Steve Wozniak: AI Will Never Be Smart Enough to Drive a Car

This guy is dead wrong. It's already here. A highway populated entirely with Teslas driving themselves at speed would see no accidents.

I disagree. One airplane landing on the freeway would Fuck everything up.
 
No disrespect meant for Woz but taking his views on modern computers would be like listening to William Shockley's views on what CPUs wont be able to do.
 
This guy is dead wrong. It's already here. A highway populated entirely with Teslas driving themselves at speed would see no accidents.

Not with the Tesla's current code set. They can't react to something like a deer suddenly appearing in the middle of the road. Too much speed difference. They use the auto cruise routine for the auto brake as well. If there is too much speed difference, the auto cruise and therefore the auto brake is programmed to ignore since you shouldn't be matching speed with something like a deer, road divider or stopped fire truck. That is why you keep reading about Teslas running into motionless stuff.
 
I agree. Because AI doesn't factor emotion into decision making, its based on probability and percentages. iRobot. Great book.
emotion is ultimately based on probability and percentage from ppl's life accumulated experiences.
 
Snow/blackice/flooding... those can be accounted for trained for, and with the right sensors I can see cars scanning analyzing how deep the snow/water/etc is and saying "yes we can make it or, no road too dangerous finding another route". If the car can figure out it can't get through those things then thats all the excuse I need to not show up wherever I'm going.

Deer running in the road, or birds swooping down. AI or no AI, you either have a safe way to avoid it (which AI or humans could do) or you just hit it. There's already more than enough incidents of humans avoiding an animal on the road and injuring or killing themselves and others.

They give licenses to a wet sponge these days, despite what you may think there are people legally driving out there that are dumb as a brick and selfish as hell who have no business being behind the wheel. I like woz, but he's dead wrong with his statement. It's not about being smart enough to drive a car, humans have proven that. I'd happily sacrifice arriving a few minutes faster for a slower/safer trip where you don't have to worry about drunks, some fat whore texting, or a soccer mom blindly cutting off 3 lanes of traffic so she can make her exit despite her piss poor planning.
Here's a good AI test that a human is capable of. Understanding the difference between:

1. A deer, bear, or other animal running parallel to the road that could dash out at any moment, thus creating a situation where the human driver should be on guard, slow down, switch lanes, etc.
2. A cow running parallel to the road behind a wire fence, posing no risk whatsoever.

Or another scenario:

1. A dense clump of grass or mud on a backroad that can be safely ignored and driven through
2. A baby in a hoodie that's crawled out onto the same road.
3. A sleeping cat that will get out of the way.

I'm not saying AI hasn't come a long way, but you can have all the computational power in the world and AI is still going to be unable to comprehend some scenarios that could kill people, or if it errs on the side of caution, makes some situations completely undriveable for the AI.

I guess the way I see it, is I remember AI chat programs from back in the early 90s. If I compare them to AI chat programs now, the difference doesn't seem to be very significant. As others pointed out, AI has taken down chess and go. Both of those have rigid rules and a finite number of combinations. AI is fantastic for those scenarios. AI is also good at balancing robot bipeds or quadrapeds, since all it has to do is account for physics numbers of gyroscopes. Driving is different, because on some level, it has to comprehend what is happening. Some things it will get every time, some things it may still not understand decades from now.
 
I had to re-read that headline, thought I saw "Steve Wozniak will never be smart enough to drive a car."

The AI doesn't have to be perfect, just better than the average human driver. AI doesn't need to be able to handle every possible random obstacle - humans often fail, so the machine just needs to fail less. I think the real chance for this to reduce road deaths is by taking control away from tired, distracted, drunk, or otherwise impaired drivers.

If the AI is statistically better, insurance rates start dropping for people who use it and a new trend is started (for better or worse, depending on which side of the fence you're on with this one).

Regardless of the technical part of problem, there are still liability issues that bother me with this whole thing. "We will drive you but we aren't responsible for when shit happens" doesn't seem like a sustainable policy.
 
Last edited:
Never say never. It will happen eventually. Machine learning has come a long way these past few years, and it's only going to keep getting faster and smarter.

I suppose you also subscribe to the belief that a it is no big deal if people die while the AI is learning?

AI is programmed by the same human beings so many people claim are the cause of the accidents today. You think those folks are smart enough to figoure out how to program Ai to drive a car, but cannot drive a car themselves? Hehe. Rhetorical question.

As a programmer with some 30+ years under my belt I am going to agree with the Woz on this one. Yes, I left myself open to the, inevitable, "you must not be very good then" comment. Meh.
 
There's a truly massive body of evidence that humans will never be smart enough to drive a car, too.
 
  • Like
Reactions: DocNo
like this
I suppose you also subscribe to the belief that a it is no big deal if people die while the AI is learning?

