Apple co-founder: 'I've really given up' on Level 5

[ajdelange]

Remembering more about the politics. We were to have the system ready in time for Slick Willy's re-election campaign. We didn't. In fact I remember the FAA program manager commenting some 20 years later that GPS had been adopted by nearly every industry - except aviation.

 

[kaffine]

The issue with driver assist systems is they are making the human driver worse. If the driver doesn't have to provide input for 99.999% of the time what are the chances he is actually going to be alert and ready to take control that 0.001% of the time? That 0.001% of the time would be the part the human can handle better than the computer however it only works if the driver is alert and ready to take over. In reality the car will end up being self driving 99.999% of the time and crash the 0.001% of the time because the driver wasn't ready to take over in time.

The faster we can get to L4 and or L5 the better it will be. A long time between reaching L3 and L4 will increase the accidents.

 

[SageBrush]

He may be pessimistic, but you do have to wonder how the safety level of a fully autonomous car will compare to the safety of a car with a robust safety net, with a human driver. Imagine when all cars avoid 99% of run-off road incidents, 99% of head-on collisions, and detect and intervene 99% of the time if the driver becomes incapacitated...etc...

That adds a couple of 9s to the human driver safety level...and the bar has potentially been raised for the autonomous vehicle (depending on what people decide is an acceptable error rate). At some point, it comes down to what is the maximum number of fatalities that is allowed...400 per year in the US...is that low enough?

It'll be interesting to see how things evolve.

I see horrible drivers every day. Drunk driving is epidemic. The level 5 bar is very low

 

[AlanSubie4Life]

I see horrible drivers every day. Drunk driving is epidemic. The level 5 bar is very low

The bar has to be much safer than good (the best) human drivers. A lower bar would not make any sense.

In the end, it will likely be judged by the number of fatalities, and people are going to likely want to see at least a tenfold improvement over whatever the state of the art is at that time.

 

[SageBrush]

The bar has to be much safer than a good human driver. A lower bar would not make any sense.

Sure it does. Just think what fraction of accidents are from impaired and/or distracted drivers.
You are thinking of the 'I want a chauffeur to be cool' crowd but public safety is something else.

 

[AlanSubie4Life]

Sure it does. Just think what fraction of accidents are from impaired and/or distracted drivers.

Sure it is a lot. But if we made the L5 bar “much safer” than a drunk driver, it is entirely possible that a full level 5 fleet would then have a fatality rate much higher than what we have now. Just the way the math (could) work out. And again, we’ll need to see a significant reduction in the total number of fatalities (probably tenfold) for L5 to be acceptable.

And back to my original point - that improvement will probably need to persist even after non-autonomous vehicles receive all of the (likely mandated) active safety improvements. Those active safety improvements will likely make drunk drivers safer, though it paradoxically will also likely make drunk driving more common due to risk compensation.

 

[SageBrush]

^^ we don't ask 'designated drivers' to be much safer than a 'good driver' because we know that the designated driver is much more likely to be a whole lot safer than the drunk at the wheel.

Mediocre L5 can fill the same role for all the impaired and distracted drivers on the road

 

[AlanSubie4Life]

^^ we don't ask 'designated drivers' to be much safer than a 'good driver' because we know that the designated driver is much more likely to be a whole lot safer than the drunk at the wheel.

Mediocre L5 can fill the same role for all the impaired and distracted drivers on the road

Again, the problem with this argument is when you substitute a mediocre L5 in place of a good human driver. And then you look at the total weighted result.

Remember that the net required result is not simple parity with where we are currently at - even Tesla makes this very clear.

 

[SageBrush]

Again, the problem with this argument is when you substitute a mediocre L5 in place of a good human driver.

Again, that is not my suggestion.

 

[mswlogo]

AP is currently a driver assist system so by definition, it is not safer than a human alone. However, AP plus an attentive human driver is safer than a human driver alone.

Umm, let me guess where you got that huge assumption from.

Elon’s misleading skewed stats of accident rates per mile on AP vs all other car rates.

 

[AlanSubie4Life]

Again, that is not my suggestion.

I guess I am not sure exactly what your proposal is, or how it would work, exactly. We already have Uber and Lyft which have improved the statistics a bit, from what I understand. A lot (not most, not the majority) of the remaining drunk driving fatalities are probably in lower density areas not easily serviced by ride sharing - and they won’t be easily serviced by mediocre L5 fleets either (no economic sense). But what they will be serviced well by would be background active safety features on all vehicles (see above), likely mandated. Again, you have to really worry about risk compensation with such features so the more they stay in the background the better. But there is definitely an avenue for dramatic safety improvement with such systems - which will then raise the bar for L5!

