It's happened. Self-driving Uber kills pedestrian

[Dash Lambda]

35 minutes ago, mr moose said:

-snip-

Humans are not safer drivers as a rule -I would never make the claim that either is universally safer as it is, though of course a highly developed autonomous car will be safer than most human drivers. I'm saying that autonomous driving would best be used in tandem with a real driver if the goal is safety, because machines aren't perfect either.

 

The problem of the human driver not reacting in time, that's on the driver -Being able to correct errors is safer than not being able to correct errors, regardless of what the individual driver does. Could you explain how that's not true?

19 minutes ago, dalekphalm said:

-snip-

That's not the point I'm making, I'm familiar (conceptually) with how they work. I'm not saying the programmers will leave things out -I'm saying that it's very hard to identify exactly how the program is interpreting its environment and there can always be situations where it behaves unpredictably or incorrectly because it doesn't 'think'. This may be due to adverse conditions, unexpected or unusual events, or even mechanical/electrical failure.

 

I'm actually a student in computer science, studying to work on machine learning. I am by no means an authority on it of course, I'm just very interested in it and I like self-driving cars -So my intent is not to say that autonomous cars aren't good, I just think we're a ways from self-driving cars capable of completely replacing humans.

[mr moose]

2 minutes ago, Dash Lambda said:

Humans are not safer drivers as a rule -I would never make the claim that either is universally safer as it is, though of course a highly developed autonomous car will be safer than most human drivers. I'm saying that autonomous driving would best be used in tandem with a real driver if the goal is safety, because machines aren't perfect either.

It reads like you are saying computers can't do what humans do therefore they are more dangerous.  Which is counter to all the evidence.

 

2 minutes ago, Dash Lambda said:

The problem of the human driver not reacting in time, that's on the driver -Being able to correct errors is safer than not being able to correct errors, regardless of what the individual driver does. Could you explain how that's not true?

 

I did, reaction time is not sufficient for a human to be able to correct a situation that a computer cannot account for. 

[mynameisjuan]

4 hours ago, Satisfoxy said:

Why are we still discussing the speed limit? You dead at 35 or 45 MPH when a SUV hits you.

Has nothing to do with the speed limit or even whether he was doing 38 and "speeding", I was just proving a pointless argument wrong. 

 

 maartendc is just trying to justify that is was not the cyclist fault at all when all evidence is pointing it was. 

[dalekphalm]

18 minutes ago, mr moose said:

It reads like you are saying computers can't do what humans do therefore they are more dangerous.  Which is counter to all the evidence.

 

I did, reaction time is not sufficient for a human to be able to correct a situation that a computer cannot account for. 

Yeah I’m mostly in agreement here. 

 

If the AI subsystem makes a mistake and doesn’t properly react to a situation (certainly possible), the likelihood that a human operator would have time to notice the mistake, decide on a course of action, and actually react in time, is essentially negligible. 

 

Even if the human simply reacted out of instinct, minimizing time “deciding” on what to do, their reaction time is so much slower than the computers that I doubt there is really any situation in which the human could affect the outcome in any beneficial way. 

 

Now I’m not saying it’s impossible. Just incredibly unlikely. 

[mr moose]

4 hours ago, Satisfoxy said:

Why are we still discussing the speed limit? You dead at 35 or 45 MPH when a SUV hits you.

true, but if the car is going 10Kph slower it's stopping distance is reduced  by at least 10metres above 40Kph. This could be the difference between being hit at 20 Km/h (still easily fatal) and not being hit at all.

[mynameisjuan]

4 minutes ago, mr moose said:

true, but if the car is going 10Kph slower it's stopping distance is reduced  by at least 10metres above 40Kph. This could be the difference between being hit at 20 Km/h (still easily fatal) and not being hit at all.

The problem is by the time even the driver saw her she was already hit. Even the computer had next to no time to respond. It sounds like she got hit at almost full speed.

[Xander Cousins]

Just now, mynameisjuan said:

The problem is by the time even the driver saw her she was already hit. Even the computer had next to no time to respond. It sounds like she got hit at almost full speed.

Kinda why your told to look both ways.... Plus tbh many pedestrains are walking onto the road thinking the vehicle will slow down for them. This morning i was gonna walk in front of a Van but decided not to when the vehicle got close the driver was too busy half way into the passanger seat foot well looking for something so if i did he wouldn't of been able to stop/ respond in time.

[Dash Lambda]

1 minute ago, mr moose said:

It reads like you are saying computers can't do what humans do therefore they are more dangerous. Which is counter to all the evidence.

I'm... Very much not. My wording is often roundabout, but I am not saying computers are more dangerous than human drivers -I'm saying only a computer is more dangerous than a computer plus a human driver.

