Regulation, not technology is holding back driverless cars
nytimes.com
nytimes.com
In any case, this single crash caused regulation to state that computers can never fly planes by themselves. This strikes me as rather unfair. If a single human crashed a plane, it does not make it illegal for humans to fly planes by themselves.
Another example is that in London, subways must be driven by a human. Even though driving a subway may be trivial (there is no way to steer), Londoners apparently do not feel comfortable being driven by a non-living thing. They want to be sure that if they die, the driver dies too, adding a level of accountability.
It seems that this sort of wide-spread mistrust of machines is driven more by socially normal paranoia than any kind of logic. I for one am rooting for machines to take over all forms of driving. There may be a few mishaps, but it will probably become hundreds of times safer eventually.
There's a great section in Neal Stephenson's Anathem where the science-y younger brother is speaking to his older sister, who's mechanically talented but not very techy. He says there's always a better way to build something, that you can always take the inefficiency out of whatever system you're working with. But she doesn't want that, because she wouldn't be able to fix it herself. That's how most people see technology, and to be honest, we see it similarly, because we can fix it, so it's not a threat. In 30 years, will we be able to fix the things that are running to world, and will we be as inclined to trust them as we do now?
I'm not saying it's logical, or appropriate, but it's a natural response to the situation. I think it's important to work with it rather than dismiss it.
Additionally, once cars become sufficiently advanced, it will be much cheaper to insure a self-driving car than to insure a human.
Once the change happens, it will be breathtaking how rapidly it occurs. Social stigmas are hard to sustain towards a large portion of the population. In 2000, if you were wearing a helmet skiing, you were crazy. Now, you're crazy not to. You used to be a loser if you were on a social network. Now you're a loser if you're not. Both these changes happened over perhaps one to three years. Critical mass may only be ~5% of the population. You only need enough to avoid individuals being singled out as odd.
Over 90% of Americans have used illegal drugs, and yet there is still a strong stigma there.
http://www.monitoringthefuture.org/pubs/monographs/vol2_2009...
(C.f. page 101, as well as the methodology.)
On page 103, it says that 82% of fifty-year-olds (who attended high school during the heaviest-drug-using period in America's recent history) have used marijuana at least once in their lives. But only 10% of that same group have used it in the last year.
Among current 18-year-olds, only 42% have ever tried marijuana even once. And among most age groups under the age of 30, only between 23-34% have used marijuana within the last year.
However, I don't see evidence that the study you posted counts underage drinking in among illicit drug use.
I said 'have used' in my original comment, which is what I meant. (In my reply I was thinking lifetime prevalence, i.e. around 85% 'use' marijuana at some point during their lives. Poorly phrased, my bad.)
"Among current 18-year-olds, only 42% have ever tried marijuana even once."
The age of first use of marijuana is highly variable. By age 25 around 65% of people have used marijuana, and more people keep trying throughout the next couple decades of their life. The important statistic is what percentage of people use marijuana at least once in their lives, not what percentage use it on any given day.
"On page 103, it says that 82% of fifty-year-olds (who attended high school during the heaviest-drug-using period in America's recent history) have used marijuana at least once in their lives."
The 82% figure doesn't count people who didn't make it to fall of their 12th grade year in high school, so you actually have to adjust it up a couple percentage points. Also, even if less high schoolers were using illegal drugs for a decade or two, that probably won't have much effect on the overall lifetime prevalence. It might eventually decline a few points, but probably not that much.
1977 has got to have been the historical peak of high school usage. I notice they don't mention 51 year olds or 49 year olds, they have specifically pulled out this one exact age for this statistic.
It then goes on to say that 29-30 yr olds have a 68% "adjusted" lifetime prevalence. What does that mean adjusted? It means the actual number they measured is lower (32% for 19-30 year olds), and they tweaked it upwards because they wanted a bigger number.
About this time is where I say a study has too many methodological flaws and should be disregarded.
Only 50-year-olds are mentioned because this is a longitudinal cohort they've been testing since 1977, and this year (2009) it's their turn to be reported.
The "adjusted" number is higher than the measured number because some people answered the survey ten years ago saying they'd tried marijuana in the past year but then in the current survey the same respondent said "no" when asked if they'd ever tried marijuana in their lifetime. So they're publishing both the actual numbers and the "we know you're probably lying" numbers.
It's explained the study.
"About this time is where I say a study has too many methodological flaws and should be disregarded."
Try actually reading it. It's really not that difficult to understand. The 'maturity recanting effect' section here also gives a brief summary of what they did and why:
http://www.erowid.org/psychoactives/statistics/statistics_ar...
Don't criticize studies without taking the time to understand them, otherwise you just sound like these people:
It's also charming that you cite erowid.org as a reliable unbiased source. I am familiar with the site and have read it. It is a radical pro-drug site.
Even their legal articles don't take any position on what drug policy should be, all they do is explain the case law that exists.
Hypocrisy is demonized. Sometimes rightly, sometimes wrongly.
If the car has some sort of payload (like a Semi or UPS truck), then this makes total sense. If the human is the payload, then there is no benefit at all. (Well, other than being able to multi-task -- but if you're working from your car, why not just work from home and skip the commute?)
Much like taking public transportation to work, it turns commute time into time that can be spent doing something other than paying attention to the commute.
