Uber Troubles For Uber

[LordFerret]

https://www.forbes.com/sites/davidkiley5/2018/03/19/the-first-pedestrian-fatality-with-an-autonomous-vehicle-could-set-tone-for-lawyers-and-liability/#4bc93c5233fd

So an Uber autonomous car has struck and killed a pedestrian. This brings a lot to question about AI.

It was the topic of tonight's coffee discussion;

How does AI determine when presented the situation of having to chose the life of a 7 year old vs a 70 year old... swerving and dodging that (dog, cat, squirrel, cow, ???) running out into the street vs that oncoming bus in the opposite lane... or that instant where it's your life or the life in the opposite oncoming vehicle... or the AI's own life over yours? Now imagine that AI is truly sentient; Ask that last question again.

Will we have an option in the ownership of such an autonomous vehicle to pre-set a determination, like; A) take my life instead, B) screw that, take his life ... and of course, make sure my option selection is permanently deleted the instant after such an accident, so nobody will ever know what I've picked... lawsuits and all that.

Can you picture an Uber cruising down the GSP (our Garden State Parkway, famously state-owned toll-road highway) doing the legal posted speed limit, 65 mph... while all the rest of humanity blows by doing 80mph, as is standard practice these days. Talk about road hazards.

 

His parting thought;

Quote

It came to me while I was reading the history of the life of Desiderio Erasmus.  This week Steven Hawking died.  One of his predictions was that machines will replace Mankind.  I don't know if that will happen or not, but it won't be machines that become sentient, it will be that Mathematics that becomes sentient.  So, as Prometheus was to mankind, mankind will be to mathematics.  Math will be able to do itself.

So worthy was the conversation, I had two cups of coffee and a chocolate-enrobed donut.

[Kerbart]

It will be a feast for lawyers. We have no exact knowledge of what goes on inside a neural network.Of course Uber can claim innocence with "it's not us, we don't know how its brain works" but that means admitting it sends vehicles on the road of which it doesn't know how they work. Reality is of course a bit more complicated but that's what it comes down to, really, the various parties passing the hot potato around on who messed up. It'll be an interesting court case.

[XB-70A]

This problem has been present for a long time, but the different reasoning and conceptions of ethics from all the societies make it a difficult point to deal with. However, the current over-automation will probably allow to consider new protocols.

A bit out of topic, but I fall on this video yesterday, and found a bit fun:

 

The case of these kind of client help robots is different and a bit ridiculous, not that they are running under popular OS (if I'm remembering well the MQ-9s ground controls are running under Linux),  but simply that their manufacturers are pretty slow to consider a service pack, or even a patch.

Edited March 20, 2018 by XB-70A

[tater]

Uber has had customers abducted and assaulted by drivers. A quick google shows 2 fatal crashes by Uber (meat driver) cars in one year in Milwaukee alone. One killed 3 people. How many miles were driven by Ubers in Milwaukee per fatality? How's self-driving doing in Tempe by comparison (fatalities per mile)?

That;s really the only thing that matters, deaths per mile driven. If driverless is lower, it's still better than meat drivers.

[kerbiloid]

So, AI is adult now.

[Shpaget]

There are too few details. It could be a case of the car not seeing the woman (which raises the question what was the safety guy behind the wheel doing), it could be mechanical failure of steering system where safety driver couldn't correct, or it could be a suicide by car, in which case autonomy of the car plays little role in the incident.

Edited March 20, 2018 by Shpaget

[kerbiloid]

https://www.theverge.com/2018/3/19/17140936/uber-self-driving-crash-death-homeless-arizona

As she was pushing a bicyle across the street, unlikely this was a suicide.
Maybe AI was shocked "what is this: two wheels, two legs, moves across the road?"

[YNM]

IMO the mistake ultimately falls down to the human overseer on-board during incident. (and the employer ofc.)