AI is programmed by the same human beings so many people claim are the cause of the accidents today. You think those folks are smart enough to figoure out how to program Ai to drive a car, but cannot drive a car themselves? Hehe. Rhetorical question.

As a programmer with some 30+ years under my belt I am going to agree with the Woz on this one. Yes, I left myself open to the, inevitable, "you must not be very good then" comment. Meh.

Funny how people forget that tiny little detail, and think AI is a creation of its own that's far superior to anything humans could possibly do.
 
Last edited:
As a driver, how many times do you ask yourself "What in the hell are you doing?" about other drivers and react accordingly?

Can AI ask the same question?
 
When i see "the woz" i always think, so thats how that fat kid from the goonies grew up to look :) and then shortly after my brain go to a dark place, and i have a helluva fight getting it back to the level of operational order it have.
 
Even in snow, black ice, flooding, deer running out into the highway, birds swooping down, etc.


These aren't even vaguely a problem, at least not after the move from fully autonomous self-driving cars to only partially autonomous ones…


Snow:


Roads will be plowed, and where necessary salted and graveled in a precisely orchestrated priority order by automated plows. The whole system will be overseen by an AI that will KNOW when and how roads are used in order to make it super-efficient. It will also spot potential issues, including stalled cars that have people in them, stuff in the road that needs a human to remove, etc. and refer them to human monitors and workers to deal with.

In addition, the system will have so many plows available that there will NEVER be a road that needs plowing for more than a few minutes before it is plowed. How will this be managed? Simple, the system will have 20 times as many plows available as any county now, all of which will work half again as long as any of them do now, and it will all cost half as much as any given county pays now for plowing capabilities. How will they get all those plows? If you have a good 4x4 pickup and live in an area that gets lots of snow you will be paid to let the county use your vehicle as a plow. In the weeks before the first real snow is expected the system will drive your truck to a facility at a time you don't need it (while at work, or at night). Workers there will attach a plow, then the system will drive the truck back to you. When the snow comes it will plow the roads with your truck any time you don't need it... And if you don't need your truck specifically to get around, then it will pick you up in a shared vehicle, so your truck can keep working.

The only dedicated machines needed will be salt and gravel trucks, and even those might not be needed in most places, as it shouldn’t be too hard to build a system for that which could be attached to most larger pickups. You would get paid even more to let your truck be used for that purpose. In that case you would likely have permanent mounts attached that wouldn’t get in your way really. Then only after it started snowing the trucks assigned to this purpose would be driven to a yard to have the devices attached and loaded. When the roads looked like they would be good for a while they would be driven back and detached.

If you are going over a mountain, the chains you need will already be laid down for you by human workers that are told by a computer that knows your car is coming, what size chains to lay down and where to place them. This will be trivially easy for the workers, as they will have augmented reality glasses that will literally mark the exact spots on the ground for them, with orientation, exact borders, a code for the right chain to use, and a cart that follows them around carrying the chains. As you approach the chain-up area you will be asked if you want to finish the chain up yourself, or pay for someone to attach the chains. If you do it yourself a video will show you exactly what you need to do a couple minutes before you arrive. When you arrive, your car will drive precisely onto the chains in the spot designated for it. You will get out and chain it up and head over the mountain. On the other side the reverse will happen.


Black Ice:


Even easier. All self-driving vehicles will report road conditions back to a central AI for the County… As in the slightest slip of a tire will be reported. After just a year or two the AI will know EXACTLY what weather conditions cause black ice to form and where it will happen. As soon as the conditions are close to being met those areas will be salted. All cars will continue to report, and if signs of black ice forming appear there or anywhere else it will be dealt with long before it can cause any actual harm. Cars will also be rerouted if it can't be dealt with immediately for some reason.


Flooding:


Basically exactly the same as Black Ice. First, the AI will know when flooding is likely and will be monitoring road condition reports for signs of it. As soon as it detects it has started it will automatically redirect vehicles around the flood areas and report the issue to human monitors who can dispatch people to deal with it if necessary.


Deer:


Your car will automatically slow down to the proper speed for deer-heavy areas, so it will need less time to brake than any human would. If a deer jumps out on the road it will brake at least as well as any human, and it will steer around better than any human would. In addition, it will know whether another car is coming anytime vaguely soon, so it will know if it can safely swerve into the oncoming lane. It will also have years of deer behavior and road records to rely on to know exactly when/how to brake and when/how to swerve in the safest manner possible for everyone involved, including the deer. It will even intentionally clip a deer if that is the safest possible option for all humans at risk.

Also, except in very heavily forested areas where the deer can be two steps from the road without the cars being able to see them, it will know from prior cars going through that there are deer near the road and will slow down through that area even more. It will also intentionally adjust cars coming in both directions so that at no time do they pass each other (going in opposite directions) when a deer is near the road, thus ensuring there is more swerve space. They will also make sure no tailgating is taking place.


