Comment box Scope: summary, personal viewpoint Tone: neutral Opinion: yes Sarcasm/humor: none This video from Benn Jordan (who seems to be a car enthusiast) argues that current self-driving...
Comment box
Scope: summary, personal viewpoint
Tone: neutral
Opinion: yes
Sarcasm/humor: none
This video from Benn Jordan (who seems to be a car enthusiast) argues that current self-driving technology, especially but not exclusively Tesla automation, is hopelessly unsafe relative to human drivers. He also suggests that self-driving technology in general cannot be programmed to be safer than humans. His central claim for the latter point appears to be an impression that the way a machine can efficiently navigate a world designed for humans is different than the way humans can.
I am not familiar with the video author. Not all his conclusions are convincing to me. However, the data he provides about current technology is compelling. I think he is overconfident about the ability of humans to consistently drive in a safe manner, but I agree with his conclusions that current self-driving technology is nowhere remotely close to that level of safety. People are definitely expecting too much from these vehicles and are believing fabrications about safety from executives who ought to know better.
I personally dislike Elon Musk's public persona and believe him to be a dishonest person. I feel that it is morally wrong for him to promote his vehicles as "safer than human drivers" when they evidently are not. But Jordan also blames a lack of government legislation for the proliferation of falsely marketed self-driving vehicles like Teslas. (Maybe this is unsurprising in a country that already makes virtually no effort to prevent the deaths of Vulnerable Road Users.)
Jordan's proposal that self-driving cars have their own lanes is probably not compatible with good urban design and traffic management, as in many or most cases it would significantly increase the amount of car lanes required on a street (obviously bad; far too much space redundantly dedicated to automobiles which will be empty most of the time) or would make many roads off-limit to certain kinds of vehicles (in a one-way street system in a place like Philadelphia with its single-lane streets, this would mean every one in every 4 streets oriented in a particular direction are accessible to a particular kind of vehicle, which I don't think drivers would accept). Enforcement seems difficult or impossible. Also, that proposal does not solve the technology's problems with recognizing obstacles like bicycles and pedestrians. I do not think he has any background in urban planning. His commentary on networked self-driving vehicles is not thorough and I do not think he has a background in computer science or machine learning either, but perhaps he is just keeping it simple for the video on purpose.
However I think it is important to discuss the empirical failures of self-driving vehicles and to derail the hype train, as it were, about the technology. While interesting, I agree with him that many current implementations of self-driving vehicle tech should not be legal on public roadways. It is more important to keep people safe than to make everyone into unwitting lab rats in corporate America's latest obsession.
It's high time we recognized the downsides of car culture, especially the side effects when it comes to urban design, pedestrians, and bicycles. For those who think they're only a driver, and...
It's high time we recognized the downsides of car culture, especially the side effects when it comes to urban design, pedestrians, and bicycles. For those who think they're only a driver, and never a pedestrian: we're also talking about your walk from your parking space to the store or restaurant. Whether you use walking to traverse miles of distance unassisted or park in a handicap spot 5 feet from a building entrance, we're all pedestrians sometimes.
Crazy to see people go further down the crappy urban design rabbit hole to try to force "self-driving cars" to work IRL. For the love of god, can we please just admit defeat and build some freaking trains? If we have to build special lanes for your self-driving cars, you know what you've designed? A personal train. With a huge computer in it, special expensive roads instead of rails, and a shitton of wasted space and energy. And you've made the world even harder for pedestrians to navigate, because now they have extra lanes to walk across. And now the poor people who can't afford a $100,000 self-driving car are stuck in megatraffic in the non self-driving lanes, and have to waste even more of their time in cars waiting for self-driving cars to cross at separate times.
Trains, bikes, and feet. At least for urban areas, it's all you need. Self-driving cars at best double down on an unsustainable, expensive, and dangerous lifestyle... at worst, they don't exist in reality so the crude self-driving fascimiles will extract a blood toll from the rest of us.
Honestly, building well thought out trains would fix the majority of urban design issues in North America over the long term. Development follows infrastructure. If you invest in trains and stop...
Honestly, building well thought out trains would fix the majority of urban design issues in North America over the long term.
Development follows infrastructure. If you invest in trains and stop investing in roads, people will prefer taking trains, development will coalesce around transit stops, and the problem will resolve itself after a few years.
The issue with that is that our governments don't work like China or even most places in Europe. We don't have a ton of centralized power in our governments and small pockets of people have the power to stop development that will inconvenience them in the short term. Just building trains will negatively impact a lot of people in the first few years before they see any benefit from it. Eminent domain will kick people out of their homes, a lot of people will have cars that will become worth less and less as they become less useful ways of getting around, money and space being reallocated from roads to rail means that previously used roads may be demolished, some may be mostly abandoned to disrepair, and developments that were built in completely inefficient, illogical places (read:exurbs) due to artificial market forces caused by auto subsidization and rent seeking legislation will become less convenient, less affordable places to live.
In places with political systems like ours, it's impossible to move quickly with big infrastructure projects. They need to be implemented in small chunks that can be completed relatively cheaply and quickly so that the benefits can be realized by constituents, and those successes can drive political will for the next phase.
That means that instead of a massive, citywide regional or local train network, which may be the most economical and efficient way to implement rail, something like a small two or three stop line to already well populated areas must be rolled out along with housing development agreements beforehand, bike lanes, and other positive development projects all rolled out simultaneously. Once that gets implemented and people see the benefits of it, they're more likely to vote for other, similar expansions.
Comment box Scope: comment response, information Tone: neutral Opinion: a bit Sarcasm/humor: none Actually in most US population centers, including most suburbs, there are currently active freight...
Comment box
Scope: comment response, information
Tone: neutral
Opinion: a bit
Sarcasm/humor: none
Just building trains will negatively impact a lot of people in the first few years before they see any benefit from it.
Actually in most US population centers, including most suburbs, there are currently active freight lines with dedicated, grade-separated ROW which could be co-opted (agreeably or begrudgingly), expanded (agreeably), or seized (disagreeably) to accommodate passenger rail. You do not really need to bulldoze a ton of houses to operate more passenger rail.
If you look on OpenRailwayMap you can see just how many tracks are operational in the country. You need to zoom in to see the branch lines I'm referring to (in yellow). This is not to mention all the rights of way which still exist, but which do not currently see freight or passenger service (in brown). The rights of way are the most important thing here.
If you overlay this map with a map of population density in the US, it is clear that pretty much anywhere there are people, there is already a train track not far away. Any "suburb" which is actually a suburb and not literally the middle of nowhere (i.e. rural) has freight access in the vicinity, often right next door. And actually lots of straight-up rural places have rail access too.
On the East Coast, it is rare to be more than 25 miles from a freight line. An example of this is the Allegheny National Forest in northern Pennsylvania, which is... protected land where literally nobody lives. While a true rail revolution would indeed require some amount of population redistribution around rail lines or some new rail lines to match existing population distribution, the network is largely already there.
Out West, such as in rural South Dakota, things are much more spread out, but the places where people live are typically still served by freight lines. This is because people prefer to live where they have access to amenities like businesses. Businesses prefer to operate in areas with cheap freight. There is much empty land, but in that empty land, there are few people. Suburbs are not empty; they are sub-URBS. You can't have a train right in front of everyone's house (nor would you want to), but you can have a train to pretty much every town of note.
Consider the population density of South Dakota, to continue this example. The population centers are Rapid City and Spearfish in the west, and various eastern counties centered around Sioux Falls, including the vaguely dense Brookings, Watertown, Huron, Mitchell, Yankton, and Aberdeen. I mean, half of these are towns of 15,000 people and they still have perfectly fine rail rights of way. There are other pockets of vague density around the state, like Pierre near the center. Want to know what all of these have in common? They already have active freight lines. Many of them are branch lines, which means there isn't even that much freight traffic, so passenger service would be pretty easy to schedule in without even needing additional infrastructure. But as stated previously, because the rights of way already exist and are not necessarily super constrained, it would often be feasible to expand the ROWs to include dedicated tracks for passenger service. If you live in a suburb in the US, you are closer to trains than you realize. Freight trains DOMINATE this country's landscape. They are just invisible and easily forgotten.
In terms of operating passenger service, the reason it isn't done is because it isn't profitable like freight is. The historic reason a lot of towns are still near freight lines is because freight companies used to be the ones providing passenger service (the federal government literally forced them to do so despite its unprofitability, until the mid-20th century). They don't have to do that now, but they continued operating freight, and the people stayed where they were because they had cars now. Consistent rail service would require local, state, and federal support in order to be effective. It is certainly possible. They would need to communicate with freight companies, upgrade tracks on existing ROWs in low-freight areas, install more tracks on ROWs, acquire rolling stock, create agencies to operate the trains (or assign the routes to existing agencies like Amtrak or some state services), educate a workforce to operate the trains, and build acceptable passenger stations (perhaps the most expensive part; could still be done fairly cheaply, you don't need a Moynihan everywhere). The funding would need to be consistent... like it is for literally all roads and highways. But it is totally possible.
Lately, Amtrak has been talking about reactivating service on a lot of dormant lines like these. It takes time and effort from stakeholders, but most of the time they aren't building new greenfield lines, they're utilizing existing ones. Many states, notably Virginia and North Carolina in the case of the S-Line, are also spearheading their own rail revitalization projects using abandoned freight rights of way.
Comment box Scope: information Tone: neutral Opinion: none Sarcasm/humor: none Someone else had the same question a few weeks ago. I answered them here.
This is also already happening today, just with roads.
Just building trains will negatively impact a lot of people in the first few years before they see any benefit from it. Eminent domain will kick people out of their homes
This is also already happening today, just with roads.
Yes, but the human tendency to prefer the devil you know makes that not such a big deal for people. Most people will not gawk at spending 10 billion to expand a major arterial in their city....
Yes, but the human tendency to prefer the devil you know makes that not such a big deal for people. Most people will not gawk at spending 10 billion to expand a major arterial in their city. They'll protest and scream about 1 billion for a light rail line all day though.
Good news is that's already a standard feature of cars. A car more than 10 years old is generally worth less than $10,000, from a new-purchase price over $40,000. More than 20 and it'll be worth...
a lot of people will have cars that will become worth less and less
Good news is that's already a standard feature of cars. A car more than 10 years old is generally worth less than $10,000, from a new-purchase price over $40,000. More than 20 and it'll be worth exactly what the scrap dealer will pay for it by weight.
Comment box Scope: comment response, information, personal take Tone: neutral Opinion: not much Sarcasm/humor: none Policy is necessary to discourage housing in places that are extremely...