 

[diplomat33]

Umm, let me guess where you got that huge assumption from.

Elon’s misleading skewed stats of accident rates per mile on AP vs all other car rates.

Actually, I was not thinking of Tesla's AP stats when I wrote that. But it seems to me to be basic common sense. Note I said "attentive driver". I am not talking about a careless driver who lets AP take over and is not paying attention. AP handles lane keeping, auto lane changes and adaptive cruise control very reliably now. So when you have AP that can shoulder some of those basic driving responsibilities and the driver is attentive and really paying attention to the "big picture", I think it just stands to reason that you would get safer driving than without AP.

But if you want stats, we have 4 confirmed deaths on AP and AP has been driven over 1 billion miles so far. That's an average of 1 death per 250 million miles which is 2.5 times better than the national average. So yes, I think the data would seem to indicate that AP and an attentive driver is safer than just a driver.

 

[AlanSubie4Life]

So when you have AP that can shoulder some of those basic driving responsibilities and the driver is attentive and really paying attention to the "big picture", I think it just stands to reason that you would get safer driving than without AP.

Reasonable hypothesis, but would need supporting data to know whether it is safer or not. Obviously there are equally reasonable arguments to be made for why it might be less safe.

 

[mswlogo]

Actually, I was not thinking of Tesla's AP stats when I wrote that. But it seems to me to be basic common sense. Note I said "attentive driver". I am not talking about a careless driver who lets AP take over and is not paying attention. AP handles lane keeping, auto lane changes and adaptive cruise control very reliably now. So when you have AP that can shoulder some of those basic driving responsibilities and the driver is attentive and really paying attention to the "big picture", I think it just stands to reason that you would get safer driving than without AP.

I understand where you are comping from. But almost any human will be lulled into complacency. The better AP gets the more likely that it will happen. It’s hard to stay attentive when there is nothing to do.

The only reason it works today is AP is bad enough to keep you engaged.

Self-Driving Uber Investigation Reveals Handoff Problem - CityLab

 

[diplomat33]

Reasonable hypothesis, but would need supporting data to know whether it is safer or not. Obviously there are equally reasonable arguments to be made for why it might be less safe.

As I later edited in my post, based on the 4 confirmed deaths on AP and the total mileage of AP, we can calculate an average of around 1 death per 250 million miles on AP which is 2.5 times better than the national average. I grant you that the stat is not perfect. There is some unknowns. We don't know how attentive drivers were over all those miles. Some of those miles are on AP1 and some are AP2, some are on NOA too. Also, a lot of those AP miles are probably highway miles which tend to be safer than city miles. But it is the best I can do with the information I can get on the internet since I am not privy to the detailed breakdown of AP stats. But it still seems to indicate that overall, AP is safer than the national average.

 

[mswlogo]

As I later edited in my post, based on the 4 confirmed deaths on AP and the total mileage of AP, we can calculate an average of around 1 death per 250 million miles on AP which is 2.5 times better than the national average. I grant you that the stat is not perfect. There is some unknowns. We don't know how attentive drivers were over all those miles. Some of those miles are on AP1 and some are AP2, some are on NOA too. Also, a lot of those AP miles are probably highway miles which tend to be safer than city miles. But it is the best I can do with the information I can get on the internet since I am not privy to the detailed breakdown of AP stats. But it still seems to indicate that overall, AP is safer than the national average.

The problem with that data is what are you comparing it to? We don’t have those stats. And the best stat to compare to would be miles driven on ACC (in all other vehicles) vs AP miles.

 

[AlanSubie4Life]

But it is the best I can do with the information I can get on the internet since I am not privy to the detailed breakdown of AP stats. But it still seems to indicate that overall, AP is safer than the national average.

It is most accurate to say we just do not know. This is different than saying it is not knowable! It definitely is knowable.

 

[mswlogo]

The problem with that data is what are you comparing it to? We don’t have those stats. And the best stat to compare to would be miles driven on ACC (in all other vehicles) vs AP miles.

We don’t even know if the average is low because the demographics of the drivers that buy Tesla’s is low. Or that on average the fleet of Tesla’s on average are newer cars.

 

[Kiwiadventure]

FSD has to be 100% correct at all times other wise steering wheel, brake pedal and accelerator has to stay and the person in the drivers seat is responsible for the car 100% when the car is on public roads.

 

[Uncle Paul]

Just because Woz has given up on Level 5 FSD, does not mean that others have not.

Full Autonomous driving will come. Debate is how soon, and in what form. It will apply to automobiles, shipping, trucking and so many other factors of our lives. Some day fire fighting will be done by robots, as will soldiers and security.