I think this is exactly right, and if anything the longer term for AVs being transformational is even greater. The thing about taxis is that they're still recognizably *cars* in the way we're all used to. What I really want, eventually, is an autonomous tugboat-equivalent that can pull any kind of customized cabin-trailer I need. Mobile office. Mobile sleeper. Travel trailer that's actually safe to ride in.
Sure, you could take a robotaxi to and from the airport. But I'm hoping at some point you'll be able to get into a mobile sleeper trailer at night in LA, hire an AV to tug it to a meeting in SF the next morning, then get in a mobile office trailer and get out 8 hours later at the end of the workday in SF.
For a more sobering (and informed) perspective, have a look at Rod Brooks' predictions about self driving cars https://rodneybrooks.com/predictions-scorecard-2024-january-01/ His credentials include the creation of the Roomba and heading MIT's AI lab. His take is that self-driving cars are still vastly overhyped. The safety numbers are BS btw. Don't believe press releases.
The implications on land use are truly exciting. If the price and convenience of AV taxis significantly improves, that decreases the relative value of using land as a storing place for empty vehicles. We’ll hopefully see less parking lots and more housing units and parks and things that promote community and economic activity. It feels like I see a lot of debate in policy discussions on the labor and safety sides, and relatively less from urban development and housing sides, which is unfortunate.
Seems like a lot of people in the comments here are concerned that parking lots will just be replaced with crowded roads. The overall number of car-minutes on the road would go up, but I would expect AV companies will be able to coordinate car flows to optimize total ridership—more rides but less during peaks and less congestion; they can use levers to reallocate traffic to off-peak times and more efficiently coordinate among cars simultaneously on the roads. At the least, they are more positioned to internalize the externality that any rider has on another, eg in traffic jams. Interestingly, this suggests efficiencies from consolidating to fewer companies which one could use as an argument in favor of making this a new form of public transit.
Jack's comment covered my concerns about changes in traveling behavior due to widespread robotaxi use. That's the one I've quietly been worrying about since shortly after I started thinking about self-driving cars around 2012 (the reason I got into this is because I hate driving).
One thing that ubiquitous robotaxis should do that seems nice is to make vehicles smaller. Today, many people buy cars for maximal use (whole family traveling together with luggage), but most trips involve much less people and stuff than that. With robotaxis, you can just book whatever size vehicle you need for a particular trip.
Once crashes are much less common, people would feel better in a much smaller vehicle, even a single-rider vehicle, which should be cheaper to produce and run.
The soft obstacle is not a likely scenario, IMO. That relies on a deliberate bad choice by programmers. What's more likely is failure to recognize e.g. a child chasing a ball into the street by an empty lot where kids are playing or a dark skinned person at night. That will be blamed on, what? Insufficient training of AI or perhaps some software bug. Things that are really hard to control and are actually inevitable.
This was intended to be a reply to srynerson's long response to a post below. I can't seem to copy the text on my phone to put it where it belongs and delete this comment.
Waymo's operation is bit overrated. Their sensor tech is ahead on max speed, but they apparently don't have many cars on the road in SF - only about ~250 from what I can gather. Their full autonomous only car is going to be a Chinese import - call me bearish.
20. 20 is the magic number. Autonomous vehicles should all be mandated to never go faster than 20 mph in the city. If that is accomplished, they will be safe and trusted and successful. If they are allowed to compete with human drivers in speeding contests, they will quickly start breaking stuff and killing people and prompt a swift backlash.
"What would be better would be to hear clearer statements from taxi regulators and state transportation departments that they are going to hold driverless vehicles to some kind of objective and reasonable “safer than humans on average” standard rather than pulling the plug the first time something bad happens."
What kind of betting odds can I get that this will, in fact, not happen in the next 5 (maybe 10) years? I think I'd give you 20:1 at least.
Regulators are not typically rewarded for improving the status quo. If 20 people die in a freeway pile up tomorrow, absolutely no one will knock on the regulators' doors and say "You could have prevented this, if only you'd allowed those safer autonomous cars!"
On the other hand, if one person dies in a freak accident caused by an autonomous car, the regulator is probably getting their ass handed to them. This is probably true even if the autonomous car wasn't even really "at fault" (e.g., like a pedestrian who isn't looking jumping into fast traffic, too late for the car to dodge/brake).
