grappling with the issues in it. Everyone involved, including Matt, knows this, and has incorporated into their assumptions. Maybe some are still too optimistic, but when those people say "SDCs are imminent", they're saying "We have already or will soon solve for the last 0.1% of conditions that make autonomous driving less safe, on average, than human driving."
(Btw, like all tech, there's a parallel race happening in China. I'd give it a 50%+ chance that society-changing adoption happens there first, for political reason.)
In 2008, most traffic engineers thought AVs were 50+ years off. In 2015 tech folks thought they'd be ubiquitous in 5 years. I think we're seeing they were both overweighting their view of the world.
This is a project development timeline truism based on projects with ordinary-level human intelligence as the binding constraint. I don't think we're in Kansas anymore...
Being a long-haul truck driver is a great job for antisocial people. You only have to talk to others a couple times a day. You don’t need to be that fit either, at least compared to warehouse or construction workers. You can drive around the country being obese and socially stunted and still make $50k a year, more if you go hard. It’s hardly obvious that a full employment economy will offer the kinds of jobs that displaced truckers can fill or that whatever jobs they do find will pay as much. I’m not saying other working stiffs should be forced to pay more for shipping to subsidize the long-haul trucker lifestyle, but I do think many men will be broken when they lose their niche.
It seems super problematic to encourage and allow a workplace that so clearly encourages obesity. Do truckers actually want to be obese? It’s like saying coal mining is great because you get to be in the dark all day and get black lung.
I think this is a pretty fair response to the truck driver job loss debate. It’s an industry that breeds comorbidities like no other. I feel like romanticizing its existence doesn’t do anyone any favors.
I remember well Durant’s line about Sparta “where nine-tenths of the men were slaves, and not even the masters were free.”. I suspect the act of keeping fellow human beings in cages is inherently degrading and stressful. However, I want that to be true, so don’t trust me on this point. Still, the morbidities are third world in scale.
More likely "being in close contact daily with the kind of human beings who must be kept caged for the sake of society at large" is the stressful part.
I don't know the answer to this question, but the book 'The Secret Life of Groceries' has some very disturbing insights into the stress associated with many truckers' jobs.
Trucking doesn’t really encourage obesity— fit truckers get laid much more often than fat ones and are less likely to have to pay for it. Trucking is merely a workplace where one can be obese without creating major performances issues.
This is the ultimate sedentary occupation (including possibly the getting laid part depending on you know, venue) so obesity / CVD is not a huge surprise.
There could be other factors (age, sex, demographics, etc), but that is so great of an increase over baseline that the job probably is contributing at least somewhat.
It could also be a selection effect. If my BMI is 32, I probably can’t get a job as a carpenters apprentice because I couldn’t handle it physically (unless I’m a body builder or something). I can’t get a desk job unless I can play nice in an (often feminized) office environment. I can get a warehouse job, but can’t keep it unless I can stand on my feet and/or walk briskly all day. So trucking is a nice, medium skilled niche where I can make a solid blue collar wage, maybe even aspire to become an owner-operator, spend lots of time alone, etc.
I don't think that you'll get much sympathy among left-leaning audiences when the 'victims' are primarily men, blue collar, and relative asocial or antisocial. And also primarily overweight/obese.
It's one of those weird niche areas where the current right and left is inverted w.r.t. concern for workers.
Depends how far left you lean. Biden is about as pro-worker and has as much or more sympathy for blue-collar work as any President since... Grant? And at the extreme end of the re-distributionist left you will get UBI proposals that could facilitate an acceptable glide-path from trucking to retirement without material deprivation. The uncomfortable place is the mid-left, which knows that we ultimately need and want the economic efficiencies of automation, but also knows that it cannot yet forge a re-distribution scheme vast enough to address the emerging problem (see, e.g., Andrew Yang 2020). I agree that trying to fill that yawning gap between the reality and the politically achievable with fuzzy faith that market magic and "human invention" will solve the problem (eventually) feels and sounds lame. But if the center-right starts (continues?) to tilt re-distributionist, everything to its left will fall in line with it.
I basically agree, but we shouldn’t be naive about the benefits and burdens of productivity increasing technologies. Working class wages did not increase during the early industrial revolution. Hand loom weavers were reduced to destitution and sometimes literally starved. If Amazon and Walmart have enough market power, it’s perfectly possible that the surplus created by autonomous trucks will merely increase their profits and pad the accounts of the rentiers who own them. In the long run, the invisible hand of the market will find an efficient equilibrium, but possibly after millions of lives are immiserated.
And let's not forget that industrialization led to lots of unrest as old livelihoods were rendered obsolete...and arguably AI could do the same thing except on a much greater scale, and a lot more rapidly.
Technical advancements do not automatically turn into corporate profits. If capitalism is working properly (because the government is doing its job to prevent monopolies) then prices fall (because that is how competition works).
That’s a rather heroic assumption. There’s a lot of market power out there and antitrust enforcement has been spotty. I do think most retail is diffuse enough that most productivity gains in shipping will be passed through to consumers as lower prices. However, this completely depends upon the structure of the market snd things could change if, eg, Amazon or Walmart buys Kroger.
If you want to take your case of Amazon, however, they are popular because they are cheap and convenient. Obviously if shipping costs fall to zero this helps them but it also helps all the brick-and-mortars and the other large by mail companies (i.e. Walmart) Clearly they would lose some market share if they didn't bother to compete. So perhaps some of it is absorbed by profits but the more they absorb in profit, the more alternatives exist which are cheaper.
So really, then, if you are buying from Amazon, you are doing it for convenience since there would be cheaper alternatives elsewhere. And, if people want to pay more for convenience, isn't that their prerogative?
You actually don't have to structure your society to follow the logic of the market. The success and persistence of NIMBYism demonstrates that even enormous economic costs can be borne, as a matter of practical politics, as long as the costs are sufficiently opaque.
Where economic growth and human flourishing are in opposition, we can choose to prioritize human flourishing.
Now, being a truck driver apparently sucks, per surveys, so I would support gradually reducing the number of truck drivers in a way that minimizes disruptions to the lives of existing truck drivers.
I do think we should in general try restructure the economy and society such that people who are not great at school and not great at customer service can have an easier time living a decent life.
But it also seems really difficult to fill those jobs. Unfortunately for me, a person who hates driving, I spend a lot of time driving between the East Coast and Michigan for family reasons. Virtually every truck on the road is advertising jobs with the company. I also met a guy during the pandemic who owned a small regional trucking company (he had previously run logistics for a large furniture manufacturer). He had so much difficulty hiring people who didn't do things like rarely showing up, or taking the hiring bonus and quit at the first opportunity, that he was considering retirement.
I'm not disputing your point, it's about the worst job I could imagine for myself but some people are able to make a good living (though others struggle with the current industry structure). And I don't agree with Matt's take, especially for this context. But it does seem as if the industry as a whole is short-staffed.
There’s a structural reason for that. Trucking is very competitive. Few barriers to entry (you can buy a working rig for under $200k), lots of competition, no economic profits. Jobs with higher barriers to entry pay more. This is why trucking pays only a bit more than warehouse work.
This has to be a "no one goes there any more, it's too crowded" thing. Trucking is very competitive and that's why trucking companies can't find enough drivers?
The market clearing wage still isn't being set, probably because, and this isn't an attack, many of the owners of said businesses either can't afford the market clearing wage _or_ they are still in a 2010's views on wages.
Karen Levy had a really good book on the trucking industry intersection with both working standards, unions, and Automation/Technology. If you find this topic interesting, I recommend Data Driven: Truckers, Technology, and the New Workplace Surveillance.
I thought one of the things we learned from the supply chain issues during the pandemic is that we are short of truck drivers as a country and that our current truck driver population is older than is ideal. It seems like some of that is the result of it being a fairly shitty job that makes it hard to have a stable family life, is extremely hard on the body, and requires works to often buy their own rig so that they are taking on huge debts that make it harder to set reasonable limits on their work hours and conditions. It felt like while the whole "nobody wants to work anymore" things was not very real "nobody under 50 wants to be a truck driver" thing was pretty real and a lot of the proposed solutions seemed to be about how to trick younger people into this industry without really changing it because that would make everything too expensive.
I am much more sympathetic to the human toll of people losing a job and especially a career that they have invested a lot of their lives to. But it also feels like if were making a list of careers that people are at risk of losing over the next 30 years due to climate change concerns an automation: coal miner, truck driver, and hotel maid are all ones that were objectively not great jobs. I think people in those displaced industries need a lot more support to change career paths and that we should assume that asking them to move out of their communities is fair or reasonable even if that seems to maximize economic efficiency. (I'd love to see us focusing bringing other manufacturing jobs and real employment opportunities to coal country as well as the rust belt.) I also suspect that given the limitations on scale and location, even if AV trucking started for short haul in warm urban areas we might run out of trucks around the time we got to the point where there was a real risk of us running out of trucking jobs.
Yeah I think this is a rare bad (as opposed to one I merely disagree) take from Matt. I think the experience of the Amazon checkout-less shops should make us far more sceptical about how personless these tech gimmicks are, and how easily they can scale. I think the reality is that to be anywhere near an acceptable level of safe, driverless cars just need far too much backroom support that it'll ever make financial sense to use them instead of just paying someone to drive
the fact that we aren't even having a conversation about how much arbitrarily safer we (royal) expect them to be in order to greenlight widespread adoption is disconcerting... as actual lives will be lost in the interim based on where we set that figure, as it will take more and more time to iteratively hone those driving algorithms with data.
I think they're going to have to be ridiculously safer than human drivers for people to accept them. People's fear of harm intensifies when they give up control, even when their actual risk declines. Think about how many people are afraid of elevators and airplanes, even though these are the safest forms of travel ever invented.
The other challenge for self driving cars is that the very things that make them safer will make a lot of people absolutely hate riding in them. I think people will find them too cautious and too slow. Think about how many road rules the average driver breaks on a typical day.
I think a contributing factor is that the failure modes of self-driving cars can often be very unusual compared to other car accidents. Like who wants to die because their car mistook a white truck for a cloud? (Yes, I'm sure this one has been fixed, but there will continue to be other strange accidents that will get overweighted in people's minds.)
Yeah I think a lot of people here just say, well, if it's safer on average than human drivers, that's OK.
But human drivers are the aggregate of all drivers, including some drunk, asleep, angry, on medication, looking at their phones etc.
Obviously, the AI could be safer on average just by being an otherwise entirely mediocre driver but none of the above.
That doesn't mean that it's safer than *me*, assuming I'm a healthy alert driver who isn't do anything wrong.
I think people will not really want to accept this technology until it's not just safer on average by not making the mistakes humans make, but until it's much safer on average and won't make the mistakes humans *would never* make.
On that last point, it's true in the short term but in the longer term, there are ways to allow self driving cars to go much faster safely while following closely in convoys. Drivers can't do that.
The Amazon example was a leadership failure, not a technological one. It's totally possible to do what they were trying to do. They spent too much time and effort believing the data labeling approach they started with would scale, when it turns out they needed entirely different approaches to the problem of detecting what goods are in the basket.
Even big well funded companies will have failed AI initiatives that have more to do with organizational dynamics than the tech itself.
the Amazon story is another example of Liberal Media's own "misinformation" problem, as the real issue with that story was Elite Educated reporters unfortunately have zero knowledge (and unfortunately zero interest) in understanding how AI training works end to end and there was a lot of naive (or worse, bad faith) pushing of an incorrect narrative that it was Indians doing all the actual work. Ultimately the thing that really killed wider expansion there is that consumers didn't really care for it in *most* cases, probably before you had to scan your credit card BEFORE you bought anything (and perhaps before buying nothing and walking out). It's fine for Amazon to scale back after leaning that resistance was there.
well, it was largely liberal / progressive aligned outlets (with their negative bias towards tech in general) and media influencers who ran with an technically incorrect narrative about the story (probably based on headlines alone in many cases) such that many of us talking about it here have the completely incorrect idea of what happened. Doesn't seem like conservatives cared all that much about the story at all because it had nothing to do with "wokeness"....so their outlets and media influence folks just ignored it.
well, the NYT is already a billion dollar tech company that makes much of it's money from getting people addicted to word games (many people who don't care about news they also serve up) that is well on it's way to having a monopoly over national / international media coverage (sooner or later...) so that "negative lens only" focus on "Big Tech" is only going to get more problematic for them...but it's quite hard to ignore that tech companies being demonstrably better at every form of advertising than regional newspapers and fancy magazines once hoarded has basically made all these reporters wear that anti-Tech bias right on their sleeves...
But the Times is a great example (along with Substack) of the trend away from advertising as the mainstay of online revenues. I think their reporters are fine with Wordle subscribers reducing the need to run MyPillow ads. As for monopoly, I am not too worried about the Times' paltry 10M subscribers taking over either the gaming world or the actual world. The console and PC gaming platforms have hundreds of millions to billions of subscribers and users worldwide, as do Facebook and Twitter as disseminators of "news content". I think any negative bias against "big tech" is a reaction to companies that have grown incredibly powerful very quickly with near-zero public scrutiny over or accountability for business practices that are wildly anti-competitive.
Well, at my local supermarket, something like half of customers use automated checkouts. In the last 10 years, this has become pretty standard where I live (Boston area). So an alternative lesson from supermarket checkouts is: full automation might be very hard, but replacing 50% of workers can happen really fast!