 

I mean, given the crash, what difference there is between a Cruise Control (which has been around for ages) and an AI that ran down someone ? It's just something that informs and ease the human driver, not replace said human driver.

 

It's also IMO a good metaphor of just how deadly cars are. Get a train !

 

2 hours ago, Shpaget said:

... or it could be a suicide by car...

You're still guilty - drivers are meant not only to watch what's directly in the front, but also what's around, even in the back and behind. You have to be cautious when driving - assuming you have a clear path even when the tiniest perceivable chance exist could be the end of your driving license.

Obviously, people committing suicide by falling off a bridge down a highway won't be anyone's mistake down the highway - that's impossible to tell and to prevent (once happening).

Spoiler

Not meant for road users nor it's addressed to the driver - but you know if you're the driver you'd still feel very, very, very guilty.

 

Edited March 20, 2018 by YNM

[Bill Phil]

It was inevitable, really. The question isn't can we stop all deaths, it's whether or not we can lower the rate at which they occur.

Does make me wonder about the safety driver though...

[kerbiloid]

And about necessity of the safety driver.
Monotonous, dull observing of robot work could seduce anybody.

[Shpaget]

12 hours ago, kerbiloid said:

https://www.theverge.com/2018/3/19/17140936/uber-self-driving-crash-death-homeless-arizona

As she was pushing a bicyle across the street, unlikely this was a suicide.
Maybe AI was shocked "what is this: two wheels, two legs, moves across the road?"

When I first read the story in the morning it was devoid of any details. Apparently not a suicide (not that I thought it was, just considered it one of many possibilities).

11 hours ago, YNM said:

You're still guilty - drivers are meant not only to watch what's directly in the front, but also what's around, even in the back and behind. You have to be cautious when driving - assuming you have a clear path even when the tiniest perceivable chance exist could be the end of your driving license.

Obviously, people committing suicide by falling off a bridge down a highway won't be anyone's mistake down the highway - that's impossible to tell and to prevent (once happening).

Some accidents just can not be avoided and are trust upon you. If someone is walking on a pavement and suddenly jumps in front of a big rig, it is not reasonable to blame the driver of the truck. Anyway, apparently this was not a suicide, but it does sound like a jaywalker which could involve limited visibility from the driver/car perspective (again, just speculating). It is one of my phobias and makes me very uncomfortable when I see people mindlessly stepping on the road while being obscured by a truck/bus/tram, even on a marked crosswalk. I always stop and peek behind such obstacle when I'm the one crossing the road, but when driving (or especially being a passenger) I always get queasy when I see such scenario.

Considering this is an autonomous car, I'd assume they have video of the accident, synced with telemetry so investigators will have somewhat easier job in this case. 

It's just a shame that best lessons are learned the hard way and that somebody lost her life.

[LordFerret]

Interesting how the focus of this discussion fixed itself on the crash and people (meat bags, just for you @tater ), when the real point I was trying to key on was the philosophical concept that Mathematics will become sentient and not the machines run by said Mathematics.

Edited March 20, 2018 by LordFerret

[YNM]

2 hours ago, Shpaget said:

Some accidents just can not be avoided and are trust upon you. If someone is walking on a pavement and suddenly jumps in front of a big rig, it is not reasonable to blame the driver of the truck.

In such cases, the enginnering means would (well, should) take over and speed limits are usually low enough (good reason for 20's plenty), or comfortable barriers could be raised.

There's no reason why someone should be able to do suicide unless something very wrong has happened / actively worked around.

2 hours ago, Shpaget said:

Anyway, apparently this was not a suicide, but it does sound like a jaywalker which could involve limited visibility from the driver/car perspective (again, just speculating).