Chapter one of Mankiw, I think: People respond to incentives.
The fact that Uber and Lyft exist is a big part of why we are not even considering buying a third car when my daughter starts driving. We only have two now because one is an old EV with limited range and we have to do a lot of long range driving for her sports. If Uber or Lyft was about 25% less expensive, I think that would be the tipping point for me to decide to sell both of our cars and invest in one longer range EV. I know my husband would use Uber or Lyft more if he didn't have to talk to drivers. Right now I still take Lyft about twice a month even though I have my own car. This is usually when I am going out for the evening. This is partly because the neighborhoods that I go out to have really expensive and/or hard to find parking and so just taking a vehicle that can drop me off saves me costs and frustration enough to make it worthwhile. Sometime it is a case of my going out to drinks for a girl's night with my neighbors. While I don't drink, carpooling in a Lyft is a low friction way of making sure none of my friends do either. If drank, I would be using it a lot more. I pretty much always take it to the airport. I actually think having Uber has made me take the bus more! Seeing out much it costs to take an Uber to downtown makes just hopping on the bus seem much more appealing. I think part of the problem with so many people having their own cars is that the costs are mostly sunk costs so that their is little motivation to use transit or walk for short trips because it won't save you money. (My husband was 100% counting on their being self-driving cars in 10 years about 16 years ago and the fact that we don't and my teen now has to learn to drive is pretty sad for him.)
Another aspect of the land use point is the number of cars that would be required in a self-driving cars world. The vast majority of cars today are not in use 97% of the time - they're parked and waiting for the owner (or renter) to need it. The number of passenger vehicles in circulation could be reduced by 90%+ if everyone could summon a self-driving car whenever they need it.
No one EVER talks about what to me is the most obvious and terrifying problem.
Literally everything gets hacked, from lazy workers leaving a laptop with WiFi IP addresses in a cafe to sophisticated groups actually finding backdoors.
Can you imagine a VIP driving at the mercy of a computer that can be hacked and controlled at will? Or alternatively, hack the car of the random guy driving alongside them to suddenly swerve left at 70mph...
And every time this is brought up the engineers dismiss it without any actual plan to address. As a biologist this is very similar to issues with mutagenicity and crispr therapy -no one looks for it because it's expensive to search and you don't want to learn the answer.
Yeah cars have computers and an older car wouldn't have any internet connectivity. The modern ones do, but if the designers were smart, then hopefully there would be no physical connection at all between the internet connected computers used for infotainment, GPS and stuff, and the chips that control safety critical functions like ABS, ESP, etc.
“ Electronic stability control modulates the engine throttle and each wheel’s brake. Like other driver aids, ESC turns on when you start your car. The technology operates in the background, monitoring steering wheel activity and speed sensors on each wheel.
When sensors determine the driver is losing control, the ESC system engages automatically to adjust for oversteer and understeer. It applies or eases the brakes of each wheel on the car to bring the vehicle back onto the intended path. ESC involves the traction control systems, adjusts engine power to reduce speed, and applies brakes if the driver gives too much gas.
Did you even read your own article??? This is about onstar. Insurance companies also have plugins that can track your driving to give reduced rates. That is just collecting mechanical information from the car. There is no wifi control of anything. You don't seem to have any idea what you are talking about...
How does Waymo (and similar companies) handle younger passengers who require car seats? If the endgame of this tech is to allow people to ditch their own cars and just use driverless taxis for transportation, would people with young children just need to lug car seats to their destinations? Would there be a special class of vehicles that contain car seats?
On crashes, insurance, bankruptcy, and negligence laws hurt AV adaption. Commercial transportation is held to the highest standard of care under negligence laws, while private car drivers are need only exercise regular due care for their driving.
It's also much easier for a jury to stick Alphabet with a $50 million verdict over a "normal" car crash "to teach greedy corporations a lesson" that their AV caused versus sticking some warehouse worker or whoever with the same. The warehouse worker is also probably judgement proof.
Yes, this +10,000. I think about it constantly as a litigator. (What follows is not legal advice!!!)