Birds swooping down:

I don’t quite understand what the concern is here. I have never had a bird swoop down in front of my car, but if an AI did, it would just ignore it. Unlike a human that would freak out, it would analyze the situation and just take the hit if that was safer than trying to swerve. If the windshield cracked the car would drive you to the nearest place that could replace the windshield, you would be taken to your destination by an Uber, and would continue to Uber around until your car just showed back up again.


Other general fixes to better deal with unexpected issues:


In addition to the above, for the safety of all, self-driving cars will also self-drive themselves in for oil changes, brake inspections, tire balancing, worn out and seasonal tire changes, necessary alignments, and any other maintenance you need. The cost of this will either be built into the purchase price, or possibly you will just have to pay for it out-of-pocket. Fortunately though, if you can’t afford it you can just let your car be used as an Uber to pay for maintenance. The point being that cars that aren’t in good condition won’t really be allowed on the roads, and maintaining condition will be easy because it will be based on sensors and automation.


When will this all happen? Not with the first self-driving cars. But as issues are identified with the fully-autonomous cars, way more additional sensors on cars, mandatory central reporting, and eventually mandatory central re-routing and control will be required for almost all vehicles. It could be anywhere between 25-50 years, but I am leaning towards no more than 30 before almost everything I described is in place… Or similar but better solutions, because I can’t think of everything. Also, not long after that basically nobody in cities, and eventually even few people in rural areas, will actually own their own car. The central AI will route Uber’s to where they are needed based on historical travel data and when you need a car you will just walk outside and it will already be there.
 
Last edited:
Not with the Tesla's current code set. They can't react to something like a deer suddenly appearing in the middle of the road.
Too be fair if you're traveling at freeway speeds and a deer suddenly jumps out in the freeway, how many people will be able to not hit it (or take out any cars around them if they swerve?) And at the end of the day isn't the whole idea that "AI driving" should at the very least be as good as a human?
 
AI would need to be able to apply the "what if" reasoning and intuition that drivers use every day when behind the wheel of their car. It is decades away from that level. It should not be applied to replacing a human driver until it does. While I am a pretty bleeding edge kinda guy, I certainly wouldn't trust my life with an AI-piloted car today.
 
These aren't even vaguely a problem, at least not after the move from fully autonomous self-driving cars to only partially autonomous ones…


Snow:


Roads will be plowed, and where necessary salted and graveled in a precisely orchestrated priority order by automated plows. The whole system will be overseen by an AI that will KNOW when and how roads are used in order to make it super-efficient. It will also spot potential issues, including stalled cars that have people in them, stuff in the road that needs a human to remove, etc. and refer them to human monitors and workers to deal with.

In addition, the system will have so many plows available that there will NEVER be a road that needs plowing for more than a few minutes before it is plowed. How will this be managed? Simple, the system will have 20 times as many plows available as any county now, all of which will work half again as long as any of them do now, and it will all cost half as much as any given county pays now for plowing capabilities. How will they get all those plows? If you have a good 4x4 pickup and live in an area that gets lots of snow you will be paid to let the county use your vehicle as a plow. In the weeks before the first real snow is expected the system will drive your truck to a facility at a time you don't need it (while at work, or at night). Workers there will attach a plow, then the system will drive the truck back to you. When the snow comes it will plow the roads with your truck any time you don't need it... And if you don't need your truck specifically to get around, then it will pick you up in a shared vehicle, so your truck can keep working.

The only dedicated machines needed will be salt and gravel trucks, and even those might not be needed in most places, as it shouldn’t be too hard to build a system for that which could be attached to most larger pickups. You would get paid even more to let your truck be used for that purpose. In that case you would likely have permanent mounts attached that wouldn’t get in your way really. Then only after it started snowing the trucks assigned to this purpose would be driven to a yard to have the devices attached and loaded. When the roads looked like they would be good for a while they would be driven back and detached.

If you are going over a mountain, the chains you need will already be laid down for you by human workers that are told by a computer that knows your car is coming, what size chains to lay down and where to place them. This will be trivially easy for the workers, as they will have augmented reality glasses that will literally mark the exact spots on the ground for them, with orientation, exact borders, a code for the right chain to use, and a cart that follows them around carrying the chains. As you approach the chain-up area you will be asked if you want to finish the chain up yourself, or pay for someone to attach the chains. If you do it yourself a video will show you exactly what you need to do a couple minutes before you arrive. When you arrive, your car will drive precisely onto the chains in the spot designated for it. You will get out and chain it up and head over the mountain. On the other side the reverse will happen.


