Federal official are investigating GM's Cruise unit over complaints that their self-driving cars are annoying because they've been designed to be super safe. This is catnip for Atrios:
I think Sam was the first person who really made this crystal clear, that even if autonomous cars sorta "work" they'll be unpleasant and people won't really want to use them and they'll be annoying to passengers and other vehicles. These aren't really "incidents" so much as "baked into how they function." This is how they make them "safe!"
I have my fun arguing with Atrios over this, but seriously, what's the issue here? The first cars traveled ten miles an hour and broke down all the time (and scared the horses). The first computer had to be programmed with switches and could barely calculate a simple ballistic. The first mobile phones were the size of a brick. The Gutenberg Bible took five years to produce and cost the equivalent of about $6,000. The first radios were drenched in static and there was nothing to listen to anyway.
So of course the first self-driving cars are annoying and limited. Everything is annoying and limited at first. But what's the reason for thinking that autonomous driving is the one special thing that will stay that way forever? I continue to stick with my guess that 2025 is a reasonable estimate for when self driving will be truly convenient and useful, but if it takes a few years longer, who cares? There's no reason not to think that, one way or another, the technology will get to human level (or better) within a few years.
If not, why not?
The question is how will self-driving cars handle a snowstorm covering the lines on the road, heavy rainstorms, and thick fog? Are autonomous vehicles viable if they work 95% of the time? 90%?
I drove a rental (a Honda, I think) that had rudimentary lane assist and would automatically brake and speed up relative to the vehicle in front of me, like a smarter version of cruise control. But there was no doubt I was doing the driving.
How well do humans handle those conditions?
My experience from living in a place where heavy rain, heavy snow, and dense fog are all fairly common., is that most human drivers handle those conditions quite well as long as they stay in practice. Such conditions are MUCH more of a problem in places where they are less common but still happen once in a while, because the drivers aren't used to them.
Heavy snow is fairly easy when the road runs through forested terrain, because the trees tell you roughly where the edge of the road is. It's much more of a problem in flat open country because simply trying to figure out where the edge of the road is without driving off it is a challenge. I don't think that an autonomous vehicle will be able to handle such conditions just using cameras any time soon, but with the addition of sonar and radar sensors autonomous vehicles should be able to actually do better than humans in heavy snow.
I’ve profited $17,000 in just four weeks by working from home comfortably part-time. I was devastated when I lost my previous business right away, but happily, I found this project, which has allowed me to get thousands of dollars from the comfort cfs06 of my home. Each person may definitely complete this simple task and earn extra money online by
visiting the next article———>>> https://brightfuture241.blogspot.com/
As someone living in a midwest blizzard for the past 3 days and who has moreover been driving around... you are right. Unfortunately it takes a couple blizzards to warm up. Lucky for us this like our 3rd real one.
The systems are already useful, at least from what I see in my Tesla beta testing the FSD software.
For starters, Tesla's highway system makes longer drives an absolute pleasure, far better than any cruise control and I would bet SuperCruise is the same.
I would say 2025 might be about right for Tesla, looking at how far they got from 2019 to 2022, and just extrapolating for three more years.
Its a software problem, to convert visual images to software than then program the car how to react. But its already solved from what I can see.
Except for a broken down horseless carriage stuck in the middle of the road, none of your examples involve actions or devices which slow down what third parties are trying to do. Self driving cars block the fucking road, the way they are programmed, causing misery for those stuck behind them.
Cars make things a lot harder for pedestrians. And over time they've made getting around on a horse A LOT harder than it used to be...
It'll happen a whole lot faster once the manufacturers get that liability exemption, the one the gun makers got. Time to spread more of that sweet campaign cash around!
Srsly, driving a car safely and efficiently is relatively easy in some circumstances, extremely complex and experience-dependent in others. That second kind of circumstance presents a very wide (but probably ultimately finite) range of inputs, some of which can call for contradictory actions, and we humans compensate for the ambiguity and the inputs we don't pick up using whatever degree of experience and judgment we have. Often we blow it, but also we often don't, and everybody knows at least one of those stories.
Wide-open stretches of mostly dry road could probably be run successfully almost all the time by AI. Gravel switchbacks on snowy or rainy mountainsides, maybe not so much. Or as hexcalibur says, snowstorms.