I think he's talking about Amazon's stores that supposedly would ring you up as you walked out the door, not something where goods had to be manually scanned at all.
(1) Environmental: doubling down on car mobility, even if electric, still puts a high floor on transportation emissions because a 5000 lb vehicle will always be a very inefficient way to move a 150 lb person.
(2) Economic: Matt is dreaming of the wonderful cheap sprawl these things will enable. But sprawl is quietly expensive, and is actually already bankrupting us. This is because it has huge infrastructure costs that are paid for by the federal government that nobody thinks very hard about until it comes time to maintain it. Doubling down on sprawling development patterns connected only by cars has, of course, a dozen other issues, from the merely dystopian aesthetics to the lack of walkability that is a major public health issue.
(3) Technological: Matt correctly points out that the urban problem is harder, but like everyone, he's still underestimating how *much* harder. It's not 10x harder, it's 1000x harder. These existing cabs have millions of dollars in route-specific training, and exist only in good weather places with fairly freeway-like infrastructure. This is very difficult to scale. And now you'll get pressure to make the urban landscape less chaotic, which is a fundamental feature of urban places. We'll probably have automated gates for crossing the streets, or maybe we'll make bicycles illegal, just to make self driving cars work.
I try to avoid hot takes, but I'm going to get this out there because otherwise it won't leave me alone...
Self-driving cars are just another manifestation of the myriad costs of sprawl. The massive R&D effort going making into them work is expensive on its face, the consequences of that effort not being spent elsewhere are expensive, the fares will be expensive relative to transit, the impact on the built environment will be expensive. American society is so desperate to cling to suburban lifestyles that it struggles mightily to engineer its way out of the downsides, costs be damned. Even if it succeeds, this doesn't feel like the right direction.
Pressure to make the urban environment friendlier to autonomous cars is an angle I hadn't thought about, and I don't think it's just scaremongering because that's exactly what has happened with human-operated cars already.
Counterpoint: now that I have kids I see that the robo taxi is unlikely to replace cars for kid transport: too much variability in car seats, lots of other specific stuff parents have in cars, the huge mess kids create.
When robo taxis are safe enough, you won't worry about putting your kids in car seats. (Admittedly, that probably won't happen in the timeline of anyone who currently has kids.)
I just had a baby as well and dealing with the bulk of stuff I just need to have in my car now has been eye opening. My car has become another storage place for things too bulky to keep in my apartment like the car seat, stroller, diaper bag (which are always huge for some reason). Not having my own car means I need to magically create storage space inside. One thing to note against your comment specifically though, is that for a good chunk of babyhood the kid can’t lift their head up/ control their body enough to sit in a regular seat no matter how safe the actual ride is. And in the unlikely case that an advanced self driving car is in a serious accident it will almost definitely be fatal to the child passenger not in a car seat compared to adult riders because of the placement of airbags.
I live in a country where car seats aren't required.
For kids under 4 or 5 years old, not having a car seat is extremely unpleasant.
They stand up, sit down, walk across the back seat, kick the back of the seat in front of them, pull on door handles, roll down windows, push and pull every single button available.
Leaving them in the back seat alone isn't tenable at all. So you have to have them sit on your lap while you try to contain all that chaos.
Even if robo taxis were very safe, they are still going to accelerate and decelerate in ways that a toddler doesn't expect and will result in them falling since they will be standing up and walking around the car. Yes, they won't actually be hurt in any meaningful way but you still have a hysterically crying toddler.
Car seats offer a massive benefit beyond just safety. They keep a toddler in one place. Once a kid gets a little older they can finally sit in one place and learn not to roll down windows, open doors, etc.
I just don't see it as likely that cars will actually become toddler proof even when autonomous when you can instead just say "use a car seat".
Furthers car dependence for non-commute trips, makes communities less walkable, pollutes the air and water with brake and tire pollution, takes energy away from other uses (thus requiring more capacity from a system that is already struggling to keep up with demand), takes away land from local agriculture and keeps people farther away from natural amenities. Gasoline is not the only problematic aspect of car dependence.
Clean air, clean water, and safe access to healthy movement aren’t aesthetic preferences, they are critical public policy concerns. It is much harder to stay fit and healthy when you can’t integrate your exercise into your every day life; designing a transportation system around sitting for hours a day is a costly mistake. But I guess aversion to early death is also just an aesthetic preference.
No it isn't but... is this a standard you apply to all policy arguments? "We should try to locate trash dumps away from cities" is per-se illegitimate?
Walkable cities, local agriculture, and distance to natural amenities are asthetic preferences. Road pollution is an actual issue, but is more of a consequence of the weight of EVs rather than self-driving technology.
Yeah, the political economy of congestion pricing should be easier for EVs. To an extent, people will feel that the big corporation that owns the robotaxis is paying, not the poor working class driver. This may be the one weird trick we need to get congestion pricing (for AVs only, which will be fine if the unit economics still appeal to riders vs. owning and driving a car)
I just read Hannah Ritchie's excellent book on how to make environmental progress, and it enforced this. So much of deforestation is due to eating beef and dairy that there's not much room left for harm due to wood from farmed trees. She also emphasizes that cement and steel are the large carbon cost in building homes.
I don’t see any way that any kind of taxi service works for long-haul commutes—the whole nature of taxi service is that it operates in a limited area, so it can drop someone off and pick up the next fare nearby. It can’t be economical if it’s got to go places where there’s only a fare one way (I know there are crazy stories about taxis/Ubers going ridiculous distances, but we all know it’s not normal).
Just buy your own self-drive car for that. And if your place has no parking, it can self-drive to the nearest garage for storage and self-drive to pick you up in the morning.
It could be economical. Especially if those long-haul rides typically take place overnight or during some other slower time.
You aren't paying the driver, so it's really just a matter of keeping the asset being productive as much as possible.
Imagine it is driving from City A to a suburb of City B. Assuming the infrastructure to refuel and/or maintain the robotaxi is in City B, it can just operate in that city afterwards for a while, until it can get a fare to go back to City A.
Yeah, we’re still at the honeymoon stage where everyone focuses on the aspects of self driving that they like. We will get more miles traveled in cars and more outlying development, although personally I think that’s a reasonable trade off.
I think there's a certain level of density in between leafy suburb and dense urban core that people hate living in because it's full of congestion with none of the benefits of transit/walkability (e.g. Cicero, IL). And I think this is what NIMBYs are fighting to prevent, and why people talk about missing middle housing.
My theory is that robo taxis will have an outsized benefit for medium-low density places (too expensive for the suburbs and too congested in downtown). That could help bridge the gap between low and high density. If every level of density is nice to live in, then there's no reason to fight densification, and I can at last have my urbanist utopia.
I'm not sure about self-driving buses saving money. I take the bus a lot, and a large part of what the driver does is kick people off who are throwing things out the window or fighting, giving people who are lost transit advice, securing people in wheelchairs and making sure they get on and off the bus safely, etc. A larger number of bus passengers are disabled or elderly or don't have technology than in the general population. We may not need drivers, but they will probably still need staff to make it a pleasant and efficient experience for passengers.
Also, a benefit in being able to have a person who is spending their full time and attention making sure conditions are ok for passengers, rather than only intervening when things get out of hand.
Yes, I thought of this issue too -- I suspect that (at least absent unionization) "conductors" or "attendants" could be hired more cheaply than drivers.
The technology may actually come later to city busses than other forms of transport.
Busses have passengers that are not wearing seatbelts, or are even standing.
As far as I'm aware, Autonomous Emergency Braking technology has been slower to be adopted for this use for that reason too. It's one thing in a car, or even an intercity coach, but bus drivers have to be be taught to plan ahead to avoid heavy braking, or they'll send people flying.
It may well be we get to the point of having reliable autonomous cars, but they're a bit "reactive", and sometimes don't plan / predict well enough ahead, but are able to avoid an accident with some last minute braking, but this tech won't be good enough for city busses with standing passengers.
places with nonzero parking costs need to be very careful. Right now it’s not possible for my car to loop the block for an hour while I shop, dropping the cost of that antisocial behavior to ~zero would totally destroy the road network even in midsize cities
The key would be to build centralized garage/lots for the car to go to and wait before coming back to pick you up. I see eliminating the need to get a close parking spot in urban areas as one of the major benefits of the technology.
I think the idea is that the car would not be personally owned in most cases - it would be part of a fleet and would go pick up someone else while you’re doing your chore.
Makes sense but I don't really buy it. Owning a self-driving car would be awesome (Siri drive my van packed with all my stuff to Utah, I'll take a plane. At 30c/mi that's competitive with rental rates for non-self-driving cars today!), and once you have one surely it would be cheaper to use it than rent?
It would be pretty easy to have automated police vehicles apply fines for excessive non-point-to-point driving. Or since parking would be in abundance, just drop parking fees to be comparable to per minute fuel costs of circling the block.
Once you make a significant percentage of cars self driving then parking lots for those cars can store 3x as many cars due to fewer lanes and tight spacing. When a car needs to leave, it advises adjacent cars and they shuffle around to "free" it. This improvement in parking density is one of the "holy grails" of autonomous vehicles.
Maybe I'm just one of those jaded folks that Matt describes in the opening of his piece, but he's written is basically the same stuff that I heard 10+ years ago when I worked in the auto insurance industry. Yes I know some self-driving cars operate, but not a material or meaningful percentage.
There are still a lot of known problems (e.g. bad weather), but I'd imagine that there are also a lot of unknown unknowns that will emerge as future problems as this stuff continues to be tested in more situations and more scenarios.
And, while I hate to sound like a boomer or a tech doomer, I would worry greatly about the scenario that my self-driving car gets hacked while driving down the freeway at 70 mph and runs itself into a guardrail or oncoming traffic or something like that.
That said, I hope I'm wrong because I have no particular emotional attachment to driving and would prefer to not have to do it.
I'm slightly leery of of drive-by-wire cars in general, I'd prefer mostly mechanical control systems. I'm really aggressively repulsed by the idea of a car with DBW systems that are accessible to over the air manipulation. I guess I assume Tesla's are probably susceptible to this currently... I suppose I need to investigate what manufacturers might be installing DBW systems that might be accessed without a hardwire. I'm already on the lookout to avoid vehicles with remote stop systems.
Throttles may be drive by wire but steering and brakes are fully mechanical. Steering is frequently power assist(ed). Otherwise, it would not require hacking but merely a blown fuse or a stalled engine to send you careening off the road.
Right, but once you're getting into things like electric wheel motors you start dealing with differential acceleration controlled by the DBW system. Things like that. Maybe some all wheel drive systems use computer control to determine which wheels get power? Anything like that should absolutely be isolated from wireless input.
Sure but there is nothing magical about electric motors. They don't need a network connection to the outside world. The Tesla has battery charge-and-protect (avoid explosions) circuitry but they don't allow that to be accessed by the outside world. It is true that there are more opportunities for problematic design choices as the percent of vehicle functions under digital control rises. However, there is no need for these to be accessible to the outside world.
I have led design teams working on ICs that control lasers in consumer environments that were 100x the power needed to char eyeballs and start fires. In my experience, it is fairly standard to have a type of redundancy where a complete set of safety systems is independent of any system which can be modified by a user or hostile third party. In our case, even opening the sealed unit would disable functionality. The only risk we felt unable to mitigate was one where a user unsoldered the laser and rewired it directly to a power supply.
All that compute power does in fact need a network connection for updates and telemetry, as evidenced by the fact that Teslas do have that capability built in for that reason.
Right, I am mostly ok with electronically controlled vehicle functions... as long as the are unnetworked. I don't trust software/encryption to prevent those being compromised. If someone at least has to physically plugin in a laptop to my car to access those systems I'm happy.
It's a danger long term. I didn't think highly of the Canadian trucker protest, but the extent to which Canada went after bank accounts and credit cards raises serious slippery slope concerns. There will always be some legitimate justification for doing stuff, but the story of government power is it never stops at the legitimate justification.
Not impossible at all, but I'd imagine it would be easier to take a control of a situation where you're actually driving and aware of your surroundings vs a self-driving car where you're sitting in the back or in the front passenger seat, not paying attention, etc.
Admittedly, I drive a 2009 model year car and I've been out of auto insurance for a while, so possible that I'm underestimating the semi-autonomous features that are in 2024 model year vehicles (though I don't think I am)
Well just to take one example, a lot of modern throttles are fly-by-wire, meaning that the computer, and not you, controls the mechanical action of putting fuel into the car. Now if someone hacks that, how are you going to control your car?
Fair enough, but I'm still not convinced that problems like this wouldn't be either more frequent and/or more severe in self-driving cars.
Unless your argument is something to the effect of: cars today could be hacked today but aren't, so it's probably not an issue for self-driving cars in the future. But I still see the risk as potentially more severe for self-driving cars
I don't see why. If someone wants to hack, they can hack. The big obstacle is that there's a lot of encryption in these systems. But an insider at GM or Tesla (or Waymo) can probably do it.
When you're the driver and (hopefully) paying attention to your surroundings, more likely you can steer/brake your way to safety vs in a self-driving car
What we discovered with automobiles was a lot of unknown unknowns, some good and some bad. I don’t think anyone realized people would start living 10-15 miles from their workplace and commuting, or that traffic jams would become a thing, or that parking lots would reshape urban development.
