Well that's debatable, and I would come out on the other side of that debate.
But it's neither here nor there, because GOFAI is specifically done with Human readable rules. And there's no one suggesting Neural Nets can be interpreted as that. That is not debatable.
You are only looking at the cons and not the pros. 6 inches further back is neither here nor there. You get a far greater variance than that for human drivers based on the model of car they drive and how long the hood (bonnet) is.
Meanwhile on the plus side, you have 360 degree coverage at all times, rather than the limited field of view of a human driver, and the blind spots caused by pillars.
You don't have the selective vision of humans, where the driver simply "didn't see" a motorcycle coming, because they were looking for cars, and forgot about motorcycles.
And you don't get the bad behaviour, where the human driver may not even stop at a STOP sign, because he thinks he knows better, assumes it's clear, or he's just an asshole.
It doesn't take a few hours to charge a car. A Tesla supercharger will give a car an 80% charge in 30-40 minutes.
You're whining about range. No doubt you were whining about electric car range too a few years ago when it was about 100 miles. Now it's anything up to 350 miles.
This is the 1st generation of Semi. It's not where the technology will stop. At the moment, the use cases will be limited. But its just a start.
You'd do well to remember the dumbass comment from Cmdr Taco when the iPod launched: "No wireless. Less space than a nomad. Lame."
That's the kind of dumbass you look when you keep declaring every new thing is shit.
They did. Tesla Model S battery can be replaced in 1 minute 35 seconds. They demonstrated it on stage back in 2013. And there was a pilot battery swap station built between LA and SF I believe. But there was no call for it. No one ever used it.
None of that reflects the post you were replying to, which was about the fact people want autopilot, just as they wanted cruise control in previous decades.
And the Gran Turismo jibe - I've been driving for 30 years. I never play driving games. So duh!
OK, so you have no evidence for motorcycles being missed 10 times as often as cars of pedestrians.
The first two were stories about collisions. So what? Human drivers often miss motocycles and hit them. You've presented no evidence that Autopilot does it more often.
Neural Nets are very specifically NOT rule based. They are trained. GOFAI was pretty much a phrase invented to label stuff that IS NOT the neural net approach. Autonomous vehicles do not need AGI. It's very much a single domain system. You don't need your autonomous car to be able to diagnose diseases for example.
There are side facing cameras in the B Pillars. This is only a few inches further back than where a human driver's eyes are, and approximately the same height.
Cruise control is better than maintaining a set speed yourself. It's a bore to act as a feedback device for a couple of hours looking at the speedo a couple of times a minute and adjusting your foot pressure to adapt.
Adaptive cruise control is better than being in heavy traffic and cycling through speeding up, slowing down and stopping ensuring that you don't run into the car in front, and yet not leaving such a big gap that people keep cutting in front of you.
Why wouldn't having the car do the steering too not be better? Tesla drivers certainly like it.
I'd much rather the car did the driving chore for me, even it I have to watch it. It's certain a better experience for me. And one I'll pay extra for when it comes within by budget.
Put it another way, the trolley problem is like Asimov's Three Laws of Robotics. Philosophically interesting, and seems so reasonable any layman would thing that all robots should be programmed with them.
But the reality is no robot production robot has ever had the laws of robotics programmed into them. A Roomba only knows that when it hits something it should turn a random direction and start again. If it's working at the top of some stairs and you don't put a barrier, either physical or light based, across the top, then it'll happily run off that cliff and destroy itself.
(I'm talking about my Roomba here which is a decade old, they may have put a cliff detector in by now. But it's still only hardware, and the same old rule for what to do when it comes to an obstacle.)
Likewise with the most advanced industrial robot. If you give it a knife or a gun, and instruct it to stab or shoot it's operator, then it will. There are no Asimov laws there. Instead, you put physical guards around the robot, that cause it to stop if broken. Just like any other industrial machine.
The Trolly Problem is the same. It's philosophically interesting, and seems so reasonable that a layman might assume that it must somehow be built into autonomous systems. But it isn't, and there's no prospect that it will be. It simply doesn't bear any relationship to what an autonomous system is doing.
I'm afraid you're suffering from the Dunning Kruger effect. You understand so little of how neural nets work, you don't even know how little you know.
As to regulatory agencies, they are interested in is demonstrable performance. Number of miles driven in tests, and how many incidents happened. Where an incident might be a collision, or a breaking of the law, such as running of a red light. Regulatory approval will simply come from a demonstration that over a large number of miles there are fewer/less serious incidents with the autonomous system, than with human drivers. That's not a guess, that's what governments are asking for and the autonomous driving companies are setting about proving right now.
They have no interest in looking at code to see whether it has codified rules. In fact autonomous riving would never, ever happen if it had to be programmed with discrete rules. It's only possible by training NNs.
Sorry, but you don't know what you are talking about. You need to study neural networks. The are not "specifically programmed" they are trained with data sets. And not only are the ways they work not specifically programmed, a programmer cannot find out in any meaningful sense how it does work.