If self-driving vehicles need dedicated lanes or roadways, that's basically a more wasteful railroad, so the degree of success hexcalibur asks about is an important point. How about this-- when AI gets to the point where you'll let your youngest child cross streets of mostly self-driving cars? When you'll take a self-driving vehicle from one major city center to another that's 200 miles away?
Ultimately, it's probably going to happen. Personally though I don't see it before the end of the decade, and in more limited ways than people are talking about. Even if Elmo likes to pretend there's no such thing, liability is real for both financial and moral reasons and it will keep development slow. Before the Civil War we learned a lot about how to engineer steam power on riverboats, but the explosions killed and maimed tons of people. Liability standards were much much looser then.
I agree with you completely. I'd guess self-driving comes around 2030-2035, with widespread adoption by the 2040s. That's a guess, and I could be wrong, but 2025 seems too close to working out the niggling issues. In CS (and most things) solving 90% of the problem is usually much easier than solving 100% of the problem.
The gun-makers exemption is an exemption from being sued because their product works as designed, Gun makers are not exempt from product liability actions if their product malfunctions; for instance, if a rifle is made with a weak breach that blows up in the shooter's face, the shooter can sue.
I would expect the same thing to apply to autonomous vehicles; no manufacturer liability if the product works as designed. Autonomous vehicles will be a success not when you would let your child cross a street with self-driving cars on it, but when you can pack your child into the car and trust it to get him to Grandma's house 200 miles away.
I like the Grandma's house test, but I think the street-crossing one is also needed. And if I understand your explanation, it's relevant-- a car that isn't designed to stop moving when a human appears in front of it is mal-designed. Or creature, depending on circumstances.
Collision avoidance, whether with other vehicles, pedestrians , animals, or fixed objects, is really basic. I would expect that's number one on the programmer's list; you really can't let a "self-driving" vehicle move if it can't do that one fairly well.
One problem is that manufacturers are advertising them as self driving cars now, They are only self driving in a very limited set of cases, and they frequently require a human operator to intervene, so someone needs to be alert and on standby at all times. This is simple fraud. The phrase "self driving" implies autonomous operation without the need for a human driver. We don't need backup operators in automatic elevators.
When automobiles were first developed, someone had to be out in front with a flag to warn people the way we handle a wide load today. There were all sorts of other restrictions in addition. Some states required an engineer on board. Since soi disant self driving cars can cause problems for other drivers, perhaps they need some way of warning human drivers that they need to take special precautions.
I wouldn't play down the problem. There is a big gap between what a human driver with modest driving skill can do and what the most advanced self driving vehicle can do.
In time...
But it has been over-promised for a long time.
It may not truly take off until the infrastructure is in place. We'll have to go from each car being completely autonomous to them using what will be built into roads plus car to car communication.
golack
Electrify the roads. Run a cable under the road. Each vehicle would then be travelling through the magnetic field that is created when electrical power flows through a line. This could serve 2 purposes. First it provides for constant contact between the vehicle and the magnetic field making communication constant. Second the magnetic waves could also assist in powering the electric vehicles - not a full charge of course but it could improve the range of the vehicle. The problem of course is keeping the power on ! In Texas this won't work............
A solution that requires digging up literally every road in the nation is not a solution. At most, it is a solution for affluent areas.
O/T, but this dude reports the Fed is actually growing more pessimistic on inflation:
https://theovershoot.co/p/the-fed-is-getting-less-sanguine
Because too many people have realized they can just raise their price and people will have to pay it and no one can actually do anything to stop them?
And, also, China might crash and burn.
The first workable steam engines were so inefficient that the only economically viable use for them was pumping water out of coal mines, where the fuel was literally RIGHT THERE. The whole industrial revolution started with one niche use of one almost-useless technology.
Sorry, but isn't that kind of upside-down? Condensing steam engines being specifically developed as an improved way to operate existing water removal systems in mines, then with further developments that improved efficiency, the tech could be adapted to other uses? IOW, was it really a case of having a technology floating around that people had big ideas for but it just wasn't efficient enough?
If the point is that initial technologies generally haven't been very efficient compared to later iterations, absolutely. And the real breakthrough in this case has to have been producing rotary motion rather than the vertical pumping motion that the mine water problem needed. That let steam power replace water power and also let it become mobile power.