Why aren't you worried about your non-self driving car getting hacked while you are driving down the freeway. Are you aware of a specific vulnerability difference in the two cases? It would obviously be easy to create the same level of immunity in the two cases if that was your goal. (edit: I see you have already engaged in this discussion....)
Yeah, admittedly, Dilan has convinced me that I've been underrating the possibility of present-day vehicles being hacked and how severely they can be hacked. I can still imagine more severe scenarios in a world of self-driving cars, but really more imagination than reality at this point
Counterpoint: it’s already possible to get much safer roads, judging by international example. Will robotaxis be an excuse to avoid solving problems we already know the answer to?
What actually is the answer that we know? It would be cost-prohibitive (especially including the psychological costs of living in a place under construction for several years) to completely transform the streetscape to match the streetscape of other countries. But we don’t know which fractional changes might get us most of the safety benefits. (We do know a lot of things that will give us small bits of those benefits, but those should already be an excuse not to do the other things.)
New York, Rhode Island, and Hawaii all have less than half the road deaths of the United States, while Mississippi, Arkansas, South Carolina, and New Mexico all have nearly double the road deaths of the United States. The streetscape in Canadian and Australian cities is much more like the former states than the latter states.
Self driving robotaxis owned and operated by big companies will probably be pushed to be safer and safer the same way airline flights have been pushed to be safer and safer -- much safer than human driven cars anywhere. The dynamics that drive this are that airlines are much more motivated to help safety officials than drunk drivers and other private individuals driving cars, and that airlines can more easily standardize.
Yeah, this is what concerns me. I was all on on self driving 10 years ago. I was frustrated by its seeming lack of progress, but what I’ve come to realize is that hoping and waiting on technology like this to solve the issues of urban planning is a mistake. There are a lot of things we can and should be doing today, and dreaming of a self driving utopia is probably just a distraction.
Given there’s not any momentum that I’m aware of to convert light rail in any American city to automated (and increase security with labor savings) despite this being a solved problem for decades, I’m skeptical we’re anything less than 20 years away from automated busses replacing bus drivers in Americas blue cities.
This is a good example of an instance where unions are clearly not our friend.
Unions here are standing athwart history and yelling stop!
I mean, they really are. Look at America's seaports. They are much less efficient than Asian and European seaports because of the lack of automation and not being operational 24/7.
Similarly, any shift to electric vehicles is going to decimate the automotive manufacturing workforce, because EVs are so much simpler overall.
The UAW's recent renegotiation 'success' be damned, it is just going to accelerate outsourcing.
Or we'll put up massive tariffs on EVs and massively overpay for American-made EVs.
I find this fascinating because I am guessing the Euro seaports also employ Unions? Has anyone tried to take a "look how much better the German union is doing compared to yours?" approach when negotiating?
Our pal David Simon is on the side of the American unions there! https://www.youtube.com/watch?v=0-k5bLQ9Epw
Unions are standing athwart the path of an oncoming train yelling "help us!" Why is it that (mostly non-union) Americans expressing outrage at poorly-executed trade policy that created economic devastation in specific industries is considered authentic and legitimate political expression, while unions trying to prevent economic devastation for their members in other economic sectors is considered greedy, grasping and counter-productive? All the embarrassment belongs to -- and all of the anger and frustration should be directed toward -- the public policy industrial complex (of which this blog and its ilk are a part), which has yet to devise or implement any credible or reliable solutions to these increasingly frequent failures to reconcile technology-driven efficiency and net economic benefit with stakeholder-specific harms. Typing shruggies and "that's sad"s while worshipping at Schumpeter's tomb doesn't really cut it. Not intellectually and certainly not politically.
It is possible to conclude in a given case both that a) the interests of the Union do not align with the interests of society as a whole and b) the Union is a perfectly reasonable organization acting in good faith.
The union doesn't have to be "greedy" or "grasping" to be counterproductive.
Also, as an aside, I don't think the tone you are using is very conducive to persuading others to adopt your position.
As to both logic and tone, please read the comments and many like them in this space. There is a LOT of reflexive, gratuitous union-bashing that does not confine itself to cases in which union interests are carefully and clearly distinguished from broader societal interests. Nor is it the case that -- whatever pretenses we all share -- the interests of "society as a whole" are somehow better known to or more accurately represented by commenters here than by unions making demands for their members. If this is meant to be a haven for consequentialist / aggregationist analysis, then the first principle is that all interests of all members of society are adverse to some degree all the time. Unions trying to capture more surplus for their members are competing with shareholders trying to capture more for their portfolios, are competing with executives trying to capture more for their salaries, are competing with consumers trying to demand lower prices. Reduced mortality from better AI driving is "paid for" with job losses and increased private costs for transportation and investment capital diverted from other productive uses.
Ultimately, nobody is against generating net positive sums from technological progress, but everybody disagrees about how best to allocate the net positive sums. Most people claiming to represent the interests of "society as a whole" are really just arguing for the net positive sum, which, as I say, is the easy part (and what, I believe, you are doing when you say union demands are counterproductive). My only "position" is that union demands should be seen as operating at the distributional level, on par with other distributional claims exercised through tax rates, spending cuts, minimum wages, executive compensation, etc. Calling unions "a few inefficient rent extractors" as one commenter here does (is that acceptable "tone"?) deliberately obscures the fundamental equivalence of all these claims and claimants, but heeding his call to develop "a coherent set of programs for wealth transfer" would make that equivalence clear in a hurry.
The tone certainly alienated me.
Any instance of a union is a good example of where unions are clearly not our friend. If you think poor people should get more stuff, you should support a coherent set of programs for wealth transfer, not a few inefficient rent extractors setting things up nicely for themselves.
That's fine for the 9 actual consistent neoliberals out there, but in the real world, the vast majority of anti-union people are also anti-government transfers.
I also, think as much as hurts neoliberals to hurt, people would rather be unionized and make more wages than be cut a pity check from the government because they're unskilled workers.
To your first point, if that is, in fact, a popular opinion, then the job is to convince folks otherwise. If it is an unpopular opinion, it would be nice to force them to own it.
To your second, "people" is way too broad. Once you account for union fees and the like, there are plenty of professions where non-union workers would probably make more. In general, Unions trade wages for job security, benefits, work/life balance, etc. Also, they solve for the lowest common denominators, meaning in any given union, the worst workers are overpaid and the best workers are underpaid. "People" is the whole bell curve, so I would disagree.
As soon as I'm convinced any sort of job has actual good ways to determine whose the "best workers" or "worst workers" outside of obvious outliers, maybe I'll start thinking about having so little solidarity with fellow workers I'd screw them over for a few bucks more a year.
If your only goal in life is to make as much as possible and screw everybody else on the way, then yes, I'm not shocked you're anti-union.
As far as convincing people, that's on the McKinsey grads and other neoliberals who stay awake at night in horror that a non-college educated auto worker, grocery cashier, etc. might be earning a little more than their supposed worth is, all as CEO's making combined, billions than their supposed worth to convince conservatives not to be ghouls.
"outside of obvious outliers"
I mean... literally every job I have worked at, there were a people that everyone could identify on both ends of the spectrum. If you are in a profession where this is hard, though, I can see how the union could benefit both employees and employers.
"If your only goal in life is to make as much as possible and screw everybody else on the way, then yes, I'm not shocked you're anti-union."
Ah, I see we've moved on the part where you've already made up your mind about everything, and I must be the bad guy because I didn't should "You go, Jesse!" Carry on, then.
I doubt you're convincing anyone with this ranting.
I think the wealth transfer is only one part of it, though—there are all sorts of employer abuses that wouldn't be corrected with just more money. But I'm a union member so I'm definitely biased here!
Yeah, I'm a lot more sympathetic to that perspective.
I think there's a range of anti-poverty programs that could prompt employees to say "take this job and shove it" to abusive workplaces, but that range probably overlaps considerably with one where labor force participation drops off a lot.
Well I'm a Southerner who is reflexively somewhat anti-union plus I had some really bad experiences with the UAW that cemented that view, so you are preaching to the choir.
But I don't want to come across as a crank on the topic, so limited my comment to a specific example where unions are very clearly harmful. =)
To be fair, pro-union people want all or most jobs to be unionized, or at least governed by a collecting bargaining agreement, as is the case in some European countries. So in theory everyone would benefit, not just a small number of unionized workers.
For sure I'm not sold on whether this would be a good thing, I wish Matt would do a piece on it actually.
That would address the issue of it being a small number getting the advantage of a union, but there's other issues to consider, too. For any given union, there are rents extracted to the benefit of the union members and the detriment of everyone else. If that were the only effect, then maybe economy-wide unionization would more or less even out. But those rents are also *inefficiently* extracted, making everyone poorer. Those countries in Europe with economy-wide unionization are much poorer than the United States.
I think it's clear that Matt is pro-union for political reasons, regardless of the economics.
When it comes to policy, Matt isn't a political hack. He holds positions because he believes the empirical evidence backs them up. He would never say "unions are good because lefties think so" - he certainly isn't afraid to critique lefties when he thinks they're wrong.
I'm skeptical of widespread unionization for the reasons you say. But yes, Matt does seem to be pro-union, so I wish he would do a piece making the case for it, beyond simply saying "the current US collective bargaining model is bad" - literally everyone already agrees on that, the only reason it persists is that the disagreements on an alternative model are too wide.
Matt has changed my mind on other issues, maybe he could do the same with unions. I would also love for someone to take a dive into why sectoral bargaining seems to work so well in Germany but so poorly for US automakers.
Good plug for a column!
I think the reason most people in the comments are not as pro-union as Democratic politicians is they're mostly college-educated well-off in demand knowledge/etc. workers who have never needed the specific backing of a union to help them or protect them.
I don't think Matt is a political hack, but I also think he is somewhat of a political realist. He recongizes that unions are a huge boost to the Democratic party, so he supports them for that reason, even if the economic case doesn't make much sense. Not because he's a party hack, but because the Democrats tend to advance causes he believes in.
American unions seem far more adversarial and far less interested in making sure their host companies are successful compared to German unions.
I think the fact that workers councils are illegal in the US is largely to blame for that.
Hasn't he already declared for strong unions with sectoral bargaining (Nordic Model)? Think it was a couple years ago so maybe worth a revisit in this era of higher inflation and interest rates.
Management and business owners are also trying to extract rents. It is neither unique to unions nor are they necessarily the most effective rent extractors at the expense of the larger population.
"Management and business owners are also trying to extract rents."
Yep, they sure are! Which is why we have anti-trust laws that explicitly make that rent-seeking behavior illegal. But we have laws that explicitly make union rent-seeking behavior protected. I think that's bad.
Look at WMATA. They have a fully automated system that they scrapped for one with drivers.
Because of a crash where automated driving wasn’t even implicated!
They're very slowly bringing it back! Probably by the end of this year!
What a farce
I don’t think we have the demonstrated technology for driverless light rail. We do have demonstrated technology for driverless fully grade separated rail, like the Vancouver skytrain and the new Honolulu system. But you have to design the system with full separation of the relevant sorts.
It would be interesting to know which, if any, cities in Japan or Korea or China or Taiwan have made any move to convert existing rail to driverless, to know how feasible this is. (I’m not aware of any such cities - I think Copenhagen and Paris have a few driverless lines, that were built driverless from the start.)
A couple of absolutely ancient Parisian lines (built before 1910!) have been automated, with no disruption in service to boot: https://www.urban-transport-magazine.com/en/conversion-without-service-interruption-paris-metro-line-4-now-fully-automated/
In the US, BART trains are fully automated for travel, with drivers only operating the doors at stations I believe that the main blocker to automating this (besides drivers' unions, ofc) is the cost of installing platform screen doors at all stations.
Paris has done a couple of conversions. But having worked on feasibility studies of similar projects in my city, the issue is quite simple - you'd need billions to retrofit (most) existing lines to make them suitable for automation, and if you had billions available for capital investment (you probably don't), the RoI would be *much* higher on other projects (including entirely new automated lines!), so no one wants to do it.
Given the ubiquity of car use in America, there will be a larger public call for increasing ease of use to car commuters with this technology. So I would be optimistic; the pressure for regulatory changes could very likely be faster and stronger. And then once people use AVs to get to work from wider sprawl (and thus more people), they'll want more reliable transit for the final commute in.
I think the idea of self-driving vehicles is very exciting. But I wonder about the use case being developed versus what I would find useful enough to pay for. Robotaxis don't seem to offer that much savings versus an Uber or Taxi itself -- the driver just isn't that much of the cost when compared to the capital costs of a Waymo.
On the other hand, it would be very, very helpful to be able to have a few drinks, hop in the car and let it drive me home rather than risk a DUI. Or plop the family in the car in DC at 11:00 PM and wake up at Grandma's house in Jacksonville at 7:00 AM, fully rested and ready to go.
It seems the technology and regulatory path to the use cases I describe above will be measured in decades rather than years, though.
Autonomously controlled vehicles will only ever get cheaper relative to human controlled ones. Thank Baumol!
Making DUI extinct and allowing red eye driving is one of the greatest potentials for automated vehicles.
I bet you scolds make it illegal to be intoxicated while occupying the “drivers seat” of an autonomous vehicle. This really is tragic, because reducing DUIs (and permitting freer and more joyous socializing in the suburbs) is a major part of the AV value proposition.