Black Ice:


Even easier. All self-driving vehicles will report road conditions back to a central AI for the County… As in the slightest slip of a tire will be reported. After just a year or two the AI will know EXACTLY what weather conditions cause black ice to form and where it will happen. As soon as the conditions are close to being met those areas will be salted. All cars will continue to report, and if signs of black ice forming appear there or anywhere else it will be dealt with long before it can cause any actual harm. Cars will also be rerouted if it can't be dealt with immediately for some reason.


Flooding:


Basically exactly the same as Black Ice. First, the AI will know when flooding is likely and will be monitoring road condition reports for signs of it. As soon as it detects it has started it will automatically redirect vehicles around the flood areas and report the issue to human monitors who can dispatch people to deal with it if necessary.


Deer:


Your car will automatically slow down to the proper speed for deer-heavy areas, so it will need less time to brake than any human would. If a deer jumps out on the road it will brake at least as well as any human, and it will steer around better than any human would. In addition, it will know whether another car is coming anytime vaguely soon, so it will know if it can safely swerve into the oncoming lane. It will also have years of deer behavior and road records to rely on to know exactly when/how to brake and when/how to swerve in the safest manner possible for everyone involved, including the deer. It will even intentionally clip a deer if that is the safest possible option for all humans at risk.

Also, except in very heavily forested areas where the deer can be two steps from the road without the cars being able to see them, it will know from prior cars going through that there are deer near the road and will slow down through that area even more. It will also intentionally adjust cars coming in both directions so that at no time do they pass each other (going in opposite directions) when a deer is near the road, thus ensuring there is more swerve space. They will also make sure no tailgating is taking place.


Birds swooping down:

I don’t quite understand what the concern is here. I have never had a bird swoop down in front of my car, but if an AI did, it would just ignore it. Unlike a human that would freak out, it would analyze the situation and just take the hit if that was safer than trying to swerve. If the windshield cracked the car would drive you to the nearest place that could replace the windshield, you would be taken to your destination by an Uber, and would continue to Uber around until your car just showed back up again.


Other general fixes to better deal with unexpected issues:


In addition to the above, for the safety of all, self-driving cars will also self-drive themselves in for oil changes, brake inspections, tire balancing, worn out and seasonal tire changes, necessary alignments, and any other maintenance you need. The cost of this will either be built into the purchase price, or possibly you will just have to pay for it out-of-pocket. Fortunately though, if you can’t afford it you can just let your car be used as an Uber to pay for maintenance. The point being that cars that aren’t in good condition won’t really be allowed on the roads, and maintaining condition will be easy because it will be based on sensors and automation.


When will this all happen? Not with the first self-driving cars. But as issues are identified with the fully-autonomous cars, way more additional sensors on cars, mandatory central reporting, and eventually mandatory central re-routing and control will be required for almost all vehicles. It could be anywhere between 25-50 years, but I am leaning towards no more than 30 before almost everything I described is in place… Or similar but better solutions, because I can’t think of everything. Also, not long after that basically nobody in cities, and eventually even few people in rural areas, will actually own their own car. The central AI will route Uber’s to where they are needed based on historical travel data and when you need a car you will just walk outside and it will already be there.


Didn't read a single word.

srs.

dSDotWR.gif
 
I suppose you also subscribe to the belief that a it is no big deal if people die while the AI is learning?

AI is programmed by the same human beings so many people claim are the cause of the accidents today. You think those folks are smart enough to figoure out how to program Ai to drive a car, but cannot drive a car themselves? Hehe. Rhetorical question.

As a programmer with some 30+ years under my belt I am going to agree with the Woz on this one. Yes, I left myself open to the, inevitable, "you must not be very good then" comment. Meh.

I subscribe to this belief:

A person is smart. People are dumb, panicky dangerous animals and you know it.

As for people dying while the AI is learning... hasn't that been an inherent part of technological progress? How many people died while we were learning to build planes? How about going to space? How many people are dying on the roads now because people are dumb, panicky dangerous animals? If the AI reduces traffic fatalities and accidents by only 10%, I would say we're off to a good start.
 
I respect The Woz a lot. In this case however I think he greatly over estimates the intelligence of his fellow man.

Human intelligence leads people too a lot of very deadly errors... such as the tendency for people to steer directly into things they intend to miss when they unexpectedly end up on a shoulder ect. Every first responder has multiple stories of finding people wrapped around trees in the middle of an otherwise empty field.

There are cases where the simple 0/1 logic of AI will be far superior and save lives.

And, there will always be cases where an AI will find an unresolvable problem. AI's don't have ethics, only subroutines. what is the AI solution for a kid that runs in front of a car? most people will swerve a car into a tree before they would hit a kid. how do you program that? How about a dog? do you run over a dog, or run into the ditch?

I think he's right for the near future, AI assisted driving should be a thing. computers react much faster, and can warn and intervene much more effectively in an emergency. but they have a long way to go before they should be turned loose on public roads.
 