I'll admit that there's a difference here. Nobody at the time seems to have predicted how transformative steam engines would eventually be. And if you'd told somebody back then that the funny device pumping water out of the coal mine would eventually give rise to machines that would lead to mass affluence previously undreamed of, they almost certainly would have thought you were nuts. The Industrial Revolution was unforeseen, and might have never happened at all, if not for the fact that 18th century Britain was so deforested that they had to use coal instead of wood for heating fuel, and had to dig below the water table sometimes to get it.
Some technologies come from successive practical iterations that lead to unexpectedly useful results, others from a deliberate development process that starts with "we know it's theoretically possible, we just need to figure out the technical details of how to make it work". Self-driving cars are obviously the latter.
No one doubts it will eventually happen, but all your driving AI posts have an undercurrent of "anyone who disagrees is a luddite" so seeing your concrete prediction be wrong will be pleasant.
My personal belief is rather than make the cars smart the roads will be smart. For example on any major highway or interstate cars will be slaved to the road AI so the vehicles can go super fast and because there are relatively fewer unexpected events and it is a situation where all cars are 100% predictable it will be significantly safer with perhaps 1% of the current accidents.
Though those rare accidents will likely be catastrophically more intense for those involved.
Sounds like a train.
People who can drive can afford to scoff at this stuff, but this is a really big deal for people who can't drive for some reason. I have a non-obvious disability that has prevented me from getting a license, and I can tell you, being dependent on other people for transportation stinks. Even if you have family members who love you and are willing to help you when you need it, you still need to plan and coordinate with them every single time you want to go anywhere. And that's if you're lucky, as I thankfully am! Without reliable family drivers, you have to either pay through the nose for a cab, or rely on public transportation which is often infrequent service, requiring careful scheduling. Those who can drive don't have to devote much thought to how they're going to get places, those who can't have to devote a lot.
Self-driving cars would solve so many problems for me and people like me, and enormously improve our quality of life. I really, really hope they become reality.
(Not sure how my comment ended up here, I didn't intend to post it as a reply.)
Thanks for bringing up this scenario!
Frankly, I could use a truly self-driving car now. And I'd put up with a lot of inconveniences, such as excessive waits at stop signs and whatever. Because the alternatives would be taking the bus, which would turn a 15 minute errand into over two hours, mostly of waiting, or a $60 Uber/Lyft "sharing". Or begging a friend, relative, or spous for a lift. There AND back, of course...
Agreed. Also, it would be a game changer for older people who can no longer drive but still want to go somewhere. And also for parents, who won't have to drive their kids to whatever practice innumerable times --- just program the destination in the car and they can go themselves.
An unintended consequence of lots of driverless cars though might be a significant increase in miles driven, road congestion, and carbon emissions. Also unemployment for a lot of folks who drive for a living.
Except better because you can easily choose a different destination with multiple routes in case the path is blocked.
the vehicles can go super fast and...with perhaps 1% of the current accidents. Though those rare accidents will likely be catastrophically more intense for those involved.
Over 6 million traffic accidents occur annually in the US. You really think the public will accept 60,000 "catastrophically intense" collisions caused by driverless vehicles?
That'll never happen.
I do think fully driverless technology will be perfected, but the emphasis will be on safety, which is why there's no way the cars will go "super fast" (nor much need for them to do so. What's wrong with 60 or 70 mph?).
Kevin claims that if automated driving cars don't work out eventually (long time frame, lol), this will be the first time in history that something hasn't worked out.
This strikes me as.....not true in the slightest.
But obviously, if you assume that everything in history has always worked out well in the end, it's hard to imagine something failing.
Speaking of things that never panned out in the automotive line, the reciprocating gasoline or diesel engine is apparently a very complex and inefficient things with pistons moving back and forth and all sorts of friction bearings. Turbine engines would seem to be more efficient, and steam turbines replaced reciprocating steam engines in ships*, but turbines never worked out for land vehicles, nor rotary (Wankel) gas engines. There are material problems with these things that have never been worked out. The old systems turned out to be more efficient after all under real conditions.
*Of course modern cruise ships use a diesel reciprocating engine to drive a generator, which then powers the electric motors which drive the propellers. Locomotives are similar.