In fairness, this isn't just being a joyless scold; there is something to be said "no system is perfect, and if the car's control system goes on the fritz suddenly, a conscious human driver should be able to take over."
Of course, this also removes the possibility of sleeping in the car (red-eye driving), which would be another major use of autonomous vehicles, so... eh... I dunno? What is the cost-benefit ratio of "I want to get drunk and have my car drive me safely home" vs. "I may fall asleep in the car and never wake up if the driving system decides to break down at the wrong time"?
In safety engineering world, the man in the loop is often a key part of a redundant fault tolerant system.
Yet I wonder how alert we can really expect the 'driver' of an autonomous car to remain, such that they are able to quickly and correctly take over when the car goes on the fritz? Especially if it is presumably a very rare occurrence.
I would wager that even fairly diligent people will let their guard slip significantly.
I would never be interested in a self driving car that presumes I'm paying attention in it's safety protocols.
I _might_ if the car's procedure was to say "I can't handle this", pull over safely to a stop, and _then_ ask me to take over. But never if I have to do it spontaneously.
And even the situation I described also assumes I've been through Driver's Ed.
An intentionally designed self-driving car would have many layers of redundant software controls/mitigations...
...but to reduce the probability of a mishap into an acceptable range, you may also claim the driver/man-in-the-loop as another layer.
You don't have to do that, but if so you have to do a bunch of extra work on the software safety side to get the probabilities down.
"Yet I wonder how alert we can really expect the 'driver' of an autonomous car to remain, such that they are able to quickly and correctly take over when the car goes on the fritz? Especially if it is presumably a very rare occurrence."
Yes, this is a point that I've literally been bringing up for at least a decade in discussions about self-driving cars. Advocates for full self-drive frequently waive off questions about irreconcilable conflicts in safety commands and other issues by saying that the car will simply throw control back to the human driver, but, when I took driving classes many, many years ago, I remember being told that a driver has less than 2 seconds on average to react to an adverse event. Obviously, if you have a self-driving car that has to perform some sort of assessment of a situation and then return control to the driver, that's going to shorten the available time to react further (even if we're just talking milliseconds), so the system is (1) going to require a driver to be just as alert and aware as normal, while (2) simultaneously putting the driver into a state of tedium by relieving them of the hundreds or thousands of tiny adjustments a driver has to make under normal driving conditions.
Hell, I worry about delayed reactions even when taking my feet away from the pedal for a few seconds while using cruise control on the interstate.
And I at least still have my hands on the wheel.
But that's why I dont think we'll really get a slow incremental improvement path to these things.
It will either be truly excellent and safe 99.999% of the time, or it won't be on the market.
Eh... most truly fault tolerant systems are built around the idea that the human is *guaranteed* to make a mistake, so the machine should work regardless of whether the human screws up.
A country that thought adults should wear masks while socializing is unlikely to understand the benefits of communal drinking. Yet they are real.
Not entirely sure those are equivalent. I listened to Senator Chris Murphy speak the other week, and he talked a lot about the crisis of social isolation, etc. and he's a fairly normie Democrat.
Mask politics are tricky, and both sides got some stuff wrong in that debate. But I don't think that wholesale makes Democrats the ant-social policy. At least, I hope not...
I'm not denying communal drinking has benefits, I'm wondering where the tradeoff is between "I want to get drunk with my friends and then pass out in my car" vs. "I want to stay awake and aware so I could take over if Murphy's Law happens on my way home."
This can be answered almost a priori. Autonomous cars are safer than sober humans. Ergo, any policy regime should definitely try to substitute autonomous cars for drunk humans.
All of this discussion is moot if the cars need a backup human driver.
Sounds like something the "morally superior" Right is more apt to do rather then the "decadent and debauched" Left. \S
Why would you want to be in the drivers seat if you're drunk? Lie down in back. "Freer and more joyous" (by which I assume you mean "drunken") socializing is already a thing in the suburbs and can be accomplished in the front passenger or any of the rear seats.
As likely, expect many of these features to be gated by a subscription cost.
Why do you say the driver just isn’t that much cost compared to the capital? If your driver needs to make $15 an hour, and the vehicle goes 10 miles per hour, for 100,000 miles, then the cost of the driver is $150,000 over the life of the vehicle. If the vehicle can get a longer life than 100,000 miles, or the driver gets more expensive, then the budget available for the capital expense to become a savings gets larger.
The challenge here is that present dollars are more expensive than future dollars, so 150k up front is going to cost you more than 150k over the lifetime of the vehicle (because of interest or equivalent financing cost paid or forgone.)
That said, labor is expensive enough and capital is cheap enough that your autonomous vehicle doesn't need to be *that* cheap up front for things to pencil out. And because robotaxis don't require as much downtime as human drivers, you can at least theoretically get good revenue-miles in fairly quickly.
Plus, labor is variable while the upfront costs are fixed, so the payoff under uncertainty is heavily skewed toward variable costs.
For taxis / ubers / robotaxis you have to denominate and compare everything in costs per mile. Robotaxis can accumulate miles and recover capital costs much faster than either taxis or ubers, and it's not yet clear that their initial capital cost will be even 2x the capital cost of those human-driven vehicles. I don't expect the expensive, multi-sensor approach of Waymo to survive unless sensor costs drop dramatically.
Great point...also capital costs are much easier to innovate down and often have savings at scale versus personnel costs (which tend to go up).
AVs probably cost way more than $150,000 today. But in 10 years they'll probably be that cheap, and in 20 years the automation parts will probably be <$10,000
I wish a self-driving car was an option for my dad, who really shouldn't be driving anymore (at least, not without a backup that can automatically take over if he gets distracted or falls asleep) but doesn't want to give it up. I'm sure a lot of older folks are in the same boat, especially as more boomers reach their 70s and soon 80s.
But unfortunately it sounds like that type of usage is pretty far away as well.
My Dad is in his mid 70s and lives in the suburbs of orange county. He and my mom frequently drive to LA to visit my sister and her grandkids. He purchased a Tesla Model 3 a few years ago without the full self driving software. There was a recent update to the software that has been such a significant inlmprovement that they gave him a free 1 month trial of full self driving. He and my mom have been using it to drive door to door from OC to LA with no issues. The software still requires supervision but even at this point he trusts it more than my mom's driving and he chooses to use it because it puts him in the role of an observing passenger and reduces driver fatigue. He is now no longer worried about needing to move out when he is no longer able to drive reliably.
This is a great use case and good example of how the current, imperfect version of FSD can still provide benefits. I have been using the same trial for the last few weeks and while I agree that "supervising" is much less taxing than driving, I am still surprised by the number of mistakes the current (12.3) version makes. I get at least one significant mistake per trip (wrong turn despite correct mapping) and one scary error every 5 trips or so (veering off-map to follow train tracks intersecting a street at grade at a 45 degree angle). It's impossible for me to know how close this version is to "real" FSD that does not require supervision at the technical level, but it does not feel "decades" away by any means.
This is a real problem that forces many older people to put themselves and others at risk. It's shocking how difficult it can be for an elderly person to get by without a car, even in a relatively compact and service-rich independent living community. We are currently going through this with my wife's 92 year old mother in SoCal. We were able to convince her to give up driving (with some help from the DMV) and have had pretty good luck with a service called "gogograndparents", which offers uber and lyft services with drivers trained to expect and cater to elderly clients, as well as a phone number to book rides for non-app savvy seniors. Not a perfect solution by any means, but might be good enough. Good luck!
I think the capital-intensive Waymo model is going to lose. The cheap Tesla model will win. Then the capital cost is a drop in the bucket compared to paying the driver. And then the transition can happen very, very quickly.
it certainly seems that lots of people (even very smart folks in the space) have just completely written them off for quite arbitrary reasons without even considering the quite real advantages the business model has in terms of cost structure / scalability...
I definitely agree that if Tesla solves self-driving in a fully generalized way, it’ll be a huge win for the company and its business model.
But the bear case is worth discussing here. This is a problem with very high execution difficulty— significantly harder than working out a patchwork of effective local solutions. And in its own way, it’s actually also going to be very capital-intensive to solve— as the evolution of language models etc has shown, incremental improvements in neural network performance tend to require increasing compute and training data inputs by orders of magnitude (eg: GPT-4 has about ten times as many parameters as GPT-3), and it’s far from clear that Tesla will get to acceptable fully-generalized performance before these scaling costs get prohibitive. Management’s record of over-promising and under-delivering in this domain many, many times also makes it a lot harder for investors to trust that they actually have a good handle on the problem and a good path to solving it. I don’t think it’s impossible that they pull it off, but there actually are good reasons to be skeptical.
The advantage of the solve-more-limited-cases Waymo approach is that we know it actually works with feasible levels of compute and training data.
Tesla actually has an order of magnitude more data to work with than anyone else (over 1.3 billiion miles of driving data, versus Waymo's 40 million).
Additionally, from a capital stand point...Tesla has been lucky to (somehow?) subsidize this by getting people to pay thousands of dollars a head to beta test the software for them...rather than rely on low wage test drivers to drive in circles on a map.
Per Elon, they are "no longer compute limited" on the training side and the most recent cadence of FSD revisions has proven that (there is a new one going out every 2 weeks, it seems), so it seems like most of that capital has already been spent.
I'm guessing the take up will probably follow a similar path to cars themselves. I get the sense in popular imagination, cars as a mass-produced product widely available to the masses was a quicker process than it really was. Henry Ford and the Model T is held up historically for a variety of reasons, but I think one misconception that's been created about the Model T is that it allowed us to go from a world where everything was horse and carriage to suddenly everyone had a car. But by the end of the 20s, still only 25% of Americans families had a car. https://www.digitalhistory.uh.edu/disp_textbook.cfm?smtid=2&psid=3396#:~:text=In%201929%2C%20a%20quarter%20of,the%20American%20way%20of%20life.
I bring this all up because I think Matt makes a pretty good case that we're in the 1918-1920 stage with driverless cars. Once they allowed on highways, it really will be the case that there is a real use case to take a driverless car over an Uber. Even if we are faraway from everyone owning a driverless car outright, I think we are on the cusp sooner than you think of driverless cars fundamentally changing the nature of urban and suburban living just like Model-T cars did in the 20s.
forget what consumers find useful: trucks is where the true savings are
Until there's competition between multiple robotaxi companies in the same market, there's no real reason for a robotaxi company to charge much less than what a human-driven taxi would cost.
“On the other hand, it would be very, very helpful to be able to have a few drinks, hop in the car and…”
…have a few more.
I remember seeing a study a few years ago where some researchers hired chauffeured cars for a group of people to simulate how people would use privately owned AVs. The big takeaways were that total VMT increased ~50%, trip volume increased 25%, and 80% of the number of new trips were deadheading (driving the car with no passengers in it). Transit usage cratered among people with access to a private AV. I’d be nervous of broad based adoption of private AVs without a major congestion tax to manage demand on our roads - the increase in VMT and deadheading would cause huge snarls of congestion without a way to manage it.
Beat me to it: this is the article. [https://jalopnik.com/zombie-miles-and-napa-weekends-how-a-week-with-chauffe-1839648416]
I too was amazed that Matt didn't consider the zombie miles problem, and that, as Nathan just said, didn't consider that cars not in parking spaces still have to be physically present somewhere. And yes, I'm guessing that Matt would just counter with congestion pricing, and that's fine, but there's a limit to how effective cars, automated or not, can travel with all the merging and intersections that occur. At some point, there's no way getting around the need for mass transit.
It seems plausible AVs empower much more aggressive sprawl outside a city with commuting to a mass transit stop (bus or light rail) for the final 30 minutes. In fact, a few zombie miles to a charging location could help solve the parking problem that would otherwise hinder this mass adoption for dropping off at a bus or train station to work.
I don't see why that would be the case. When people originally got cars and spread out into the suburbs, that did not spur the development of train stations with big parking lots so they could commute into the city... it spurred the development of suburban office parks (where most American jobs now are), and the replacement of many buildings in America's downtown cores with parking, for the jobs that remained. The jobs sprawled too.
In the UK cities have been introducing low and ultra low emission zones. You're vehicle must be more efficient than some standard or you need to pay a (hefty) fee to drive into the zone. I think it was started in London and has since been rolled out to more cities so it can't be completely politically toxic.
I could see these type of schemes ratcheting up until driving to mass transit station on the outskirts will be cheaper and more convenient than driving all the way to the centre.
I think there is enough in the centre (as opposed to in the suburban office parks) to make that worthwhile
But when the suburban office park took off, there wasn't much remote office work possible to do and the car needed to be driven by the occupant. If both of those variables change for a fraction of workers, you end up with a different calculation.
If anything the move to remote work has intensified the desire for short, simple commutes when people do go in. The pattern we've seen since COVID is that office occupancy has been hit harder in city centers than suburbs. Few people are going to want to mess around with transferring to a train to a dense job center when they could just commute to an office in or near their suburb - whether the first part of the trip is in a self driving car or not.
It seems like the zombie miles problem is a problem of sole ownership vs highly available on demand per-trip renting?
If I own an AV, it's going to do zombie miles for me, and also sit around and do nothing a lot too.
I thought the dream is more like yellow cabs in Manhattan - they're just everywhere, they're hopefully mostly doing useful work, and they're so plentiful the service seems on demand.
