How am i mistaken? I just wrote idea..other way more experienced in cybersecurity also wrote that it can be done. But from a driving POV main computer and OS should be isolated. Only communication that is required is basic information about vehicle&obstacles: like location and measurements and driving style and direction..all of it does not require breach in isolation but just sensors. Sensor abuse|misuse|damage is only way i can think atm of then to create accidents.
MercedesBenz selfdriving truck concept
BTW: Speaking about car accidents. Recently some another supercar driver in our country got into car accident even at small speeds
https://www.youtube.com/watch?v=mdXoe1TofuM
so i see no difference if cheap car or expensive car or driver human or driver robot. There will always be unexpected accidents that just happen. I guess in order to make accident free driving we must make either "Knight Rider" type cars or create new form of transportation.

Artificial Intelligence (AI) doesn't exist and won't exist in the foreseable future. What we call AI is not actual artificial intelligence, it's a regular algorithm that is marketed as advanced future technology. In the 70's people called Expert Systems AI. Nowadays some people call Artificial Neural Network (ANN) an AI. It's not. It's just a classification algorithm that can solve very narrow and specific problems. It's not a mystic silver bullet that will bring a new world on its own. AI is not right around the corner. It's not coming in 20 years either. It will probably never come and if it would it would not be in this millenium. It would require some great leaps, the kind of leap that never happened in human history to date.

Now what an ANN does is classify stuff according to fuzzy unknown rules. There is no AI programmer to tell the rule, there is training instead. The network is initialy in a random state. We present the network with good answers and bad answers and tell it to train itself so it gives the same answers. The result is a trained network. Since the initial state was random, the result is always different. No network will give 100% good answers. With good training you might end up with a good network that will give you like 95% good results. Some networks won't ever learn because the initial random state was not lucky. They will get stuck bellow 50% and never improve. More training doesn't get you better results. Your good network will plateau at around 98% and more training will decrease its efficiency.
Nobody is able to tell what the networkis doing, or how it reaches its conclusions. If we knew we wouldn't use ANN in the first place, but a proper algorithm instead. As you can see ANN do have a common trait with humans, they make errors and we don't know how.

Now about the self driving cars. If your dream is a car that can be released into the wild and be expected to find its way in an unknown environment, ... well forget it, it won't happen. Computer vision is largely based on ANN and it fails a lot. Self driving cars can't drive themselves when it's snowing. They can't tell the road from the snow. What we call AI is actually pretty dumb. It's not safer than humans. A human can tell a pedestrian from a picture of a pedestrian, he understands what the road is. The AI can not, the AI is good at mathematics but it has 0 instinct, I mean, it's literally dumber than a fly. It might work in controlled environment like an airport with roads specifically crafted for the car but that's it. You won't see them in the wild without human supervision, ever.

Bravo! An excellent post. But what you don't allow for is the ambitions of car manufacturers and the stupidity of the general public. Never mind that these systems are not actually intelligent. Everyone thinks they are, so they will indeed be "released into the wild and be expected to find its way in an unknown environment". The expectation won't be met, but that is another story.

But that is not what Google, Yahoo and other company people see it. Here is presentation about progress. It is coming in foreseable future if they keep it up same pace. Plenty of progress is made. Just a matter of time.

With all my respect, I disagree with you and with all my respect you proved the point of the poster just before you indeed.
The video is great marketing material but it fails to expose the limits of his self driving car and their scale.
It's not just a small glitch to fix, it's a huge leap to take in order to meet his expactations. As you can see in the video, computers don't see well. It works relatively well in a controlled environment with perfect road condition and well understood, predefined hazards. It won't work so well when the road marks are not so good, when the traffic light aren't working properly, when the weather conditions are not perfect or when the roadworks aren't properly marked with standard plots the AI is programmed to recognize as such.
It's dumb. You don't need to be a cyber security expert to fool the car. It just takes painting whites lines on the ground to trick it.