Steam engines, gas turbines and internal combustion engines are all heat engines. The laws of thermodynamics tell us the maximum possible efficiency of a heat engine is governed by the difference between the hot reservoir (the combustion chamber or the boiler) and the cold reservoir (the outside environment). The maximum working temperature of steam turbines is limited by the tendency of the working fluid (water) to dissociate into hydrogen and oxygen at higher temperatures. The resultant hydrogen results in hydrogen embrittlement of the metal working parts. In terms of efficiency, gas turbines do better as they can run at higher temperatures. Piston engines, particularly diesels, do better yet as the maximum temperature is cyclic rather than continuous as in a turbine, so that the working parts are cooler than the maximum temperature of the working fluid, allowing higher peak temperatures of the working fluid.
With a few limited exceptions, such as the Abrams tank, gas turbines have not proved to be the best solution for land vehicles. That's not the same thing as "gas turbines have not worked out". They are used extensively in aircraft, and in limited marine applications.
The technical problem is clearly a very hard one, otherwise they'd have figured it out years ago, but just because they haven't solved it yet doesn't mean they never will. AI is improving; ChatGPT has an IQ of 83, and I'm pretty sure humans with IQs in that range can and do get driver's licenses. Also, unlike certain other things that are "theoretically possible" like time travel or bending spacetime to move faster than light, it doesn't require any really stupendous breakthroughs to achieve; just better iterations of technologies that already exist. I won't pretend to know how much longer it will take, but I am reasonably confident that self-driving cars will be available to consumers in my lifetime.
2025? Right, sure. And flying cars by 2030!
Seriously, what's even the need or desire this fills?
See shapeofsociety's comment, above.
Just off the top of my head...
1) Ideally it would be safer than human drivers. Lots of people die from car accidents, decreasing that would be great.
2) Obviously some people can't drive, this would help them.
3) You don't think somebody with a long commute would like to be able to take a nap on the way to work, or watch Netflix on the way back?
4) Parking is easier. Not everybody going downtown needs to PARK downtown. You can just tell your car to go park 10 minutes away or wherever there's parking available. And you don't have to wait while looking for parking.
5) Say I need to drive my car to work in the morning and then my wife needs it to run some errands at noon, and then I need it back in late afternoon to drive home from work. Wouldn't it be great if the car can go back home after I go to work for my wife to use it, and then pick me up afterwards?
By that argument, the Segway ought to be ubiquitous by now. So it doesn't always work.
I'm looking forward to cars that obey the speed limit, come to complete stops at stop signs, and won't start if there is snow on the road.
Yeah, that will probably take some kind of AI, since we humans don't seem to be able to handle it. It's kind of like "responsible ownership of guns": a theoretical possibility, but so many failures that cynicism is warranted.
Oh, and if it isn't obvious, by "artificial intelligence", I do NOT mean "a program that thinks like a person does."
I don't think 2025 is a reasonable estimate (not for the average driver), and I'm in the "if it takes longer (and it will), who cares?" camp.
In any case, expect self-driving cars to have limitations. For example, some will never be able to:
1. Navigate a trip to Mastodon
2. Detect the ElonJet on its radar
I think you are right about when it comes. Hopefully it won't be something like fusion, which has been just a decade away for several decades now. But when it arrives matters a lot, because, for instance, I sure as heck want it to arrive before I am too old to safely drive myself around.
Kevin is never getting off this horse. JFC, dude, just admit you were wrong and move on. Driverless cars will work...after we completely redesign all of our roadway infrastructure, which is decades away at minimum. They will never work on the roads we have now, especially in places where road markings are routinely obscured by snow, as my street at this very moment as I look out my living room window.
How do human drivers operate in snow or whatever? Driving is not nuclear physics or great art - almost anybody can do it, even in bad conditions. There is a finite number of things that drivers can be aware of. There is no reason to assume that programmers - not robots - can't break down the decision-making processes and eventually incorporate this into computer algorithms. There also have to be sensors to collect all the information that human drivers get - and maybe more. The information - primarily video now - has to be processed in real time by the on-board computer. This was impossible a few years ago, but computers continually improve and so to video cameras and other possible sensors.