No disrespect meant for Woz but taking his views on modern computers would be like listening to William Shockley's views on what CPUs wont be able to do.

Right, dude's been dead for nearly 30 years, you think if he was still alive he would not have kept up? implying that someone that was a pioneer in a field can't figure shit out because he's old? wow. The ghost of Stephen Hawking would like a word with you sir.
 
These aren't even vaguely a problem, at least not after the move from fully autonomous self-driving cars to only partially autonomous ones…


Snow:


Roads will be plowed, and where necessary salted and graveled in a precisely orchestrated priority order by automated plows. The whole system will be overseen by an AI that will KNOW when and how roads are used in order to make it super-efficient. It will also spot potential issues, including stalled cars that have people in them, stuff in the road that needs a human to remove, etc. and refer them to human monitors and workers to deal with.

In addition, the system will have so many plows available that there will NEVER be a road that needs plowing for more than a few minutes before it is plowed. How will this be managed? Simple, the system will have 20 times as many plows available as any county now, all of which will work half again as long as any of them do now, and it will all cost half as much as any given county pays now for plowing capabilities. How will they get all those plows? If you have a good 4x4 pickup and live in an area that gets lots of snow you will be paid to let the county use your vehicle as a plow. In the weeks before the first real snow is expected the system will drive your truck to a facility at a time you don't need it (while at work, or at night). Workers there will attach a plow, then the system will drive the truck back to you. When the snow comes it will plow the roads with your truck any time you don't need it... And if you don't need your truck specifically to get around, then it will pick you up in a shared vehicle, so your truck can keep working.

The only dedicated machines needed will be salt and gravel trucks, and even those might not be needed in most places, as it shouldn’t be too hard to build a system for that which could be attached to most larger pickups. You would get paid even more to let your truck be used for that purpose. In that case you would likely have permanent mounts attached that wouldn’t get in your way really. Then only after it started snowing the trucks assigned to this purpose would be driven to a yard to have the devices attached and loaded. When the roads looked like they would be good for a while they would be driven back and detached.

If you are going over a mountain, the chains you need will already be laid down for you by human workers that are told by a computer that knows your car is coming, what size chains to lay down and where to place them. This will be trivially easy for the workers, as they will have augmented reality glasses that will literally mark the exact spots on the ground for them, with orientation, exact borders, a code for the right chain to use, and a cart that follows them around carrying the chains. As you approach the chain-up area you will be asked if you want to finish the chain up yourself, or pay for someone to attach the chains. If you do it yourself a video will show you exactly what you need to do a couple minutes before you arrive. When you arrive, your car will drive precisely onto the chains in the spot designated for it. You will get out and chain it up and head over the mountain. On the other side the reverse will happen.


Black Ice:


Even easier. All self-driving vehicles will report road conditions back to a central AI for the County… As in the slightest slip of a tire will be reported. After just a year or two the AI will know EXACTLY what weather conditions cause black ice to form and where it will happen. As soon as the conditions are close to being met those areas will be salted. All cars will continue to report, and if signs of black ice forming appear there or anywhere else it will be dealt with long before it can cause any actual harm. Cars will also be rerouted if it can't be dealt with immediately for some reason.


Flooding:


Basically exactly the same as Black Ice. First, the AI will know when flooding is likely and will be monitoring road condition reports for signs of it. As soon as it detects it has started it will automatically redirect vehicles around the flood areas and report the issue to human monitors who can dispatch people to deal with it if necessary.


Deer:


Your car will automatically slow down to the proper speed for deer-heavy areas, so it will need less time to brake than any human would. If a deer jumps out on the road it will brake at least as well as any human, and it will steer around better than any human would. In addition, it will know whether another car is coming anytime vaguely soon, so it will know if it can safely swerve into the oncoming lane. It will also have years of deer behavior and road records to rely on to know exactly when/how to brake and when/how to swerve in the safest manner possible for everyone involved, including the deer. It will even intentionally clip a deer if that is the safest possible option for all humans at risk.

Also, except in very heavily forested areas where the deer can be two steps from the road without the cars being able to see them, it will know from prior cars going through that there are deer near the road and will slow down through that area even more. It will also intentionally adjust cars coming in both directions so that at no time do they pass each other (going in opposite directions) when a deer is near the road, thus ensuring there is more swerve space. They will also make sure no tailgating is taking place.


Birds swooping down:

I don’t quite understand what the concern is here. I have never had a bird swoop down in front of my car, but if an AI did, it would just ignore it. Unlike a human that would freak out, it would analyze the situation and just take the hit if that was safer than trying to swerve. If the windshield cracked the car would drive you to the nearest place that could replace the windshield, you would be taken to your destination by an Uber, and would continue to Uber around until your car just showed back up again.


