4 ms·

I am fairly confident that the autonomous car business model is pretty dead. Electric cars - that's great ! But self driving is waaaay harder. And self drivin

I am fairly confident that the autonomous car business model is pretty dead. Electric cars - that's great ! But self driving is waaaay harder. And self driving has a inbuilt expectation of nearly perfect. It's not that if we swapped to self driving cars today we would halve car deaths (which would be something like 500k people pa - which would be a huge very important thing) but it would not be perceived as robots save 500k people a year, it would be instead "robots kill the other 500k per year" Liability, insurance, legal minefields and plain old marketing would never allow cars ,that perform as well as the best performing cars do today, to be on the roads in "every day" conditions. My conjecture is we end up building AV only roads. initially one lane of a highway, then ring roads round cities and major warehousing hubs, then across urban areas. Walled off in some way they simply become railways with benefits. At that point every business model ever written with Self Driving in the title goes in the bin. I am not saying the tech is useless - frankly it's fucking awesome that this is happening in my lifetime. But fucking awesome tech and workable business model aren't always the same thing. Not sure where I am going with the rant but I sure hope we get more out of the billions spent here than Teflon.

hakfoo 5y ago
I still have hope, because we have two different mentalities at play. The individual still thinks in terms of "I'm an above-average driver, I'm better than a silly bleep-bloop box." They'll resist self-driving en masse if it's not 1000% perfect. Insurers, however, will see self-drivers as predictable, never drunk, always following the speed limit, and in aggregate preferrable to humans. They will be able to see "3% better" as meaningful even if it's not perfect, and price accordingly. Eventually, you'll pay a rapidly increasing penalty rate for having a steering wheel. This is where the avalanche comes from. I also figure self-driving could unlock new features-- in particular I'm imagining the only way we get road speeds much over 120kph is self-driving, as the human response time becomes a limiting factor. And of course the in-dashboard wet bar. This will feed back into the consumer market even if it's only closed markets at first: why can't I go 250kph on the highways considering the self-driving airport shuttle does?
failwhaleshark 5y ago
Nope. You're just not patient enough. ACs will come, but very gradually. They're already (mostly) here but it will take another decade or two to be completely A.
- treeman79 5y ago
  So between “not invented” and “not proven impossible yet”. https://xkcd.com/678/ https://xkcd.com/678/
  - failwhaleshark 5y ago
    Lol, I guess. ACs aren't the same as the Moller Skycar. It's a very, very complicated "DARPA Grand Challenge" where the goal is to not get sued by Ralph Nader into oblivion because the algorithm went MCAS and killed grandma. Waymo or someone will need to do racing and the Gumball 3000 first before people will accept anything more than semi-A (glorified cruise control and lane following).
erikpukinskis 5y ago
Doesn't "I'm fairly confident" kind of imply you're not actually confident?
- Doctor_Fegg 5y ago
  It’s common British English idiom.
  - Sohcahtoa82 5y ago
    In my experience, it's common in American English as well. Not sure what the parent commenter is on about.
    - developer93 5y ago
      their point is understatement, "I'm fairly confident" = "I fully believe"
propogandist 5y ago
you can skirt all regulation and set false expectations by simply calling your driver assist feature “autopilot”
AbrahamParangi 5y ago
You'll know self driving is almost here when it works in simulation at non-interactive framerates.
separateform 5y ago
Railways with benefits aren't good enough and would be an even worse business model.
lvs 5y ago
I think the issue is bigger than just self-driving. AI/ML is wildly oversold as an engineering solution to real world problems. In the best case, it's solving easy problems under narrowly defined conditions. It's not really solving hard problems robustly under real conditions.
ajross 5y ago
I see this rather differently. In fact AI is already as good as human drivers, statistically. Only a few years back, we could point at people getting killed by software bugs like the Tesla path-under-semi-trailer issue (not a lot, but it did happen and they were real bugs). That's not happening any more. All we have left is laughing at stuff like this, where the visualization (not even the autopilot!) gets confused by seeing real (!) traffic lights on a truck, so it paints them in space, but then has to re-recognize them because they are moving. At some point, the luddites will just run out of ammunition. It's sort of happening already.
- lstamour 5y ago
  > AI is already as good as human drivers But the AI drives slowly and gets confused easily. Regular drivers routinely have to go around self-driving cars. Not to say they won't improve, but it seems like current AI is assistive to the point where it might be harmful when drivers rely on it in speeds and situations where they shouldn't. I'm sure it will keep improving, but I feel like this is one of those situations where the amount of data and training required, and the amount of iteration on the software required to handle edge cases is not impossible but is exceptionally difficult.
  - ajross 5y ago
    > But the AI drives slowly and gets confused easily. Probably better to say that the AI drives slower than you would and gets confused differently. Real drivers die every day to terrible mistakes that look obvious in hindsight. Most of them probably thought AI was terrible too. The bar here is much lower than you think it is, and frankly existing automation has already crossed it.
  - visarga 5y ago
    > the amount of iteration on the software required to handle edge cases is not impossible but is exceptionally difficult. If you want to fix this one exception (false positive) you will introduce unwanted false negatives. That's how precision and recall work, there's a trade-off. So I am not sure it's useful to fix it, how many times does this happen? Will fixing it introduce more frequent bugs?
exporectomy 5y ago
It doesn't need to be almost perfect. Waymo, for instance has remote human assistance when it gets stuck, and even a human in a normal car drive to the scene and to get it unstuck if it's really bad. There could well be a point where those humans plus the tech are cheaper than a traditional full-time-per-customer human taxi driver. You might say people don't want those delays but we tolerate delays in normal commuting. An unattended bag on a tube station stopping trains, a car accident blocking traffic, traffic jams blocking traffic, mechanical breakdowns, etc. As long as they're infrequent enough, it should be OK for riders.
- bigyikes 5y ago
  How does remote assistance work? My imagination says latency would make that really unsafe.
  - lifeisstillgood 5y ago
    I believe that's the promise of 5G
    - stjohnswarts 5y ago
      Uh oh.
  - exporectomy 5y ago