And it is fine. We both disagree with each other. But i do have my own skin driving experience in Europe so i do know how it is for human being to drive and it is possible to create specif to task(not universal but specific to profession) artificial intelligence in near future. I was just saying that our kids if not we ourselves during our lifetime could see autonomos robotic devices including self driving cars that have learned to drive on snow and have learned how to see stuff. Driving is not rocket science. Given enough time things repeat and for this reason it can be learned.
Only argument against nonhuman task doers is about taking away jobs is valid
If we imagine humans themselves having computer(brain) in skull then same logic applies. When i did not have drivers license i thought it is very hard to achieve but when started to take and drive myself it became very easy and repetitive cause nothing much needs to be learned..even in harsh conditions it all comes down to formulas. Just need learn those formulas. Next time you drive your car think about it : you also drive car using preinserted driving lesson formulas as automatic muscle memory responses etc. either by learning in school or from real life events but they are same regardless drive Audi or BMW or Volvo or whatever..

I have considerable cybersecurity skill, taught some to my Son who is now employed protecting US Utility installations like General Electric from cyber attack but be that as it may what is more important is that I own hundreds of live OS CDs, DVDs and Thumbdrives as well as a fairly sophisticated home network and I have "broken in" to all of them. It's really not hard.

As for isolation, now that perhaps you see that what is called AI is as of yet quite limited, isolation is not an option. Self-driving vehicles need to get traffic and weather conditions updates not to mention simple upgrades since it is a rapidly changing field.

I happen to agree with you that they will be deployed fairly soon and that they will very likely be constrained in ways that demonstrate a better track record than human drivers but it will be a very long time before they can just go anywhere, anytime.

Incidently I just drived home and thought about it. There were indeed some cases where the AI would have been confused. For instance, a street corner near my home where you have absolutely no visibility. They installed a mirror at the corner so you can see if the street is clear before entering. I am pretty sure no AI in their current state would understand the mirror. I also had to literally drive on the pavement in order to let room for an emergency vehicle from the fire department. When I reached my home, I had to drive back because the street is not large enough for 2 vehicles. I drived back, let him pass and then entered the street.
Anyway there may be some room in the middle. I can see autonomous vehicles allowed in specific places under specific conditions. Self driving trucks that only take some highways specifically adapted to them. Some cities with electronic markers specifically placed for autonomous vehicles could allow them do drive in that area.
And anyway even if it doesn't look possible to me, I don't want to break the dream of the dreamers. Even if they don't succeed, I commend them for trying, there is a lot to be learned from it whatever the outcome.

That was not my point. My point was this quote:
If there are measures to "break in" there should be measures to prevent "breach" or delay enough so that it not matter.

In USA there is already self-drive taxi service.
https://www.youtube.com/watch?v=WBkgs4u5tW0
Why would they need traffic and weather updates and constant upgrades if they have base location that is isolated and can insert such data safely?
Why are you using exceptional situations as examples? Human drivers in those cases also get confused. But they also can be teached cause even then you apply formula how to react and besides self-drive cars have sensors than not require mirror or other info givers for human drivers. But i was not speaking about AI in current state but near future when they are enough skilled and sensors enough developed to allow it. I agree that today there is no actual KITT(knight rider car from TV was first autonomos vehicle i know so it stays in my mind everytime there is chat about self driving cars) car on street but tommorow there could be at least beginning of such vehicles.
As mentioned before. Only argument when it comes to robotics taking over that works is not if it is possible cause it is possible but if it should be allowed and implemented cause it takes away human labour. Our limited human brains work very similar to computers. If you think about raising any child|kid|newborn then you will see that it is very similar to teaching AI how to do stuff.

