Newco Shift:
The received wisdom in the Valley is that the technology for self-driving cars is already here — we just have to wait a few years while the slowpokes in Washington get with the program. Within five years, we’ll all be autopiloted around — free to spend our otherwise unproductive driving time answering email, Snapchatting, or writing code.
Except, come on, there’s no way that’s gonna happen. Not in five years, anyway.
It’s the messy human bits which will slow it all down.
The optimism for self-driving cars is great. It will drive innovation and excitement. But it must be tempered with reality.
I don’t think anyone was stating that in five years ALL cars will be self driving. But, I’d wager quite a few more will be, certainly many many more than there are now.
It’ll be interesting to say the least to see where the tech is in five years.
Now I’ll have a David Bowie song stuck in my head all day.
The five minutes it takes to get any where I need to go on a regular basis won’t leave me a lot of time for such tasks as email or writing code……
Gosh darn all those burdensome, nitpicky “safety” “regulations.” /s
I’m guessing there are at least a few Silicon Valley VCs who’ve been boasting in that vein, trying to spur demand for their interests.
I’m never impressed by these arguments that the Trolley Problem will be a significant component of the safety approach self-driving cars will have. At a conceptual level it’s possible to imagine that the cars will engage in some kind of high-concept freeze-frame scenario analysis when something unexpected occurs, but I’ve never seen anyone establish that the engineering is actually going to work that way.
I really expect them to just slam on the brakes and try to swerve around obstacles on a second-by-second basis. It’s going to be a bottom-up approach rather than top-down, reactive rather than planned, a few moments of chaos and then the insurance company will pay the settlement. Same as human drivers except the self-driving cars will probably try not to exceed 25mph around kids.
I still think it will happen in increments (we see emergency stop and other features being advertised today) until we finally get to the last bit and allow the ability to actually drive the car to happen. 5 years? Nah. 15? Likely.
That does seem likely. But as partial self-driving becomes more common, we will find out whether it results in insufficiently attentive drivers unable to react quickly to emergency situations, leading to more accidents. Something like this happened with aircraft, and while I don’t remember much detail there, the current thinking seems to be to keep the pilot involved more, not less.
This is the thinking that has caused some self-driving programs to focus on complete, all-at-once self-driving exclusively.
interesting point. The commercials do seem to highlight this – there’s one with this woman singing about how she’s awesome but then an alarm rings as she drifts into the oncoming lane and my reaction is “pay attention, stop effing SINGING is the answer!”
The problems of self-driving range from relatively simple to enormously difficult. I believe we are currently in a state where enough of the simpler problems have been solved that self-driving appears to be a realistic near-future possibility, while the more difficult problems will take much longer to solve than most people currently realize.
One of my favorite examples is that of driving in snowy and icy conditions. Road markings disappear, and with them any sort of lane discipline. Traction ranges from quite decent to almost completely absent, and drivers have to evaluate and respond to each different type of snow, ice, slush and miscellaneous muck on a second-by second basis, often using very-subtle visual cues such as the precise nature of reflections off the surface. Factors such as the camber of the road and even strong winds begin to greatly affect maneuvering. Other drivers occasionally lose control, and the ability to predict when this is most likely (are you on a main road approaching an intersection with a slick downhill stop required of cross traffic?) also becomes critical.
This is just one extremely challenging environment that will take, I suspect, much more than five years of advancement to master even on a non-production “test” basis, several more years to enter production, several more years to become common, and decades beyond that to reach near-universal adoption.
If it’s up to me I’ll still be driving my older cars, which don’t have computers interfacing with the driver and with every part of the drivetrain; and that, unlike today’s cars, I can actually see out of.
Did you read the part of the article where it states that Mercedes has already programmed their experimental self-driving cars with a specific response to the trolley problem? (To maximize the safety of the car’s driver even at the expense of other lives).
If truly self-driving cars ever do exist, they will certainly need to make decisions such as this.
You do bring up another problem with self-driving cars, though, which is that if the cars are programmed to assume worst-case scenarios with regard to other road-users, then they will by extension be forced to drive in an extremely conservative manner. Remember the Google car that caused an accident by moving into an oncoming lane to go around a stopped bus? This is not a major problem for a human driver but could be a catastrophic problem for the possibility of self-driving, especially in chaotic urban situations.
Any given self-driving technology may be forced to decide between an intractable level of conservatism that would make efficient transit impossible; and the acknowledgment that perfect safety is not attainable, with all the problems that implies.
I keep reading these naysayer articles about how government will be so slow to adopt self-driving vehicles, much less have regulations in place quickly, and yet our top government city, Washington DC, is part of a worldwide initiative to speed this along.
http://dcinno.streetwise.co/2017/01/11/d-c-will-take-global-lead-on-self-driving-cars/
In fact, the Olli Bus is already in trial on the roads in DC.
http://www.theverge.com/2016/6/16/11952072/local-motors-3d-printed-self-driving-bus-washington-dc-launch
https://localmotors.com/olli/
Freight trucks are already self-driving in Nevada, as well. And Florida has self-driving/remote-controlled safety vehicles for construction in operation in cities there.
So, while ubiquitous self-driving might be a bit off, we are in fact seeing automated vehicles come along quickly for many specific scenarios. I think 5 years is a perfectly fine time frame to start seeing them everywhere and start getting people used to the idea of their existence as the rules and all are worked out for civilian use.
I read that, but while I’m sure that’s a design principle for them it sounds a little like PR. It doesn’t mean there’s, like, some kind of Trolley Calculator Module that’s predicting what will happen in future seconds during the chaos of a crash avoidance situation. (And it’s not that anyone is explicitly claiming the existence of such a module, but I feel like the articles about this tend to be written along those lines).
I wouldn’t be surprised to see some vague heuristics like “prefer to hit the larger person”, “never drive off a cliff, since really, who knows who’s down there anyway” etc, but not any deep analysis. Like with human drivers, the worst crashes will occur when the car has incorrectly processed the scene, so it’s hard to believe that some of the philosophy getting thrown around would be very useful for safety.
But what do I know? I guess it’s a little frustrating that we’re at a stage where so much of the reporting is a little hypothetical, since the people who really know a lot about self-driving-car decision-making are busy right now.
I imagine a true self-driving car would have to be constantly evaluating multiple risks, just as a driver does. If one of those risks begins to peak, such as a pedestrian beginning to step off a curb who would be hit if the car didn’t make an avoidance maneuver, the software would need to continually re-evaluate the various other risks in real time. It seems to me this would inevitably become a decision mechanism and would need to, one way or another, handicap those various risks in something like what we think of as moral terms.