Slashdot Mirror
Why Self-Driving Cars Should Never Be Fully Autonomous (roboticstrends.com)
An anonymous reader writes: David Mindell, an MIT professor, says self-driving cars should never be fully autonomous. "There's an idea that progress in robotics leads to full autonomy. That may be a valuable idea to guide research but when automated and autonomous systems get into the real world, that's not the direction they head. We need to rethink the notion of progress, not as progress toward full autonomy, but as progress toward trusted, transparent, reliable, safe autonomy that is fully interactive: The car does what I want it to do, and only when I want it to do it." Mindell writes, "Google's utopian autonomy is a more brittle, less functional solution than a rich, human-centered automation."
safe autonomy that is fully interactive: The car does what I want it to do, and only when I want it to do it."
Sure, if I own the car it should do only what I want it to when I want it to, but why should I own a car at all? I use a car only a few times a month, driving maybe 5000 miles/year total. Why should I spend $30,000 on a depreciating asset and devote 200 sq ft of space towards housing it.
I want to call a car and have it come when I want it, take me where I want to go, then go away until I need it again.
There are already systems that will warn you if you're drifting out of your lane, and systems that will warn you/apply brakes if you're in danger of collision. And of course systems that will plot a route for you and give you step by step directions to your destination have been around for quite awhile at this point.
If the goal isn't full autonomy then it doesn't really seem like we need to do much more research and development. How boring will it be to be "driving" a car that can do 99% of the driving by itself but insists on you paying attention (at least intermittently) to do the remaining 1%, instead of kicking back and enjoying your time doing something else?
(And note that anything less than full automation will provide little benefit to the biggest commercial interest, long distance trucking. Having to pay a person to ride along and babysit the automation doesn't save anything over just making that person drive in the first place.)
This Space Intentionally Left Blank
“The notion of ceding control of something as fundamental to life as driving to a big, opaque corporation - people are not comfortable with that,” -- David Mindell
I'm not sure I agree with that. Sounds similar to someone 150 years ago saying "The notion of ceding control of something as fundamental to life as growing and hunting the food to feed my family to a big, opaque corporation - people are not comfortable with that".
People get comfortable with a great number of things if you make their life significantly better even while asking them to give up a little control.
-- All that is necessary for the triumph of evil is that good men do nothing. -- Edmund Burke
The guy in the article has just written a book about robots and automation. It was released today.
The purpose of this article, and the connection with driverless cars, is to draw attention to the book, so he can sell more copies.
If he had brought up reasonable points, I could accept that, but the points he brings up in the article seem rather lame to me.
"First they came for the slanderers and i said nothing."
From the article:
“[Full automation is] just proven to be a loser of an approach in a lot of other domains,” Mindell says. “I’m not arguing this from first principles. There are 40 years’ worth of examples.”
For how many of those 40 years have today's sensors, computing hardware, and AI been available?
It's possible that fully automated driving will turn out to be hard like commercial fusion power, or like commercial space travel. I think it's more likely, though, that it will turn out to be hard like speech recognition or cheap, lightweight flying drones -- each popularly regarded to be "a few years away" for decades, until suddenly it was here, courtesy of a few research advances and a great deal of exponential improvement in computer hardware.
I don't think we'll have self-driving cars right away. Instead, we'll have cars with "Enhanced Cruise Control." You get into a lane on a highway, hit Enhanced Cruise Control, and your car will stay in that lane (turning left or right as needed) keeping to the speed you set but slowing down if needed (e.g. if the car in front of you brakes). For long car trips, this would mean that a bulk of your trip would be automated. You'd still need a driver there to take control once you wanted to leave the highway and you might not be able to use this during bad weather (just like you wouldn't put cruise control on during a snowstorm), but it would be one step towards autonomous cars.
As the software gets more refined and the edge cases are dealt with better, the car will be able to handle more driving situations. For example, "automatically stop at red lights" or "keep going straight unless the driver indicates otherwise." Eventually, cars driving themselves will be the norm and human drivers will be the exception.
My sci-fi novel, Ghost Thief, is now available from Amazon.com.
Incorrect in principal and practice. It's like the angry bear vs the two people. You don't have to be faster than the bear, just faster than the other human being. The cars' don't need to be perfect - they just need to beat a human mind that is NOT an expert. If the car by itself can do better than a human without the AI, than it is sufficiently good to replace the current model that is human without the AI.
The idea that we need to achieve the maximum possible result of human+AI ignores the current situation's inherent problems of poor drivers, the elderly, drunk drivers, children, etc. etc. etc.
Parents of teenagers, children of the elderly, alcoholics and their loved ones ALL are VERY comfortable with the idea of having the car drive, not the person. They will provide the demand and market. Once their demand is met, then simple continuing research will eventually make EVERYBODY comfortable with letting the AI drive. If you are OK with the AI drive your teenager, your grandma, and your drunk cousin Joe - knowing they might be in the car next to you, then you will be OK with letting them drive you.
His comparison of other modes of transportation such as space, submarine and airplane, is also flawed.
The main reason we never automated those is that their need for accuracy was much much higher than we have for automobiles and up until recently, computers have not had the real ability to beat a trained, expert human. But in cars, they don't have to beat an expert, just a licensed and impaired human - drunk, young, elderly for example.
A better comparison is to look at welding. Originally people welded. Then robots came along and were better. Automated welding has taken over a large proportion of welding, we don't have humans over-riding them. Why? Because the robots are better at it than humans in most cases.
excitingthingstodo.blogspot.com
Wait: TV sets which self-align vertical and horizontal hold.