Exemplary
Comment box
Scope: comment response, information, personal take
Tone: neutral
Opinion: not much
Sarcasm/humor: none
Policy is necessary to discourage housing in places that are extremely inefficient to access. Such housing is bad for the environment and expensive for government to maintain access to. Market forces can react to such policy, which may be in the form of taxation, to "naturally" (lol) change where new homeowners choose to settle over some decades. I don't think this is a simple issue though.
Passenger rail is an essential part of a functional suburban transportation network for reasons I've discussed elsewhere, but just as important are buses. Though they are a form of rubber-tire automobile with all associated externalities (space, pollution, noise, etc.), they have the benefits of being:
Public and therefore accessible to anyone who needs them (unlike private vehicles)
Reliant on trained, professional drivers (rather than untrained, unprofessional laypeople drivers in cars)
Operational on specific, fixed, predetermined routes which can be carefully designed for their presence and which drivers (or automation tech) can predictably practice on before operating passenger service
Often fairly inexpensive to operate on a per-revenue-mile basis relative to rail, all things considered
Relatively straightforward to deploy in an area with existing car infrastructure. Even Bus Rapid Transit (BRT) requires relatively little infrastructure as it can utilize existing highway/arterial rights of way
Less technical to operate and maintain than trains, which are more complicated and expensive machines and require a somewhat more specialized workforce
Rail is valuable because trains have significantly higher capacity than even articulated buses. This is why New York City's upcoming Interborough Express in suburban parts of the city (like 70th St & 66th Pl in Queens, or 40th St & Ave H in Brooklyn... neither are "urban" despite being in NYC city limits) will be rail. Many suburban places could feasibly support such services. But in suburban areas which don't necessarily have the population density to support light or heavy rail outside of a few corridors, buses running fixed routes are a legitimate and fairly safe way to get people around.
Street-running light rail has some of the same benefits as buses insofar as it is still a professionally operated service and does not require expensive grade separation, instead utilizing existing ROW. But it requires somewhat more infrastructure. And when you consider things like "trackless trolleys" (rubber-tire buses running on electrified overhead wiring) and battery-powered light rail, the distinction starts to blur a little.
There are a lot of streets/roads in the US where private vehicle use should be restricted nearly entirely in favor of buses. There are many other streets/roads where private vehicle use should be prohibitively slow (for safety reasons) to the point that taking a bus or train is always the faster option. Obviously, bus or train service also needs to be frequent enough to get people where they need to go in an efficient manner.
Add more buses. Make their times more reliable. I would definitely take the bus more if there were more/better routes and there were more buses so I didn't have to plan my time four hours in...
Add more buses. Make their times more reliable. I would definitely take the bus more if there were more/better routes and there were more buses so I didn't have to plan my time four hours in advance if I wanted to get somewhere slightly off the bus route.
I went into this with a pretty open mind. Elon is a grandiose idiot. I think FSD is essentially a scam. I've owned my Model 3 for almost 5 years now, clocked in 60K miles of which around 20K are...
I went into this with a pretty open mind. Elon is a grandiose idiot. I think FSD is essentially a scam. I've owned my Model 3 for almost 5 years now, clocked in 60K miles of which around 20K are with autopilot (almost entirely on interstates).
I did not find this video particularly convincing and the creator did not do his due diligence. At 15:00, he mentions 150 million miles driven with FSD that Elon announced in an investor call of April 2023, and then compares it to 736 crashes of which 17 involved fatalities to demonstrate a rate of 11.3 deaths per 100 million miles to show that Teslas and FSD isn't actually that safe. Except he mixed up autopilot accidents with FSD (originally reported by WaPo here). Every Tesla has autopilot. Only like 400K have the full FSD beta. I have no idea how many miles have been driven with autopilot but it's a lot more than FSD, so combining both of those to get this death rate both wrong and deceptive, especially since this is one of his "data" linchpins for "Teslas are more dangerous, actually" points.
I think it will be quite some time before we have real, legitimate level 4+ self driving cars. Like with many software solutions, a 70% solution was quick to create and works "well enough" for a lot of scenarios, but that last 30% is very, very difficult and people won't accept a self driving car that doesn't perform better than they can by pretty much every metric in every scenario. Technology will get there eventually, so declaring cars will never drive us around is very likely wrong. I don't know what a good solution to this kind of behavior is and I more or less agree with the conclusion of his video, that this tech isn't ready and probably shouldn't be tested on public roads.
I like autopilot and thinks it works well for what it is when applied to the right situations. I only use it for stop and go traffic on any road, and interstate driving when conditions are pretty clear like light to no rain, day and night, which turns out to be the vast majority of my interstate driving.
Yep. I wish Elon would be removed by the board as I think he does more harm to the company with his public idiocy these days. However my Tesla FSD does a really good job driving me places, and is...
Yep. I wish Elon would be removed by the board as I think he does more harm to the company with his public idiocy these days. However my Tesla FSD does a really good job driving me places, and is getting better. That said, I am always alert because it's software and can make mistakes. My job is to make sure it doesn't, and that's made really clear by the UI. It's basically riding coach to a pretty good student driver. I think being against this is only a smart idea if you believe something societally catastrophic is going to happen, otherwise the tech will eventually get to L4 given enough time.
For some reason, I feel like it's exceedingly unlikely that the board that tried to pay him like $56B or something because he's so essential will want to remove him. Maybe there's some way he...
For some reason, I feel like it's exceedingly unlikely that the board that tried to pay him like $56B or something because he's so essential will want to remove him. Maybe there's some way he could get removed because of all his misleading comments, though.
Saw this a week or so ago, and think it's mostly a good breakdown of where things stand. Unfortunately I do think there's some flaws in how he presents his arguments, but at the end of the day his...
Saw this a week or so ago, and think it's mostly a good breakdown of where things stand. Unfortunately I do think there's some flaws in how he presents his arguments, but at the end of the day his facts mostly line up.
There's a lot of pie in the sky absurdity around automated driving, but what has been especially disappointing is watching someone like Musk overhype a tech, which I get, and then no one seeming to care that they've completely missed deadline after deadline? Like that should maybe be a sign to manage expectations?
Comment box Scope: comment response, personal viewpoint Tone: neutral Opinion: yes Sarcasm/humor: none Yes, my takeaway -- more than just the empirical data suggesting that self-driving Teslas are...
Comment box
Scope: comment response, personal viewpoint
Tone: neutral
Opinion: yes
Sarcasm/humor: none
Yes, my takeaway -- more than just the empirical data suggesting that self-driving Teslas are 10x more dangerous than human drivers -- is indeed that the hype around self-driving tech is wholly unrecognized to be a problem among a certain subset of the population. People just gobble up the false promises.
Jordan quotes some stats indicating the safety that L1 and L2 automation tech can provide when a driver is still expecting to control the vehicle, like lane assist tech, or automatic speed reductions when the vehicle detects something ahead. But this idea that Teslas can "drive themselves" and still see those safety benefits seems to lull people into a false sense of security. When they are no longer expecting to to control the vehicle "in general," those extant safety benefits are canceled out by the driver's inattention to the vehicle's erratic behavior.
I think a lot of people, including some very smart people I know, have fallen prey to a billionaire grifter's narrative that full automation tech is actually on-par with human driving in terms of safety, or that it's close enough to call "safe." It is not. Frankly I just don't think most people have considered how ridiculously complex streetscapes are. I think that many people have been conditioned to believe that technology (and nothing else) is the solution to literally every single problem, and that a more technological solution is basically always the better one. (Even Jordan is at least partially an example here: his suggestion about networked vehicles is a techy idea to a problem that does not consider... concrete. Or actually any part of road layout, traffic management, other than cars technologically (magically) talking to other cars.) Vague tech-hyping is a narrow perspective which clouds our judgment.
The reason this video stood out to me is that this technocratic perspective of the world is (in my experience) particularly pervasive among upper-middle class/professional class types: people who are educated (not in tech specifically, but in something), at least kind of well-off and likely to buy fancy new technology, and politically progressive, and who should know better given their education than to take someone like Elon Musk at his word. i.e. the market for Teslas. As you say, Tesla has repeatedly failed to deliver what Musk has claimed and yet the widespread acceptance of self-driving tech among people who can afford Teslas is weirdly high. At least some of them have also specifically invested in the stock and thus have a perverse incentive to hype it up, as Jordan says. I think there's a lot of criticism of Tesla and self-driving cars in general, but a lot of that is tied politically to electric vehicles (because Teslas are electric) and denial of climate change and thus I would consider that critique irrelevant in this discussion; and isn't rigorous or academic to begin with.
I think that a lot of less educated and younger and/or more impressionable people have also fallen prey to the grifting, especially people who see Musk as this brilliant billionaire who definitely pulled himself up by his bootstraps and didn't have parents who owned an emerald mine or anything. Many of these people are the exact OPPOSITE of the group I just described. It is weird that they have remained so steadfast for so long.
Another commenter mentioned that many people who work with software are skeptical of self-driving tech, which I think is true for some such people I know. But there is a weirdly passive attitude, across the board, toward this unproven and evidently dangerous technology. The passive attitude is especially weird when you think about how terrified some of these same people are of, like, flying (which is at least an order of magnitude safer than driving on a per-mile basis). Just look at how much press Boeing is getting.
As someone who is very interested in urban planning and public transportation, it is troublesome to hear people talking about "the end of transit" as though self-driving vehicles can somehow replace traditional transit, e.g. robotaxis somehow being more efficient than trains or buses. This kind of attitude discourages investment in public transportation infrastructure (as well as infrastructure for pedestrians/anyone not in an automobile) because it implicitly reinforces car-centric attitudes about the way society should be designed. No amount of car automation will ever (and I mean, literally, EVER) solve the majority of traffic-related problems experienced by cities; nor will it magically solve income inequality; nor do I imagine it will truly and wholly stop all traffic fatalities; etc etc. Because automated cars are still cars, they still have most of the externalities of cars. Hype around self-driving tech forgets that a world designed primarily for cars rather than people is uniquely bad for the environment, loud and hazardous to health (in ways beyond being physically struck by a vehicle), and expensive to maintain relative to its social benefits.
One of the interesting things to me that has often come up in these automation debates is planes. They're often used as an example of why it should be possible, when really they are a clear...
One of the interesting things to me that has often come up in these automation debates is planes.
They're often used as an example of why it should be possible, when really they are a clear example of why it likely isn't.
Planes are in almost 100% known environments with miles of space between them. When taking off or landing EVERYTHING is supposed to be known and scheduled by a team of humans. ONCE all that is done, yes the mechanical actions of taking off/landing are mostly automated, even to some advantage in situations where it's instruments only.
However, if you're the kind of person who reads the aviation accident investigations and the discussions around them, say about air france and their major crash, you'd know that there's a huge discussion of "too much information and too much automation" being unsafe. When you have 40 alerts going off, and they conflict with your instruments, you start conditioning yourself to ignore alerts.