I can understand completely why trucking companies, taxi companies, etc. would love self-driving vehicles. Under the doctrine of respondeat superior, a business is normally liable for the negligence of its employees while they are performing their job duties. Therefore, as long as a self-driving car is even marginally better safetywise on average than a moderately competent human driver, the company will probably come out ahead on savings for the cost of lawyers and judgments in negligent driving cases.
However, what I cannot understand -- and have never seen an article even try to address -- is why *manufacturers* would want to make self-driving cars, because it seems like it's going to deeply drag them into any litigation where their systems are potentially at issue.
For context right now, in many (if not most) car crash cases, the manufacturer isn't even named as a defendant and, if they are named, there's a good chance they get out on summary judgment or for a modest settlement, or if they are actually stuck going all the way to trial, they are likely to be assigned a very small portion of the fault for the crash. This is because in normal car crash litigation, the manufacturer is only going to have liability relating to the design and manufacture of the car, not anything about the operation of it. So, if the manufacturer can accurately testify, "We designed the braking system so that if the car was traveling X miles per hour and the brakes were applied for Y seconds, the car would come to a complete stop in Z feet, assuming normal road conditions," then as long as those figures are reasonable based on industry standards and available technology there shouldn't be any liability.
Those sorts of figures are subject to (at this point) well-understood and highly objective engineering, materials sciences, and physics analysis, such that it is difficult for a plaintiff's expert to make a lot of hay out of them. The fact that in the particular case at issue "Z feet" from the point where the brakes were first applied happened to be on the opposite side of the entire kindergarten class of St. Swithen's Academy for Children of Highly Motivated Personal Injury Attorneys is completely irrelevant -- that's the driver's problem, not the manufacturer's! The manufacturer had no knowledge (and could not be expected to have any knowledge) of the actual road conditions, obstacles, etc., let alone how fast the vehicle was actually traveling or how long the brakes were applied. All of that stuff is "externalized" to the driver.
But, with a self-driving car, that situation now becomes very decisively the manufacturer's problem and it involves a large amount of stuff that you can't even begin to hope to resolve on summary judgment because it's going to involve a bunch of questions about the reasonability of different assumptions that went into the programming of the car when considered under the actual circumstances of the collision. E.g., through its software, the manufacturer has to decide (and will be held at least partially responsible for) when the brakes were applied, how long they were applied for, whether any plausible evasive maneuvers might have avoided the crash, etc.
I mean, just imagine the first time a self-driving car splatters a toddler without even braking and, in the subsequent lawsuit, the engineers testify that the car was programmed to disregard anything it deemed to be a "soft" obstacle estimated to be less than 50 lbs so as to maximize driver safety from possible rear-end collisions that could be caused by sudden braking. If that happens in a state without wrongful death damages caps, you're going to see the judgment from orbit!
I think this also means the consequences of the tail risk - i.e. a horrendous accident where an autonomous vehicle smashes a school bus and kills 10 children - are much more commercially consequential than when caused by a negligent human driver. Because, okay, you can say that insurance pays out either way. And that's true, but in the latter case the human driver is put in jail and never drives again, and no one cares. The underwriter shrugs and acknowledges the risk was captured in the spread among hundreds of millions of drivers. In the autonomous vehicle case, the manufacturer has to try to renew their insurance premium and finds it has massively increased. And there are probably not going to be more than 5-10 AV manufacturers in the world for a very long time (perhaps only 2-3), and designs will probably be very similar - so insurance premia for all of them will jump up together, immediately.
I'm pretty certain I was replying to this when I posted at the root level. The awful software of substack won't allow me to copy the text to put it here and I'm not willing to retype it on my phone. Oh how hard it is to do what should be really really simple in software...... Did I mention that software is often do much harder than it should be? Because that was sort of a point of my post earlier which was intended to be a reply to this.
I think you're not following my point because in your hypothetical Amazon is the employer, not the manufacturer of the vehicle. Employers are normally liable for the negligence of their employees while performing their job duties, although in your particular hypothetical that's actually not super clear because you've added the drunk part, which complicates the analysis, and then the knowledge part that complicates it further. (If I were Amazon's lawyer, I would probably tell them to write a very generous check because the whole thing would otherwise go to a jury and there would be a lot of bad publicity.) The *manufacturer* of the Amazon delivery truck, however, would be expected to have zero liability for a crash caused by the driver's intoxication.
