Would you rather have the car kill the most humans?
In an emergency situation the car would steer for optimum stability and brake for optimum deceleration.
No ethical decisions are involved.
Crane
All things considered, I’d rather not have the AI making the decisions. At least, no AI I can envision. There have been no fundamental breakthroughs in machine intelligence in a long, long time - the processors have simply gotten so fast that they can do a brute-force simulation of cogitation. (E.g., most chess computers of recent generation, which “play” by doing exhaustive pattern analysis.) Which is not the same thing as real judgment.
For routine control, sure, machines can outperform human judgment tirelessly. Crises can’t be mathematically mapped no matter how many ziga-flops the processor churns.
The OP presents the problem from the position of the designer. No, corporation would design a system that accepts legal and financial liability for “ethical” decisions.
A more realistic issue is what happens when the system is compromised. Modern
ABS systems constantly run diagnostics. When any error is detected, the system reverts to manual.
I was driving a car when the left front spindle failed. The front of the car dropped shearing off the brake controls and sensors. There was no steering or braking assistance.
Should automated automobiles remain active during partial failure?
Crane
Pilots usually have some time to take over - car accidents happen much faster usually. And even highly trained pilots screw up sometimes. We’re not talking about them, we’re talking about Joe Slob who is busy texting or daydreaming when the crisis happens.
To answer Critical1’s question - I’ve never had to avoid someone running from the cops in 45 years of driving. Life is not like an action movie. In any case, it is fairly simple to pull over, and your car might get alerted about this by other cars, and pull over long before the moron shows up.
We don’t anticipate every conceivable emergency either. If the car is programmed to deal with 99.9% of emergencies (relatively simple since the programmers will get data on emergencies experienced) it will do better than most of us do. And most of the rest can be handled using basic principles, which is what we do. Maybe you spend your spare time anticipating your response if a UFO lands on the highway ahead of you, but most of us have better things to do with our time.
Fault tolerance is a well researched subject. The first thing is that the car itself should be drivable under most failure conditions. But one of the most important components of fault tolerance is anticipating failures. A good, well instrumented car should detect that your spindle was about to fail and complain about it, just like my Prius detects low air pressure in my tires and complains about it without me having to check all the time.
I could see cars with critical problems refuse to go anywhere but the repair shop.
Okay, you’re gone somewhere beyond tech to gosh-wow, verging on near-fantasy.
It’s one thing to have active monitoring on critical systems, especially moving, wearing or power-using ones. The idea that you could wire up a car so that any of fifty passive parts are monitored for potential failure is… well, NASA dreams of that kind of safety net. And doesn’t have it.
For one thing, your future-Prius would weigh about 6,000 pounds, not only from the weight of the sensors and wiring, but from the collateral engineering that would support, say, a spindle strain sensor.
Today’s cars could weigh 1000-1500 pounds less, using modern materials and manufacturing techniques… if we hadn’t crammed them full of smog gear, safety gear and every electronic gizmo imaginable. Study the lesson.
I have an older version of this, the first generation, 2 -3 years out of date version of this:
And I think it works well. It doesn’t take over steering (yet), but it does what it was designed to do.
My older version has a few issues with false lane sway warnings when encountering strips of road tar, and it doesn’t always recognize lanes when the lane stripe paint is missing. Biggest problem for my older camera system is when I am driving toward the sun low on the horizon. The system can’t handle all that sunlight (and I find it difficult too).
My older version has already recognized a lone person standing some distance away in my path and has alerted and applied brakes before I did ( I was about to, but it did it first )
When I think of 10-15 years down the road (hah) when this kind of system will be common, as well as many cars being able to communicate their presence and stats to other cars, at the speed of light, I think we will all be much safer.
How many does it take to kill you?
I can see North Korea, or Iran, or anyone for that matter hacking the self-driving programs, creating a giant demolition derby of our freeway system.
You keep focusing on the one-in-a-million accidents. It’s the one-in-a-hundred accidents you want to watch out for. Drunk drivers are COMMON. Idiot drivers are COMMON. If we get those off the roads through driverless cars, we will all be a LOT safer.
Kind of an “if a frog had wings” wish, though. You’re talking about trillions - literally - in roadway improvements and changes to accommodate any significant fraction of US roads, and the last 10-20% are going to be a huge part of the cost.
Trying to divide the road system between automated and manual drive is not going to solve enough of the problem to make it worthwhile. So you’re safe on the highway - and get creamed by the drag-racer or drunk on the half-mile between the offramp and your driveway. Win or lose? Especially given that the per-vehicle-mile crash and death rates are far lower for multilane, controlled-access roads than your basic surface streets?
At best, I think it’s a century-long solution to be implemented one city center and one major connector at a time, and the bootstrap to automated-only roads is going to be a very slow one. Just to WAG out, would you buy a $60k car with auto-drive that is probably not the best choice for manual-drive roads? When only your 10-mile commute is auto-drive, and everything else is manual?
I’d think that *trillions *is overstating it a bit. The Google Car hasn’t required upgrades to California’s existing roads. Certainly they’ll need guidance at first, but over time, the machines will get better, the GPS will get tighter and the maps of the existing roadways will get rock solid.
I’d pay 60k for a reasonable car that could drive itself, even if it needed a nudge once in awhile.