7 minutes ago, mr moose said:

I did, reaction time is not sufficient for a human to be able to correct a situation that a computer cannot account for. 

You said having human drivers to correct errors would be more dangerous than not having them. That does not explain how it's more dangerous.

As for how it's safer, I'm not talking about the human having better reaction time. I'm talking about the human understanding their environment and noticing if, say, their car is drifting into the wrong lane on a road where the snow is obscuring the lines, or there's a traffic event the car doesn't recognize that affects its route, or maybe -in an extreme scenario- the car just randomly stops on some train tracks. You're assuming the only problems will be crashes that a human driver wouldn't have been able to avoid anyway -That's not the case, we can't make anything perfect. We can make them very good, but not perfect.

[mynameisjuan]

Just now, Alex Colson said:

Kinda why your told to look both ways.... Plus tbh many pedestrains are walking onto the road thinking the vehicle will slow down for them. This morning i was gonna walk in front of a Van but decided not to when the vehicle got close the driver was too busy half way into the passanger seat foot well looking for something so if i did he wouldn't of been able to stop/ respond in time.

Exactly, you're taught that as a kid. 

 

I have to drive through a campus road every day and college students dont look at all (most hope to get hit for insurance, which is true) and I have to drive extra careful. Thats why I installed a train horn... It usually teaches them to not just cross while looking at their phones or you know, actually use a cross walk. 

[marldorthegreat]

So there has been 1 person killed by a self driving car in how long? There is probably someone killed by a normal car every second. Self driving cars need not be perfect, they just need to be safer than human driven cars. 

[79wjd]

7 hours ago, maartendc said:

Why is the AI in the car allowed to go above the speed limit AT ALL though? Very concerning to me.

Because it's incredibly dangerous to not keep up with the flow of traffic. I would seriously hope that any self driving cars would maintain the average road speed, not the nonsensical and irrelevant speed limit.

 

On top of that, speedometers are notoriously inaccurate, so most people don't have a clue how fast they're actually going in relation to the posted speed limit. 

[mynameisjuan]

3 hours ago, marldorthegreat said:

So there has been 1 person killed by a self driving car in how long? There is probably someone killed by a normal car every second. Self driving cars need not be perfect, they just need to be safer than human driven cars. 

Come on you should know people never take perspective into account. I mean they (google, telsa, uber) have been testing for what, 3 or 4 years or more. thats 120,000-160,000 deaths since then on average. 1 death and suddenly the shit hits the fan. 

 

3 hours ago, djdwosk97 said:

Because it's incredibly dangerous to not keep up with the flow of traffic. I would seriously hope that any self driving cars would maintain the average road speed, not the nonsensical and irrelevant speed limit.

 

On top of that, speedometers are notoriously inaccurate, so most people don't have a clue how fast they're actually going in relation to the posted speed limit. 

Modern automatic cruise control in a lot of new cars will keep the flow of traffic even if above its set speed. Driving at or below on a highway or freeway is most likely to cause accidents due to how people slow down more and more resulting in usually a dead stop. Common in big cities. 

[Commodus]

1 minute ago, marldorthegreat said:

So there has been 1 person killed by a self driving car in how long? There is probably someone killed by a normal car every second. Self driving cars need not be perfect, they just need to be safer than human driven cars. 

I checked some stats.  In the US, about 37,000 people die in car crashes every year, or about 101 people every day.  Just cutting that in half would be worth it, and it'd likely be much better than that.

 

The true objection, I suspect, is that those crashes that did occur would be out of the passengers' control.  We're oddly fine with our cars crashing so long as we're the ones doing the crashing.

[79wjd]

50 minutes ago, mynameisjuan said:

Modern automatic cruise control in a lot of new cars will keep the flow of traffic even if above its set speed. Driving at or below on a highway or freeway is most likely to cause accidents due to how people slow down more and more resulting in usually a dead stop. Common in big cities. 

Don't forget people passing you (and potentially cutting you off). It's really just an all around awful idea to follow a speed limit strictly.

[mr moose]

2 hours ago, Dash Lambda said:

I'm... Very much not. My wording is often roundabout, but I am not saying computers are more dangerous than human drivers -I'm saying only a computer is more dangerous than a computer plus a human driver.

You said having human drivers to correct errors would be more dangerous than not having them. That does not explain how it's more dangerous.

As for how it's safer, I'm not talking about the human having better reaction time. I'm talking about the human understanding their environment and noticing if, say, their car is drifting into the wrong lane on a road where the snow is obscuring the lines, or there's a traffic event the car doesn't recognize that affects its route, or maybe -in an extreme scenario- the car just randomly stops on some train tracks. You're assuming the only problems will be crashes that a human driver wouldn't have been able to avoid anyway -That's not the case, we can't make anything perfect. We can make them very good, but not perfect.