Subway systems which run completely automatically (granted with all sorts of staff there to pull the Panic switch and/or make travellers feel more 'safe')
elevators which automatically go to requested floors.
Heck, IP packets which automatically make it to their intended destination.
I think this guy has no idea where and how software is running automated control systems.
https://app.box.com/WitthoftResume Code: https://github.com/cellocgw
Because people are irrational. We fear exotic deaths more than we do mundane deaths. Travel by airliner is much safer than travel by car, yet every time there's a plane crash we have a huge investigation for the purpose of figuring out what went wrong so we can prevent it from happening again. But tens of thousands of people die in car accidents, and all they get is a brief police report stating it was an accident without really delving into the cause. Why? Because death by airplane crash is more exotic than death by car crash. The same thing plagues nuclear power. Death by radiation is exotic, death by falling or getting lung cancer from soot inhalation is not. So we scrutinize and heavily regulate everything to do with nuclear power when it's already the safest power source we've ever invented in deaths per MWh of power generated, while turning a blind eye to deaths by coal (pollution), wind, and solar (primarily falls during maintenance - their diffuse nature means there's a lot more maintenance to do).
You double down on this if the accident was in your control vs out of your control. If you could've done something to prevent the accident (was driving a car) but failed to so, you say "Oopsie, I won't make that mistake again. No give me my keys back." If someone else could've done something to prevent the accident (driving a bus or piloting a plane) but failed to do so, you sue the bastard for everything he's got and try to get him banned so he never drives/flies again.
The combination of these two means autonomous cars have to become a helluva lot better than human drivers before they'll be accepted. Dying because of a typo in a line of code counts as really exotic, and the press will have a field day with it the first time it happens. And the makers of the autonomous cars will need huge insurance policies to deal with the extra liability they'll incur, since it'll likely be bigger than the sum total of all private auto liability insurance policies today (a few percent of the purchase price of the vehicle every year it's in operation).
More like it is so rare AND so many people die that the news organizations play it over and over and Over and OVER and OVER!!!
Now imagine if those news organizations gave that same coverage to every single car crash (with a fatality).
The news would be nothing but car crashes.
And people would start to be terrified of driving anywhere.
Also TFA is incredibly stupid. His examples are meaningless in this context. An autonomous car SHOULD be able to stop itself and turn control over to a human when it encounters something it cannot handle.
And, over time, those cars WILL become more popular because the people who use them will pay lower insurance rates. That is because any accident they are in SHOULD be the fault of another driver OR the programming.
Look at the airports around Thanksgiving. They will be packed with people. Because people see the value in flying. Even when they give up control to someone else and it could result in an "exotic" death. The same with autonomous cars.
I think one of the biggest problem with autonomous vehicles is directing them where you want to go. Let's say you are in a crowded parking lot and you want the car to park in the 3rd spot, 4 rows over. How do you instruct the vehicle efficiently to do that without taking control of the steering yourself? That's not an easy thing to articulate clearly. Worse, how do you tell it where to go when you don't clearly know the final destination yourself? Sometimes you don't have an address or the destination is very large like an airport.
I think autonomous vehicles might do well on major roads but I think the problem of giving specific instructions is going to be a LOT harder than many people think.
Also TFA is incredibly stupid. His examples are meaningless in this context. An autonomous car SHOULD be able to stop itself and turn control over to a human when it encounters something it cannot handle.
I don't think you (and many other people) have really thought this one through.
If one is riding along in an autonomous car, they are not going to be paying attention to their surroundings. They're going to be talking to others, texting, surfing the web, daydreaming, or even nodding off, if not entirely asleep. They might look around once in a while but they're not going to maintain the kind of focus on the road that would give them any kind of real situational awareness. It's just not going to happen. There's no reason to pay attention when your car is driving itself. And don't kid yourself. People may say "Yeah, I'd stay focused on the road." But they're full of crap. Anyone behind the wheel is not going to pay anywhere near as much attention to the road when the car is driving itself as they would when they're driving the car. If you have to pay attention to the road, why have the car drive itself in the first place?
Suddenly, the car runs into a problem that it can't deal with so it hands over control to the driver. But this "driver" has no idea what's really going on. Hell, it will probably take a fair number of seconds for the person to realize that the car has handed over to them in the first place. Reaction time will be slowed substantially by the fact that they have been lulled into a false sense of security. They are entirely unprepared to take over at that moment. Suddenly being forced to change focus is difficult and time consuming. And a lot can happen in the many seconds it takes for the average person to realize what's going on well enough to do something about it.
But here's the thing. The only reason the car would hand over control is that it is in a situation where it doesn't know what to do and that usually means there's something significantly wrong RIGHT NOW! Given that, do you really think it's a good idea for the computer to just say "screw it, I'm out" and suddenly dump control over to a passenger who has no idea what's going on around him?
No, the idea that an autonomous car would have an option to hand over control to a passenger in the car is ludicrous on it's face. There's no way in hell anyone would design a car to do that (for the consumer audience) once they spent a few minutes understanding the problem. For research cars that get used in controlled environments? Sure. But the first average consumer caught napping when their car handed off control would sue the hell out of the manufacturer if they survived the crash. Or the family of the person would sue if they didn't. Either way, that kind of liability is just not something a manufacturer would ever consider taking on when they're building for the public.
We're not going to see autonomous cars handing off control to the driver. Not once they hit the consumer market. Or at least not long after they hit the consumer market.
You're making the same mistake that TFA makes.
No one is saying that a car IN MOTION should cease autonomous operation.
What I said was that the car should STOP and then turn over control to a human when it encounters a problem it cannot handle.