Slashdot Mirror
House Passes Bill To Speed Deployment of Self-driving Cars (go.com)
The House voted Wednesday to speed the introduction of self-driving cars by giving the federal government authority to exempt automakers from safety standards not applicable to the technology, and to permit deployment of up to 100,000 of the vehicles annually over the next several years. From a report: The bill was passed by a voice vote. State and local officials have said it usurps their authority by giving to the federal government sole authority to regulate the vehicles' design and performance. States would still decide whether to permit self-driving cars on their roads. Automakers have complained that a patchwork of laws states have passed in recent years would hamper deployment of the vehicles, which they see as the future of the industry. Self-driving cars are forecast to dramatically lower traffic fatalities once they are on roads in significant numbers, among other benefits. Early estimates indicate there were more than 40,000 traffic fatalities last year. The National Highway Traffic Safety Administration says 94 percent of crashes involve human error.
And a sizeable fraction of the remainder would be eliminated by automatic monitoring of car performance. (brake failure due to neglect say).
Then there are those accidents that are 'unavoidable' (debris falling on road say).
But become not-unavoidable if you have an AI with reflexes beyond a trained stunt/rally driver who has a week to prepare.
If the self driving car is not at fault in the accident (the vast majority of present day cases), then the self driving car has tons of data both in visible light and other parts of the spectrum to show everything that happened prior to the crash.
It is an inevitability, once statistics catch up with it, that a self driving car will be the cause of a major accident. I doubt that this can ever be a criminal trial, because no criminal intent is involved at any level of the design or implementation of the self driving car. It's an accident.
As more self driving car accidents occur, the self driving cars will get better and better at avoiding them (unlike puny humans). If for no other reason than the designers will make improvements based on all of the data from each accident.
In court, the lawyers can argue about how the self driving car came to the decision to run over a group of people whose skin color it did not like. There won't be any NDAs. The owner of the technology will file a motion to keep the technology under seal. It will be discussed in court, but in a closed courtroom, with court members bound to secrecy about the technology. This is nothing new.
BTW, I'm all for requiring safety standards of automakers. (OMG! regulation!) As long as you can quantify it in a way that is clear in the law. You can't have laws that are so vague that you can unintentionally violate them. There needs to be a bright line.
The line cannot be that no accidents can occur -- because self driving cars are already safer than cars driven by puny humans.
I'll see your senator, and I'll raise you two judges.
That will make dying hurt less right?
No, but the ~90% reduction in auto-related fatalities will.
Linux, you magnificent bastard, I read the fucking manual!
Damn! Will wonders never cease? Look, all of you self-driving car Luddites can just stay the hell away from them. For the rest of us, though, it makes PERFECT sense to have National standards for these cars in the same way that we currently have National standards for all current vehicles. Otherwise, you'd have an amalgam of incompatible state-based standards that would severely hinder development and deployment. (California's stricter emissions standards being the only exception I'm aware of.) What's wrong with that?
I'm always curious when self-driving discussions appear. I'm "somewhat" informed on this topic, and am relatively neutral; but, I can't help but believe that tech folks are a bit too optimistic about the benefits of "eliminating human error." For example, I see in these types of discussions the example of debris on the road. Theoretically, most human drivers have the ability to see such debris and determine a course of action, and most of the time they choose correctly and avoid disaster.
On the other hand, it would only take a single bug in an AI "debris subroutine" running in a whole bunch of self-driving cars to choose the wrong course of action 100% of the time. Such a bug would *probably* only be identified after enough failures were accurately recorded to piece together a pattern that could point to it (i.e., an incomplete test plan didn't catch it, a code review didn't catch it, differences between virtual test worlds vs. the real world hid the defect, etc.).
I guess if someone could convince me that it is possible to write 100% bug-free code, I would feel better about this. However, what I perceive as the somewhat naive optimism of technical folks is somewhat terrifying in this context.
When I drive in Atlanta I'm wrapped in a 2.5 ton steel machine that is on a full steel frame. It's a 2003 Mercury Grand Marquis and in one accident a toyota splattered itself to pieces on the side of it and 1,000 dollars later my car was fine. I'm still afraid, not because of me
Well, you probably should be because you're clearly poor at making choices. The "splattering" you so deride is caused by things called "crumple zones". Those absorb the energy of the impact so that the car splatters, not you. That person's car going "splatter" likely saved *you* from serious injuries. You're basically gambling that most people have safer cars than you and that one of those unsafe car drivers won't be the one to hit you.
What's amazing is you clearly know how dangerous the roads are, yet knowingly drive a car that doesn't have good crash test ratings.
SJW n. One who posts facts.