Does a driverless car only take me to a prearranged destination, or can I change my mind mid-route?
Can I command "take a left at the next corner" or "drive around the block"?
I see personal value in the ability to hop in my car, point it in the direction I want to go...and go! And I like being in control. I went out of my way to buy a car with a stick shift - I don't want an automatic transmission, let alone a completely automatic car.
I also prefer being in control, and I despise surveillance. I use Google Maps with "location services" (GPS) turned OFF. I enjoy reading (and using) a map. I don't want (or need) a map that reads me.
In the case of Tesla, I can envision some interesting integration opportunities between the robotaxi business and the EV charging business. Specifically, one could imagine big level 2 charging hubs set up, with the expectation that a driver pulls in, plugs in, and gets whisked away on a robotaxi to their actual destination. Then, a few hours later, they take the robotaxi back to the charging hub to pick up the car on the way home. At least, in theory, if the robotaxis were to ever become cheap enough, this could become a solution for how apartment dwellers with EVs quickly and affordably charge their cars.
(Of course, if robotaxis are this cheap, it begs the question of why people need to own cars at all; the answer to that question would be a combination of zero wait time for most trips, being able to leave stuff in the car, and not being geographically constrained based on where the robotaxis are certified to drive).
If we could build a driverless car that can handle any conditions - snow, random pedestrian behavior, police officer waving cars around an accident site - then I think it would have to be basically human-level intelligence. In that case, we've built AGI, and that can be put to much more transformative uses than just driving cars.
I think this is exactly right, and if anything the longer term for AVs being transformational is even greater. The thing about taxis is that they're still recognizably *cars* in the way we're all used to. What I really want, eventually, is an autonomous tugboat-equivalent that can pull any kind of customized cabin-trailer I need. Mobile office. Mobile sleeper. Travel trailer that's actually safe to ride in.
Sure, you could take a robotaxi to and from the airport. But I'm hoping at some point you'll be able to get into a mobile sleeper trailer at night in LA, hire an AV to tug it to a meeting in SF the next morning, then get in a mobile office trailer and get out 8 hours later at the end of the workday in SF.
For a more sobering (and informed) perspective, have a look at Rod Brooks' predictions about self driving cars https://rodneybrooks.com/predictions-scorecard-2024-january-01/ His credentials include the creation of the Roomba and heading MIT's AI lab. His take is that self-driving cars are still vastly overhyped. The safety numbers are BS btw. Don't believe press releases.
The implications on land use are truly exciting. If the price and convenience of AV taxis significantly improves, that decreases the relative value of using land as a storing place for empty vehicles. We’ll hopefully see less parking lots and more housing units and parks and things that promote community and economic activity. It feels like I see a lot of debate in policy discussions on the labor and safety sides, and relatively less from urban development and housing sides, which is unfortunate.
Seems like a lot of people in the comments here are concerned that parking lots will just be replaced with crowded roads. The overall number of car-minutes on the road would go up, but I would expect AV companies will be able to coordinate car flows to optimize total ridership—more rides but less during peaks and less congestion; they can use levers to reallocate traffic to off-peak times and more efficiently coordinate among cars simultaneously on the roads. At the least, they are more positioned to internalize the externality that any rider has on another, eg in traffic jams. Interestingly, this suggests efficiencies from consolidating to fewer companies which one could use as an argument in favor of making this a new form of public transit.
Jack's comment covered my concerns about changes in traveling behavior due to widespread robotaxi use. That's the one I've quietly been worrying about since shortly after I started thinking about self-driving cars around 2012 (the reason I got into this is because I hate driving).
One thing that ubiquitous robotaxis should do that seems nice is to make vehicles smaller. Today, many people buy cars for maximal use (whole family traveling together with luggage), but most trips involve much less people and stuff than that. With robotaxis, you can just book whatever size vehicle you need for a particular trip.
Once crashes are much less common, people would feel better in a much smaller vehicle, even a single-rider vehicle, which should be cheaper to produce and run.