Right, as soon as I read the words "you don't need a parking space" I knew there was trouble, because that means that the cars are taking up space on the roads instead. There aren't secret out-of-everybody's-way parking alcoves that AVs would disappear to during downtime; they'd just keep roving around, and we have solid research on TNCs (Uber and Lyft and so on) that that's exactly what happens.
They can of course double, triple, quadruple... park since they can just wake up and shuffle themselves around when one needs to enter service. So while "no parking space" may be an exaggeration, it could still be transformational to land use.
We would end up building those out of the way parking spots, but it would take time. Eventually having a big garage every few blocks would be Waymo efficient than the current system of street parking.
Wow
If these are zero emission vehicles (electric yes, but tires also produce emissions, that's still a problem that needs solving) and they're cruising slowly and safely, I'm not sure deadheading is a big problem.
Two objects cannot occupy the same space at the same time. Deadheading is going to make congestion worse, if nothing else.
For robo taxis, I wouldn’t think so. If demand is high, ie a time of high traffic, there won’t be much deadheading. Pickups and drop offs would be close. If demand is low there would be some deadheading, but then it wouldn’t matter as much. Deadheading is definitely an unintended consequence of this tech that needs to be monitored.
In the robotaxi vision, existing roads become the new parking lots. During peak, the robos are all driving on the road, during lull, those robos that are still working can easily and safely navigate around the "waiting" robos that are parked in the same roads. But this vision is not fully compatible with a "mixed" system of robos and human drivers, so you will need some version of private vehicle / human driver bans to make it work. This is hugely inefficient land use when compared to transit's footprint, but as a society we have consistently chosen the convenience (and personal efficiency) of door-to-door transport over land use efficiency, so I think that will probably continue.
"Liked" in part for use of "deadheading" -- a term I unselfconsciously use on occasion at work (a law firm) and then have to deal with blank looks from everyone else.
For anyone who does the New York Times spelling bee, it’s useful to know that DEADHEADED is always present whenever HEAD is.
That's one of the gimmes along with the set of "acai, acacia, tact, attic, tactic, tacit, cacti" whenever you have ACIT, which is weirdly often!
One of the biggest business model challenges for autonomous vehicle operators will probably be figuring out how to maximize utilization/avoid deadheading-type situations. I think that in practice, robotaxis will be most economical if they also serve as delivery or small-load short-range freight vehicles some of the time, and I also suspect that if they become common enough, some businesses will adopt models and locations to take advantage of the cheaper leg of the trip. (Analogous to people containerizing recycling and soybeans to ship from the US to China.)
I like AV grocery deliveries coupled with food waste / compost pickups. Laundry and dry-cleaning pickup / drop-off is obvious. Amazon should be picking up all its packing waste after each drop off (if it isn't already), offering some kind of discount/credit for packing returned in re-usable condition. Some percentage of robo-SUVs that take commuters into city centers could be redeployed as package delivery vehicles in the city during working hours, then re-assigned for the commute out, then used as suburban meal delivery vehicles and DUI-avoidance taxis in the evening.
I wonder if some kind of standard containerization system for small-scale freight would facilitate this.
It would - along with convergent (if not fully standard) vehicle dimensions. Time for "The Box v. 2.0"...
I have heard the argument that we should ban self-driving cars because people should be taking public transit and self driving cars will displace public transit. I suspect that public transit could hold on to some riders if it is safe, reliable, fast, clean and pleasant, but that means improving the current transit system.
Correct - this is the real issue and congestion management taxation / TOU charges are the only viable solution. Transportation has a huge peak simultaneous use problem that AVs do not make better and will likely make worse, at least in the near term. In the early days of this discussion (MY's piece today is basically Tony Seba from 2020) a lot of emphasis was put on increasing throughput and reducing congestion by using AVs capable of moving very fast and very safely (so safe that they would not have to stop at lights, for example). Another strain of the discussion was a combination of systems to reduce traffic and deadheading with multi-rider vehicles, including AV carpools for school pickup / dropoff (a huge peaking problem in many suburbs) and smallish AV buses / vans for routes underserved by existing transit. TOU charges and VMT taxes obviously reinforce the latter.
But really, nothing works politically until there is widespread acceptance that scarcity economics apply to road-space in many places at some times of the day.
Seems like converting public transit to smallish AV buses / vans for routes underserved by existing transit is a big opportunity to increase bus frequency on a given route, and to have more routes
An additional issue is parking.
I live in an area in Asia with a relatively high number of people with private drivers.
When you drive your own car, you generally park relatively close to your destination. This pressure allows parking garages to exist.
When you have a driver (or an autonomous vehicle, presumably) the driver avoids paid parking and simply drives to some neighbourhood where they can park for free.
If you're commuting to work you could probably tell your autonomous car to find a free parking spot within 10-15 miles. You could even have it move every 2 hours or whatever to avoid those kinds of rules.
Yay, that's just what we need: roads clogged with empty cars that move themselves every two hours to evade parking limits.
Yes... but what about the parking?
edit: I think the logic would be that if AVs become common, most people would not own one, because the economics would be insanity. It'd be like owning a plane. Why would an averagely wealthy person own a plane, with its enormous capital costs, instead of just buying plane trips when you need one?
And obviously the economics are different. If VMT and trip volume has a direct impact on marginal costs, then they probably don't go up (i.e. if you don't own the AV, there is no economy of scale).
You could use the money made from congestion pricing to underwrite autonomous buses, providing free transportation for those who choose it.
I dunno. Can't deal with snow and ice and requires absolutely massive capital investment to expand aren't really convincing me that the technical problems are going to be imminently solved and it is underhyped.
The transition will happen slowly and then all at once. As soon as one company (probably Tesla) has cheap robot taxis, everyone else will copy them to get cheap robot taxis. Just how no one could do what ChatGPT could, and after ChatGPT was released, there were half a dozen clones within a six months.
Humans aren’t very good at snow and ice, I’m sure the self driving cars can at least rival human drivers.
I wouldn't be sure of that at all. It wouldn't surprise me if the artificial neural networks used for self-driving cars were trained on conditions largely without snow and ice, and bear in mind that artificial neural networks have only passing similarities to the natural ones and don't necessarily have the same capacity for generalizations as natural ones. There's also the question of what sensors are being used on the cars. If they are using ones more easily confounded by snow and ice, that could be an issue, too.
Humans use sensors that are confounded by snow and ice, but most of us have enough common sense to pull over and scrape the ice off the windshield if it gets to that point.
I'm thinking more of situations where the electronic sensors would be confounded in places where humans' sensors would not.
Also, you bring up an interesting point. Humans often can tell when their sensors are producing input that gives them a very uncertain view of their surroundings, and they'll respond by taking that uncertainty into account, say, by (as in your example) pulling over. Dealing with uncertainty is a very open question in machine learning, and it’s far too easy for an AI to deal with uncertain inputs by being confidently wrong, even in conditions where a human would say "I don't know".
I don’t think solving the problem of, for example, being the first vehicle driving on a road with six inches of newly fallen snow, is only training. But that certainly strikes me as a place where non-human sensors could be engineered to have an advantage.
I think you misunderstood. "Training" is a term of art in machine learning, and if one is talking about supervised machine learning in particular, it's basically nonlinear regression.
Think of it as a kind of curve fitting, where you have a bunch of sample points-- pairs of inputs and outputs -- generated from some process, a function with lots and lots of parameters, and you try to adjust those parameters so that (1) the function gets as close to those points as possible, and (2) the function can approximate the original process, that is, given an input, generate close to the same output(s) that the original process would generate. Doing that second part *reliably* is hard.
I don't know, human overconfidence is a huge problem in driving and almost every other activity. At least AI can be "tuned" - I get a very different ride when I set Tesla FSD to "Assertive" mode than when I set it to "Chill"...
I'd say that humans and AIs tend to be overconfident in different ways. A human wouldn't mistake a staticky picture of a panda for a gibbon, but an AI might (and one already had). It can be difficult to predict how an AI will go wrong, as it can do so in ways that seem arbitrary and don't fit with human intuition.
I think the bigger issue is weird reflections off every surface around you, particularly if you are sensing distances by emitting laser or radar or something else humans don’t usually use.
If that’s the primary problem I don’t see any reason why engineers won’t find a fix.
Humans who learned to drive in areas with significant snow and ice are very good at driving in it. Automated vehicles that learned to drive in warm climes (Waymo in San Francisco, Ford in Miami) probably could learn if they were instead tested in wintry places (Minneapolis or Waterville, Maine, say).
A large portion of both the US and global population lives where snow and ice just aren’t a problem. But importantly if there is a good toehold of commercial viability the remaining technical problems will start to be addressed much faster.
At the likely pace of driverless car development and implementation, by the time we should worry about their ability to deal with snow and ice we won't have to worry about their ability to deal with snow and ice.
Driving through flooded streets (hi, Ken from MIA!) and on melting asphalt -- that may be another thing.
This is exactly right, and Matt just breezing by it was silly. You literally cannot use AVs in the much of the eastern half of the country for most of the year! My city has over 180 days with precipitation, what are those vehicles supposed to do?
This.
Matt, I generally like your articles, but sometimes you display an obliviousness to the obvious.
You talk about cab & truck drivers losing their jobs as "sad". It isn't "sad", it's a bloody crisis for the individuals involved, particularly if they didn't just lose their job, but that entire class of jobs has been automated away. I'm glad you'll benefit from saving a few cents on your Amazon deliveries (as will millions of others) but that doesn't obviate the harm to the people who's trade has vanished.
Perhaps you'll suggest they learn to code? Do I need to spell out the history here?
Much like free trade, the benefits are broadly distributed (and likely a net positive for the economies involved), but the pain is sharply concentrated. And one reason why we've become more reactionary is the complete ignorance of "knowledge workers" such as yourself have of the trades. If you don't have a realistic plan for how millions of suddenly-endangered taxi & truck drivers have to make a living, particularly older ones, don't be surprised by the viciousness of their reaction, particularly at the ballot box.
Just to quibble - being a truck driver is a career, and it's very tough to lose a career. Being an Uber driver is by design just a random job, and Uber drivers can pick up other random jobs. And unfortunately all the cab drivers already lost their careers to Uber drivers 10 years ago.
If you look at the mobile aloha project at Stanford, we have Rosie the Robot you can build today, for around 15K (project includes a BOM to build yourself).
It isn't just drivers. Hotel maids are now automatable. Robots are being trained right now that have massive increases in dexterity - partly from advances in reasoning capability of LLMs and especially as those get implemented in smaller, faster and less expensive to run models.
Of course, most email jobs that involve taking information from one thing and passing it back and forth are on the way out too. GPT4 class AI can do most of that, and do it just as poorly as a 40th percentile mediocre middle manager, which in many contexts, is probably good enough for government work. The automation will hit most levels.
I personally don't see this going down without a means to distribute the productivity gains in some manner from AI, but that is a political question, not an economic one.
Agreed. I think it's important to take a broader view of the poltical landscape and coalition politics.
America is currently a threat from a fascist movement. The conservative movement is inspired by Viktor Orban in Hungary, who managed to change election rules to lock himself into power and use the government to reward his allies an punish his opponents.
If the Dems are seen as being ok with throwing truck drivers, bus drivers, etc. to the wolves, that just reinforces the image as elitists who don't care about working people and will cause us them to lose more working class support. The loss to society from the end of democracy would greatly outweigh the efficiency gains from driverless vehicles, so we need to prioritize coalitional politics here.
I feel like the time horizon gap between economical autonomous trucking and "I don't need my car anymore because I can reliably schedule an autonomous taxi to and from work every day" is a really big one. Actually I'm kinda skeptical that second one ever pencils out.
The idea that there would be a massive fleet of cheap automated taxis available for rush hour seems silly.
Yes and no, right? Some places have enormous fleets of non-automated taxis available for rush hour, and it seems like only a matter of time before the self-driving versions are strictly superior to that.
“Some places have enormous fleets of non-automated taxis available for rush hour…”
And the residents in those places have already decided if the trade off of paying to commute by taxi is worth not having to own a car. I don’t see what changes when it comes to driverless taxis, unless those are dramatically cheaper to hire.
Yes, the expectation is that they will be substantially cheaper and safer (although that probably doesn't matter as much to the passenger).
"they will be substantially cheaper"
We will see.
Right, it seems like the "self driving" part is a relatively small hurdle to that sort of paradigm. Seems you'd have a lot more challenges redefining the "car" part to even begin to make that plausible.
Uber is already that. Replacing existing Uber vehicles with FSD-enabled Teslas feels like the easy part of the problem. "Cheap" is relative. Robotaxis will under-price (non-robo) Ubers enough to take market share, then raise prices to meet the market demand.
As we've seen recently, Uber was only 'cheap' because of massive VC subsidization.
Sure, but I think the fraction of workers who commute by Uber every day is exceedingly small.