    They provide some sort of information to the computer driver. They're not directly steering the car in a normal driver control loop.
ctdonath 5y ago
Au contraire, seems we're mostly there. It's incredible having watched self-driving, and neural networks, go from new concepts and barely functioning to "order here" and public use (tell me the numerous YouTube FSD videos aren't what they are). Insofar as there are still serious edge cases to address, they're being solved. Time and again I've watched "ain't happening" technology become the preferred norm practically overnight. Eagerly awaiting my FSD CT, and making long trips without having to micro-manage every foot across thousands of miles.
axguscbklp 5y ago
I have been saying for years that full self driving is extremely unlikely to be developed any time in the near future because it is simply too hard of a problem for any reasonable near-future extension of currently available technology. I am gratified to see that the conventional wisdom has recently come to also see things that way. I am not surprised that journalists bought the hype - I assume that half of them have no idea what they are writing about and/or are just advertisers in disguise. But it does surprise me that so many experienced corporate professionals and investors got taken in. I guess maybe part of it is that they just are not as smart as I once thought they were and maybe another part of it is that some of them used the hype to grab a bunch of cash with both hands, cash that the people who put it up will now never see again.
fullshark 5y ago
> My conjecture is we end up building AV only roads. initially one lane of a highway, then ring roads round cities and major warehousing hubs, then across urban areas. Walled off in some way they simply become railways with benefits. That would be awesome still and make people want to have that technology...how does that murder Tesla's business model? Waymo would easily pivot to licensing their tech to car manufacturers.
mdorazio 5y ago
Disclaimer: I work on the business side of the autonomous team at a major OEM. The autonomous car business model is very much alive and well, it has simply shifted in two key areas that cause general consumers to misinterpret what is happening. 1) Use cases (ODD). Many people think "if I can't buy an autonomous car, it's not a real thing". But most manufacturers and AV developers other than Tesla aren't even trying to tackle that use case. Instead, the focus is on autonomous vehicles operating very profitably in specific locations and operating models that work. Heavy trucking on "easy" routes, ridesharing in well-mapped areas of the Southwest US, shuttles on popular metro and airport routes, etc. These use cases represent very large revenue opportunities and the tech development is progressing well. 2) Timing. Too many pundits and CEOs looking to generate PR buzz made ridiculous claims over the last few years about when autonomous vehicles would be ready. This set false expectations among the general public which has now soured perceptions since autonomous tech is not easy. Fully driverless is already ready in some very specific scenarios (ex. Waymo in Chandler) and is on track to roll out to more ODDs over the next 5-7 years. I can pretty much guarantee that you will see a significant number of fully driverless vehicles in suitable areas by 2030. No, they won't drive in upstate New York in a blizzard, but no one actually cares about that from a business perspective. The other issues you raise - liability, insurance, legal roadblocks, etc. are mostly non-issues that already have solutions. The only one that's a continuing problem is the fragmented nature of legislation across different states. But there is very heavy lobbying going on right now to rectify this by the time it's actually needed.
- lifeisstillgood 5y ago
  Thank you for the comments. I am actually writing a book (The Software Mind) covering, well, quite a lot of ground, and would be interested in getting a real perspective on this area - can I reach out to you? Edit: I would be interested in real experts correcting my poor understanding of important subjects - can we talk ?
  - mdorazio 5y ago
    Sure thing, but I might not be able to comment on some things due to NDAs. Shoot me an email - dorazio at gmail
moduspol 5y ago
> My conjecture is we end up building AV only roads. initially one lane of a highway, then ring roads round cities and major warehousing hubs, then across urban areas. Walled off in some way they simply become railways with benefits. Close, but I think it's far more probable that we just whitelist roads / intersections / routes known to work well and be reliable--whether with a general-purpose or location-specific algorithm. There's still a ton of value shuttling people to and from the airport and local hotels. Then there'll be incentive for local restaurants to also be included in the available destinations, so there'll be pressure to both improve local roads / intersections to make them more easily navigable by self-driving cars, and on self-driving cars to get better at navigating them. With time, the locations / routes that self-driving cars can reliably reach will expand until it covers the vast majority of desired destinations. We actually are following the same pattern with the horses -> cars transition. There are still plenty of places that cars can't reach and horses can. Over decades, though, we just kind of paved roads everywhere anyone really wanted to go. Nowadays it's more or less a given that if you want people to be able to reach somewhere, it'll need to have a road. We'll see the same thing with self-driving cars. There'll still be roads that self-driving cars avoid and humans drive on, but they'll become less and less relevant as the value of self-driving cars becomes more apparent.
tsjq 5y ago
it is very unfortunate that the first & really good electric vehicle is a highly complicated tech gadget. and this sets a bad benchmark for the other companies which want to step into this space. why not a simple electric vehicle : just different fuel and engine , and all the rest same as ICE vehicles . wouldn't that be easier / simpler to deal with ? sadly, in India too, the major electric 2wheeler Ather, is kinda going the same high-tech way as Tesla. and very expensive !
- hakfoo 5y ago
  It's likely market positioning factors. A big part of Tesla's success was making electric vehicles that were covetable, rather than being obvious "we made this for compliance reasons or to get a government contract" products. This means targeting a luxury market segment. The way they chose to wow luxury consumers was with high performance and a lot of technological gimmickry.
- XargonEnder 5y ago
  I agree with your sentiment that I'd like to see EVs made for normal people like me, but I'm grateful that people with lots of money fund R&D on stuff like Tesla's to help bring the price down to where I can get it one day. Kinda reminds me of smart phone evolution.
stjohnswarts 5y ago
I don't need autopilot until it is 100% (or 99.999%) but I'll be damned is cruise control "auto pilot" in a stop and go commute is nice and works well enough. that takes a bit of the grind off the drive to work.