Other general fixes to better deal with unexpected issues:


In addition to the above, for the safety of all, self-driving cars will also self-drive themselves in for oil changes, brake inspections, tire balancing, worn out and seasonal tire changes, necessary alignments, and any other maintenance you need. The cost of this will either be built into the purchase price, or possibly you will just have to pay for it out-of-pocket. Fortunately though, if you can’t afford it you can just let your car be used as an Uber to pay for maintenance. The point being that cars that aren’t in good condition won’t really be allowed on the roads, and maintaining condition will be easy because it will be based on sensors and automation.


When will this all happen? Not with the first self-driving cars. But as issues are identified with the fully-autonomous cars, way more additional sensors on cars, mandatory central reporting, and eventually mandatory central re-routing and control will be required for almost all vehicles. It could be anywhere between 25-50 years, but I am leaning towards no more than 30 before almost everything I described is in place… Or similar but better solutions, because I can’t think of everything. Also, not long after that basically nobody in cities, and eventually even few people in rural areas, will actually own their own car. The central AI will route Uber’s to where they are needed based on historical travel data and when you need a car you will just walk outside and it will already be there.

Snow: Your puny 4x4 can't plow real snow (real snow uses road construction size equipment) and there's no way in hell a municipality will double anything.

Birds swooping: You're thinking crow, we're thinking turkeys. Those fat fucks cross the road by gliding at a 45 from a tall tree on one side to the other and they'll do more than crack the windshield if you hit them at speed.

Honestly, if you are programming cars based on your city driving experience it's going to be much worse than what we have now.
 
And, there will always be cases where an AI will find an unresolvable problem. AI's don't have ethics, only subroutines. what is the AI solution for a kid that runs in front of a car? most people will swerve a car into a tree before they would hit a kid. how do you program that? How about a dog? do you run over a dog, or run into the ditch?

I think he's right for the near future, AI assisted driving should be a thing. computers react much faster, and can warn and intervene much more effectively in an emergency. but they have a long way to go before they should be turned loose on public roads.

Sounds like a Will Smith movie. :) Perhaps we need better AI nannies to solve that issue.

I joke but AI can be programmed with whatever ethics we like. AI can be trained to know the difference between a kid and a bag of garbage or whatever else could end up on the road that the car would prefer to run over instead of steering into something solid.

AI driving is always going to be about the numbers. Over 40 thousand people die in the US alone every year in some form of car accident. If AI takes over say 50% of the driving drops that to 20-25 thousand but in 10 cases a year perhaps you could argue the AI choose poorly and killed a driver or a pedestrian (I say could argue because its also possible a human behind the wheel would have made the same mistake... people do hit kids all the time) how is that not an improvement.

I guess I'm saying people are going to die in car accidents as long as we have cars. It really doesn't matter who is driving them. Right now in the US there are 40 thousand deaths a year... and 2.5 Million injuries. Those are all caused by human error every single one of them. The majority of them can be directly tied to drivers not paying enough attention... its why 1/3 of all car accidents happen within a mile of victims homes. AI won't have that problem... its logical to believe that based on that alone AI should be able to reduce accidents by 1/3, so 1.7 million or so injuries avoided and 27 million aprox deaths.

So for every pissed off Cop with a Robot Limb with an ax to grind with AI... there will be a few million other people at home who would have otherwise had funerals. :)
 
Sounds like a Will Smith movie. :) Perhaps we need better AI nannies to solve that issue.

I joke but AI can be programmed with whatever ethics we like. AI can be trained to know the difference between a kid and a bag of garbage or whatever else could end up on the road that the car would prefer to run over instead of steering into something solid.

AI driving is always going to be about the numbers. Over 40 thousand people die in the US alone every year in some form of car accident. If AI takes over say 50% of the driving drops that to 20-25 thousand but in 10 cases a year perhaps you could argue the AI choose poorly and killed a driver or a pedestrian (I say could argue because its also possible a human behind the wheel would have made the same mistake... people do hit kids all the time) how is that not an improvement.

I guess I'm saying people are going to die in car accidents as long as we have cars. It really doesn't matter who is driving them. Right now in the US there are 40 thousand deaths a year... and 2.5 Million injuries. Those are all caused by human error every single one of them. The majority of them can be directly tied to drivers not paying enough attention... its why 1/3 of all car accidents happen within a mile of victims homes. AI won't have that problem... its logical to believe that based on that alone AI should be able to reduce accidents by 1/3, so 1.7 million or so injuries avoided and 27 million aprox deaths.

So for every pissed off Cop with a Robot Limb with an ax to grind with AI... there will be a few million other people at home who would have otherwise had funerals. :)

Choosing between killing the driver or a pedestrian isn't a mistake, it's an ethical dilemma. it's one of those things that will have to be worked out.
 