Further, even with 2 people in the cockpit, there's an issue of complacency on long automated trips. "QUICK YOU HAVE 2 SECONDS TO MAKE A LIFE OR DEATH DECISION" after minutes or hours of nothing is something humans are notoriously bad at, and it's darkly hilarious how these car automation systems say "well in an emergency or situation it doesn't understand, it'll alert the driver and they'll have X seconds to respond", as if somehow the vast majority of the population is going to do better than 2 trained pilots. And that's ignoring that in all likely hood they'll have waaaay less than X to do anything because unlike planes you don't have miles of space to react in.
At the end of the day I think one of the most interesting points brought up in the video is that the best automated system probably doesn't try to mesh with human needs, but at that point we're back to "Oh shit it's trains isn't it", or some similar issue.
The training those pilots get is also on that specific model of airplane, and understanding how all of those automated systems work on a deep level. I'd be willing to bet that you could ask a...
The training those pilots get is also on that specific model of airplane, and understanding how all of those automated systems work on a deep level.
I'd be willing to bet that you could ask a pilot to draw you a high level system architecture of how the autopilot system on their assigned aircraft works, and they'd be able to do it to a fairly accurate level.
In cars though, the whole point of these automated driving systems is that the drivers are explicitly not supposed to know how they work. Auto manufacturers go to great lengths to not only market the fact that their cars are basically just magical boxes that take care of all the hard stuff for you, their self driving systems are probably the most viciously guarded piece of intellectual property that they own at this point.
Yeah, this bugs me. I do think that automated transportation is possible, even on a personal level, but trying to get it to work on public roads alongside humans seems to be incredibly reckless....
At the end of the day I think one of the most interesting points brought up in the video is that the best automated system probably doesn't try to mesh with human needs, but at that point we're back to "Oh shit it's trains isn't it", or some similar issue.
Yeah, this bugs me. I do think that automated transportation is possible, even on a personal level, but trying to get it to work on public roads alongside humans seems to be incredibly reckless. Heck, when it comes to public transport, there are tons of fully automated systems in service right now! The difference is that those were, for the most part, built in tandem with the infrastructure to support them. In contrast, seeing companies trying to sell autonomous cars that work on existing infrastructure feels like a scam because it’s basically a pipe dream. It’s no better than what Elizabeth Holmes was doing with Theranos.
Video games can help teach in this vein. Fortnite matches typically run 30ish minutes beginning to end. If you drop far away from everybody else, there's a decent chance you encounter nobody for...
QUICK YOU HAVE 2 SECONDS TO MAKE A LIFE OR DEATH DECISION" after minutes or hours of nothing is something humans are notoriously bad at
Video games can help teach in this vein.
Fortnite matches typically run 30ish minutes beginning to end. If you drop far away from everybody else, there's a decent chance you encounter nobody for 25 minutes and then are steamrolled by the final 5.
If you drop at the beginning, and continually chase gunfire, you'll play better thanks to that constant stream of adrenaline.
This is most apparent in group play, where a bit of coordination and attentiveness is required.
I have driven multiple cars with lane assist. They all do fine and dandy when conditions are calm and the roads are well-painted and relatively flat. Introduce hills, faded overlapping lines,...
I have driven multiple cars with lane assist.
They all do fine and dandy when conditions are calm and the roads are well-painted and relatively flat.
Introduce hills, faded overlapping lines, variable lane widths, 30+ mph winds, and especially any combo of those...things get wonky real fast.
I think you've got the nail on the head here. Ill admit that when I was a young nerd, I often fell prey to that mindset as well. With every problem, I often thought that there was always a...
I think that many people have been conditioned to believe that technology (and nothing else) is the solution to literally every single problem, and that a more technological solution is basically always the better one.
I think you've got the nail on the head here. Ill admit that when I was a young nerd, I often fell prey to that mindset as well. With every problem, I often thought that there was always a technological solution, and that if only those luddites got onboard, we could computerize and automate all of our problems away.
After a long career in technology, and especially after I made the transition to cybersecurity, I've realized that in most cases, the opposite is actually true.
Simpler solutions are better than complex ones, all other things being equal. Any increase in complexity in a system needs to be carefully considered with it's drawbacks, and that complexity needs to be defended and justified.
Complexity for complexity's sake, or to look more "modern", or to save some engineer a few hours designing a more elegant, simple solution is endemic to most of the consumer landscape unfortunately. With each increase in complexity though, the points of potential failure are multiplied, potential security weaknesses are multiplied, and human's mental model of the system is hampered, making troubleshooting and evaluating much more costly and less effective.
There are a lot of great advances in technology in automobiles. Fuel injection is incredible. Carbauration was always a clunky, terrible solution for fuel mixing. Various sensors around cars can make diagnoses much easier.
But there are antipatterns there too. Infotainment systems with custom touchscreen OSes are buggy and prone to crashes on every car I've ever been in. Things like movable headlights are pointless show offs that don't do much. My car has electronic suspension that is supposed to dynamically stiffen or loosen based on how I'm driving. Do I ever notice it? Absolutely not. Will it help me avoid a crash ever? I very much doubt it. Does it require probably thousands of dollars of sensors, computers, and actuators? I would be willing to bet it does.
Peak infotainment system was when cars included an aux-in port built-in, thus rendering them futureproof against any future audio technology. (half joking)
Infotainment systems with custom touchscreen OSes are buggy and prone to crashes on every car I've ever been in.
Peak infotainment system was when cars included an aux-in port built-in, thus rendering them futureproof against any future audio technology. (half joking)
My suspicion is that the passive attitude from those who should know better stems from their knowledge of the atrocious safety record of human drivers. They (incorrectly) assume that it should be...
My suspicion is that the passive attitude from those who should know better stems from their knowledge of the atrocious safety record of human drivers. They (incorrectly) assume that it should be relatively easy to beat the dogshit capabilities of the average human at operating heavy machinery.
Thanks for sharing this, it's an informative video with a lot of facts to back up my gut apprehension about autonomous vehicles. The more I understand about how these technologies work, the less I...
Thanks for sharing this, it's an informative video with a lot of facts to back up my gut apprehension about autonomous vehicles. The more I understand about how these technologies work, the less I trust them. I say that as a software engineer by trade, and my less geeky friends and family have raised their eyebrows when I've mentioned my hesitancy to them. They expected me, the "tech guy," to be all-in on self-driving vehicles, but I don't want anything to do with the things. Anecdotally it seems like it's the less technically inclined folks who are more trusting of it, which doesn't make a lot of sense to me.
The other day I was driving through a construction zone in a pedestrian-heavy area with speed bumps and crosswalks. There were traffic cones indicating a lane switch into the oncoming lane, signaled by a flagger. Straight ahead through the cones was a giant gaping hole. Honestly as I approached it I could barely sort out where I was supposed to go and when. There's not a computer system on the planet that could navigate it correctly. Best case scenario, the car just stops in the middle of the road. Worst case, you end up in the hole (along with the flagger your car dragged beneath it).
I visited Maui last year and drove the Road to Hana. It's a stunningly beautiful drive. It also demanded my full attention, with its blind hairpin curves, dozens of single-lane bridges, and frequent occurrences of tourists and their cars wherever they felt like parking on the shoulder for some sightseeing. The bridges were the craziest part, often I had to engage in some nonverbal communication with oncoming cars just to negotiate who went when. At one point I had to quickly throw the car in reverse to avoid a collision. And I haven't even mentioned the locals who drive way too fast (and the expected courtesy of slower drivers pulling off the road when they come up behind so they can pass), or the oversized oncoming work trucks crossing the center line around blind corners.
Anyway, I'll have confidence in self-driving cars when they can seamlessly handle both of the above situations without killing anyone (or giving up and just stopping the vehicle in the middle of the street). If the tech ever matures to that point, I don't expect it to happen for several decades at least.
If you'll indulge me for a moment, I want to talk about alternative approaches to vehicular autonomy. Things like some sort of mesh network all cars participate in or the swarming behavior shown in the video are really interesting but they're impossible for one reason: uncontrollable variables in a dynamic environment. You can get all the new cars to play nicely together, but they still have to share the space with the old cars that don't. (The designated lanes idea in the video would help but not solve this fully.) You have non-vehicles that enter the roadway unpredictably — wildlife, pedestrians, construction, rogue soccer balls, whatever. You can't plan for every contingency.
At the same time, we still don't have flying cars. Maybe this is their moment. What if when they finally arrived, they all came equipped with a standardized self-driving system that communicates with neighboring vehicles? The biggest old argument I heard against them was the danger a bad driver could pose driving drunk at cruising altitude or whatever. But if every single one of them is autonomous from day one? Maybe not so bad. With no manually driven cars sharing the airspace, and of course no pedestrians or construction to contend with, maybe no real hazards apart from birds... I dunno, I think it's compelling. It'd be a hell of a lot faster to get from A to B, I know that much at least. If Not Just Bikes has taught me anything, it's that roads suck... I'm kinda ready to just be done with them altogether.
I don't mean to be overly negative in my response, but I want to point out a few things. There are plenty of other hazards to aircraft than just birds. Many are weather related: lightning, icing...
At the same time, we still don't have flying cars. Maybe this is their moment. What if when they finally arrived, they all came equipped with a standardized self-driving system that communicates with neighboring vehicles? The biggest old argument I heard against them was the danger a bad driver could pose driving drunk at cruising altitude or whatever. But if every single one of them is autonomous from day one? Maybe not so bad. With no manually driven cars sharing the airspace, and of course no pedestrians or construction to contend with, maybe no real hazards apart from birds... I dunno, I think it's compelling.
I don't mean to be overly negative in my response, but I want to point out a few things. There are plenty of other hazards to aircraft than just birds. Many are weather related: lightning, icing (interferes with aerodynamic properties of lift surfaces), downdrafts, etc. But you also have dealing with mechanical and electrical failures during flight.
Aircraft systems have an very high amount of redundancy -- typically three of everything (you can look up Triple Modular Redundancy if you want to know more). This is necessary because you can't merely "pull over" -- you have to keep flying. The 777 was the first fully by-wire aircraft (electronic actuation, rather an mechanical/hydraulic) and even the computer networks are tripled up and routed through different parts of the fuselage so that an intrusion is unlikely to sever all three. The weight cost of this is not trivial and becomes much more significant as you scale down the size of the aircraft.
Incidentally, the redundancy question is one that I haven't seen satisfactory answers for from the autonomous car industry yet either, but it would be a problem for them too, eventually. Doubling up on sensors means adding more of the most expensive parts, plus there are interference problems with redundant LIDAR and RADAR. They haven't been bitten by this because the technology had not achieved a level of automation where it can be deployed at scale, which is where those long tail failures start to show up.