The economically viable model for this does not involve these vehicles waiting around all day, which is what commuter vehicles do 90% of every day and what Ubers do most of the day. Robotaxis can be cheap when capital costs can be offset with additional economically productive activity. This would entail using them for many more jobs than just ferrying one human to work and sitting in a parking lot doing nothing for 8 hours, or worse, incurring costs for staying parked (typical non-AV commuter vehicle) or for just driving around (typical Uber). The additional jobs could be as simple as providing back-up electrical capacity while parked and plugged in (or off-taking mid-day solar curtailments in CA), or as complex as running around delivering packages. The efficiency comes from improved logistics and automated price signaling ($2.50 / hour for backing the grid right now v. $1.50 net for delivering that package v. $0.04 per kWh for offtaking 50kWhs over the next hour), which is software driven so ultimately cheap and scalable.
Maybe. But that’s ultimately only viable if the vehicles have a comparative advantage in those tasks. I don’t see why a machine capable of safely self-driving on crowded roads while keeping passengers comfortable and/or entertained would do those other tasks as cheaply as machines purpose-built as specialists. Sure, it’d be better to make some income from contributing in tiny ways to grid reliability and the like, but that doesn’t mean that income will offset the costs rathe than just make the losses smaller.
I'm not sure what paradigm you are using to compare the alternatives. If the question is "how much would I pay to commute in a robotaxi rather than buying a commuting car, paying for parking, etc.", then it's pretty easy to calculate the breakeven point. Net capital costs for purpose-built robotaxis are going to shake out in the 3-4 cents per mile range, "fuel" cost in the 5-7 cents per mile range and maintenance costs (almost entirely tire replacement) in the 4-5 cent/mile range, for a total of 15-16 cents per mile or $15 per 100 miles traveled. That is a very low cost base from which to deploy a wide range of services from 15 cent/mile delivery to 5 cent per kWh battery based services. These "cars" are gig economy robots entering an existing market for some of the services they can provide and an emerging market (especially on the energy side) for other services they can provide. Every element of their cost base (AI and logistics software, high capacity batteries, electricity) is currently on or will soon be on a declining cost curve (except -- again and weirdly, tires). There will be a market for these services at the margin and beyond.
How much will “rush hour” continue to be a thing as work from home continues to expand?
It’s contracting, not expanding. E.g.,
https://fortune.com/2024/04/11/remote-work-return-to-office-miami-new-york-wall-street-80-percent/
https://tech.co/news/top-10-cities-back-office
I don’t have the subscription for the first article. The second article doesn’t show anything about trends, just current occupancy rates. I suppose for peak congestion purposes, it doesn’t matter whether a higher fraction of people are doing office commutes, or if total white collar employment in these cities has gone up - what matters for number of cars on the road at peak is the number of people going to offices at peak.
But for the “rush hour” issue for self-driving cars, what really matters is the *ratio* of peak to off-peak travel, to figure out what fraction of vehicles sit around unused and depreciating the rest of the day.
I have a (pretty expensive) CoStar subscription - their news feed has a monthly story about office occupancy levels from data aggregated by Kastle, a company that sells building access control systems. Those reports have shown very definite trends of more people going to the office. The trends are not uniform: The 2021 increases were much larger than subsequent ones, big drop in December 2021 from the Delta wave, large dips around holidays, etc. But the general trend was large increases Q2 - Q4 2021 and then a leveling off with much smaller gains overall through today. I’ll see if I can find a story later and grab a fair-use screenshot.
None of this is to say remote work won’t increase over the long term, only that we haven’t seen that yet.
https://500px.com/photo/1091495870/screenshot-2024-04-25-at-6.07.31-pm-copy-by-ken-in-mia
I will note that the title of the chart is referring to this detail:
"Office attendance averaged 51.2% of pre-pandemic levels in 10 large cities tracked by Kastle Systems for the week ended March 27, down slightly from the prior week’s 51.3% and remaining below the peak 53% reached in the final week of January."
I don't read too much into that. As you can see, it's a pretty messy chart looking at it week to week, but the general trend appears to be a levelling off after climbing out of the Delta wave.
If you're looking for more data on the topic, CoStar, Kastle, and Placer.ai are the best sources I know of. None are cheap.
>I am still not a technical person
Oh, indeed.
This article is *aggressively* incorrect in being completely blind to the fact that the first 90% of the problem takes the first 90% of the development time, and the next 10% of the problem takes 90% of the development time, and the last 0.1% takes the most development time of all.
Catch y'all for the evening shift.
What is the technical issue that makes operating in Houston, Dallas, San Antonio, Orlando, Tucson, San Diego, Sacramento, Las Vegas so different from operating in Austin, San Francisco, and Los Angeles? Just expanding to the full city of all of these cities would be pretty transformative.
I will say that if you get 99.9% of the stuff solved, they should already be far superior to human drivers at that point.
The 2 most likely failure modes would be:
1) Sometimes you have to call in and have someone remote drive the car around the place where it just gave up
2) Some areas will not be suitable for these cars - but almost nobody needs to go those areas, and those that do will either have a car they drive themselves or rent one.
This is true, but already baked into the tech timelines. The 99.9% was solved years ago, all progress and projections on AV are about solving for the 0.1%.
What do you mean here? I know Sharty was being vague, but you seem to have something more specific in mind. What is the 99.9.% that's solved in your mind?
In my mind, the jump from L2 autonomy (driver assist functions) to L3 autonomy (driving is *mostly* automated, but requires full human attention still) was the 90% mark, but we're just *barely* there now. L4 is still years away.
The 90% etc. framework that Sharty used, and I continued is indeed vague (90% of trips? Miles? Driving hours? Regions?) We could spend a lot of time trying to define it better and then debating progress and projections. But Waymo is currently running L4 in areas of AZ and CA - the tech is here, at least in certain climates. I'd say they've solved for at least 99.9% of the technical needs in those regions, and the proof is they're doing it! (Cruise might've solved for 99.5%, and that wasn't enough...again this is really vague and debatable language.)
If Waymo was told they had an a) unlimited capital expenditure budget, and b) that regulators and public perception would only hold them accountable to being at par with human-level safety metrics, we'd be at L4 in much of the country right now. That's what Matt means by expansion being primarily blocked by economic and political hurdle right now.
So I disagree with Sharty's contention - that this article is "aggressively incorrect" in not treating the timelines right. It's correct that the last 0.1% is much harder than the first 90%, but incorrect in stating that this fact isn't already baked into the assumptions of this article and of every company and regulator that's grappling with the issues in it. Everyone involved, including Matt, knows this, and has incorporated into their assumptions. Maybe some are still too optimistic, but when those people say "SDCs are imminent", they're saying "We have already or will soon solve for the last 0.1% of conditions that make autonomous driving less safe, on average, than human driving."
(Btw, like all tech, there's a parallel race happening in China. I'd give it a 50%+ chance that society-changing adoption happens there first, for political reason.)
The traffic engineers with whom I work suggest that this is the correct take on AVs.
In 2008, most traffic engineers thought AVs were 50+ years off. In 2015 tech folks thought they'd be ubiquitous in 5 years. I think we're seeing they were both overweighting their view of the world.
I think those traffic engineers were way too optimistic.
I think Matt would've benefitted from talking to experts before posting this take
This is a project development timeline truism based on projects with ordinary-level human intelligence as the binding constraint. I don't think we're in Kansas anymore...
Being a long-haul truck driver is a great job for antisocial people. You only have to talk to others a couple times a day. You don’t need to be that fit either, at least compared to warehouse or construction workers. You can drive around the country being obese and socially stunted and still make $50k a year, more if you go hard. It’s hardly obvious that a full employment economy will offer the kinds of jobs that displaced truckers can fill or that whatever jobs they do find will pay as much. I’m not saying other working stiffs should be forced to pay more for shipping to subsidize the long-haul trucker lifestyle, but I do think many men will be broken when they lose their niche.
It seems super problematic to encourage and allow a workplace that so clearly encourages obesity. Do truckers actually want to be obese? It’s like saying coal mining is great because you get to be in the dark all day and get black lung.
I think this is a pretty fair response to the truck driver job loss debate. It’s an industry that breeds comorbidities like no other. I feel like romanticizing its existence doesn’t do anyone any favors.
But think of the effect on the country music industry if we end trucking as a romantic way of life... the dominos just keep falling...
Prison guards have it much worse. And they have shorter life expectancies than Pakistanis last time I checked.
JFC, that's bad. Why? Alcoholism? Getting injured in fights?
I remember well Durant’s line about Sparta “where nine-tenths of the men were slaves, and not even the masters were free.”. I suspect the act of keeping fellow human beings in cages is inherently degrading and stressful. However, I want that to be true, so don’t trust me on this point. Still, the morbidities are third world in scale.
More likely "being in close contact daily with the kind of human beings who must be kept caged for the sake of society at large" is the stressful part.
I don't know the answer to this question, but the book 'The Secret Life of Groceries' has some very disturbing insights into the stress associated with many truckers' jobs.
Hasn’t there been a shortage and aging of the truck driver population?
Huge shortage. AVs will be a blow to the Sikh community however.
Trucking doesn’t really encourage obesity— fit truckers get laid much more often than fat ones and are less likely to have to pay for it. Trucking is merely a workplace where one can be obese without creating major performances issues.
This is the ultimate sedentary occupation (including possibly the getting laid part depending on you know, venue) so obesity / CVD is not a huge surprise.
I was thinking that the truckers that I've seen didn't seem to be more obese than usual, but apparently the rates are really high:
https://www.cdc.gov/niosh/topics/truck/infotext.html
There could be other factors (age, sex, demographics, etc), but that is so great of an increase over baseline that the job probably is contributing at least somewhat.
It could also be a selection effect. If my BMI is 32, I probably can’t get a job as a carpenters apprentice because I couldn’t handle it physically (unless I’m a body builder or something). I can’t get a desk job unless I can play nice in an (often feminized) office environment. I can get a warehouse job, but can’t keep it unless I can stand on my feet and/or walk briskly all day. So trucking is a nice, medium skilled niche where I can make a solid blue collar wage, maybe even aspire to become an owner-operator, spend lots of time alone, etc.
Understood and agreed.
I don't think that you'll get much sympathy among left-leaning audiences when the 'victims' are primarily men, blue collar, and relative asocial or antisocial. And also primarily overweight/obese.
It's one of those weird niche areas where the current right and left is inverted w.r.t. concern for workers.
That's a huge blind spot that left-leaning audiences have.
I consider myself left-leaning/liberal, but my parents and grandma taught me to respect anyone who does an honest day's work.
Depends how far left you lean. Biden is about as pro-worker and has as much or more sympathy for blue-collar work as any President since... Grant? And at the extreme end of the re-distributionist left you will get UBI proposals that could facilitate an acceptable glide-path from trucking to retirement without material deprivation. The uncomfortable place is the mid-left, which knows that we ultimately need and want the economic efficiencies of automation, but also knows that it cannot yet forge a re-distribution scheme vast enough to address the emerging problem (see, e.g., Andrew Yang 2020). I agree that trying to fill that yawning gap between the reality and the politically achievable with fuzzy faith that market magic and "human invention" will solve the problem (eventually) feels and sounds lame. But if the center-right starts (continues?) to tilt re-distributionist, everything to its left will fall in line with it.
I mean, you're not wrong, but that's the way the world works. A lot of people liked being blacksmiths, too.
I basically agree, but we shouldn’t be naive about the benefits and burdens of productivity increasing technologies. Working class wages did not increase during the early industrial revolution. Hand loom weavers were reduced to destitution and sometimes literally starved. If Amazon and Walmart have enough market power, it’s perfectly possible that the surplus created by autonomous trucks will merely increase their profits and pad the accounts of the rentiers who own them. In the long run, the invisible hand of the market will find an efficient equilibrium, but possibly after millions of lives are immiserated.
And let's not forget that industrialization led to lots of unrest as old livelihoods were rendered obsolete...and arguably AI could do the same thing except on a much greater scale, and a lot more rapidly.
Technical advancements do not automatically turn into corporate profits. If capitalism is working properly (because the government is doing its job to prevent monopolies) then prices fall (because that is how competition works).
That’s a rather heroic assumption. There’s a lot of market power out there and antitrust enforcement has been spotty. I do think most retail is diffuse enough that most productivity gains in shipping will be passed through to consumers as lower prices. However, this completely depends upon the structure of the market snd things could change if, eg, Amazon or Walmart buys Kroger.
If you want to take your case of Amazon, however, they are popular because they are cheap and convenient. Obviously if shipping costs fall to zero this helps them but it also helps all the brick-and-mortars and the other large by mail companies (i.e. Walmart) Clearly they would lose some market share if they didn't bother to compete. So perhaps some of it is absorbed by profits but the more they absorb in profit, the more alternatives exist which are cheaper.
So really, then, if you are buying from Amazon, you are doing it for convenience since there would be cheaper alternatives elsewhere. And, if people want to pay more for convenience, isn't that their prerogative?
My local Walmart is 5 miles away, so every time I go there, it costs about $5.60 in mileage, gas and depreciation.
Once you build in the transportation cost of driving to a physical store. Amazon is often cheaper and that’s assigning zero value to your time.
You actually don't have to structure your society to follow the logic of the market. The success and persistence of NIMBYism demonstrates that even enormous economic costs can be borne, as a matter of practical politics, as long as the costs are sufficiently opaque.
Where economic growth and human flourishing are in opposition, we can choose to prioritize human flourishing.
Now, being a truck driver apparently sucks, per surveys, so I would support gradually reducing the number of truck drivers in a way that minimizes disruptions to the lives of existing truck drivers.
I do think we should in general try restructure the economy and society such that people who are not great at school and not great at customer service can have an easier time living a decent life.