The funny thing is, you don't even have to be smart to drive a car, and most people still fail at it daily. No, I don't care that you haven't been in an accident. You make mistakes. I know I did make mistakes, even if they didn't result in accidents. I've narrowly avoided quite a few accidents, because I wasn't alert enough. An AI would never loose concentration and always pays 100% undivided attention to the driving itself.
 
Choosing between killing the driver or a pedestrian isn't a mistake, it's an ethical dilemma. it's one of those things that will have to be worked out.

No doubt. When the AI gets smart enough to know if its a kid or an evil fat man standing on the trolly track someone is going to have to give it some form of ethical choice routine.

Of course we could just let the AI figure that out for itself... what is the worse that could happen. :)

 
Snow: Your puny 4x4 can't plow real snow (real snow uses road construction size equipment) and there's no way in hell a municipality will double anything.

Birds swooping: You're thinking crow, we're thinking turkeys. Those fat fucks cross the road by gliding at a 45 from a tall tree on one side to the other and they'll do more than crack the windshield if you hit them at speed.

Honestly, if you are programming cars based on your city driving experience it's going to be much worse than what we have now.
Yeah somebody else gets it. Also don't forget muddy, snowy, or icy roads on hills where the ONLY way you'll get up the hill on the other side is to go full speed going down to maintain your momentum.
 
Yeah somebody else gets it. Also don't forget muddy, snowy, or icy roads on hills where the ONLY way you'll get up the hill on the other side is to go full speed going down to maintain your momentum.

And you guys are still missing the point. Most accidents and road deaths occur during normal road conditions!

I think a lot of the tension around this is in the idea that the auto drive should be able to handle every possible situation better than a human would - it doesn't need to do this to be a statistically better driver overall.

I don't see this being much use up in the dirt anyways. Statistically, its not needed there. The cities need this tech, that's where the population is.
 
  • Like
Reactions: DocNo
like this
AI is already smart enough to drive a car but only if it's not exposed to human stupidity.
 
Too be fair if you're traveling at freeway speeds and a deer suddenly jumps out in the freeway, how many people will be able to not hit it (or take out any cars around them if they swerve?) And at the end of the day isn't the whole idea that "AI driving" should at the very least be as good as a human?

Judging by the number of dead deer and damaged cars, not many. But IMO, that is where Tesla and others have failed. Instead of jumping direct to 'Autopilot', why not start with IR sensors that can detect deer and others by the side of the road and project target markers on some kind of HUD and let the human figure it out. Do this for a few years while they hone the target identification algorithms. When you can tell the difference between a deer, human, bicycle, human pushing a bicycle, firetruck in the middle of the road, a bouncing wind driven construction barrel, etc, then move on to Autobrake and Autopilot. The test Uber car that hit the bicyclist saw her in plenty of time to simply steer around but instead kept trying to fully ID the target until the collision. A human doing what she should have been doing instead of watching her tablet would likely instinctively steered around the large moving shape without needing to know exactly what she was avoiding.
 
And you guys are still missing the point. Most accidents and road deaths occur during normal road conditions!

I think a lot of the tension around this is in the idea that the auto drive should be able to handle every possible situation better than a human would - it doesn't need to do this to be a statistically better driver overall.

I don't see this being much use up in the dirt anyways. Statistically, its not needed there. The cities need this tech, that's where the population is.
No, we're not missing the point, we're realizing that the standard it's going to be held to will be as high as that of a human. If a human fucks up and kills somebody due to incompetence, they go to jail. If an AI vehicle fucks up and kills somebody due to a bad decision, ALL AI cars risk being banned due to how our politics and legal system works.

It may not matter if the AI does a better job on AVERAGE of protecting passengers. If it does a WORSE job of protecting passengers than most human drivers in ANY SPECIFIC SITUATION, that could threaten them as a whole.
 
No, we're not missing the point, we're realizing that the standard it's going to be held to will be as high as that of a human. If a human fucks up and kills somebody due to incompetence, they go to jail. If an AI vehicle fucks up and kills somebody due to a bad decision, ALL AI cars risk being banned due to how our politics and legal system works.

It may not matter if the AI does a better job on AVERAGE of protecting passengers. If it does a WORSE job of protecting passengers than most human drivers in ANY SPECIFIC SITUATION, that could threaten them as a whole.

That's not a technical problem, more of a legal liability issue - which I think is actually the bigger problem in this whole mess. Abstracting responsibility typically ends badly, but the buck has to stop somewhere - and I completely agree that the current liability waivers are flawed. To be perfectly clear, I do not like the idea of giving up control to a machine which I have no clue how it works - but I also have no clue what is going on in most people's heads while they're driving, and I tend to trust a well built machine more than unreliable humans - present company included :)

But what I'm getting at with the system only needing to be better on average is the way market forces will dictate this stuff on us. You are right - there will be specific situations where the auto drive will do a worse job - but if there are more situations where it does as good or better, this stuff will be forced on us - like seat belt laws. If you choose to forgo using the driver "assists" - be prepared to pay more on your insurance.
 