No it doesn't have to decide, any more than a human does. Just as a human does it only has to react (or not). It's equivalent.
We're in the world of training, neural nets and fuzzy logic here, where there is no programmer that knows the specific rules by which the system is acting. Just as the human conscious does not know the reasons for which the subconscious reacts. We can only guess.
You could have the developers make moral judgement on a series of these trolley problem scenarios, assigning different weights for different categories of people killed if you want. Then train based on that data set. But it would be a choice to do that training. And not a required choice, as it's not something that human drivers do.
The trolley problem is not real. It's a philosopher and psychologists plaything. It does not represent what happens in an emergency situation at all. When there's an imminent collision, people clearly don't think and weigh up alternative outcomes and make a choice before operating the controls. They simply react. Like 99% of driving the conscious mind that can make such high level choices isn't being used at all. Driving is simply a behaviour that comes from the subconscious.
It's seems likely that in an emergency situation the subconcious is simply doing what it normally does when driving. Try at all costs not to hit anything. So if there's two things that might be hit and no possibility of avoiding hitting either, it's likely simply the one that is futher away that gets hit, as the result of the driver swerving to avoid the more imminent threat. But it's only a guess. Like an artificial neural net, you can't translate trained behaviour into simple rules.
Cost per mile will undoubtably be greater for a non-owned fleet vehicle. But without the fixed costs. No purchase cost, insurance, road tax, servicing, repairs. And for the next generation, no time or money spent learning to drive.
Things like this are generational change. I was reading the other week that fewer young people are even bothering to get a driving license these days. And that's before autonomous vehicles are even available.
It's like stick shift in the American market. Old timers always swore there were going to stick to them. After all, autos were expensive, inefficient, and not as fast. But these days, younger drivers don't even know how to drive a stick shift. They learned to drive in an auto and it's been autos ever since.
So whilst you and many others will want to keep on owning your own car, as your generation dies off, you will not be replaced with a car owning generation.
Slashdot Mirror
PMSL! drinkypoo thinks the tesla battery is glued on.
Oh look, a conspiraloon. How entertaining!
Well that's debatable, and I would come out on the other side of that debate.
But it's neither here nor there, because GOFAI is specifically done with Human readable rules. And there's no one suggesting Neural Nets can be interpreted as that. That is not debatable.
You are only looking at the cons and not the pros. 6 inches further back is neither here nor there. You get a far greater variance than that for human drivers based on the model of car they drive and how long the hood (bonnet) is.
Meanwhile on the plus side, you have 360 degree coverage at all times, rather than the limited field of view of a human driver, and the blind spots caused by pillars.
You don't have the selective vision of humans, where the driver simply "didn't see" a motorcycle coming, because they were looking for cars, and forgot about motorcycles.
And you don't get the bad behaviour, where the human driver may not even stop at a STOP sign, because he thinks he knows better, assumes it's clear, or he's just an asshole.
LynwoodRooster the Luddite.
It doesn't take a few hours to charge a car. A Tesla supercharger will give a car an 80% charge in 30-40 minutes.
You're whining about range. No doubt you were whining about electric car range too a few years ago when it was about 100 miles. Now it's anything up to 350 miles.
This is the 1st generation of Semi. It's not where the technology will stop. At the moment, the use cases will be limited. But its just a start.
You'd do well to remember the dumbass comment from Cmdr Taco when the iPod launched: "No wireless. Less space than a nomad. Lame."
That's the kind of dumbass you look when you keep declaring every new thing is shit.
They did. Tesla Model S battery can be replaced in 1 minute 35 seconds. They demonstrated it on stage back in 2013. And there was a pilot battery swap station built between LA and SF I believe. But there was no call for it. No one ever used it.
None of that reflects the post you were replying to, which was about the fact people want autopilot, just as they wanted cruise control in previous decades.
And the Gran Turismo jibe - I've been driving for 30 years. I never play driving games. So duh!
OK, so you have no evidence for motorcycles being missed 10 times as often as cars of pedestrians.
The first two were stories about collisions. So what? Human drivers often miss motocycles and hit them. You've presented no evidence that Autopilot does it more often.
Neural Nets are very specifically NOT rule based. They are trained.
GOFAI was pretty much a phrase invented to label stuff that IS NOT the neural net approach.
Autonomous vehicles do not need AGI. It's very much a single domain system. You don't need your autonomous car to be able to diagnose diseases for example.
Chris Lattner wasn't there long enough to get started. We don't know why be backed out.
Personally I could never see why a compiler guy was being hired as head of one of the most complicated AI projects anyway. Different field.
Nobody wants? I can't wait!
I suppose you imagine no one wanted cruise control either.
Evidence?
There are side facing cameras in the B Pillars. This is only a few inches further back than where a human driver's eyes are, and approximately the same height.