In addition to redundant design, current aviation is workable because you have two (or more) highly trained, professional pilots who are further certified on the particular aircraft they are flying who can work with ground control (and sometimes even the aircraft engineers) to resolve issues with the plane, route around bad weather, etc. The problems that show up if one tries to delegate handling all these unexpected cases to a computer would be the same for aviation as they have been for the current group of autonomous vehicles. There are too many weird situations to effectively plan for them all with our current methods. Trying to rely on pilots at the same level of capability as car drivers would be unworkable for a variety of reasons: too little training, too little responsibility for things going wrong, too much variation in physical and mental ability, no redundancy.
Maintenance is another difficulty. Ask anyone who owns a private plane, and you'll find out that the parts and the mechanics have to be specially certified, which is very expensive. There are reasons that private jets are so expensive to operate.
Even though the sky is pretty empty, that's mainly because there aren't that many planes, and they are all going to a relatively small number of destinations. A "near miss" for aviation means coming within a few hundred feet to another aircraft, and even at that distance, it is far from a solved problem. Most of the problems happen at the airport, and that environment requires close manual monitoring by highly trained humans to avoid collisions and manage traffic. If the number of incoming aircraft at airports were, say, 10x number of smaller vehicles, even the current system would be intractable, both in collision avoidance and in runway availability. If one starts having to manage collisions and near misses for far more vehicles in airspace above population centers, it would be even more challenging, and the failures would be high profile and horrific.
All that is to say, there is no free lunch when it comes to autonomy in any domain. The world is complicated, humans are fallible, and each domain has its own difficult challenges.
Meanwhile, at Boeing: What if we added a computer that can pitch the nose of the plane based on information from a single sensor, and didn't tell the pilots how it works?
The problems that show up if one tries to delegate handling all these unexpected cases to a computer would be the same for aviation
Meanwhile, at Boeing: What if we added a computer that can pitch the nose of the plane based on information from a single sensor, and didn't tell the pilots how it works?
Shit like this still blows my mind. How it was ever allowed in the air relying on a single sensor is just unacceptable beyond belief and the fact that more heads didn't roll for that alone shows...
Shit like this still blows my mind. How it was ever allowed in the air relying on a single sensor is just unacceptable beyond belief and the fact that more heads didn't roll for that alone shows how toothless the FAA is becoming.
It's right up there with aribus' insane 'oh if both pilots put in different inputs on the stick, then it'll just accept the first one and ignore the second WITH NO FEEDBACK' that lead to another 200 deaths.
That Airbus crash was the source of my most frustrating ever online discussion. Someone claiming to be an engineer kept insisting that it simply wasn’t possible to have a switch in the cockpit...
That Airbus crash was the source of my most frustrating ever online discussion. Someone claiming to be an engineer kept insisting that it simply wasn’t possible to have a switch in the cockpit that toggled control between the pilot and copilot, and that the current system where the two pilots agree between themselves as to which one is actually flying the airplane at any given time was the only way it could be. I find this extremely hard to believe.
Comment box Scope: comment response, personal viewpoint Tone: neutral Opinion: yes Sarcasm/humor: none To me, this is the root of the issue. Anytime technology is being deployed to solve a novel...
Comment box
Scope: comment response, personal viewpoint
Tone: neutral
Opinion: yes
Sarcasm/humor: none
To me, this is the root of the issue. Anytime technology is being deployed to solve a novel problem (which will be the case until the end of time), and anytime the strategic decisions of that deployment are being made by fallible human beings (also a constant), there are bound to be misapplications of the technology due to misalignment between humans (who necessarily have a subjective perspective), our subjective understanding of the environment, and our subjective understanding of the tech.
This is especially the case when those fallible human beings are 1) not engineers or otherwise technical people, and 2) incentivized to release technology and not to prioritize safety. That can theoretically be solved, but practically speaking, every company or government agency will always contain non-technical perspectives and motivations (valid or not). Safety checklists for maintenance and good software QA don't mean much if executives, governments, or other forces are creating a company culture that precludes those checks.
Probably some amount of additional regulation can solve some number of these issues, but it has its limits. Some domains are narrow enough to deploy automated systems with a realistic amount of confidence, but car automation on public streets is probably not one of them given the inherently high number of potential human-computer misalignments in such a complex environment.
Yes, well that is a problem too. I think more than anything, the Boeing debacle(s) show how precarious safety critical infrastructure is without strong regulatory oversight. I wish the lesson...
Yes, well that is a problem too.
I think more than anything, the Boeing debacle(s) show how precarious safety critical infrastructure is without strong regulatory oversight. I wish the lesson learned from this would be the walking back of regulatory capture at the FAA and the establishment of similar oversight in various tech industries, but, as the man says, "if wishes were fishes, we'd all cast nets."
I agree that self-driving vehicles that purely rely on sensors (eg visual and radar) are unlikely to reach the safety level of human driving. Waymo does pretty well, but it's definitely not...
I agree that self-driving vehicles that purely rely on sensors (eg visual and radar) are unlikely to reach the safety level of human driving. Waymo does pretty well, but it's definitely not perfect and requires an intense amount of programming specific to its fenced-in area in order to be effective. Full, safe autonomy will likely require vehicles and infrastructure to be able to communicate their intentions to each other, and that will only be effective if it's everywhere. That level of infrastructure probably won't exist in our lifetime, but I wouldn't say "never". It's a shame, because self-driving cars are one of the few transportation technology improvements that I'd actually be excited about.
Comment box Scope: comment response, personal viewpoint Tone: neutral, tech-skeptical Opinion: yes Sarcasm/humor: none "Never" is a strong word, yeah. I am continually surprised by the way the...
Comment box
Scope: comment response, personal viewpoint
Tone: neutral, tech-skeptical
Opinion: yes
Sarcasm/humor: none
"Never" is a strong word, yeah. I am continually surprised by the way the world changes. But communication between multiple self-driving vehicles doesn't inherently address the safety of people who aren't in vehicles. If the car is running pedestrians over now, being able to communicate with another car doesn't change that.
Communication between autonomous vehicles and fixed infrastructure is interesting and has some capacity to improve safety. For example, there already exist sensors that detect pedestrians, bikes, or cars in a given space and adjust traffic light signals accordingly. These systems are common in the Netherlands and some other European countries but nonexistent in most of North America. In theory, these systems could communicate with cars to inform the cars that the sensor has detected, say, a pedestrian who might cross the street (which maybe the car can't see). That could be useful for the automobile, which can't otherwise intuit or anticipate human behavior.
But like you say, it would require a huge amount of quite expensive fixed infrastructure to be effective, which costs a lot to maintain. That cost alone makes this unfeasible in most municipalities, many of whom are already going bankrupt trying to maintain a road network whose maintenance costs they vastly underestimated. The benefits would also be absent anywhere this infrastructure has not been installed. The problem is that it's very hard to design a computerized system to pre-emptively detect every aspect of human behavior; because a human has to program it to predict humans, and it has to work in context. For instance, pedestrians often do not use designated crosswalks in a given location because it is inconvenient, dangerous, or unpleasant to do so and instead cross streets at locations where a traffic sensor wouldn't have been placed by an engineer, and therefore wouldn't detect them. In that case, the system structurally breaks down because it assumes that the car has information that it doesn't have (because it isn't receiving information from the sensors, because the sensors reflect idealized data and not real-world data).
Additionally, the more we rely on software communicating with other software (not even made by the same company) to make decisions about safety, the more we are relying on extremely thorough software engineering QA to keep us safe. There are lots of people much smarter than me working on solutions to these problems, but I have never met a software engineer who consistently writes mathematical proofs demonstrating the correctness of their algorithms. Outside of academia, it does not happen. I don't think software needs to be literally perfect, but a single bad deployment rolled out to hundreds of thousands or millions of vehicles or other detection systems could conceivably kill or seriously injure a lot of people.
Maybe I am just old-fashioned, but I would rather the approach taken to safety emphasizes the removal of vehicles from most public spaces; an extraordinary reduction in vehicle speed anywhere pedestrians, cyclists, and other Vulnerable Road Users are expected to be present; and a preference for concrete and other physical protections in lieu of computer magic.
The communication wouldn't be limited to other cars. It would also need to talk to infrastructure like traffic signals, which pedestrians would also need to obey of obvious safety reasons. It...
But communication between multiple self-driving vehicles doesn't inherently address the safety of people who aren't in vehicles. If the car is running pedestrians over now, being able to communicate with another car doesn't change that.
The communication wouldn't be limited to other cars. It would also need to talk to infrastructure like traffic signals, which pedestrians would also need to obey of obvious safety reasons. It would also supplement, and not replace proximity sensors.
But like you say, it would require a huge amount of quite expensive fixed infrastructure to be effective, which costs a lot to maintain.
For infra-based communication (not car to car), it would be a matter of retrofitting traffic and pedestrian signals with a signal emitter that tells vehicles what the state of the intersection is. This would be something that could be phased in over time as traffic signals and their controls are replaced.
The problem is that it's very hard to design a computerized system to pre-emptively detect every aspect of human behavior; because a human has to program it to predict humans, and it has to work in context. For instance, pedestrians often do not use designated crosswalks in a given location because it is inconvenient, dangerous, or unpleasant to do so and instead cross streets at locations where a traffic sensor wouldn't have been placed by an engineer, and therefore wouldn't detect them.
I agree that there needs to be better pedestrian infrastructure, but at the end of the day, this problem comes down to a human behavior problem and pedestrians would need to consider their safety by not illegally jaywalking. Much like trying to cross train tracks, it's pretty much your own damn fault if you perform an illegal crossing and get hit by something that has the right of way.
Maybe I am just old-fashioned, but I would rather the approach taken to safety emphasizes the removal of vehicles from most public spaces; an extraordinary reduction in vehicle speed anywhere pedestrians, cyclists, and other Vulnerable Road Users are expected to be present; and a preference for concrete and other physical protections in lieu of computer magic.
I don't necessarily disagree with this in urban settings, but I'll also admit some bias here in favor of the existence personal vehicles since I have no interest in living in a city and walking everywhere.
Comment box Scope: comment response, personal take Tone: neutral, a little peeved - but not at you specifically! More at... society Opinion: yes Sarcasm/humor: none This is true but I think you...
Comment box
Scope: comment response, personal take
Tone: neutral, a little peeved - but not at you specifically! More at... society
Opinion: yes
Sarcasm/humor: none
I agree that there needs to be better pedestrian infrastructure, but at the end of the day, this problem comes down to a human behavior problem and pedestrians would need to consider their safety by not illegally jaywalking.
