> One of the world's top self driving car engineers reckons we might never get Level 5.
I have some minor-level experience with self-driving car software, algorithms, etc. The majority of this is from various MOOCs I have participated in over the years, as well as research papers, books, and other things I have read and consumed. In short, I am not an expert, but I am not unfamiliar with the technology, either.
I personally think we'll never see actual widespread usage of self-driving vehicles outside of a very few narrow and carefully controlled (and regulated) cases. At least, I suspect, within my lifetime (which honestly, I'll be lucky for another 30 years or so).
My reasoning is that people will only trust perfection when it comes to riding in a self-driving vehicle. They will only be willing to use one if they can be assured that it will Never Crash, or be crashed into. They have no problem driving a car themselves, or being surrounded by other people driving cars. They have no problem with those crashing and even killing people - maybe even themselves (though they tell themselves fairy tales of it-will-never-happen-to-me to soothe over the reality). But introduce a machine into the equation...
...and that machine has to be Perfect. It cannot make any mistakes. It must avoid issues and be Safe 100% of the time, no exceptions.
In other words, people want the impossible from a machine, but will give utmost allowances to themselves and others as "humans".
I think a lot of this has to do with assignment of blame. When they crash or are crashed into - there is someone to assign blame to; themselves, the other driver, etc. Someone they can yell at, figuratively or literally.
A self-driving car? No one to yell at. No one to assign blame. Nothing that will feel bad for its error or failure to avoid something.
People can't handle that. They don't want a self-driving vehicle that has a safety factor of say, "seven 9s" - it has to be 100% safe or nothing. Because even if it makes a mistake only once out of a million miles of driving, that is still not safe enough. They want the unobtainable - a perfect machine, a machine that will never fail. Nothing like that can or will ever exist (basic laws of thermodynamics prevent it, for one thing).
Even though they themselves, or even the most professional of professional drivers - can't even come close to approaching this level. It both madness, and understandable at the same time.
> My reasoning is that people will only trust perfection when it comes to riding in a self-driving vehicle.
From what I've seen I think your model of human behavior is flawed. Aircraft have been mostly fly-by-wire for decades and software flaws have led to some crashes (most recently those of the 737 Max), but that hasn't stopped people from flying in planes.
People have already been killed by self-driving cars, but that hasn't stopped the testing programs. Tesla automation (though far from full self-driving) has led to a number of fatal accidents yet people still buy Tesla's and I haven't heard of any public outcry to ban them from the roads.
I too am skeptical that we'll be seeing full self-driving cars anytime soon, but not for the reason you give.
I think the real issue is that AI-like technology has a 90-10 problem. It's relatively easy to get to 90% of human level performance but that last 10% is much much harder - unpredictably harder. Unfortunately self-driving cars are an area where 90% isn't good enough, even 99% isn't good enough. You probably need to get to something like 5 or 6 9's to be acceptable for broad general use and no AI-like technology has yet gotten anywhere close to that.
EDIT: I didn't mean to imply that the 737 Max is fly-by-wire, but the MCAS system that caused the crashes is a comparable technology.
The usual practice is for autopilots to be switched off when the plane is within 200 feet of the ground. (Or a similar distance.) I think a self driving car would be a much easier sell if it never came within 200 feet of anything it could hit.
The way I see it, we will have the manufacturers of the cars to take the blame and to ask for compensation. They will probably take out insurance for each car they sell to pay out any of these claims with the premium that depends on the accident rate, added to the price of the car. This would be better even if the accident rate is the same as the current human one. I think we would see self-driving cars of limited capability become commonplace in the near future because of the extra safety it offers us especially from blame if an accident does occur. Also, the complexity of developing self-driving systems is because we are trying to adapt them to current-day roads if we make changes to the roads by adding lanes, stationary sensors to help self-driving cars we could make accidents are very rare.
I have some minor-level experience with self-driving car software, algorithms, etc. The majority of this is from various MOOCs I have participated in over the years, as well as research papers, books, and other things I have read and consumed. In short, I am not an expert, but I am not unfamiliar with the technology, either.
I personally think we'll never see actual widespread usage of self-driving vehicles outside of a very few narrow and carefully controlled (and regulated) cases. At least, I suspect, within my lifetime (which honestly, I'll be lucky for another 30 years or so).
My reasoning is that people will only trust perfection when it comes to riding in a self-driving vehicle. They will only be willing to use one if they can be assured that it will Never Crash, or be crashed into. They have no problem driving a car themselves, or being surrounded by other people driving cars. They have no problem with those crashing and even killing people - maybe even themselves (though they tell themselves fairy tales of it-will-never-happen-to-me to soothe over the reality). But introduce a machine into the equation...
...and that machine has to be Perfect. It cannot make any mistakes. It must avoid issues and be Safe 100% of the time, no exceptions.
In other words, people want the impossible from a machine, but will give utmost allowances to themselves and others as "humans".
I think a lot of this has to do with assignment of blame. When they crash or are crashed into - there is someone to assign blame to; themselves, the other driver, etc. Someone they can yell at, figuratively or literally.
A self-driving car? No one to yell at. No one to assign blame. Nothing that will feel bad for its error or failure to avoid something.
People can't handle that. They don't want a self-driving vehicle that has a safety factor of say, "seven 9s" - it has to be 100% safe or nothing. Because even if it makes a mistake only once out of a million miles of driving, that is still not safe enough. They want the unobtainable - a perfect machine, a machine that will never fail. Nothing like that can or will ever exist (basic laws of thermodynamics prevent it, for one thing).
Even though they themselves, or even the most professional of professional drivers - can't even come close to approaching this level. It both madness, and understandable at the same time.