Yes, but most of them existed in pre-democratic societies where their voting power and preferences did not matter. Truckers are different that way...
But it also seems really difficult to fill those jobs. Unfortunately for me, a person who hates driving, I spend a lot of time driving between the East Coast and Michigan for family reasons. Virtually every truck on the road is advertising jobs with the company. I also met a guy during the pandemic who owned a small regional trucking company (he had previously run logistics for a large furniture manufacturer). He had so much difficulty hiring people who didn't do things like rarely showing up, or taking the hiring bonus and quit at the first opportunity, that he was considering retirement.
I'm not disputing your point, it's about the worst job I could imagine for myself but some people are able to make a good living (though others struggle with the current industry structure). And I don't agree with Matt's take, especially for this context. But it does seem as if the industry as a whole is short-staffed.
There’s a structural reason for that. Trucking is very competitive. Few barriers to entry (you can buy a working rig for under $200k), lots of competition, no economic profits. Jobs with higher barriers to entry pay more. This is why trucking pays only a bit more than warehouse work.
This has to be a "no one goes there any more, it's too crowded" thing. Trucking is very competitive and that's why trucking companies can't find enough drivers?
The market clearing wage still isn't being set, probably because, and this isn't an attack, many of the owners of said businesses either can't afford the market clearing wage _or_ they are still in a 2010's views on wages.
“can’t afford” can be very a squishy term
Karen Levy had a really good book on the trucking industry intersection with both working standards, unions, and Automation/Technology. If you find this topic interesting, I recommend Data Driven: Truckers, Technology, and the New Workplace Surveillance.
I thought one of the things we learned from the supply chain issues during the pandemic is that we are short of truck drivers as a country and that our current truck driver population is older than is ideal. It seems like some of that is the result of it being a fairly shitty job that makes it hard to have a stable family life, is extremely hard on the body, and requires works to often buy their own rig so that they are taking on huge debts that make it harder to set reasonable limits on their work hours and conditions. It felt like while the whole "nobody wants to work anymore" things was not very real "nobody under 50 wants to be a truck driver" thing was pretty real and a lot of the proposed solutions seemed to be about how to trick younger people into this industry without really changing it because that would make everything too expensive.
I am much more sympathetic to the human toll of people losing a job and especially a career that they have invested a lot of their lives to. But it also feels like if were making a list of careers that people are at risk of losing over the next 30 years due to climate change concerns an automation: coal miner, truck driver, and hotel maid are all ones that were objectively not great jobs. I think people in those displaced industries need a lot more support to change career paths and that we should assume that asking them to move out of their communities is fair or reasonable even if that seems to maximize economic efficiency. (I'd love to see us focusing bringing other manufacturing jobs and real employment opportunities to coal country as well as the rust belt.) I also suspect that given the limitations on scale and location, even if AV trucking started for short haul in warm urban areas we might run out of trucks around the time we got to the point where there was a real risk of us running out of trucking jobs.
Yeah I think this is a rare bad (as opposed to one I merely disagree) take from Matt. I think the experience of the Amazon checkout-less shops should make us far more sceptical about how personless these tech gimmicks are, and how easily they can scale. I think the reality is that to be anywhere near an acceptable level of safe, driverless cars just need far too much backroom support that it'll ever make financial sense to use them instead of just paying someone to drive
I suspect driverless cars are, right now, safer than human drivers.
the fact that we aren't even having a conversation about how much arbitrarily safer we (royal) expect them to be in order to greenlight widespread adoption is disconcerting... as actual lives will be lost in the interim based on where we set that figure, as it will take more and more time to iteratively hone those driving algorithms with data.
I think they're going to have to be ridiculously safer than human drivers for people to accept them. People's fear of harm intensifies when they give up control, even when their actual risk declines. Think about how many people are afraid of elevators and airplanes, even though these are the safest forms of travel ever invented.
The other challenge for self driving cars is that the very things that make them safer will make a lot of people absolutely hate riding in them. I think people will find them too cautious and too slow. Think about how many road rules the average driver breaks on a typical day.
I think a contributing factor is that the failure modes of self-driving cars can often be very unusual compared to other car accidents. Like who wants to die because their car mistook a white truck for a cloud? (Yes, I'm sure this one has been fixed, but there will continue to be other strange accidents that will get overweighted in people's minds.)
https://www.theguardian.com/technology/2016/jun/30/tesla-autopilot-death-self-driving-car-elon-musk
Yeah I think a lot of people here just say, well, if it's safer on average than human drivers, that's OK.
But human drivers are the aggregate of all drivers, including some drunk, asleep, angry, on medication, looking at their phones etc.
Obviously, the AI could be safer on average just by being an otherwise entirely mediocre driver but none of the above.
That doesn't mean that it's safer than *me*, assuming I'm a healthy alert driver who isn't do anything wrong.
I think people will not really want to accept this technology until it's not just safer on average by not making the mistakes humans make, but until it's much safer on average and won't make the mistakes humans *would never* make.
On that last point, it's true in the short term but in the longer term, there are ways to allow self driving cars to go much faster safely while following closely in convoys. Drivers can't do that.
And massively safer than teenage drivers.
The Amazon example was a leadership failure, not a technological one. It's totally possible to do what they were trying to do. They spent too much time and effort believing the data labeling approach they started with would scale, when it turns out they needed entirely different approaches to the problem of detecting what goods are in the basket.
Even big well funded companies will have failed AI initiatives that have more to do with organizational dynamics than the tech itself.
Are you suggesting that a bunch of workers in India are driving these cars from some cubical farm?
Leaked picture secretly taken in an unmarked Waymo-owned building in Delhi: https://www.fxguide.com/wp-content/uploads/2018/03/RPO-TRLR-0163.jpg
"the experience of the Amazon checkout-less shops should make us far more sceptical about how personless these tech gimmicks are"
Oh so you misread that story
I read that story quite carefully and I don’t see anything wrong with that sentence
the Amazon story is another example of Liberal Media's own "misinformation" problem, as the real issue with that story was Elite Educated reporters unfortunately have zero knowledge (and unfortunately zero interest) in understanding how AI training works end to end and there was a lot of naive (or worse, bad faith) pushing of an incorrect narrative that it was Indians doing all the actual work. Ultimately the thing that really killed wider expansion there is that consumers didn't really care for it in *most* cases, probably before you had to scan your credit card BEFORE you bought anything (and perhaps before buying nothing and walking out). It's fine for Amazon to scale back after leaning that resistance was there.
What does this have to do with liberals, media and elite or otherwise? Sheesh.
well, it was largely liberal / progressive aligned outlets (with their negative bias towards tech in general) and media influencers who ran with an technically incorrect narrative about the story (probably based on headlines alone in many cases) such that many of us talking about it here have the completely incorrect idea of what happened. Doesn't seem like conservatives cared all that much about the story at all because it had nothing to do with "wokeness"....so their outlets and media influence folks just ignored it.
Being "anti-tech" and being "anti tech monopoly power" are two completely different things.
well, the NYT is already a billion dollar tech company that makes much of it's money from getting people addicted to word games (many people who don't care about news they also serve up) that is well on it's way to having a monopoly over national / international media coverage (sooner or later...) so that "negative lens only" focus on "Big Tech" is only going to get more problematic for them...but it's quite hard to ignore that tech companies being demonstrably better at every form of advertising than regional newspapers and fancy magazines once hoarded has basically made all these reporters wear that anti-Tech bias right on their sleeves...
But the Times is a great example (along with Substack) of the trend away from advertising as the mainstay of online revenues. I think their reporters are fine with Wordle subscribers reducing the need to run MyPillow ads. As for monopoly, I am not too worried about the Times' paltry 10M subscribers taking over either the gaming world or the actual world. The console and PC gaming platforms have hundreds of millions to billions of subscribers and users worldwide, as do Facebook and Twitter as disseminators of "news content". I think any negative bias against "big tech" is a reaction to companies that have grown incredibly powerful very quickly with near-zero public scrutiny over or accountability for business practices that are wildly anti-competitive.
Well, at my local supermarket, something like half of customers use automated checkouts. In the last 10 years, this has become pretty standard where I live (Boston area). So an alternative lesson from supermarket checkouts is: full automation might be very hard, but replacing 50% of workers can happen really fast!
I use self checkout counters for most my shopping.
I think he's talking about Amazon's stores that supposedly would ring you up as you walked out the door, not something where goods had to be manually scanned at all.
Ugh I kinda hate this on multiple fronts:
(1) Environmental: doubling down on car mobility, even if electric, still puts a high floor on transportation emissions because a 5000 lb vehicle will always be a very inefficient way to move a 150 lb person.
(2) Economic: Matt is dreaming of the wonderful cheap sprawl these things will enable. But sprawl is quietly expensive, and is actually already bankrupting us. This is because it has huge infrastructure costs that are paid for by the federal government that nobody thinks very hard about until it comes time to maintain it. Doubling down on sprawling development patterns connected only by cars has, of course, a dozen other issues, from the merely dystopian aesthetics to the lack of walkability that is a major public health issue.
(3) Technological: Matt correctly points out that the urban problem is harder, but like everyone, he's still underestimating how *much* harder. It's not 10x harder, it's 1000x harder. These existing cabs have millions of dollars in route-specific training, and exist only in good weather places with fairly freeway-like infrastructure. This is very difficult to scale. And now you'll get pressure to make the urban landscape less chaotic, which is a fundamental feature of urban places. We'll probably have automated gates for crossing the streets, or maybe we'll make bicycles illegal, just to make self driving cars work.
Self driving cars are still overrated.
I try to avoid hot takes, but I'm going to get this out there because otherwise it won't leave me alone...
Self-driving cars are just another manifestation of the myriad costs of sprawl. The massive R&D effort going making into them work is expensive on its face, the consequences of that effort not being spent elsewhere are expensive, the fares will be expensive relative to transit, the impact on the built environment will be expensive. American society is so desperate to cling to suburban lifestyles that it struggles mightily to engineer its way out of the downsides, costs be damned. Even if it succeeds, this doesn't feel like the right direction.
Pressure to make the urban environment friendlier to autonomous cars is an angle I hadn't thought about, and I don't think it's just scaremongering because that's exactly what has happened with human-operated cars already.
+100000 to all this.
Counterpoint: now that I have kids I see that the robo taxi is unlikely to replace cars for kid transport: too much variability in car seats, lots of other specific stuff parents have in cars, the huge mess kids create.
Yeah,
Trying to get an Uber with a car seat is “grinds my gears” level of annoying.
And it becomes exponentially harder the more children you have. Have fun finding an Uber when traveling with a 3-year-old and a 5-year-old.
Not great for America's below-replacement fertility!
What do you do with the car seats when you reach your destination at the supermarket or mall or movie theatre?
Normally people can just leave them in the car.
That’s a good point that I hadn’t thought of.
When robo taxis are safe enough, you won't worry about putting your kids in car seats. (Admittedly, that probably won't happen in the timeline of anyone who currently has kids.)
I just had a baby as well and dealing with the bulk of stuff I just need to have in my car now has been eye opening. My car has become another storage place for things too bulky to keep in my apartment like the car seat, stroller, diaper bag (which are always huge for some reason). Not having my own car means I need to magically create storage space inside. One thing to note against your comment specifically though, is that for a good chunk of babyhood the kid can’t lift their head up/ control their body enough to sit in a regular seat no matter how safe the actual ride is. And in the unlikely case that an advanced self driving car is in a serious accident it will almost definitely be fatal to the child passenger not in a car seat compared to adult riders because of the placement of airbags.
I live in a country where car seats aren't required.
For kids under 4 or 5 years old, not having a car seat is extremely unpleasant.
They stand up, sit down, walk across the back seat, kick the back of the seat in front of them, pull on door handles, roll down windows, push and pull every single button available.
Leaving them in the back seat alone isn't tenable at all. So you have to have them sit on your lap while you try to contain all that chaos.
Even if robo taxis were very safe, they are still going to accelerate and decelerate in ways that a toddler doesn't expect and will result in them falling since they will be standing up and walking around the car. Yes, they won't actually be hurt in any meaningful way but you still have a hysterically crying toddler.
Car seats offer a massive benefit beyond just safety. They keep a toddler in one place. Once a kid gets a little older they can finally sit in one place and learn not to roll down windows, open doors, etc.
I just don't see it as likely that cars will actually become toddler proof even when autonomous when you can instead just say "use a car seat".
I’m worried that this will increase sprawl as people sleep through a 2 hour commute.
If commutes becomes pleasant and cars use clean energy, then what exactly is the problem with sprawl?
Furthers car dependence for non-commute trips, makes communities less walkable, pollutes the air and water with brake and tire pollution, takes energy away from other uses (thus requiring more capacity from a system that is already struggling to keep up with demand), takes away land from local agriculture and keeps people farther away from natural amenities. Gasoline is not the only problematic aspect of car dependence.
Those are asthetic preferences.
Not wanting to breathe in particulate pollution from car tires is an aesthetic preference?
Road pollution is more of a consequence of the heavy weights of EVs than self-driving technology.
And with autonomous vehicles, you get more miles driven (due to deadheading), hence more pollution.