This is true but I think you are not considering the ridiculous lack of sidewalks across the United States. Nor are you considering the ridiculous number of ADA code violations committed by private contractors/construction companies, police officers (such as illegal vehicle parking), etc. on a regular basis. I walk by several such violations every single day. There are many situations where it is physically impossible for a pedestrian not to enter the road in a "disallowed" place. Many such pedestrians are wheelchair users, who are particularly vulnerable. Many VRUs are also disabled in some way (visual, auditory, sensory, overall mental capacity, etc).
Obviously, the correct solution is to fine violators of the ADA to an extreme extent to ensure that this does not happen. But that does not solve the problem for the wheelchair user who has to enter the street in a "disallowed" location before the ADA violation has been addressed by government. They still have to go into the street in a way that is unpredictable for an automated system.
In many cases a tree branch or other obstacle will also prevent Vulnerable Road Users from using sidewalks or bike lanes "as intended." This is objectively not their fault.
The complexity of streets means that an automated car will always be in a novel and untested situation with pedestrians sometimes. This means they will always be putting VRUs at risk. It isn't acceptable to blame pedestrians for being "at fault" for technically breaking a rule when the fault is obviously with the systems that require them to do so.
There was recently a story here, or on Reddit, in which a wheelchair user received a traffic citation for rolling in the road. The police officer did not understand that the lack of sidewalks made this violation physically impossible to avoid. Situations like this are unbelievably common, but they are invisible to drivers, because drivers are not Vulnerable Road Users (by definition). Drivers are always given purpose-built infrastructure. VRUs are not.
Regardless, even if a pedestrian is in the wrong for crossing a street at the wrong time, they do not deserve to die. It should be physically impossible for a car to go fast enough on a road to kill a pedestrian who steps into the wrong place of the road at the wrong time, whether that was on purpose of by accident. Many drivers have a callous attitude toward pedestrians in this regard which I feel is unhelpful and a bit cruel.
I don't think you're trying to be callous/cruel here, but a lot of otherwise well-meaning people are indifferent toward the sanctity of human life when they arbitrarily decide that someone was "in the wrong." This is a very narrow-minded way of looking at safety because it assumes 1) that everyone has the same objective definitions of right and wrong behavior, 2) that everyone is making a conscious decision to break the rules, and 3) that the rules are even optimized to begin with. These assumptions make it easy to assign moral judgment about a person (or make actuarial insurance decisions), but are unhelpful when the goal is to save ALL lives.
I'll also admit some bias here in favor of the existence personal vehicles since I have no interest in living in a city and walking everywhere.
I meant full-sized automobiles specifically. IMO there is nothing wrong with a personal vehicle that happens to be an electric bicycle or a speed-limited golf cart (like in Peachtree City, GA). I don't think full-sized cars should be allowed within city limits, with some but quite few exceptions. Obviously, access should be maintained for emergency vehicles, but the general population should not be driving around two-ton cars with the laughable amount of training they have.
Outside of cities, I have the same opinion for residential areas. I just do not think you should be allowed to drive a full-sized vehicle in a place where children might be expected to play. They are kids and they do not act rationally. They don't go outside anymore because parents know (whether they admit this is the main reason or not) that they're afraid of traffic killing their child. Empirically it's rather rare to kill someone, even a child, with a golf cart that goes 10mph. But it happens constantly with full-sized automobiles simply on account of their obscene mass.
Personally I see little issue with building parking garages near arterial roads at the edge of town and using that to store everyone's full-sized automobiles, and using your legs, wheelchairs, bicycles, electric bicycles, and golf carts (they can be enclosed, I know it gets cold some places, yes yes yes) to get everywhere locally. Because those local places are where PEOPLE are likely to be, it makes sense to design spaces for people... and not for massive hunks of steel.
The exact implementation of the principles I'm espousing here is obviously going to differ from place to place, but the point is that there should be dramatically more pedestrianized zones all over the United States, as well as dramatically more half-pedestrianized zones where you can drive a little golf cart but not a massive SUV. However that needs to happen -- and it wouldn't be as hard as we think -- it ought to be done.
Comment box Scope: information Tone: neutral Opinion: none Sarcasm/humor: none Jordan refers to Waymo at around 3:00 but does not elaborate on their technology. He states that "in my opinion, no...
Comment box
Scope: information
Tone: neutral
Opinion: none
Sarcasm/humor: none
Jordan refers to Waymo at around 3:00 but does not elaborate on their technology. He states that "in my opinion, no legitimate [automation] stage 4 vehicle exists on public roads," in reference to claims from companies like Waymo which claim to have reached this level of automation.
The video is mostly about Tesla and Openpilot, both of which he tests anecdotally; and especially Tesla, which he criticizes extensively.
Comment box
This video from Benn Jordan (who seems to be a car enthusiast) argues that current self-driving technology, especially but not exclusively Tesla automation, is hopelessly unsafe relative to human drivers. He also suggests that self-driving technology in general cannot be programmed to be safer than humans. His central claim for the latter point appears to be an impression that the way a machine can efficiently navigate a world designed for humans is different than the way humans can.
I am not familiar with the video author. Not all his conclusions are convincing to me. However, the data he provides about current technology is compelling. I think he is overconfident about the ability of humans to consistently drive in a safe manner, but I agree with his conclusions that current self-driving technology is nowhere remotely close to that level of safety. People are definitely expecting too much from these vehicles and are believing fabrications about safety from executives who ought to know better.
I personally dislike Elon Musk's public persona and believe him to be a dishonest person. I feel that it is morally wrong for him to promote his vehicles as "safer than human drivers" when they evidently are not. But Jordan also blames a lack of government legislation for the proliferation of falsely marketed self-driving vehicles like Teslas. (Maybe this is unsurprising in a country that already makes virtually no effort to prevent the deaths of Vulnerable Road Users.)
Jordan's proposal that self-driving cars have their own lanes is probably not compatible with good urban design and traffic management, as in many or most cases it would significantly increase the amount of car lanes required on a street (obviously bad; far too much space redundantly dedicated to automobiles which will be empty most of the time) or would make many roads off-limit to certain kinds of vehicles (in a one-way street system in a place like Philadelphia with its single-lane streets, this would mean every one in every 4 streets oriented in a particular direction are accessible to a particular kind of vehicle, which I don't think drivers would accept). Enforcement seems difficult or impossible. Also, that proposal does not solve the technology's problems with recognizing obstacles like bicycles and pedestrians. I do not think he has any background in urban planning. His commentary on networked self-driving vehicles is not thorough and I do not think he has a background in computer science or machine learning either, but perhaps he is just keeping it simple for the video on purpose.
However I think it is important to discuss the empirical failures of self-driving vehicles and to derail the hype train, as it were, about the technology. While interesting, I agree with him that many current implementations of self-driving vehicle tech should not be legal on public roadways. It is more important to keep people safe than to make everyone into unwitting lab rats in corporate America's latest obsession.
It's high time we recognized the downsides of car culture, especially the side effects when it comes to urban design, pedestrians, and bicycles. For those who think they're only a driver, and never a pedestrian: we're also talking about your walk from your parking space to the store or restaurant. Whether you use walking to traverse miles of distance unassisted or park in a handicap spot 5 feet from a building entrance, we're all pedestrians sometimes.
Crazy to see people go further down the crappy urban design rabbit hole to try to force "self-driving cars" to work IRL. For the love of god, can we please just admit defeat and build some freaking trains? If we have to build special lanes for your self-driving cars, you know what you've designed? A personal train. With a huge computer in it, special expensive roads instead of rails, and a shitton of wasted space and energy. And you've made the world even harder for pedestrians to navigate, because now they have extra lanes to walk across. And now the poor people who can't afford a $100,000 self-driving car are stuck in megatraffic in the non self-driving lanes, and have to waste even more of their time in cars waiting for self-driving cars to cross at separate times.
Trains, bikes, and feet. At least for urban areas, it's all you need. Self-driving cars at best double down on an unsustainable, expensive, and dangerous lifestyle... at worst, they don't exist in reality so the crude self-driving fascimiles will extract a blood toll from the rest of us.
How do you fix suburban American/Canadian design with just a few trains?
Honestly, building well thought out trains would fix the majority of urban design issues in North America over the long term.
Development follows infrastructure. If you invest in trains and stop investing in roads, people will prefer taking trains, development will coalesce around transit stops, and the problem will resolve itself after a few years.
The issue with that is that our governments don't work like China or even most places in Europe. We don't have a ton of centralized power in our governments and small pockets of people have the power to stop development that will inconvenience them in the short term. Just building trains will negatively impact a lot of people in the first few years before they see any benefit from it. Eminent domain will kick people out of their homes, a lot of people will have cars that will become worth less and less as they become less useful ways of getting around, money and space being reallocated from roads to rail means that previously used roads may be demolished, some may be mostly abandoned to disrepair, and developments that were built in completely inefficient, illogical places (read:exurbs) due to artificial market forces caused by auto subsidization and rent seeking legislation will become less convenient, less affordable places to live.
In places with political systems like ours, it's impossible to move quickly with big infrastructure projects. They need to be implemented in small chunks that can be completed relatively cheaply and quickly so that the benefits can be realized by constituents, and those successes can drive political will for the next phase.
That means that instead of a massive, citywide regional or local train network, which may be the most economical and efficient way to implement rail, something like a small two or three stop line to already well populated areas must be rolled out along with housing development agreements beforehand, bike lanes, and other positive development projects all rolled out simultaneously. Once that gets implemented and people see the benefits of it, they're more likely to vote for other, similar expansions.
Comment box
Actually in most US population centers, including most suburbs, there are currently active freight lines with dedicated, grade-separated ROW which could be co-opted (agreeably or begrudgingly), expanded (agreeably), or seized (disagreeably) to accommodate passenger rail. You do not really need to bulldoze a ton of houses to operate more passenger rail.
If you look on OpenRailwayMap you can see just how many tracks are operational in the country. You need to zoom in to see the branch lines I'm referring to (in yellow). This is not to mention all the rights of way which still exist, but which do not currently see freight or passenger service (in brown). The rights of way are the most important thing here.
If you overlay this map with a map of population density in the US, it is clear that pretty much anywhere there are people, there is already a train track not far away. Any "suburb" which is actually a suburb and not literally the middle of nowhere (i.e. rural) has freight access in the vicinity, often right next door. And actually lots of straight-up rural places have rail access too.
On the East Coast, it is rare to be more than 25 miles from a freight line. An example of this is the Allegheny National Forest in northern Pennsylvania, which is... protected land where literally nobody lives. While a true rail revolution would indeed require some amount of population redistribution around rail lines or some new rail lines to match existing population distribution, the network is largely already there.