hakfoo 5y ago

I still have hope, because we have two different mentalities at play. The individual still thinks in terms of "I'm an above-average driver, I'm better than a silly bleep-bloop box." They'll resist self-driving en masse if it's not 1000% perfect. Insurers, however, will see self-drivers as predictable, never drunk, always following the speed limit, and in aggregate preferrable to humans. They will be able to see "3% better" as meaningful even if it's not perfect, and price accordingly. Eventually, you'll pay a rapidly increasing penalty rate for having a steering wheel. This is where the avalanche comes from. I also figure self-driving could unlock new features-- in particular I'm imagining the only way we get road speeds much over 120kph is self-driving, as the human response time becomes a limiting factor. And of course the in-dashboard wet bar. This will feed back into the consumer market even if it's only closed markets at first: why can't I go 250kph on the highways considering the self-driving airport shuttle does?

failwhaleshark 5y ago

Nope. You're just not patient enough. ACs will come, but very gradually. They're already (mostly) here but it will take another decade or two to be completely A.

treeman79 5y ago
So between “not invented” and “not proven impossible yet”. https://xkcd.com/678/ https://xkcd.com/678/
- failwhaleshark 5y ago
  Lol, I guess. ACs aren't the same as the Moller Skycar. It's a very, very complicated "DARPA Grand Challenge" where the goal is to not get sued by Ralph Nader into oblivion because the algorithm went MCAS and killed grandma. Waymo or someone will need to do racing and the Gumball 3000 first before people will accept anything more than semi-A (glorified cruise control and lane following).

erikpukinskis 5y ago

Doesn't "I'm fairly confident" kind of imply you're not actually confident?