Clean air, clean water, and safe access to healthy movement aren’t aesthetic preferences, they are critical public policy concerns. It is much harder to stay fit and healthy when you can’t integrate your exercise into your every day life; designing a transportation system around sitting for hours a day is a costly mistake. But I guess aversion to early death is also just an aesthetic preference.
No it isn't but... is this a standard you apply to all policy arguments? "We should try to locate trash dumps away from cities" is per-se illegitimate?
Walkable cities, local agriculture, and distance to natural amenities are asthetic preferences. Road pollution is an actual issue, but is more of a consequence of the weight of EVs rather than self-driving technology.
Congestion is an externality, this has exciting potential to make it massively worse absent (politically toxic) road pricing
I'm curious if the political toxicity on congestion pricing won't apply to AVs as people want to pick on them in the early days of development.
Yeah, the political economy of congestion pricing should be easier for EVs. To an extent, people will feel that the big corporation that owns the robotaxis is paying, not the poor working class driver. This may be the one weird trick we need to get congestion pricing (for AVs only, which will be fine if the unit economics still appeal to riders vs. owning and driving a car)
As ZFC says roads and congestion are underpriced. Also 100% clean energy is a while away. Agree sprawl would be better than unaffordable homes.
Individual decisions have to be eliminated to achieve the utopian vision.
I think massive environmental destruction for more needless single family homes is bad, actually.
There are more trees on earth than stars in the galaxy.
The environmental problem with cars is carbon emissions, not deforestation.
I just read Hannah Ritchie's excellent book on how to make environmental progress, and it enforced this. So much of deforestation is due to eating beef and dairy that there's not much room left for harm due to wood from farmed trees. She also emphasizes that cement and steel are the large carbon cost in building homes.
Exactly...this would be one way we could solve the affordable housing problem if we could build on nice land currently too far away from job centers.
I don’t see any way that any kind of taxi service works for long-haul commutes—the whole nature of taxi service is that it operates in a limited area, so it can drop someone off and pick up the next fare nearby. It can’t be economical if it’s got to go places where there’s only a fare one way (I know there are crazy stories about taxis/Ubers going ridiculous distances, but we all know it’s not normal).
The reason that’s a problem with taxis and Ubers is because the driver lives in the original city, so they have to go back there.
It’s not a huge problem for rental cars, where you can do one way trips and you often see rental cars with license plates from different states.
As long as there isn’t a net flow out of a certain location, it would be fine for autonomous taxis to just remain in their new destination.
I’d expect they might just charge an extra fee if you’re going really far away.
Just buy your own self-drive car for that. And if your place has no parking, it can self-drive to the nearest garage for storage and self-drive to pick you up in the morning.
In my perfect world, suburban commutes would be taxi/bike + commuter rail + taxi/bike
It could be economical. Especially if those long-haul rides typically take place overnight or during some other slower time.
You aren't paying the driver, so it's really just a matter of keeping the asset being productive as much as possible.
Imagine it is driving from City A to a suburb of City B. Assuming the infrastructure to refuel and/or maintain the robotaxi is in City B, it can just operate in that city afterwards for a while, until it can get a fare to go back to City A.
Yeah, we’re still at the honeymoon stage where everyone focuses on the aspects of self driving that they like. We will get more miles traveled in cars and more outlying development, although personally I think that’s a reasonable trade off.
I think there's a certain level of density in between leafy suburb and dense urban core that people hate living in because it's full of congestion with none of the benefits of transit/walkability (e.g. Cicero, IL). And I think this is what NIMBYs are fighting to prevent, and why people talk about missing middle housing.
My theory is that robo taxis will have an outsized benefit for medium-low density places (too expensive for the suburbs and too congested in downtown). That could help bridge the gap between low and high density. If every level of density is nice to live in, then there's no reason to fight densification, and I can at last have my urbanist utopia.
Interesting take, food for thought
Work from home has already been doing that too.
I'm not sure about self-driving buses saving money. I take the bus a lot, and a large part of what the driver does is kick people off who are throwing things out the window or fighting, giving people who are lost transit advice, securing people in wheelchairs and making sure they get on and off the bus safely, etc. A larger number of bus passengers are disabled or elderly or don't have technology than in the general population. We may not need drivers, but they will probably still need staff to make it a pleasant and efficient experience for passengers.
There could still be a benefit in hiring someone that doesn't need to be trained in driving a bus, though.
Also, a benefit in being able to have a person who is spending their full time and attention making sure conditions are ok for passengers, rather than only intervening when things get out of hand.
I agree; I am just not sure that it would save money like Matt states.
I think it would save money in situations were transit needs both drivers and security but only then.
Yes, I thought of this issue too -- I suspect that (at least absent unionization) "conductors" or "attendants" could be hired more cheaply than drivers.
The technology may actually come later to city busses than other forms of transport.
Busses have passengers that are not wearing seatbelts, or are even standing.
As far as I'm aware, Autonomous Emergency Braking technology has been slower to be adopted for this use for that reason too. It's one thing in a car, or even an intercity coach, but bus drivers have to be be taught to plan ahead to avoid heavy braking, or they'll send people flying.
It may well be we get to the point of having reliable autonomous cars, but they're a bit "reactive", and sometimes don't plan / predict well enough ahead, but are able to avoid an accident with some last minute braking, but this tech won't be good enough for city busses with standing passengers.
Yes but then we could hire someone who could do security full time!
Yes, that would be great! But Matt was selling it as a money saving measure, which I don't think it would be.
places with nonzero parking costs need to be very careful. Right now it’s not possible for my car to loop the block for an hour while I shop, dropping the cost of that antisocial behavior to ~zero would totally destroy the road network even in midsize cities
The key would be to build centralized garage/lots for the car to go to and wait before coming back to pick you up. I see eliminating the need to get a close parking spot in urban areas as one of the major benefits of the technology.
I think the idea is that the car would not be personally owned in most cases - it would be part of a fleet and would go pick up someone else while you’re doing your chore.
Makes sense but I don't really buy it. Owning a self-driving car would be awesome (Siri drive my van packed with all my stuff to Utah, I'll take a plane. At 30c/mi that's competitive with rental rates for non-self-driving cars today!), and once you have one surely it would be cheaper to use it than rent?
Yeah, congestion pricing for AVs only would be a real political possibility (paid for by the big companies that own the taxis)
It would be pretty easy to have automated police vehicles apply fines for excessive non-point-to-point driving. Or since parking would be in abundance, just drop parking fees to be comparable to per minute fuel costs of circling the block.
Why would parking be in abundance? You're making it easier to visit dense areas (no driving in traffic stress) without changing the built environment.
Fewer cars parked because taxis prefer to pick up riders rather than park
Once you make a significant percentage of cars self driving then parking lots for those cars can store 3x as many cars due to fewer lanes and tight spacing. When a car needs to leave, it advises adjacent cars and they shuffle around to "free" it. This improvement in parking density is one of the "holy grails" of autonomous vehicles.
Maybe I'm just one of those jaded folks that Matt describes in the opening of his piece, but he's written is basically the same stuff that I heard 10+ years ago when I worked in the auto insurance industry. Yes I know some self-driving cars operate, but not a material or meaningful percentage.
There are still a lot of known problems (e.g. bad weather), but I'd imagine that there are also a lot of unknown unknowns that will emerge as future problems as this stuff continues to be tested in more situations and more scenarios.
And, while I hate to sound like a boomer or a tech doomer, I would worry greatly about the scenario that my self-driving car gets hacked while driving down the freeway at 70 mph and runs itself into a guardrail or oncoming traffic or something like that.
That said, I hope I'm wrong because I have no particular emotional attachment to driving and would prefer to not have to do it.
What makes you think it is impossible to hack a modern computerized car with a driver in it?
I'm slightly leery of of drive-by-wire cars in general, I'd prefer mostly mechanical control systems. I'm really aggressively repulsed by the idea of a car with DBW systems that are accessible to over the air manipulation. I guess I assume Tesla's are probably susceptible to this currently... I suppose I need to investigate what manufacturers might be installing DBW systems that might be accessed without a hardwire. I'm already on the lookout to avoid vehicles with remote stop systems.
Throttles may be drive by wire but steering and brakes are fully mechanical. Steering is frequently power assist(ed). Otherwise, it would not require hacking but merely a blown fuse or a stalled engine to send you careening off the road.
Right, but once you're getting into things like electric wheel motors you start dealing with differential acceleration controlled by the DBW system. Things like that. Maybe some all wheel drive systems use computer control to determine which wheels get power? Anything like that should absolutely be isolated from wireless input.
Sure but there is nothing magical about electric motors. They don't need a network connection to the outside world. The Tesla has battery charge-and-protect (avoid explosions) circuitry but they don't allow that to be accessed by the outside world. It is true that there are more opportunities for problematic design choices as the percent of vehicle functions under digital control rises. However, there is no need for these to be accessible to the outside world.
I have led design teams working on ICs that control lasers in consumer environments that were 100x the power needed to char eyeballs and start fires. In my experience, it is fairly standard to have a type of redundancy where a complete set of safety systems is independent of any system which can be modified by a user or hostile third party. In our case, even opening the sealed unit would disable functionality. The only risk we felt unable to mitigate was one where a user unsoldered the laser and rewired it directly to a power supply.
All that compute power does in fact need a network connection for updates and telemetry, as evidenced by the fact that Teslas do have that capability built in for that reason.
Right, I am mostly ok with electronically controlled vehicle functions... as long as the are unnetworked. I don't trust software/encryption to prevent those being compromised. If someone at least has to physically plugin in a laptop to my car to access those systems I'm happy.
I can't wait until the Feds lock me out of my car and all public transportation because I have been unpersoned.
It's a danger long term. I didn't think highly of the Canadian trucker protest, but the extent to which Canada went after bank accounts and credit cards raises serious slippery slope concerns. There will always be some legitimate justification for doing stuff, but the story of government power is it never stops at the legitimate justification.
Not impossible at all, but I'd imagine it would be easier to take a control of a situation where you're actually driving and aware of your surroundings vs a self-driving car where you're sitting in the back or in the front passenger seat, not paying attention, etc.
Admittedly, I drive a 2009 model year car and I've been out of auto insurance for a while, so possible that I'm underestimating the semi-autonomous features that are in 2024 model year vehicles (though I don't think I am)
Well just to take one example, a lot of modern throttles are fly-by-wire, meaning that the computer, and not you, controls the mechanical action of putting fuel into the car. Now if someone hacks that, how are you going to control your car?
Fair enough, but I'm still not convinced that problems like this wouldn't be either more frequent and/or more severe in self-driving cars.
Unless your argument is something to the effect of: cars today could be hacked today but aren't, so it's probably not an issue for self-driving cars in the future. But I still see the risk as potentially more severe for self-driving cars
I don't see why. If someone wants to hack, they can hack. The big obstacle is that there's a lot of encryption in these systems. But an insider at GM or Tesla (or Waymo) can probably do it.
When you're the driver and (hopefully) paying attention to your surroundings, more likely you can steer/brake your way to safety vs in a self-driving car
Well in most cars you could put the car into neutral and then brake.
Not in an electric car though - the motor is always connected to the drivetrain.
Hackers!
https://m.youtube.com/watch?v=piI9vJ9-UZ0&t=2s&pp=ygUPSGFja2VycyB0cmFpbGVy
What we discovered with automobiles was a lot of unknown unknowns, some good and some bad. I don’t think anyone realized people would start living 10-15 miles from their workplace and commuting, or that traffic jams would become a thing, or that parking lots would reshape urban development.
Why aren't you worried about your non-self driving car getting hacked while you are driving down the freeway. Are you aware of a specific vulnerability difference in the two cases? It would obviously be easy to create the same level of immunity in the two cases if that was your goal. (edit: I see you have already engaged in this discussion....)
Yeah, admittedly, Dilan has convinced me that I've been underrating the possibility of present-day vehicles being hacked and how severely they can be hacked. I can still imagine more severe scenarios in a world of self-driving cars, but really more imagination than reality at this point
Counterpoint: it’s already possible to get much safer roads, judging by international example. Will robotaxis be an excuse to avoid solving problems we already know the answer to?
What actually is the answer that we know? It would be cost-prohibitive (especially including the psychological costs of living in a place under construction for several years) to completely transform the streetscape to match the streetscape of other countries. But we don’t know which fractional changes might get us most of the safety benefits. (We do know a lot of things that will give us small bits of those benefits, but those should already be an excuse not to do the other things.)
How is the streetscape different between Australia or Canada and the US? They are essentially identical. Yet America is over 2x the road deaths.
New York, Rhode Island, and Hawaii all have less than half the road deaths of the United States, while Mississippi, Arkansas, South Carolina, and New Mexico all have nearly double the road deaths of the United States. The streetscape in Canadian and Australian cities is much more like the former states than the latter states.
Self driving robotaxis owned and operated by big companies will probably be pushed to be safer and safer the same way airline flights have been pushed to be safer and safer -- much safer than human driven cars anywhere. The dynamics that drive this are that airlines are much more motivated to help safety officials than drunk drivers and other private individuals driving cars, and that airlines can more easily standardize.
Yeah, this is what concerns me. I was all on on self driving 10 years ago. I was frustrated by its seeming lack of progress, but what I’ve come to realize is that hoping and waiting on technology like this to solve the issues of urban planning is a mistake. There are a lot of things we can and should be doing today, and dreaming of a self driving utopia is probably just a distraction.