No I didn't, I said human drivers are more dangerous than computer I was contesting that having human drivers as a backup does not change that.

2 hours ago, mynameisjuan said:

The problem is by the time even the driver saw her she was already hit. Even the computer had next to no time to respond. It sounds like she got hit at almost full speed.

 

Interesting, I might say that if neither the driver nor computer could see her then either the crossing is in the wrong spot or the speed limit is too high.

[Cole5]

How many people were killed the same day by drunk or distracted drivers? I Bet it was way more than 1 

[mynameisjuan]

1 hour ago, mr moose said:

Interesting, I might say that if neither the driver nor computer could see her then either the crossing is in the wrong spot or the speed limit is too high.

It was a highway though. Many two lanes highways here where 45 is reasonable. Any speed is too high though when people jump out in traffic. 

22 minutes ago, Cole5 said:

How many people were killed the same day by drunk or distracted drivers? I Bet it was way more than 1 

And the funny thing is if AI is in all cars, the idiots that drive drunk, it would be able to minimize the innocent people killed by them. But AI is bad right?

[dalekphalm]

3 hours ago, Dash Lambda said:

I'm... Very much not. My wording is often roundabout, but I am not saying computers are more dangerous than human drivers -I'm saying only a computer is more dangerous than a computer plus a human driver.

You said having human drivers to correct errors would be more dangerous than not having them. That does not explain how it's more dangerous.

As for how it's safer, I'm not talking about the human having better reaction time. I'm talking about the human understanding their environment and noticing if, say, their car is drifting into the wrong lane on a road where the snow is obscuring the lines, or there's a traffic event the car doesn't recognize that affects its route, or maybe -in an extreme scenario- the car just randomly stops on some train tracks. You're assuming the only problems will be crashes that a human driver wouldn't have been able to avoid anyway -That's not the case, we can't make anything perfect. We can make them very good, but not perfect.

I'd challenge that your stance has some merit, but not in the context you are presenting.

 

You are saying: Computers plus humans means safer driving.

 

I would posit that, under very specific, and rare circumstances, yes, you're probably right. I would further posit that most of the time, a computer is going to be safer than a human.

 

Most accidents are quite simply avoidable, and were caused by people driving like an idiot - tailgating, changing lanes without checking, running a red light, etc. These are all things that a computer can avoid much easier than a human. They have (generally) superior sensors that can see farther, often through cloud/smoke/fog/rain/snow/bush/trees/night, etc. They can think faster. And they can react almost instantly. In these types of common situations, a computer has a huge advantage over a human.

 

I would further posit that a human trying to "help" would actually interfere in the computer's capable response, in these common situations. The human would either take control and make the wrong (or simply less good) choice, or they would simply slow down the reaction time of the computer, even if they chose the same action the computer was going to take anyway.

 

So, I suggest this:

 

9/10: a computer is going to be safer, by itself, at driving, than a human, or a human plus a computer.

1/10: a human will be safer, because they can anticipate or use human intuition to come up with a reactive response on-the-fly to a situation that a computer has trouble dealing with

 

So, based on this, I can still say that a computer, by itself, is overall, on average, safer than a human, or human plus computer.

 

Even if in that 1 out of 10 event, where an accident occurs that might have otherwise been prevented, there were 9 others that were prevented.

 

Even if we can reduce the number of accidents by 50%, that would save literally more than 18,000 lives. In 2016 alone, 37,461 died in car accidents, in the US alone.

https://www.usatoday.com/story/money/cars/2018/02/15/national-safety-council-traffic-deaths/340012002/

 

Computers, even if they make occasional mistakes, will make that number shrink.

 

And, the computers are getting better.

 

And, in this particular case, it looks like Uber's computer didn't make a mistake.

[mr moose]

15 minutes ago, mynameisjuan said:

It was a highway though. Many two lanes highways here where 45 is reasonable. Any speed is too high though when people jump out in traffic. 

And the funny thing is if AI is in all cars, the idiots that drive drunk, it would be able to minimize the innocent people killed by them. But AI is bad right?

Ahh,  I don't know why, but for some reason I had it in my mind the accident happened at a pedestrian crossing.  My apologies.

[Zodiark1593]

12 hours ago, dalekphalm said:

Indeed, the Volvo systems can mitigate potential collisions. But there is no evidence that such mitigations would have prevented the accident. It might have made the accident less severe, perhaps. But also, we don't know whether the systems were even enabled or not.

 

The Volvo systems might well have been fully enabled.

 

Let's wait until the facts before we start claiming things like the Brake Assist + a human driver would have changed the outcome in any meaningful way.