The soft obstacle is not a likely scenario, IMO. That relies on a deliberate bad choice by programmers. What's more likely is failure to recognize e.g. a child chasing a ball into the street by an empty lot where kids are playing or a dark skinned person at night. That will be blamed on, what? Insufficient training of AI or perhaps some software bug. Things that are really hard to control and are actually inevitable.
This was intended to be a reply to srynerson's long response to a post below. I can't seem to copy the text on my phone to put it where it belongs and delete this comment.
Waymo's operation is bit overrated. Their sensor tech is ahead on max speed, but they apparently don't have many cars on the road in SF - only about ~250 from what I can gather. Their full autonomous only car is going to be a Chinese import - call me bearish.
20. 20 is the magic number. Autonomous vehicles should all be mandated to never go faster than 20 mph in the city. If that is accomplished, they will be safe and trusted and successful. If they are allowed to compete with human drivers in speeding contests, they will quickly start breaking stuff and killing people and prompt a swift backlash.
"What would be better would be to hear clearer statements from taxi regulators and state transportation departments that they are going to hold driverless vehicles to some kind of objective and reasonable “safer than humans on average” standard rather than pulling the plug the first time something bad happens."
What kind of betting odds can I get that this will, in fact, not happen in the next 5 (maybe 10) years? I think I'd give you 20:1 at least.
Regulators are not typically rewarded for improving the status quo. If 20 people die in a freeway pile up tomorrow, absolutely no one will knock on the regulators' doors and say "You could have prevented this, if only you'd allowed those safer autonomous cars!"
On the other hand, if one person dies in a freak accident caused by an autonomous car, the regulator is probably getting their ass handed to them. This is probably true even if the autonomous car wasn't even really "at fault" (e.g., like a pedestrian who isn't looking jumping into fast traffic, too late for the car to dodge/brake).
Chapter one of Mankiw, I think: People respond to incentives.
The fact that Uber and Lyft exist is a big part of why we are not even considering buying a third car when my daughter starts driving. We only have two now because one is an old EV with limited range and we have to do a lot of long range driving for her sports. If Uber or Lyft was about 25% less expensive, I think that would be the tipping point for me to decide to sell both of our cars and invest in one longer range EV. I know my husband would use Uber or Lyft more if he didn't have to talk to drivers. Right now I still take Lyft about twice a month even though I have my own car. This is usually when I am going out for the evening. This is partly because the neighborhoods that I go out to have really expensive and/or hard to find parking and so just taking a vehicle that can drop me off saves me costs and frustration enough to make it worthwhile. Sometime it is a case of my going out to drinks for a girl's night with my neighbors. While I don't drink, carpooling in a Lyft is a low friction way of making sure none of my friends do either. If drank, I would be using it a lot more. I pretty much always take it to the airport. I actually think having Uber has made me take the bus more! Seeing out much it costs to take an Uber to downtown makes just hopping on the bus seem much more appealing. I think part of the problem with so many people having their own cars is that the costs are mostly sunk costs so that their is little motivation to use transit or walk for short trips because it won't save you money. (My husband was 100% counting on their being self-driving cars in 10 years about 16 years ago and the fact that we don't and my teen now has to learn to drive is pretty sad for him.)
Another aspect of the land use point is the number of cars that would be required in a self-driving cars world. The vast majority of cars today are not in use 97% of the time - they're parked and waiting for the owner (or renter) to need it. The number of passenger vehicles in circulation could be reduced by 90%+ if everyone could summon a self-driving car whenever they need it.
No one EVER talks about what to me is the most obvious and terrifying problem.
Literally everything gets hacked, from lazy workers leaving a laptop with WiFi IP addresses in a cafe to sophisticated groups actually finding backdoors.
Can you imagine a VIP driving at the mercy of a computer that can be hacked and controlled at will? Or alternatively, hack the car of the random guy driving alongside them to suddenly swerve left at 70mph...
And every time this is brought up the engineers dismiss it without any actual plan to address. As a biologist this is very similar to issues with mutagenicity and crispr therapy -no one looks for it because it's expensive to search and you don't want to learn the answer.
Your* car is online now and could be hacked. They could command the stability control to turn you in any direction, stop, accelerate, etc.
* any car built in the last 5-10 years.
My car is just over 10 years old but I still don't get what you mean.