You should be careful near pavements (sidewalk). Seriously, there are quite a lot of deaths/injuries/incidents down my country by courtesy of someone crossing the road by moving from behind another vehicle, and someone (often a motorcycle) overtaking something else on the left (we're left-hand traffic, right-hand drive) by closing the "safe gap" (where you'd have the time to stop if someone sprung out). True that there needs to be some rules for pedestrians (like, why would anyone walks down a 4-lane highway ? Why should someone try to access the middle of a large and busy traffic island in rural areas ?), but where pedestrians are an inseparable needs for transport, be that private (ie. parking building) or public transport, you need to cater for them properly, and not "just doing enough to meet whatever standard's in place".

 

10 minutes ago, LordFerret said:

when the real point I was trying to key on was the philosophical concept that Mathematics will become sentient and not the machines run by said Mathematics.

Because they'll never be. Or at least in this case that wasn't expected to be. The fact a human overseer existed proven it that way around.

And let's not start on the true nature of these machines - do they understand concepts ? Are they really smart ?

 

Edited March 20, 2018 by YNM

[LordFerret]

11 minutes ago, YNM said:

Because they'll never be.

You don't know that. A human overseer present proves / disproves nothing.

 

 

13 minutes ago, YNM said:

And let's not start on the true nature of these machines

Why not? It's my post, isn't it? Do I attempt to dictate what you discuss in your posts?

 

My discussion has nothing to do with machines or neural networks. It has to do with Mathematics itself, and the concept of it having the unlocked potential of sentience.

[tater]

52 minutes ago, LordFerret said:

Interesting how the focus of this discussion fixed itself on the crash and people (meat bags, just for you @tater ), when the real point I was trying to key on was the philosophical concept that Mathematics will become sentient and not the machines run by said Mathematics.

I see this as pretty distant from artificial general intelligence. If the topic was on the safety issue, and goal alignment, we might agree, actually. There are more ways to get it wrong than right, by a wide margin.

On topic to Uber, my point was that short of a total accounting of deaths due to human Uber drivers per mile traveled vs the single death to self-driving, we can’t make any judgment about the safety of the system.

[YNM]

1 hour ago, LordFerret said:

You don't know that. A human overseer present proves / disproves nothing.

Well at least in this case it's never going to, it's already killed. Others could rise in the future.

1 hour ago, LordFerret said:

Why not? It's my post, isn't it? Do I attempt to dictate what you discuss in your posts?

Ah, let's elevate then XD

 

What about temporary diversion signs ? Could the AI in the car know the diversion ? (they come in various forms, god knows when they'd be on the gps)

52 minutes ago, tater said:

On topic to Uber, my point was that short of a total accounting of deaths due to human Uber drivers per mile traveled vs the single death to self-driving, we can’t make any judgment about the safety of the system.

True for this single class, untrue for the whole system.

 

 

Interestingly, the investigation seems to have stopped on "there's no way to tell of this pedestrian".

But, the car in question was travelling 38 mph (so probably indicated 40-45) down an urban single carriageway road.

 

Seriously !? Was that legal ?!

You Americans really are having it with cars. I hope something really knocks you all off.

Edited March 21, 2018 by YNM

[tater]

42 minutes ago, YNM said:

Well at least in this case it's never going to, it's already killed. Others could rise in the future.

Uber’s human drivers have already killed, as have traditional taxi drivers. Should taxis and ride sharing be banned?

44 minutes ago, YNM said:

Seriously !? Was that legal ?!

You Americans really are having it with cars. I hope something really knocks you all off.

I don’t hope anything knocks you off. I don’t really want most people knocked off.

Self driving cars will become the norm, and will be safer than meat drivers.

This will save net lives. 10s of thousands die every year in the US to auto accidents, and almost all are driver error.

[LordFerret]

41 minutes ago, YNM said:

What about temporary diversion signs ? Could the AI in the car know the diversion ? (they come in various forms, god knows when they'd be on the gps)

Like us, for the AI in the car a diversion temporary or otherwise should be something it recognizes... unless maybe the AI is picking up human traits and ignoring signs - as I see on the road all the time.

As for them being on the GPS, they are. I myself have a Garmin Nuvi in my car, and it has the ability to interrupt my cruise with detours based on traffic, construction, weather, etc., oft times annoying at that, as well as incorrect... because on my local roads, I know better.