Doctor_Fegg 5y ago
It’s common British English idiom.
- Sohcahtoa82 5y ago
  In my experience, it's common in American English as well. Not sure what the parent commenter is on about.
  - developer93 5y ago
    their point is understatement, "I'm fairly confident" = "I fully believe"

propogandist 5y ago

you can skirt all regulation and set false expectations by simply calling your driver assist feature “autopilot”

AbrahamParangi 5y ago

You'll know self driving is almost here when it works in simulation at non-interactive framerates.

separateform 5y ago

Railways with benefits aren't good enough and would be an even worse business model.

lvs 5y ago

I think the issue is bigger than just self-driving. AI/ML is wildly oversold as an engineering solution to real world problems. In the best case, it's solving easy problems under narrowly defined conditions. It's not really solving hard problems robustly under real conditions.

ajross 5y ago

I see this rather differently. In fact AI is already as good as human drivers, statistically. Only a few years back, we could point at people getting killed by software bugs like the Tesla path-under-semi-trailer issue (not a lot, but it did happen and they were real bugs). That's not happening any more. All we have left is laughing at stuff like this, where the visualization (not even the autopilot!) gets confused by seeing real (!) traffic lights on a truck, so it paints them in space, but then has to re-recognize them because they are moving. At some point, the luddites will just run out of ammunition. It's sort of happening already.

lstamour 5y ago
> AI is already as good as human drivers But the AI drives slowly and gets confused easily. Regular drivers routinely have to go around self-driving cars. Not to say they won't improve, but it seems like current AI is assistive to the point where it might be harmful when drivers rely on it in speeds and situations where they shouldn't. I'm sure it will keep improving, but I feel like this is one of those situations where the amount of data and training required, and the amount of iteration on the software required to handle edge cases is not impossible but is exceptionally difficult.
- ajross 5y ago
  > But the AI drives slowly and gets confused easily. Probably better to say that the AI drives slower than you would and gets confused differently. Real drivers die every day to terrible mistakes that look obvious in hindsight. Most of them probably thought AI was terrible too. The bar here is much lower than you think it is, and frankly existing automation has already crossed it.
- visarga 5y ago
  > the amount of iteration on the software required to handle edge cases is not impossible but is exceptionally difficult. If you want to fix this one exception (false positive) you will introduce unwanted false negatives. That's how precision and recall work, there's a trade-off. So I am not sure it's useful to fix it, how many times does this happen? Will fixing it introduce more frequent bugs?

exporectomy 5y ago

It doesn't need to be almost perfect. Waymo, for instance has remote human assistance when it gets stuck, and even a human in a normal car drive to the scene and to get it unstuck if it's really bad. There could well be a point where those humans plus the tech are cheaper than a traditional full-time-per-customer human taxi driver. You might say people don't want those delays but we tolerate delays in normal commuting. An unattended bag on a tube station stopping trains, a car accident blocking traffic, traffic jams blocking traffic, mechanical breakdowns, etc. As long as they're infrequent enough, it should be OK for riders.

bigyikes 5y ago
How does remote assistance work? My imagination says latency would make that really unsafe.
- lifeisstillgood 5y ago
  I believe that's the promise of 5G
  - stjohnswarts 5y ago
    Uh oh.
- exporectomy 5y ago
  They provide some sort of information to the computer driver. They're not directly steering the car in a normal driver control loop.

ctdonath 5y ago

Au contraire, seems we're mostly there. It's incredible having watched self-driving, and neural networks, go from new concepts and barely functioning to "order here" and public use (tell me the numerous YouTube FSD videos aren't what they are). Insofar as there are still serious edge cases to address, they're being solved. Time and again I've watched "ain't happening" technology become the preferred norm practically overnight. Eagerly awaiting my FSD CT, and making long trips without having to micro-manage every foot across thousands of miles.

axguscbklp 5y ago

I have been saying for years that full self driving is extremely unlikely to be developed any time in the near future because it is simply too hard of a problem for any reasonable near-future extension of currently available technology. I am gratified to see that the conventional wisdom has recently come to also see things that way. I am not surprised that journalists bought the hype - I assume that half of them have no idea what they are writing about and/or are just advertisers in disguise. But it does surprise me that so many experienced corporate professionals and investors got taken in. I guess maybe part of it is that they just are not as smart as I once thought they were and maybe another part of it is that some of them used the hype to grab a bunch of cash with both hands, cash that the people who put it up will now never see again.