They have computers but how are they connected to the internet? There's no central control of my breaking
Yeah cars have computers and an older car wouldn't have any internet connectivity. The modern ones do, but if the designers were smart, then hopefully there would be no physical connection at all between the internet connected computers used for infotainment, GPS and stuff, and the chips that control safety critical functions like ABS, ESP, etc.
There is central control of your braking that’s how your stability control works. As for connectivity:
https://www.nytimes.com/2024/04/23/technology/general-motors-spying-driver-data-consent.html
Ford, Honda, Toyota they all have their version.
Differential control braking has been around forever - it is literally the basis of the court case in My Cousin Vinny.
What does that have to do with the internet?
https://www.kbb.com/car-advice/electronic-stability-control/
“ Electronic stability control modulates the engine throttle and each wheel’s brake. Like other driver aids, ESC turns on when you start your car. The technology operates in the background, monitoring steering wheel activity and speed sensors on each wheel.
When sensors determine the driver is losing control, the ESC system engages automatically to adjust for oversteer and understeer. It applies or eases the brakes of each wheel on the car to bring the vehicle back onto the intended path. ESC involves the traction control systems, adjusts engine power to reduce speed, and applies brakes if the driver gives too much gas.
Did you even read your own article??? This is about onstar. Insurance companies also have plugins that can track your driving to give reduced rates. That is just collecting mechanical information from the car. There is no wifi control of anything. You don't seem to have any idea what you are talking about...
Again - https://www.kbb.com/car-advice/electronic-stability-control/
How does Waymo (and similar companies) handle younger passengers who require car seats? If the endgame of this tech is to allow people to ditch their own cars and just use driverless taxis for transportation, would people with young children just need to lug car seats to their destinations? Would there be a special class of vehicles that contain car seats?
On crashes, insurance, bankruptcy, and negligence laws hurt AV adaption. Commercial transportation is held to the highest standard of care under negligence laws, while private car drivers are need only exercise regular due care for their driving.
It's also much easier for a jury to stick Alphabet with a $50 million verdict over a "normal" car crash "to teach greedy corporations a lesson" that their AV caused versus sticking some warehouse worker or whoever with the same. The warehouse worker is also probably judgement proof.
Yes, this +10,000. I think about it constantly as a litigator. (What follows is not legal advice!!!)
I can understand completely why trucking companies, taxi companies, etc. would love self-driving vehicles. Under the doctrine of respondeat superior, a business is normally liable for the negligence of its employees while they are performing their job duties. Therefore, as long as a self-driving car is even marginally better safetywise on average than a moderately competent human driver, the company will probably come out ahead on savings for the cost of lawyers and judgments in negligent driving cases.
However, what I cannot understand -- and have never seen an article even try to address -- is why *manufacturers* would want to make self-driving cars, because it seems like it's going to deeply drag them into any litigation where their systems are potentially at issue.
For context right now, in many (if not most) car crash cases, the manufacturer isn't even named as a defendant and, if they are named, there's a good chance they get out on summary judgment or for a modest settlement, or if they are actually stuck going all the way to trial, they are likely to be assigned a very small portion of the fault for the crash. This is because in normal car crash litigation, the manufacturer is only going to have liability relating to the design and manufacture of the car, not anything about the operation of it. So, if the manufacturer can accurately testify, "We designed the braking system so that if the car was traveling X miles per hour and the brakes were applied for Y seconds, the car would come to a complete stop in Z feet, assuming normal road conditions," then as long as those figures are reasonable based on industry standards and available technology there shouldn't be any liability.
Those sorts of figures are subject to (at this point) well-understood and highly objective engineering, materials sciences, and physics analysis, such that it is difficult for a plaintiff's expert to make a lot of hay out of them. The fact that in the particular case at issue "Z feet" from the point where the brakes were first applied happened to be on the opposite side of the entire kindergarten class of St. Swithen's Academy for Children of Highly Motivated Personal Injury Attorneys is completely irrelevant -- that's the driver's problem, not the manufacturer's! The manufacturer had no knowledge (and could not be expected to have any knowledge) of the actual road conditions, obstacles, etc., let alone how fast the vehicle was actually traveling or how long the brakes were applied. All of that stuff is "externalized" to the driver.