Out West, such as in rural South Dakota, things are much more spread out, but the places where people live are typically still served by freight lines. This is because people prefer to live where they have access to amenities like businesses. Businesses prefer to operate in areas with cheap freight. There is much empty land, but in that empty land, there are few people. Suburbs are not empty; they are sub-URBS. You can't have a train right in front of everyone's house (nor would you want to), but you can have a train to pretty much every town of note.
Consider the population density of South Dakota, to continue this example. The population centers are Rapid City and Spearfish in the west, and various eastern counties centered around Sioux Falls, including the vaguely dense Brookings, Watertown, Huron, Mitchell, Yankton, and Aberdeen. I mean, half of these are towns of 15,000 people and they still have perfectly fine rail rights of way. There are other pockets of vague density around the state, like Pierre near the center. Want to know what all of these have in common? They already have active freight lines. Many of them are branch lines, which means there isn't even that much freight traffic, so passenger service would be pretty easy to schedule in without even needing additional infrastructure. But as stated previously, because the rights of way already exist and are not necessarily super constrained, it would often be feasible to expand the ROWs to include dedicated tracks for passenger service. If you live in a suburb in the US, you are closer to trains than you realize. Freight trains DOMINATE this country's landscape. They are just invisible and easily forgotten.
In terms of operating passenger service, the reason it isn't done is because it isn't profitable like freight is. The historic reason a lot of towns are still near freight lines is because freight companies used to be the ones providing passenger service (the federal government literally forced them to do so despite its unprofitability, until the mid-20th century). They don't have to do that now, but they continued operating freight, and the people stayed where they were because they had cars now. Consistent rail service would require local, state, and federal support in order to be effective. It is certainly possible. They would need to communicate with freight companies, upgrade tracks on existing ROWs in low-freight areas, install more tracks on ROWs, acquire rolling stock, create agencies to operate the trains (or assign the routes to existing agencies like Amtrak or some state services), educate a workforce to operate the trains, and build acceptable passenger stations (perhaps the most expensive part; could still be done fairly cheaply, you don't need a Moynihan everywhere). The funding would need to be consistent... like it is for literally all roads and highways. But it is totally possible.
Lately, Amtrak has been talking about reactivating service on a lot of dormant lines like these. It takes time and effort from stakeholders, but most of the time they aren't building new greenfield lines, they're utilizing existing ones. Many states, notably Virginia and North Carolina in the case of the S-Line, are also spearheading their own rail revitalization projects using abandoned freight rights of way.
What is this "comment box" thing you're doing?
Comment box
Someone else had the same question a few weeks ago. I answered them here.
This is also already happening today, just with roads.
Yes, but the human tendency to prefer the devil you know makes that not such a big deal for people. Most people will not gawk at spending 10 billion to expand a major arterial in their city. They'll protest and scream about 1 billion for a light rail line all day though.
Good news is that's already a standard feature of cars. A car more than 10 years old is generally worth less than $10,000, from a new-purchase price over $40,000. More than 20 and it'll be worth exactly what the scrap dealer will pay for it by weight.
Comment box
Policy is necessary to discourage housing in places that are extremely inefficient to access. Such housing is bad for the environment and expensive for government to maintain access to. Market forces can react to such policy, which may be in the form of taxation, to "naturally" (lol) change where new homeowners choose to settle over some decades. I don't think this is a simple issue though.
Passenger rail is an essential part of a functional suburban transportation network for reasons I've discussed elsewhere, but just as important are buses. Though they are a form of rubber-tire automobile with all associated externalities (space, pollution, noise, etc.), they have the benefits of being:
Rail is valuable because trains have significantly higher capacity than even articulated buses. This is why New York City's upcoming Interborough Express in suburban parts of the city (like 70th St & 66th Pl in Queens, or 40th St & Ave H in Brooklyn... neither are "urban" despite being in NYC city limits) will be rail. Many suburban places could feasibly support such services. But in suburban areas which don't necessarily have the population density to support light or heavy rail outside of a few corridors, buses running fixed routes are a legitimate and fairly safe way to get people around.
Street-running light rail has some of the same benefits as buses insofar as it is still a professionally operated service and does not require expensive grade separation, instead utilizing existing ROW. But it requires somewhat more infrastructure. And when you consider things like "trackless trolleys" (rubber-tire buses running on electrified overhead wiring) and battery-powered light rail, the distinction starts to blur a little.
There are a lot of streets/roads in the US where private vehicle use should be restricted nearly entirely in favor of buses. There are many other streets/roads where private vehicle use should be prohibitively slow (for safety reasons) to the point that taking a bus or train is always the faster option. Obviously, bus or train service also needs to be frequent enough to get people where they need to go in an efficient manner.
Add more buses. Make their times more reliable. I would definitely take the bus more if there were more/better routes and there were more buses so I didn't have to plan my time four hours in advance if I wanted to get somewhere slightly off the bus route.
I went into this with a pretty open mind. Elon is a grandiose idiot. I think FSD is essentially a scam. I've owned my Model 3 for almost 5 years now, clocked in 60K miles of which around 20K are with autopilot (almost entirely on interstates).
I did not find this video particularly convincing and the creator did not do his due diligence. At 15:00, he mentions 150 million miles driven with FSD that Elon announced in an investor call of April 2023, and then compares it to 736 crashes of which 17 involved fatalities to demonstrate a rate of 11.3 deaths per 100 million miles to show that Teslas and FSD isn't actually that safe. Except he mixed up autopilot accidents with FSD (originally reported by WaPo here). Every Tesla has autopilot. Only like 400K have the full FSD beta. I have no idea how many miles have been driven with autopilot but it's a lot more than FSD, so combining both of those to get this death rate both wrong and deceptive, especially since this is one of his "data" linchpins for "Teslas are more dangerous, actually" points.
I think it will be quite some time before we have real, legitimate level 4+ self driving cars. Like with many software solutions, a 70% solution was quick to create and works "well enough" for a lot of scenarios, but that last 30% is very, very difficult and people won't accept a self driving car that doesn't perform better than they can by pretty much every metric in every scenario. Technology will get there eventually, so declaring cars will never drive us around is very likely wrong. I don't know what a good solution to this kind of behavior is and I more or less agree with the conclusion of his video, that this tech isn't ready and probably shouldn't be tested on public roads.
I like autopilot and thinks it works well for what it is when applied to the right situations. I only use it for stop and go traffic on any road, and interstate driving when conditions are pretty clear like light to no rain, day and night, which turns out to be the vast majority of my interstate driving.
Yep. I wish Elon would be removed by the board as I think he does more harm to the company with his public idiocy these days. However my Tesla FSD does a really good job driving me places, and is getting better. That said, I am always alert because it's software and can make mistakes. My job is to make sure it doesn't, and that's made really clear by the UI. It's basically riding coach to a pretty good student driver. I think being against this is only a smart idea if you believe something societally catastrophic is going to happen, otherwise the tech will eventually get to L4 given enough time.
For some reason, I feel like it's exceedingly unlikely that the board that tried to pay him like $56B or something because he's so essential will want to remove him. Maybe there's some way he could get removed because of all his misleading comments, though.
Saw this a week or so ago, and think it's mostly a good breakdown of where things stand. Unfortunately I do think there's some flaws in how he presents his arguments, but at the end of the day his facts mostly line up.
There's a lot of pie in the sky absurdity around automated driving, but what has been especially disappointing is watching someone like Musk overhype a tech, which I get, and then no one seeming to care that they've completely missed deadline after deadline? Like that should maybe be a sign to manage expectations?
Comment box
Yes, my takeaway -- more than just the empirical data suggesting that self-driving Teslas are 10x more dangerous than human drivers -- is indeed that the hype around self-driving tech is wholly unrecognized to be a problem among a certain subset of the population. People just gobble up the false promises.
Jordan quotes some stats indicating the safety that L1 and L2 automation tech can provide when a driver is still expecting to control the vehicle, like lane assist tech, or automatic speed reductions when the vehicle detects something ahead. But this idea that Teslas can "drive themselves" and still see those safety benefits seems to lull people into a false sense of security. When they are no longer expecting to to control the vehicle "in general," those extant safety benefits are canceled out by the driver's inattention to the vehicle's erratic behavior.
I think a lot of people, including some very smart people I know, have fallen prey to a billionaire grifter's narrative that full automation tech is actually on-par with human driving in terms of safety, or that it's close enough to call "safe." It is not. Frankly I just don't think most people have considered how ridiculously complex streetscapes are. I think that many people have been conditioned to believe that technology (and nothing else) is the solution to literally every single problem, and that a more technological solution is basically always the better one. (Even Jordan is at least partially an example here: his suggestion about networked vehicles is a techy idea to a problem that does not consider... concrete. Or actually any part of road layout, traffic management, other than cars technologically (magically) talking to other cars.) Vague tech-hyping is a narrow perspective which clouds our judgment.
The reason this video stood out to me is that this technocratic perspective of the world is (in my experience) particularly pervasive among upper-middle class/professional class types: people who are educated (not in tech specifically, but in something), at least kind of well-off and likely to buy fancy new technology, and politically progressive, and who should know better given their education than to take someone like Elon Musk at his word. i.e. the market for Teslas. As you say, Tesla has repeatedly failed to deliver what Musk has claimed and yet the widespread acceptance of self-driving tech among people who can afford Teslas is weirdly high. At least some of them have also specifically invested in the stock and thus have a perverse incentive to hype it up, as Jordan says. I think there's a lot of criticism of Tesla and self-driving cars in general, but a lot of that is tied politically to electric vehicles (because Teslas are electric) and denial of climate change and thus I would consider that critique irrelevant in this discussion; and isn't rigorous or academic to begin with.
I think that a lot of less educated and younger and/or more impressionable people have also fallen prey to the grifting, especially people who see Musk as this brilliant billionaire who definitely pulled himself up by his bootstraps and didn't have parents who owned an emerald mine or anything. Many of these people are the exact OPPOSITE of the group I just described. It is weird that they have remained so steadfast for so long.
Another commenter mentioned that many people who work with software are skeptical of self-driving tech, which I think is true for some such people I know. But there is a weirdly passive attitude, across the board, toward this unproven and evidently dangerous technology. The passive attitude is especially weird when you think about how terrified some of these same people are of, like, flying (which is at least an order of magnitude safer than driving on a per-mile basis). Just look at how much press Boeing is getting.
As someone who is very interested in urban planning and public transportation, it is troublesome to hear people talking about "the end of transit" as though self-driving vehicles can somehow replace traditional transit, e.g. robotaxis somehow being more efficient than trains or buses. This kind of attitude discourages investment in public transportation infrastructure (as well as infrastructure for pedestrians/anyone not in an automobile) because it implicitly reinforces car-centric attitudes about the way society should be designed. No amount of car automation will ever (and I mean, literally, EVER) solve the majority of traffic-related problems experienced by cities; nor will it magically solve income inequality; nor do I imagine it will truly and wholly stop all traffic fatalities; etc etc. Because automated cars are still cars, they still have most of the externalities of cars. Hype around self-driving tech forgets that a world designed primarily for cars rather than people is uniquely bad for the environment, loud and hazardous to health (in ways beyond being physically struck by a vehicle), and expensive to maintain relative to its social benefits.