How do human drivers operate in snow? With a lot of improvisation. Those two lanes gradually become one lane as the plows shove snow to the right side. That now-one lane meanders a bit, following the tracks from cars that passed through earlier. There is the run of elevation changes. There is one block on my usual commute that I avoid in icy conditions, because I know from experience that it is just enough of a grade to be a problem. For an extreme case, my commute includes one spot that dips down to cross a creek, then up again the other side. Usually it is merely cause for caution even in snow, but during the snowpocalypse it was a question of getting back up the other side, and this on a moderately heavily used secondary road. Oh, and with a curve at the bottom. The solution was that the drivers lined up. The guy at the head of the line would gingerly make the trip down, make the curve, then gun it to get back up. Once he had completed the maneuver, the next in line would do the same. I, and everyone ahead of me, made it through successfully. I don't know about the ones behind me. One car not making it would gum up the whole works. But with only one at a time risking it, at least it wouldn't result in a massive pile up.
"...If not, why not?"
1. Because of the dwindling supply of competent engineers, especially (but not limited to) software engineers.
2. Because of the complexity of the associated software, which makes it impossible to test or maintain. (This is not a matter of any lines-of-code metric, but of the skill of organizing and managing complexity in a large code base; the number of persons who possess, or who understand the importance of, that skill is so small that it might as well be zero.)
As for 2025, to quote somebody: "How about never? Is never good for you?"
So begins the set up for the inevitable “schedule” slip.
What this kind of failure shows is that cars are not really "smart" - they do not have human-like intelligence. If a teenager on a learning permit had an incident of unnecessary stopping, she/he would probably learn from it and adjust behavior. For the cars to correct this behavior the information has to get back to the programmers and hopefully and eventually they modify the driving algorithm. That modification can then be transmitted to all the cars.
The decisions involved in driving are only a small part of what human intelligence must cope with, so cars are not really being designed to think on their own - they are just fed increasingly elaborate algorithms to cope with driving decisions. This process has been going on for years and may eventually succeed in producing completely self-driving cars that are safer than human-driven cars. It is also possible that snags may be encountered that the programming and sensors can't overcome.
The continual hype that self-driving cars will be available next year or whenever, which Kevin has participated in to some extent, has been wrong. I think that Atrios's main point, that public transportation is a better alternative than individual cars for everybody all the time, is probably justified. This has to be done by the government through taxes, so what we get from the free-market system is continual promise of transcendental improvement of private vehicles. (Apart from Elon's magic tunnels, which so far have turned out to be just more Teslas).
1) Yeah, I agree that Kevin kind of tends to overlook Atrios's main point that you have to decide now what to be investing in. Given that people need to get around now, and there's a lot of proven technology that already does it well in densely-populated areas, investing in the known is what makes sense. I think he looks at self-driving cars or flying cars or vacuum tunnels or whatever as rich-guy toys at best, or just bald-faced open grifts, that shouldn't be getting public money until they're way farther along, if ever.
2) We need a vaportech time unit for stuff like this, like we have the Scaramucci for tenure in high government jobs. What does Elmo always promise, it'll be operational two years from now? So two years can be one Elon.
For true driverless capability, my opening bid is 5 Elons.
I'm not for or against the idea of driverless cars, although I kind of enjoy driving my car and I probably will never be able to afford a fully operational self-driving vehicle anyway. But those who argue that such vehicles can't "learn" from their mistakes because programmers would have to get access to the "mistake logs" and fix the code more or less in real time and that would simply take too damn long should consider that chatGPT is already reviewing and correcting computer code almost instantaneously, and even making suggestions how to improve the underlying premises of the code to make it more efficient. If the car can fix its own programming in real time maybe it's reasonable to move up the projected timeline for delivery of safe working models.
I should have added that chatGPT is primitive AI, and will probably get much better quickly.
The point is not "can it be done?" It probably can be done, though likely well after 2025. The question to ask is: What the hell are self driving cars good for? How much investment into developing the thing is justified? I think that more money than reasonable has already been wasted on the project.
My suspicion is that self driving has become a shiny object that everybody wants regardless of its value--especially in a car-obsessed country like the US.
I can understand the fascination of it. If I were a programmer I would also be attracted by it. But as to "bang for the buck" the project just sucks--even if it eventually succeeds.
What the hell are self driving cars good for?