But, with a self-driving car, that situation now becomes very decisively the manufacturer's problem and it involves a large amount of stuff that you can't even begin to hope to resolve on summary judgment because it's going to involve a bunch of questions about the reasonability of different assumptions that went into the programming of the car when considered under the actual circumstances of the collision. E.g., through its software, the manufacturer has to decide (and will be held at least partially responsible for) when the brakes were applied, how long they were applied for, whether any plausible evasive maneuvers might have avoided the crash, etc.
I mean, just imagine the first time a self-driving car splatters a toddler without even braking and, in the subsequent lawsuit, the engineers testify that the car was programmed to disregard anything it deemed to be a "soft" obstacle estimated to be less than 50 lbs so as to maximize driver safety from possible rear-end collisions that could be caused by sudden braking. If that happens in a state without wrongful death damages caps, you're going to see the judgment from orbit!
I think this also means the consequences of the tail risk - i.e. a horrendous accident where an autonomous vehicle smashes a school bus and kills 10 children - are much more commercially consequential than when caused by a negligent human driver. Because, okay, you can say that insurance pays out either way. And that's true, but in the latter case the human driver is put in jail and never drives again, and no one cares. The underwriter shrugs and acknowledges the risk was captured in the spread among hundreds of millions of drivers. In the autonomous vehicle case, the manufacturer has to try to renew their insurance premium and finds it has massively increased. And there are probably not going to be more than 5-10 AV manufacturers in the world for a very long time (perhaps only 2-3), and designs will probably be very similar - so insurance premia for all of them will jump up together, immediately.
I'm pretty certain I was replying to this when I posted at the root level. The awful software of substack won't allow me to copy the text to put it here and I'm not willing to retype it on my phone. Oh how hard it is to do what should be really really simple in software...... Did I mention that software is often do much harder than it should be? Because that was sort of a point of my post earlier which was intended to be a reply to this.
How would that be any different than a kid getting killed by a drunk Amazon driver where there are emails discussing the driver’s drinking problem?
I think you're not following my point because in your hypothetical Amazon is the employer, not the manufacturer of the vehicle. Employers are normally liable for the negligence of their employees while performing their job duties, although in your particular hypothetical that's actually not super clear because you've added the drunk part, which complicates the analysis, and then the knowledge part that complicates it further. (If I were Amazon's lawyer, I would probably tell them to write a very generous check because the whole thing would otherwise go to a jury and there would be a lot of bad publicity.) The *manufacturer* of the Amazon delivery truck, however, would be expected to have zero liability for a crash caused by the driver's intoxication.
Oh right - presumably a self driving semi would come with a $X,XXX/month subscription to cover liability, upgrades, tech support, etc.
Does a driverless car only take me to a prearranged destination, or can I change my mind mid-route?
Can I command "take a left at the next corner" or "drive around the block"?
I see personal value in the ability to hop in my car, point it in the direction I want to go...and go! And I like being in control. I went out of my way to buy a car with a stick shift - I don't want an automatic transmission, let alone a completely automatic car.
And, while we're at it, get off my lawn!
You pull up google maps and touch where you want go. Then get back to your glass of wine and your book.
I also prefer being in control, and I despise surveillance. I use Google Maps with "location services" (GPS) turned OFF. I enjoy reading (and using) a map. I don't want (or need) a map that reads me.
In the case of Tesla, I can envision some interesting integration opportunities between the robotaxi business and the EV charging business. Specifically, one could imagine big level 2 charging hubs set up, with the expectation that a driver pulls in, plugs in, and gets whisked away on a robotaxi to their actual destination. Then, a few hours later, they take the robotaxi back to the charging hub to pick up the car on the way home. At least, in theory, if the robotaxis were to ever become cheap enough, this could become a solution for how apartment dwellers with EVs quickly and affordably charge their cars.
(Of course, if robotaxis are this cheap, it begs the question of why people need to own cars at all; the answer to that question would be a combination of zero wait time for most trips, being able to leave stuff in the car, and not being geographically constrained based on where the robotaxis are certified to drive).
If we could build a driverless car that can handle any conditions - snow, random pedestrian behavior, police officer waving cars around an accident site - then I think it would have to be basically human-level intelligence. In that case, we've built AGI, and that can be put to much more transformative uses than just driving cars.