No, we're not missing the point, we're realizing that the standard it's going to be held to will be as high as that of a human. If a human fucks up and kills somebody due to incompetence, they go to jail. If an AI vehicle fucks up and kills somebody due to a bad decision, ALL AI cars risk being banned due to how our politics and legal system works.

It may not matter if the AI does a better job on AVERAGE of protecting passengers. If it does a WORSE job of protecting passengers than most human drivers in ANY SPECIFIC SITUATION, that could threaten them as a whole.

Laws are going to have to change, simple as that. Trying to hold the AIs to the same laws as humans isn't going to work because their function is fundamentally different.
 
Snow, Ice, seriously guys do you know what happens in snow and ice with human drivers? 100 car pile ups and dead people that's what. The automonous car might actually drive at the safe speed and the drivers might actually not mind it as they can surf the web.
What about pedestrians? Again about 6000 of them get killed each year by human drivers.
Deer? Again animal related accidents cause about 25000 injuries

I get the problems wozniak is worried about but the way I view self driving cars is just like most other things in life, when a new technology or lifestyle comes we humans use our intelligence to adapt. I don't see autonomous cars as some amazing cure all but I do think they will do way better than humans in the long run. They have one major avantage they never get tired or distracted or impatient. Lets say a human comes across a train track what do they do? The human says hmm since I was a wee child I was taught that this is a dangerous thing I should pay attention and stay away from it. Some humans disregard this and get killed but for the most part humans take extra caution around train tracks. I just see autonomous cars as a compromise now humans will start giving the road a little more respect and the best part is if they get ran over no one will take it personally any more lol. They wont be saying wow that driver was a real asshole for not watching out for me jetting out into the road, they will say well you weren't careful near an automated car.
 
  • Like
Reactions: DocNo
like this
I suppose you also subscribe to the belief that a it is no big deal if people die while the AI is learning?

AI is programmed by the same human beings so many people claim are the cause of the accidents today. You think those folks are smart enough to figoure out how to program Ai to drive a car, but cannot drive a car themselves? Hehe. Rhetorical question.

As a programmer with some 30+ years under my belt I am going to agree with the Woz on this one. Yes, I left myself open to the, inevitable, "you must not be very good then" comment. Meh.

Funny how people forget that tiny little detail, and think AI is a creation of its own that's far superior to anything humans could possibly do.

The thing both of you miss should be so obvious it's a no-brainer. Something doesn't need to be better in every regard to be consistently better in most regards. It's much easier to get a good set of safer, consistent behaviors built into a machine than a human.
 
  • Like
Reactions: DocNo
like this
The thing both of you miss should be so obvious it's a no-brainer. Something doesn't need to be better in every regard to be consistently better in most regards. It's much easier to get a good set of safer, consistent behaviors built into a machine than a human.

Oh please.

The thing you miss which should be so obvious it's a no-brainer, is this. You going to be happy becuase you daughter gets killed by this machine because it got into a situation where either a) it did not know what to do, or b) chose to kill her because it was statistically safer to do so to protect the driver?

In this litigious society, you are being very naive if you think people are going to just sit by and watch machines kill people where there were far more options available that the machine never considered.

There is no way a Mother is going to sit back and grin saying, "It's okay, I can make another baby." after hers was killed by a car, that chose that path because it would statistically do less damage to the driver. How many of those are acceptable to you, as it is statistically a very probable scenario? Volvo engineers have already said, if the choice is to hit a human or a wall, the car will chose the human, every time (even when there may be other possibilities the car could not consider due to artificial limits placed on handling of the car).

What about being the passenger in a car that chose to kill someone to prevent you from being hurt? You are going to be okay with that? I would rather hit that wall and take my chances with the air bags.
 
Last edited:
And you guys are still missing the point. Most accidents and road deaths occur during normal road conditions!

I think a lot of the tension around this is in the idea that the auto drive should be able to handle every possible situation better than a human would - it doesn't need to do this to be a statistically better driver overall.

I don't see this being much use up in the dirt anyways. Statistically, its not needed there. The cities need this tech, that's where the population is.

Except the only way this tech works is if you are all in. Once you train drivers to turn off their brains and let the computer handle it they will not be prepared to take over when it's necessary. And country roads is exactly where you want this tech because highway hypnosis is worse on dark dirt-tracks in the snow. City people shouldn't need to drive at all, they have the bandwidth to go full telecommuting but governments never somehow want to commit to that even though it would be the greenest option. I wonder which vested interest is behind that?
 
How about we just properly train drivers instead of giving anyone with a pulse a license? Or how about have real consequences for doing bad things in a car?
 
Back
Top