As for the accident in the article, the pedestrian was walking a bicycle across the street. For the pedestrian, this is proper safety protocol. For the Uber AI... what pedestrian?... there was nothing there. Maybe. We've no idea of what the AI was 'thinking'. There's no monitoring that in these vehicles, as far as I know.

[YNM]

15 minutes ago, tater said:

I don’t hope anything knocks you off. I don’t really want most people knocked off.

I mean, knock something that induces the thought of "no, having most of us needs to drive / use passenger car isn't right".

15 minutes ago, tater said:

Uber’s human drivers have already killed, as have traditional taxi drivers. Should taxis and ride sharing be banned?

The individual driver who has killed should (so, loses the driving license and driving job).

Now, I don't know are our minds the same, but it most likely is for AIs. Please conclude on your own.

17 minutes ago, tater said:

Self driving cars will become the norm, and will be safer than meat drivers.

This will save net lives. 10s of thousands die every year in the US to auto accidents, and almost all are driver error.

Still non-zero though.

18 minutes ago, LordFerret said:

For the Uber AI... what pedestrian?... there was nothing there. Maybe. We've no idea of what the AI was 'thinking'. There's no monitoring that in these vehicles, as far as I know.

... and something we'll never be sure of.

[tater]

Being a better driver than a person is a low bar. Most drivers in fatal accidents have not been in a fatal accident before.

[YNM]

11 minutes ago, tater said:

Being a better driver than a person is a low bar. Most drivers in fatal accidents have not been in a fatal accident before.

True that. If anything, a fatal accident should never involve someone previously involved in another such case.

 

But that's not my point. My point was that drivers and cars are only part of the equation; To really make sure accident rates are significantly lowered there needs to be changes : how the roads are like (street design and rules), how people are using it (traffic type and amount) and how people are driving through it (driving behaviour). It doesn't matter who drives, where hunks of metal meets hunks of flesh, accidents could occur; the question is how lethal such accidents would be.

Edited March 21, 2018 by YNM

[tater]

3 minutes ago, YNM said:

True that. If anything, a fatal accident should never involve someone previously involved in another such case.

 

But that's not my point. My point was that drivers and cars are only part of the equation; To really make sure accident rates are significantly lowered there needs to be changes : how the roads are like, how people are using it (traffic type and amount) and how people are driving through it (traffic behavior). It doesn't matter who drives, where hunks of metal meets hunks of flesh, accidents could occur; the question is how lethal such accidents would be.

Traffic deaths have dropped for years, almost entirely as a result of automotive engineering improvements, and trauma care.

Driver error is not going away. The bulk of horrific driving I see every single day involves people looking at smart phones. I used to complain about people holding cell phones to their heads, but that is nothing compared to seeing people looking at their crotch 50% of the time they are driving. That will not stop, it's been illegal to have a phone out (hands-free is allowed) in a car here for years, and the % of people on phones in the car is huge. Maybe not a majority at any given moment, but a large number. The only solution to this is to remove the driver.

[kerbiloid]

Afaik, almost all aircraft accidents with numerous victims of last 10-15 years were man-made.
Almost all of them would be by default prevented if AI was ruling the plane.

So, AI is anyway lesser evil than a human crew.

P.S.
If make cars AI, nobody can steal a car.

Edited March 21, 2018 by kerbiloid

[LordFerret]

1 hour ago, tater said:

The only solution to this is to remove the driver.

Before removing the driver, I'd remove the phone. Apparently this is being done, and it's working.

https://www.nhtsa.gov/press-releases/usdot-releases-2016-fatal-traffic-crash-data

[kerbiloid]

2 minutes ago, LordFerret said:

Before removing the driver, I'd remove the phone.

Impossible because the phone is much faster than the car.
So, rejecting the phone in a car is just a kind of personal underclocking. Will not last long.