fullshark 5y ago

> My conjecture is we end up building AV only roads. initially one lane of a highway, then ring roads round cities and major warehousing hubs, then across urban areas. Walled off in some way they simply become railways with benefits. That would be awesome still and make people want to have that technology...how does that murder Tesla's business model? Waymo would easily pivot to licensing their tech to car manufacturers.

mdorazio 5y ago

Disclaimer: I work on the business side of the autonomous team at a major OEM. The autonomous car business model is very much alive and well, it has simply shifted in two key areas that cause general consumers to misinterpret what is happening. 1) Use cases (ODD). Many people think "if I can't buy an autonomous car, it's not a real thing". But most manufacturers and AV developers other than Tesla aren't even trying to tackle that use case. Instead, the focus is on autonomous vehicles operating very profitably in specific locations and operating models that work. Heavy trucking on "easy" routes, ridesharing in well-mapped areas of the Southwest US, shuttles on popular metro and airport routes, etc. These use cases represent very large revenue opportunities and the tech development is progressing well. 2) Timing. Too many pundits and CEOs looking to generate PR buzz made ridiculous claims over the last few years about when autonomous vehicles would be ready. This set false expectations among the general public which has now soured perceptions since autonomous tech is not easy. Fully driverless is already ready in some very specific scenarios (ex. Waymo in Chandler) and is on track to roll out to more ODDs over the next 5-7 years. I can pretty much guarantee that you will see a significant number of fully driverless vehicles in suitable areas by 2030. No, they won't drive in upstate New York in a blizzard, but no one actually cares about that from a business perspective. The other issues you raise - liability, insurance, legal roadblocks, etc. are mostly non-issues that already have solutions. The only one that's a continuing problem is the fragmented nature of legislation across different states. But there is very heavy lobbying going on right now to rectify this by the time it's actually needed.

lifeisstillgood 5y ago
Thank you for the comments. I am actually writing a book (The Software Mind) covering, well, quite a lot of ground, and would be interested in getting a real perspective on this area - can I reach out to you? Edit: I would be interested in real experts correcting my poor understanding of important subjects - can we talk ?
- mdorazio 5y ago
  Sure thing, but I might not be able to comment on some things due to NDAs. Shoot me an email - dorazio at gmail

moduspol 5y ago

> My conjecture is we end up building AV only roads. initially one lane of a highway, then ring roads round cities and major warehousing hubs, then across urban areas. Walled off in some way they simply become railways with benefits. Close, but I think it's far more probable that we just whitelist roads / intersections / routes known to work well and be reliable--whether with a general-purpose or location-specific algorithm. There's still a ton of value shuttling people to and from the airport and local hotels. Then there'll be incentive for local restaurants to also be included in the available destinations, so there'll be pressure to both improve local roads / intersections to make them more easily navigable by self-driving cars, and on self-driving cars to get better at navigating them. With time, the locations / routes that self-driving cars can reliably reach will expand until it covers the vast majority of desired destinations. We actually are following the same pattern with the horses -> cars transition. There are still plenty of places that cars can't reach and horses can. Over decades, though, we just kind of paved roads everywhere anyone really wanted to go. Nowadays it's more or less a given that if you want people to be able to reach somewhere, it'll need to have a road. We'll see the same thing with self-driving cars. There'll still be roads that self-driving cars avoid and humans drive on, but they'll become less and less relevant as the value of self-driving cars becomes more apparent.

tsjq 5y ago

it is very unfortunate that the first & really good electric vehicle is a highly complicated tech gadget. and this sets a bad benchmark for the other companies which want to step into this space. why not a simple electric vehicle : just different fuel and engine , and all the rest same as ICE vehicles . wouldn't that be easier / simpler to deal with ? sadly, in India too, the major electric 2wheeler Ather, is kinda going the same high-tech way as Tesla. and very expensive !

hakfoo 5y ago
It's likely market positioning factors. A big part of Tesla's success was making electric vehicles that were covetable, rather than being obvious "we made this for compliance reasons or to get a government contract" products. This means targeting a luxury market segment. The way they chose to wow luxury consumers was with high performance and a lot of technological gimmickry.
XargonEnder 5y ago
I agree with your sentiment that I'd like to see EVs made for normal people like me, but I'm grateful that people with lots of money fund R&D on stuff like Tesla's to help bring the price down to where I can get it one day. Kinda reminds me of smart phone evolution.

stjohnswarts 5y ago

I don't need autopilot until it is 100% (or 99.999%) but I'll be damned is cruise control "auto pilot" in a stop and go commute is nice and works well enough. that takes a bit of the grind off the drive to work.