Nobody knows for sure what the future is made of.
The debate about Artificial Intelligence already happened in the 60's and people like Hubert Dreyfus have been proved right.
See here : https://en.wikipedia.org/wiki/Hubert...l_intelligence.
Must read : https://en.wikipedia.org/wiki/Artifi...l_intelligence
The state of the art in AI is inspired from a subset of our brain but we are still very, very, VERY far from even understanding how it works, let alone replicate it.
We vaguely understand how neurons and synapses work and that's what we are trying to simulate in current "AI" implementations but it appears that it is actually a very small part of the brain.
We recently discovered Glia cells. See https://en.wikipedia.org/wiki/Glia. They seem to be essential in our thinking process and they outnumber our neurons by a LOT.
What we are able to do right now it to simulate very simplified neuron networks with mathematics, but in insignificant numbers. The human brain has 100 billions neurons working all in parallel at the same time in non discrete time. We can simulate a few thousands simplified very slow neurons in discrete time and we don't understand Glia cells. That's less than a fly's brain.
Artificial Intelligence is not there yet and is not coming any time soon unfortunately, but that should not stop us from researching and trying. Even if we are not ever able to reach the Graal, the path to it is probably more interesting than the Graal itself. And we are humans, we have to dream, that's what we do. Dream on guys.

You contradict yourself. If nobody knows for sure what the future is made of then noone even people who write posts in this forum including me and you can claim that it will never happen. People who are already working on this issue have evidence that it can happen any time soon. Maybe not for us but for our scions or grandscions. But i disagree that it can't happen or will require at least 1000 years to achieve. It can happen even after 100 years if not even quicker.
https://www.youtube.com/results?sear...l+intelligence
Same kind of thinking would apply if concept of fridge or car or other nowaday item would be introduced to medieval or other past people. They would think it is science fiction. But not for us.

Those who live in the UK may have seen this morning's BBC Click which, most unusually, was taken up with a single story. For non-BBC watchers, Click is a 15-minute update on technology news which is broadcast as part of the weekend's morning news bulletin.

The story was the tragedy in Tempe, Ariz., where a pedestrian was killed by a self-driving Uber car, and the extra information that it provided made me very angry. The road at the time was empty. There was no traffic in the opposite lane, so the car could easily and safely have swerved to avoid this woman. The road is well-lit at night, so the video that Uber released showing her suddenly appearing out of pitch darkness was obviously doctored. There was a "security driver" at the wheel who was supposed to take over if something like this happened, but she was not watching the road because she trusted the technology. Finally, the car had automatic braking installed (as all Volvos do) and this should have stopped the car as soon as a collision situation was detected, but Uber had disabled it because it interfered with their AI system.

It looks as if the driver will be charged with negligent manslaughter or whatever the American equivalent is. In other words, she will be the sacrificial lamb thrown to the angry relatives to appease them. Uber will not be charged with anything, even though it was their car, their AI driving it, and they who disabled the car's built-in safety features. Apparently the state government does not want to discourage other self-drive technology companies from using its roads (and its citizens!) for testing out this technology.

You can find the full story here.

What can i say if not iDiotism. However that accident did not happen this year. And what i agree with, cause i know driving basics, is why would they either put incompetent driver behind wheel or disable safety systems that work(there is video how Volvo braking saved school boy.) or simply allow car that is not working properly on road..

As for dreaming about auto pilot cars. KITTs will always win.

What's that got to do with anything? If it was your mother that got run over, would it be any comfort to you that it didn't happen this year? Anyway, it takes time for the truth about events to be established.
I don't know why. The point is they did. And the woman at the wheel was not necessarily an incompetent driver. What she was was a passenger. She wasn't watching the road because the AI was driving and she trusted it. Everybody always said that having a human being around to take over in emergencies would be a kind of backstop, but that isn't what happened in this case.

LOL? People learn context before reply to me. I do not insult in my posts i just describe how i feel sometimes. The whole situation is iDiotic and should not happen in first place due to reasons already mentioned. Also while i agree that date not make difference for victim(s) it makes difference for self driving vehicle world cause from that day to today improvements exist to not repeat such accidents. But in that situation it actually shows that cause of accident was not self-driving car but human error(s) - one human error is driver other human error is company people who disabled what should be in all self driving cars : "just in case" safety systems. In other words if that accident would have happened without safety driver and safety systems online then it would be fault of self driving car providers.