One of the interesting things to me that has often come up in these automation debates is planes.
They're often used as an example of why it should be possible, when really they are a clear example of why it likely isn't.
Planes are in almost 100% known environments with miles of space between them. When taking off or landing EVERYTHING is supposed to be known and scheduled by a team of humans. ONCE all that is done, yes the mechanical actions of taking off/landing are mostly automated, even to some advantage in situations where it's instruments only.
However, if you're the kind of person who reads the aviation accident investigations and the discussions around them, say about air france and their major crash, you'd know that there's a huge discussion of "too much information and too much automation" being unsafe. When you have 40 alerts going off, and they conflict with your instruments, you start conditioning yourself to ignore alerts.
Further, even with 2 people in the cockpit, there's an issue of complacency on long automated trips. "QUICK YOU HAVE 2 SECONDS TO MAKE A LIFE OR DEATH DECISION" after minutes or hours of nothing is something humans are notoriously bad at, and it's darkly hilarious how these car automation systems say "well in an emergency or situation it doesn't understand, it'll alert the driver and they'll have X seconds to respond", as if somehow the vast majority of the population is going to do better than 2 trained pilots. And that's ignoring that in all likely hood they'll have waaaay less than X to do anything because unlike planes you don't have miles of space to react in.
At the end of the day I think one of the most interesting points brought up in the video is that the best automated system probably doesn't try to mesh with human needs, but at that point we're back to "Oh shit it's trains isn't it", or some similar issue.
The training those pilots get is also on that specific model of airplane, and understanding how all of those automated systems work on a deep level.
I'd be willing to bet that you could ask a pilot to draw you a high level system architecture of how the autopilot system on their assigned aircraft works, and they'd be able to do it to a fairly accurate level.
In cars though, the whole point of these automated driving systems is that the drivers are explicitly not supposed to know how they work. Auto manufacturers go to great lengths to not only market the fact that their cars are basically just magical boxes that take care of all the hard stuff for you, their self driving systems are probably the most viciously guarded piece of intellectual property that they own at this point.
Yeah, this bugs me. I do think that automated transportation is possible, even on a personal level, but trying to get it to work on public roads alongside humans seems to be incredibly reckless. Heck, when it comes to public transport, there are tons of fully automated systems in service right now! The difference is that those were, for the most part, built in tandem with the infrastructure to support them. In contrast, seeing companies trying to sell autonomous cars that work on existing infrastructure feels like a scam because it’s basically a pipe dream. It’s no better than what Elizabeth Holmes was doing with Theranos.
Video games can help teach in this vein.
Fortnite matches typically run 30ish minutes beginning to end. If you drop far away from everybody else, there's a decent chance you encounter nobody for 25 minutes and then are steamrolled by the final 5.
If you drop at the beginning, and continually chase gunfire, you'll play better thanks to that constant stream of adrenaline.
This is most apparent in group play, where a bit of coordination and attentiveness is required.
I'm certain I've seen people unironically use this logic to explain how speeding (as in 20 over) helps them focus and drive safer in the wild.
That's just a stunning lack of understanding of the physics at play.
I have driven multiple cars with lane assist.
They all do fine and dandy when conditions are calm and the roads are well-painted and relatively flat.
Introduce hills, faded overlapping lines, variable lane widths, 30+ mph winds, and especially any combo of those...things get wonky real fast.
I think you've got the nail on the head here. Ill admit that when I was a young nerd, I often fell prey to that mindset as well. With every problem, I often thought that there was always a technological solution, and that if only those luddites got onboard, we could computerize and automate all of our problems away.
After a long career in technology, and especially after I made the transition to cybersecurity, I've realized that in most cases, the opposite is actually true.
Simpler solutions are better than complex ones, all other things being equal. Any increase in complexity in a system needs to be carefully considered with it's drawbacks, and that complexity needs to be defended and justified.
Complexity for complexity's sake, or to look more "modern", or to save some engineer a few hours designing a more elegant, simple solution is endemic to most of the consumer landscape unfortunately. With each increase in complexity though, the points of potential failure are multiplied, potential security weaknesses are multiplied, and human's mental model of the system is hampered, making troubleshooting and evaluating much more costly and less effective.
There are a lot of great advances in technology in automobiles. Fuel injection is incredible. Carbauration was always a clunky, terrible solution for fuel mixing. Various sensors around cars can make diagnoses much easier.
But there are antipatterns there too. Infotainment systems with custom touchscreen OSes are buggy and prone to crashes on every car I've ever been in. Things like movable headlights are pointless show offs that don't do much. My car has electronic suspension that is supposed to dynamically stiffen or loosen based on how I'm driving. Do I ever notice it? Absolutely not. Will it help me avoid a crash ever? I very much doubt it. Does it require probably thousands of dollars of sensors, computers, and actuators? I would be willing to bet it does.
Peak infotainment system was when cars included an aux-in port built-in, thus rendering them futureproof against any future audio technology. (half joking)
By movable headlights, are you talking about the popup novelties on a Miata or the ones that angle their beams to follow steering?
My suspicion is that the passive attitude from those who should know better stems from their knowledge of the atrocious safety record of human drivers. They (incorrectly) assume that it should be relatively easy to beat the dogshit capabilities of the average human at operating heavy machinery.
Thanks for sharing this, it's an informative video with a lot of facts to back up my gut apprehension about autonomous vehicles. The more I understand about how these technologies work, the less I trust them. I say that as a software engineer by trade, and my less geeky friends and family have raised their eyebrows when I've mentioned my hesitancy to them. They expected me, the "tech guy," to be all-in on self-driving vehicles, but I don't want anything to do with the things. Anecdotally it seems like it's the less technically inclined folks who are more trusting of it, which doesn't make a lot of sense to me.
The other day I was driving through a construction zone in a pedestrian-heavy area with speed bumps and crosswalks. There were traffic cones indicating a lane switch into the oncoming lane, signaled by a flagger. Straight ahead through the cones was a giant gaping hole. Honestly as I approached it I could barely sort out where I was supposed to go and when. There's not a computer system on the planet that could navigate it correctly. Best case scenario, the car just stops in the middle of the road. Worst case, you end up in the hole (along with the flagger your car dragged beneath it).
I visited Maui last year and drove the Road to Hana. It's a stunningly beautiful drive. It also demanded my full attention, with its blind hairpin curves, dozens of single-lane bridges, and frequent occurrences of tourists and their cars wherever they felt like parking on the shoulder for some sightseeing. The bridges were the craziest part, often I had to engage in some nonverbal communication with oncoming cars just to negotiate who went when. At one point I had to quickly throw the car in reverse to avoid a collision. And I haven't even mentioned the locals who drive way too fast (and the expected courtesy of slower drivers pulling off the road when they come up behind so they can pass), or the oversized oncoming work trucks crossing the center line around blind corners.
Anyway, I'll have confidence in self-driving cars when they can seamlessly handle both of the above situations without killing anyone (or giving up and just stopping the vehicle in the middle of the street). If the tech ever matures to that point, I don't expect it to happen for several decades at least.
If you'll indulge me for a moment, I want to talk about alternative approaches to vehicular autonomy. Things like some sort of mesh network all cars participate in or the swarming behavior shown in the video are really interesting but they're impossible for one reason: uncontrollable variables in a dynamic environment. You can get all the new cars to play nicely together, but they still have to share the space with the old cars that don't. (The designated lanes idea in the video would help but not solve this fully.) You have non-vehicles that enter the roadway unpredictably — wildlife, pedestrians, construction, rogue soccer balls, whatever. You can't plan for every contingency.
At the same time, we still don't have flying cars. Maybe this is their moment. What if when they finally arrived, they all came equipped with a standardized self-driving system that communicates with neighboring vehicles? The biggest old argument I heard against them was the danger a bad driver could pose driving drunk at cruising altitude or whatever. But if every single one of them is autonomous from day one? Maybe not so bad. With no manually driven cars sharing the airspace, and of course no pedestrians or construction to contend with, maybe no real hazards apart from birds... I dunno, I think it's compelling. It'd be a hell of a lot faster to get from A to B, I know that much at least. If Not Just Bikes has taught me anything, it's that roads suck... I'm kinda ready to just be done with them altogether.
I don't mean to be overly negative in my response, but I want to point out a few things. There are plenty of other hazards to aircraft than just birds. Many are weather related: lightning, icing (interferes with aerodynamic properties of lift surfaces), downdrafts, etc. But you also have dealing with mechanical and electrical failures during flight.
Aircraft systems have an very high amount of redundancy -- typically three of everything (you can look up Triple Modular Redundancy if you want to know more). This is necessary because you can't merely "pull over" -- you have to keep flying. The 777 was the first fully by-wire aircraft (electronic actuation, rather an mechanical/hydraulic) and even the computer networks are tripled up and routed through different parts of the fuselage so that an intrusion is unlikely to sever all three. The weight cost of this is not trivial and becomes much more significant as you scale down the size of the aircraft.
Incidentally, the redundancy question is one that I haven't seen satisfactory answers for from the autonomous car industry yet either, but it would be a problem for them too, eventually. Doubling up on sensors means adding more of the most expensive parts, plus there are interference problems with redundant LIDAR and RADAR. They haven't been bitten by this because the technology had not achieved a level of automation where it can be deployed at scale, which is where those long tail failures start to show up.
In addition to redundant design, current aviation is workable because you have two (or more) highly trained, professional pilots who are further certified on the particular aircraft they are flying who can work with ground control (and sometimes even the aircraft engineers) to resolve issues with the plane, route around bad weather, etc. The problems that show up if one tries to delegate handling all these unexpected cases to a computer would be the same for aviation as they have been for the current group of autonomous vehicles. There are too many weird situations to effectively plan for them all with our current methods. Trying to rely on pilots at the same level of capability as car drivers would be unworkable for a variety of reasons: too little training, too little responsibility for things going wrong, too much variation in physical and mental ability, no redundancy.
Maintenance is another difficulty. Ask anyone who owns a private plane, and you'll find out that the parts and the mechanics have to be specially certified, which is very expensive. There are reasons that private jets are so expensive to operate.
Even though the sky is pretty empty, that's mainly because there aren't that many planes, and they are all going to a relatively small number of destinations. A "near miss" for aviation means coming within a few hundred feet to another aircraft, and even at that distance, it is far from a solved problem. Most of the problems happen at the airport, and that environment requires close manual monitoring by highly trained humans to avoid collisions and manage traffic. If the number of incoming aircraft at airports were, say, 10x number of smaller vehicles, even the current system would be intractable, both in collision avoidance and in runway availability. If one starts having to manage collisions and near misses for far more vehicles in airspace above population centers, it would be even more challenging, and the failures would be high profile and horrific.