I'm sorry - but when did I agree with banning drivers? We are definitely not at the point where autonomous vehicles should take over 100%, and all human drivers should be banned. I happen to enjoy driving, though I wouldn't go so far as to call myself a gear head or grease monkey.

 

Furthermore, the idea that AI has to be better than humans in every possible way? No. They don't. They just have to be better than humans on average. And frankly, they already are.

 

If net AI accidents/deaths = lower than net human accidents/deaths, that's all the proof I need.

 

Autonomous vehicles don't need to be perfect. They just need to be better drivers than your average person. Yes there are situations where autonomous vehicles could fail, but do you have any idea how often humans screw up and cause accidents? It's a ridiculous and insanely high amount.

If self driving cars can take the majority of the inept (those that don't take driving seriously) off the road, coexistence should be plenty feasible in the long term.

 

11 hours ago, maartendc said:

Why is the AI in the car allowed to go above the speed limit AT ALL though? Very concerning to me.

 

How do these cars know the speed limit anyway? Do they read signs? Do they go by stored data based on GPS, the way your navigation system indicates the speed limit? In that case, that data can always be outdated because speed limits change regularly.

 

Google street view photos can be outdated. Could be taken 5 years ago.

If one tried to obey the 55 mph speed limit out on Highway 49 here, they'd very quickly get a very long line of pissed off drivers tailgating one another, and some making illegal (and sometimes quite dangerous) passes. Realistic speed is at least 65, with a fair amount edging past 70.

 

In general, speed limits tend to be treated as very loose "guidelines" than actual rules.

[Jito463]

On 3/19/2018 at 4:05 PM, corsairian said:

Isn't that just from the movie iRobot?

*sigh* Look up Isaac Asimov and Philip K. Dick.  You're welcome.

On 3/19/2018 at 7:11 PM, Trik'Stari said:

Edit: I am a former delivery driver. The number of people who just blatantly and blindly walk out into traffic that I saw, is/was absurd.

Just this afternoon I had to slow practically to a stop because some idiot decided to saunter across the road while I was trying to drive.  I once had a guy ride his bicycle right out in front me, and nearly cause an accident (when I had the right of way), then proceed to flip me off.  Some people are just dumb.

15 hours ago, mynameisjuan said:

If we replaced all traffic with AI we would lower collision rates to such an amount it would essentially be eliminating them altogether. 

You can take my stick-shift vehicle from my cold, dead hands.

9 hours ago, chilicheeseburger said:

Anyway, I can't really wrap my head around the fact why they would have that X-shape pavement there... (where the accident happened). Even if they put up signs telling people not to use it - which just makes the pavement completely useless, that was an accident waiting to happen, self driving car or not.

It looks to me like the pedestrian walk path was already there when they built the highway, and for whatever reason they chose not to break up the pavement.

6 hours ago, marldorthegreat said:

So there has been 1 person killed by a self driving car in how long? There is probably someone killed by a normal car every second.

And there are how many self-driving vehicles compared to how many human driven ones?

6 hours ago, Commodus said:

The true objection, I suspect, is that those crashes that did occur would be out of the passengers' control.  We're oddly fine with our cars crashing so long as we're the ones doing the crashing.

It's because liberty and personal responsibility go hand-in-hand.  You can't have one without the other.  I refuse to give up my liberty, and so I accept my personal responsibility when driving my vehicle.  We also recognize that sometimes others will screw up.  It's the nature of liberty.

[Jito463]

3 hours ago, CCWong said:

The issue with your liberty argument is that you are effectively denying liberty to all other commuters you encounter by putting them at greater risk, infringing on their rights to life.

That has to be the weakest argument I've ever heard.  There's no infringement on right to life, and I challenge you to prove otherwise.  Do people have car accidents?  Yes.  Do people die from car accidents?  Yes.  Is my choosing to drive automatically mean taking someone else's life?  Absolutely not.  You made an enormous leap of logic there, and one that isn't consistent with reality.

[maartendc]

19 hours ago, mynameisjuan said:

Has nothing to do with the speed limit or even whether he was doing 38 and "speeding", I was just proving a pointless argument wrong. 

 

 maartendc is just trying to justify that is was not the cyclist fault at all when all evidence is pointing it was. 

Let me just ask you this: Whose side are you really on in the coming robot - human war?

[Matu20]

On 3/20/2018 at 3:02 AM, ElfFriend said:

Oh so exactly like what car manufactures did a century ago?

 

I dont get your points.

[Matu20]

On 3/20/2018 at 5:00 AM, djdwosk97 said:

People don't kill people; cars kill people.

So if you step in front of the train and get ran over, trains are at fault?