Really? Have you looked into the costs of what’s considered absolutely essential highway and road infrastructure - just restoring and rebuilding the roads to safe driving condition across the country? The t-word comes into it.
Now add whatever extensions, physical and electronic, to enable large-scale automated driving on these road that are now safe and reliable enough for same.
You can make ANYTHING work in a limited trial under good conditions. Extrapolating from those limited circumstances to something closer to unlimited application breaks many a brilliant idea.
Having to drive it manually half the time is more than a nudge - or are you willing to pay high for the car, high for road and infrastructure costs in your region, and then live only within “converted” areas? Just to get auto-drive part of the time?
this has nothing to do with self driving cars.
Sensors are cheap and light. The car network they’d use to communicate with the processor is already there. A lot of it is build in already to let the processor running the engine optimize fuel mix and the like.
Things don’t just fail, usually. For instance, an increase in vibration might indicate that something is loose. I’ve seen enough presentations from automotive engineers to know that they take this kind of thing very seriously, and design so that whenever possible the car fails in a way that it can be driven, with difficulty perhaps, to the shoulder. You do not want your customer’s car to become a brick in the middle of a crowded freeway where traffic is moving at 60 mph. That you don’t hear about cars doing this indicates the engineers do a good job.
I’m more familiar with fault tolerance inside of ICs and in telecom systems, to be honest.
I don’t think most electronic gizmos - which are options, after all, weigh more than the AM/FM radio in one of my earlier cars. The computers in the engine replace heavier mechanical equivalents. And memory is cheap and programs don’t weigh anything.
None of this has much to do with smart cars. Dumb cars will get the benefits of this also.
Airline pilots are so far away from your every-day car driver that it’s meaningless to compare. Commercial pilots get way more training, there is (almost) always a co-pilot, and they have ATC to help guide them. Even so, commercial planes are getting more an more automated because the computers are getting safer.
I don’t trust your judgement. Drivers can be tired, distracted, or drunk. Maybe they’re in a hurry or got blinded by the sun. They cut each other off or misread signs until the last second. This won’t happen with computer-driven cars. It is easily conceivable that automated cars could drive using much more information than you could ever process. For example, if cars communicate their speed and position to each other your car would know to slow down sooner while going around a blind corner. Or they could warn each other of ice.
Our infrastructure is terrible, granted. What of it would not need to be upgraded even without self-driving cars? It is not like potholes only affect Google cars.
My commute, from door to door, would all be under good conditions. I don’t think the Google cars only start being self-driving when they get on the freeway. I suspect 90% of people, at least, would never have to use the steering wheel except, maybe, to pull in their driveway. And my Roomba can dock by itself - I suspect a car could also.
Few drivers in the Bay Area spend half their time off the freeway. I might because going to work I use back roads for much of my trip to avoid congestion. I wouldn’t if I could read while the car drove.
I commute about 15 miles one way, and at most 2 miles of it is off the freeway. And those with longer commutes who’d be even more willing customers have a higher percentage of their commutes on freeways. That’s plenty of a market.
You remind me of electric car skeptics. Sure, some people commute long distances, or spend a large percentage of their time on long trips. Those people are not good candidates for electric cars. But lots more people commute shorter distances. That’s why four people in my department own electric cars already, and the big problem around here is congestion around charging stations.
And I still don’t understand why you’d have to convert anything to be self-driving friendly.
I’m thinking of the last four accidents where I was involved or which I witnessed. None of them would have happened if both cars had been computer-driven. I suspect almost everyone would say the same thing.
How many lightning strikes does it take to kill you?
Do you organize your life around not being struck by lightning?
IF we can get self-driving cars to the point where they can handle everyday road conditions significantly better than the average driver, I’ll be ready to turn over my keys - and I’m one of those people who actually likes driving, but not as much as I would enjoy reading a book or solving a kakuro puzzle, or arguing on the Dope. Which is what I could do while the car drove itself to work.
A few months ago, my brakes failed. Pressed down on the brake pedal, and had nothing. Fortunately, I was going <5 mph on a level street. I used the curb to bring me to a full stop. It didn’t even occur to me until afterwards that I could have pulled on the emergency brake, which is scary because only a minute before, I’d been going downhill at 30 mph towards a T-intersection.
The veterans’ memorial on the other side of the intersection would have kept me from going into the Chesapeake Bay if the brakes had given out then, and if I’d managed to not hit cross-traffic first. But it would have been a pretty awful event, no matter what, if the brakes had failed then.
I think it’s a pretty safe bet that a self-driving car wouldn’t forget in a pinch that it had a backup braking system. It would remember, and it would use it. Gimme that self-driving car.
The question I have is, of the technical obstacles in the way of getting self-driving cars out of meticulously mapped Mountain View (down to every last manhole, curb cut, and traffic sign, IIRC - way more detailed than Google Maps, thank you) and into the real world, are they all solvable by the continued application of Moore’s Law, or do some of them require solutions that are immune to ever-faster computers with more terabytes? To me, that’s the real question that will determine whether we see self-driving cars in 10 years, or whether it’ll be more like 30.
I’d like to see self-driving cars in 10 years (9 would be even better) that are affordable to us upper middle class types, because the Firebug is 7 years old, which means he’ll be eligible to drive in 9 years, teenage boys aren’t the greatest drivers in the world, and my son is flakier than average. But at this point, I don’t expect self-driving cars to get here that soon, alas.