All that is to say, there is no free lunch when it comes to autonomy in any domain. The world is complicated, humans are fallible, and each domain has its own difficult challenges.
Meanwhile, at Boeing: What if we added a computer that can pitch the nose of the plane based on information from a single sensor, and didn't tell the pilots how it works?
Shit like this still blows my mind. How it was ever allowed in the air relying on a single sensor is just unacceptable beyond belief and the fact that more heads didn't roll for that alone shows how toothless the FAA is becoming.
It's right up there with aribus' insane 'oh if both pilots put in different inputs on the stick, then it'll just accept the first one and ignore the second WITH NO FEEDBACK' that lead to another 200 deaths.
That Airbus crash was the source of my most frustrating ever online discussion. Someone claiming to be an engineer kept insisting that it simply wasn’t possible to have a switch in the cockpit that toggled control between the pilot and copilot, and that the current system where the two pilots agree between themselves as to which one is actually flying the airplane at any given time was the only way it could be. I find this extremely hard to believe.
Comment box
To me, this is the root of the issue. Anytime technology is being deployed to solve a novel problem (which will be the case until the end of time), and anytime the strategic decisions of that deployment are being made by fallible human beings (also a constant), there are bound to be misapplications of the technology due to misalignment between humans (who necessarily have a subjective perspective), our subjective understanding of the environment, and our subjective understanding of the tech.
This is especially the case when those fallible human beings are 1) not engineers or otherwise technical people, and 2) incentivized to release technology and not to prioritize safety. That can theoretically be solved, but practically speaking, every company or government agency will always contain non-technical perspectives and motivations (valid or not). Safety checklists for maintenance and good software QA don't mean much if executives, governments, or other forces are creating a company culture that precludes those checks.
Probably some amount of additional regulation can solve some number of these issues, but it has its limits. Some domains are narrow enough to deploy automated systems with a realistic amount of confidence, but car automation on public streets is probably not one of them given the inherently high number of potential human-computer misalignments in such a complex environment.
Yes, well that is a problem too.
I think more than anything, the Boeing debacle(s) show how precarious safety critical infrastructure is without strong regulatory oversight. I wish the lesson learned from this would be the walking back of regulatory capture at the FAA and the establishment of similar oversight in various tech industries, but, as the man says, "if wishes were fishes, we'd all cast nets."
I agree that self-driving vehicles that purely rely on sensors (eg visual and radar) are unlikely to reach the safety level of human driving. Waymo does pretty well, but it's definitely not perfect and requires an intense amount of programming specific to its fenced-in area in order to be effective. Full, safe autonomy will likely require vehicles and infrastructure to be able to communicate their intentions to each other, and that will only be effective if it's everywhere. That level of infrastructure probably won't exist in our lifetime, but I wouldn't say "never". It's a shame, because self-driving cars are one of the few transportation technology improvements that I'd actually be excited about.
Comment box
"Never" is a strong word, yeah. I am continually surprised by the way the world changes. But communication between multiple self-driving vehicles doesn't inherently address the safety of people who aren't in vehicles. If the car is running pedestrians over now, being able to communicate with another car doesn't change that.
Communication between autonomous vehicles and fixed infrastructure is interesting and has some capacity to improve safety. For example, there already exist sensors that detect pedestrians, bikes, or cars in a given space and adjust traffic light signals accordingly. These systems are common in the Netherlands and some other European countries but nonexistent in most of North America. In theory, these systems could communicate with cars to inform the cars that the sensor has detected, say, a pedestrian who might cross the street (which maybe the car can't see). That could be useful for the automobile, which can't otherwise intuit or anticipate human behavior.
But like you say, it would require a huge amount of quite expensive fixed infrastructure to be effective, which costs a lot to maintain. That cost alone makes this unfeasible in most municipalities, many of whom are already going bankrupt trying to maintain a road network whose maintenance costs they vastly underestimated. The benefits would also be absent anywhere this infrastructure has not been installed. The problem is that it's very hard to design a computerized system to pre-emptively detect every aspect of human behavior; because a human has to program it to predict humans, and it has to work in context. For instance, pedestrians often do not use designated crosswalks in a given location because it is inconvenient, dangerous, or unpleasant to do so and instead cross streets at locations where a traffic sensor wouldn't have been placed by an engineer, and therefore wouldn't detect them. In that case, the system structurally breaks down because it assumes that the car has information that it doesn't have (because it isn't receiving information from the sensors, because the sensors reflect idealized data and not real-world data).
Additionally, the more we rely on software communicating with other software (not even made by the same company) to make decisions about safety, the more we are relying on extremely thorough software engineering QA to keep us safe. There are lots of people much smarter than me working on solutions to these problems, but I have never met a software engineer who consistently writes mathematical proofs demonstrating the correctness of their algorithms. Outside of academia, it does not happen. I don't think software needs to be literally perfect, but a single bad deployment rolled out to hundreds of thousands or millions of vehicles or other detection systems could conceivably kill or seriously injure a lot of people.
Maybe I am just old-fashioned, but I would rather the approach taken to safety emphasizes the removal of vehicles from most public spaces; an extraordinary reduction in vehicle speed anywhere pedestrians, cyclists, and other Vulnerable Road Users are expected to be present; and a preference for concrete and other physical protections in lieu of computer magic.
The communication wouldn't be limited to other cars. It would also need to talk to infrastructure like traffic signals, which pedestrians would also need to obey of obvious safety reasons. It would also supplement, and not replace proximity sensors.
For infra-based communication (not car to car), it would be a matter of retrofitting traffic and pedestrian signals with a signal emitter that tells vehicles what the state of the intersection is. This would be something that could be phased in over time as traffic signals and their controls are replaced.
I agree that there needs to be better pedestrian infrastructure, but at the end of the day, this problem comes down to a human behavior problem and pedestrians would need to consider their safety by not illegally jaywalking. Much like trying to cross train tracks, it's pretty much your own damn fault if you perform an illegal crossing and get hit by something that has the right of way.
I don't necessarily disagree with this in urban settings, but I'll also admit some bias here in favor of the existence personal vehicles since I have no interest in living in a city and walking everywhere.
Comment box
This is true but I think you are not considering the ridiculous lack of sidewalks across the United States. Nor are you considering the ridiculous number of ADA code violations committed by private contractors/construction companies, police officers (such as illegal vehicle parking), etc. on a regular basis. I walk by several such violations every single day. There are many situations where it is physically impossible for a pedestrian not to enter the road in a "disallowed" place. Many such pedestrians are wheelchair users, who are particularly vulnerable. Many VRUs are also disabled in some way (visual, auditory, sensory, overall mental capacity, etc).
Obviously, the correct solution is to fine violators of the ADA to an extreme extent to ensure that this does not happen. But that does not solve the problem for the wheelchair user who has to enter the street in a "disallowed" location before the ADA violation has been addressed by government. They still have to go into the street in a way that is unpredictable for an automated system.
In many cases a tree branch or other obstacle will also prevent Vulnerable Road Users from using sidewalks or bike lanes "as intended." This is objectively not their fault.
The complexity of streets means that an automated car will always be in a novel and untested situation with pedestrians sometimes. This means they will always be putting VRUs at risk. It isn't acceptable to blame pedestrians for being "at fault" for technically breaking a rule when the fault is obviously with the systems that require them to do so.
There was recently a story here, or on Reddit, in which a wheelchair user received a traffic citation for rolling in the road. The police officer did not understand that the lack of sidewalks made this violation physically impossible to avoid. Situations like this are unbelievably common, but they are invisible to drivers, because drivers are not Vulnerable Road Users (by definition). Drivers are always given purpose-built infrastructure. VRUs are not.
Regardless, even if a pedestrian is in the wrong for crossing a street at the wrong time, they do not deserve to die. It should be physically impossible for a car to go fast enough on a road to kill a pedestrian who steps into the wrong place of the road at the wrong time, whether that was on purpose of by accident. Many drivers have a callous attitude toward pedestrians in this regard which I feel is unhelpful and a bit cruel.
I don't think you're trying to be callous/cruel here, but a lot of otherwise well-meaning people are indifferent toward the sanctity of human life when they arbitrarily decide that someone was "in the wrong." This is a very narrow-minded way of looking at safety because it assumes 1) that everyone has the same objective definitions of right and wrong behavior, 2) that everyone is making a conscious decision to break the rules, and 3) that the rules are even optimized to begin with. These assumptions make it easy to assign moral judgment about a person (or make actuarial insurance decisions), but are unhelpful when the goal is to save ALL lives.
I meant full-sized automobiles specifically. IMO there is nothing wrong with a personal vehicle that happens to be an electric bicycle or a speed-limited golf cart (like in Peachtree City, GA). I don't think full-sized cars should be allowed within city limits, with some but quite few exceptions. Obviously, access should be maintained for emergency vehicles, but the general population should not be driving around two-ton cars with the laughable amount of training they have.
Outside of cities, I have the same opinion for residential areas. I just do not think you should be allowed to drive a full-sized vehicle in a place where children might be expected to play. They are kids and they do not act rationally. They don't go outside anymore because parents know (whether they admit this is the main reason or not) that they're afraid of traffic killing their child. Empirically it's rather rare to kill someone, even a child, with a golf cart that goes 10mph. But it happens constantly with full-sized automobiles simply on account of their obscene mass.
Personally I see little issue with building parking garages near arterial roads at the edge of town and using that to store everyone's full-sized automobiles, and using your legs, wheelchairs, bicycles, electric bicycles, and golf carts (they can be enclosed, I know it gets cold some places, yes yes yes) to get everywhere locally. Because those local places are where PEOPLE are likely to be, it makes sense to design spaces for people... and not for massive hunks of steel.
The exact implementation of the principles I'm espousing here is obviously going to differ from place to place, but the point is that there should be dramatically more pedestrianized zones all over the United States, as well as dramatically more half-pedestrianized zones where you can drive a little golf cart but not a massive SUV. However that needs to happen -- and it wouldn't be as hard as we think -- it ought to be done.
I'm not going to watch this, but does it say anything about Waymo? I think each company needs to be looked at separately.
Comment box
Jordan refers to Waymo at around 3:00 but does not elaborate on their technology. He states that "in my opinion, no legitimate [automation] stage 4 vehicle exists on public roads," in reference to claims from companies like Waymo which claim to have reached this level of automation.
The video is mostly about Tesla and Openpilot, both of which he tests anecdotally; and especially Tesla, which he criticizes extensively.