Cruise control is better than maintaining a set speed yourself. It's a bore to act as a feedback device for a couple of hours looking at the speedo a couple of times a minute and adjusting your foot pressure to adapt.
Adaptive cruise control is better than being in heavy traffic and cycling through speeding up, slowing down and stopping ensuring that you don't run into the car in front, and yet not leaving such a big gap that people keep cutting in front of you.
Why wouldn't having the car do the steering too not be better? Tesla drivers certainly like it.
I'd much rather the car did the driving chore for me, even it I have to watch it. It's certain a better experience for me. And one I'll pay extra for when it comes within by budget.
"Grow up"? This is a technical discussion, not a playground.
I'm sorry, but your opinion far exceeds your knowledge.
Put it another way, the trolley problem is like Asimov's Three Laws of Robotics. Philosophically interesting, and seems so reasonable any layman would thing that all robots should be programmed with them.
But the reality is no robot production robot has ever had the laws of robotics programmed into them. A Roomba only knows that when it hits something it should turn a random direction and start again. If it's working at the top of some stairs and you don't put a barrier, either physical or light based, across the top, then it'll happily run off that cliff and destroy itself.
(I'm talking about my Roomba here which is a decade old, they may have put a cliff detector in by now. But it's still only hardware, and the same old rule for what to do when it comes to an obstacle.)
Likewise with the most advanced industrial robot. If you give it a knife or a gun, and instruct it to stab or shoot it's operator, then it will. There are no Asimov laws there. Instead, you put physical guards around the robot, that cause it to stop if broken. Just like any other industrial machine.
The Trolly Problem is the same. It's philosophically interesting, and seems so reasonable that a layman might assume that it must somehow be built into autonomous systems. But it isn't, and there's no prospect that it will be. It simply doesn't bear any relationship to what an autonomous system is doing.
I'm afraid you're suffering from the Dunning Kruger effect. You understand so little of how neural nets work, you don't even know how little you know.
As to regulatory agencies, they are interested in is demonstrable performance. Number of miles driven in tests, and how many incidents happened. Where an incident might be a collision, or a breaking of the law, such as running of a red light. Regulatory approval will simply come from a demonstration that over a large number of miles there are fewer/less serious incidents with the autonomous system, than with human drivers. That's not a guess, that's what governments are asking for and the autonomous driving companies are setting about proving right now.
They have no interest in looking at code to see whether it has codified rules. In fact autonomous riving would never, ever happen if it had to be programmed with discrete rules. It's only possible by training NNs.
Sorry, but you don't know what you are talking about. You need to study neural networks. The are not "specifically programmed" they are trained with data sets. And not only are the ways they work not specifically programmed, a programmer cannot find out in any meaningful sense how it does work.
No it doesn't have to decide, any more than a human does. Just as a human does it only has to react (or not). It's equivalent.
We're in the world of training, neural nets and fuzzy logic here, where there is no programmer that knows the specific rules by which the system is acting. Just as the human conscious does not know the reasons for which the subconscious reacts. We can only guess.
You could have the developers make moral judgement on a series of these trolley problem scenarios, assigning different weights for different categories of people killed if you want. Then train based on that data set. But it would be a choice to do that training. And not a required choice, as it's not something that human drivers do.
The trolley problem is not real. It's a philosopher and psychologists plaything. It does not represent what happens in an emergency situation at all. When there's an imminent collision, people clearly don't think and weigh up alternative outcomes and make a choice before operating the controls. They simply react. Like 99% of driving the conscious mind that can make such high level choices isn't being used at all. Driving is simply a behaviour that comes from the subconscious.
It's seems likely that in an emergency situation the subconcious is simply doing what it normally does when driving. Try at all costs not to hit anything. So if there's two things that might be hit and no possibility of avoiding hitting either, it's likely simply the one that is futher away that gets hit, as the result of the driver swerving to avoid the more imminent threat. But it's only a guess. Like an artificial neural net, you can't translate trained behaviour into simple rules.
Which is Tesla? Because it's been doing a pretty good job of both.
Cost per mile will undoubtably be greater for a non-owned fleet vehicle. But without the fixed costs. No purchase cost, insurance, road tax, servicing, repairs. And for the next generation, no time or money spent learning to drive.
Things like this are generational change. I was reading the other week that fewer young people are even bothering to get a driving license these days. And that's before autonomous vehicles are even available.
It's like stick shift in the American market. Old timers always swore there were going to stick to them. After all, autos were expensive, inefficient, and not as fast. But these days, younger drivers don't even know how to drive a stick shift. They learned to drive in an auto and it's been autos ever since.
So whilst you and many others will want to keep on owning your own car, as your generation dies off, you will not be replaced with a car owning generation.
Well you might be one of the half dozen people who buy it. But probably you won't, just like nearly everyone else.
Has anyone told that to Tesla owners, because the don't seem to know. How lucky they have you talking from a position of ignorance.