That's what I've been wondering also. What is the problem that self-driving cars are supposed to solve? The problem of driving? Driving is hardly drudgery. Many of us find driving to be a fairly pleasurable activity. But if windshield time is taking you away from other activities -- reading, napping, whatever -- you have options: train, bus, taxi, Uber. So-called self-driving cars are not the solution.
As Tesla says:
"Autopilot, Enhanced Autopilot and Full Self-Driving Capability are intended for use with a fully attentive driver, who has their hands on the wheel and is prepared to take over at any moment."
FSD is a luxury feature offering better-than-cruise-control functionality. But it's still a ridiculously long way from relieving drivers of the usual driver responsibilities.
If the promise of self-driving cars is greater safety, that seems to be on the other side of a period of worse safety while the technology bugs gets worked out. Truck fleet owners may think they'll save money when they don't need truck drivers anymore, but that's not happening soon. For the same reason airline piloting is highly automated but planes still have a human pilot and copilot on board every flight.
Another point: Privacy.
The Tesla-turned-Twitter guy seems upset that anyone could know the location of his plane or the city that he's in. Also this week, the rumor is Twitter users will need to pony up a monthly fee or Twitter will get real-time user location data. Something to keep in mind: for many cars now, and for every car in the future, your car company will know your exact location every minute of the day.
I'm hardly a Luddite. Technology is cool and more cool things are coming. But a lot of the hype is pure fantasy and that doesn't help in the meantime.
One problem it would solve is mobility for older people, most of whom really do not want to rely on Uber or busses.
Another problem is safety and drunk driving --- I suspect we could save 10s of thousands of lives annually n the roads.
Another problem is convenience for parents, who spend ungodly amounts of time driving their kids around but who could instead just drop them into the self-driving car which would deliver them to soccer practice just fine.
And your blithe dismissal that people who want more time to read or relax instead of worrying about traffic while commuting is off; most people are not going to uber to work on a regular bases and mass transit is not suitable for most American commuters.
Self-driving cars give everybody their own chauffeur, a luxury presently reserved to the rich, and to highly placed employees of rich organizations.
Frankwilhoit: How about never is Bob Mankoff. A New Yorker cartoon I believe and also a book of his.
The only self driving system that tries to drive the way humans do is Tesla's - and it is doing very well
Its NOT slowing traffic (except by obeying the speed limits) - I have heard it described as driving like a "super cautious maniac"
Its inherent advantages are huge -
looking in ALL direction -
exact placement on the road -
reactions 20 times as fast as the best human -
control of the brakes on individual wheels
no distractions
no "forgetting"
We haven't got the vaguest idea how well Tesla FSD is doing compared to Cruz and Waymo as it is closely held proprietary data. All the indications are that they all are pretty good at freeway driving in good conditions, but these are the sort of conditions that human drivers can handle almost subconsciously, distractions and all.
Some people are just very invested in the notion that human intelligence will be unique forever. I think 2025 is a little optimistic for handling all possible driving conditions, but there's no doubt that we will get there.
Suburban America is a clinic in poor planning & road design.
Is a cautious vehicle AI going to refuse to make the left turn out of McDonald's onto the typically congested 5-lane commercial strip for 20+ minutes?
Turning freeway driving into a passive activity is useful. It also makes long commutes a lot easier to swallow -- that's bad.
A must-read that takes apart a number of the positives from driverless cars:
https://www.bloomberg.com/news/features/2022-12-07/the-real-toll-of-the-ride-hailing-and-self-driving-revolutions
Excellent read, +1
Back when self-driving cars first entered the discourse, when we were assured they were just a few years away, the discussion had the air of "We have solved 95% of the problem. How hard could the last 5% be?) Now here we are, years later, having the same discussion. If there is any change at all to it, it is now "We have solved 984 of the problem. How hard could the last 2% be?) So it goes.
2025? Let us define our terms. 2025 is plausible if we limit the discussion to small areas that are mapped to a very high level of detail, those maps being constantly updated, then sure. Some of that already exists. But we need to further refine the discussion to admit that we are talking about dense, affluent, white regions with clement weather. Is that "truly convenient and useful"? It is if you live in that region, but in the general sense it is utterly useless. The other plausible application in that time frame is self-driving trucks on freeways, with dedicated facilities just off the freeway for the endpoints. That would not surprise me in the least. But some George Jetson experience for the masses? In 2025? I'll the the over on that.