This would also imply, that my premium could raise, because that kind of robot, recently crashed a bit too often. At the same time, I might be able to reduce my premium by installing a software or hardware update.
[1] I have a license for years but never owned a car. AFAIK it is not just the drivers record but also the statistical likely hood of a crash, according to gender, age in car. Though, cars today are not rated as 'unsafe' because they are buggy but because they are popular with reckless drivers.
We do not leave in the same world. I mean, really, I am in East asia, not the same world. People do not ski much here, and if they do it is without helmet.
I know it is a bit tiring to recall this but it is a good habit to think about the reality of the generalization that one write. What I like on HN is that it is usually not too provincial. Sorry for the nit-thing.
The Dockland Light Railway in London is driver-less, although there is a staff member on each train to operate the doors and check tickets:
But it's not the same - as one Frenchman put it: "it's like an elevator, only it goes horizontally instead of vertically".
That is not the case with driverless cars - not that I'm against them.
The use cases put forth in the article are quite compelling, but we still haven't been able to eradicate "elevator operators" ("ascensoristas") here in Uruguay, it will be far more difficult to remove drivers' unions !
I will wage anyone that if all the cars on the road were computer driven there would be no where near 42,636 people killed per year.
I can then imagine the newspaper headlines the few times per year that computers do end up, for whatever reason, being responsible for a small handful of deaths: "COMPUTER GLITCH KILLS FOUR PASSENGERS IN HORRIFIC CAR ACCIDENT!"
Never mind that human "glitches" are responsible for dozens of such accidents every day at present. For whatever reason, I have a strong feeling that headlines such as those will scare people worse than the way things are now.
Presumably, it would be paid for by insurance companies. It could be rolled out as a general public awareness effort (i.e. packaged with generic reminders about wearing seatbelts, maintaining proper tire-pressure, etc.), combined with localized elements that included maps highlighting particularly deadly times and places where drivers should be especially vigilant.
Once the figure is widely accepted, the companies who would be offering the technology would have an established base against which they could run ads that credit themselves with lives saved. Assuming that these claims are backed up by public data assembled by trustworthy parties, even the most vociferous "if it bleeds it leads" news agencies would have a hard time terrorizing people BACK to the days of a 9/11 every month.
Also, while I hate to be cynical, there's an upside to corporate mass media, in that a lot of ad dollars spent by pro robo-car interests WILL make news agencies think twice about scaring the daylights out of people just for a ratings bump. After all, the whole point of ratings is to boost the price of ad sales.
I realize that this may seem like a lot of effort, but no industry takes a longer view than the insurance industry. If there's one group that has the time, money, and motivation to make the push, it's the guys who are paying costs associated with all the current grief and mayhem. Moreover, our not-so-clean political system does mean that legislators can be easily persuaded once insurance companies are comfortable that the technology is truly secure.
My guess is that any push would need Federal backing, and would be focused - initially - on the Interstate highway system, where much less reactive driving takes place.
A little siren goes off, warning you to break. Navigation tells you where to go, and how.
An override button appears, allowing you to give the computer control in an emergency. Eventually, the button becomes a switch, and the switch gets left on.
"Autodrive" licenses (similar to automatic licenses) are introduced, allowing people to pass their driving test in full auto mode, provided they only drive cars in full auto.
I did not say that human-driving ought not be morally blameworthy but, since I enjoy driving, I am understandably worried about a time when something that brings me pleasure will become unethical.
If we blame the car makers, there may not be any car makers because nobody will want to take on the liability.
If we don't blame anyone then there is no incentive to improve the situation and no way to hold people accountable for sincere negligence.
We would have to accept that a consumer product will occasionally kill us, and come up with some standard to judge whether the manufacturer "tried hard enough" to prevent it.
I can't think of any precedent for that. There are plenty of products that can kill us, but only if we misuse them. A driverless car will have a few deadly design flaws that we only learn about from experience.
Maybe this should be publicly acknowledged up front and a legal framework established. That seems better than pretending it's completely safe, causing a shitstorm when something inevitably goes wrong.
During the period where both human and computer driven cars share the road, I suspect the human driven cars will get a speed advantage because of this.
Personally, I'll be more than happy to add ten minutes to my commute if it means I don't have to drive. I think a lot of people will be with me on that one.
One guy presses a large "Go" button to leave the station, which is automatically stopped at, and there's also an emergency stop button.
The other guy has an "open / close door" control.
She was carrying a very expensive violin. She got off the train and the doors closed between her and the violin, with the strap around her shoulder. Because the strap was thin, it didn't activate the door's obstruction detector.
Because the violin was so expensive, she didn't want to lose it, so instead of crawling out of the strap, she stood there and waited for the conductor to notice her. One of his jobs was to verify visually that there were no problems in the station before the train started moving -- but he didn't do that. She was dragged under the train and the wheels sliced off her legs.
Point being, those people are likely serving an important safety function. The conductor didn't do his job on the Chicago train and so a woman lost her legs. But a system with no human conductor would be an unsafe system by design.
Perhaps, yes. But in this example, with no human conductor, the woman would have just sacrificed her violin instead of relying on the unreliable human conductor.
Hell, my garage door 20 years ago did it.
Also (as a separate issue) I thought the governments of the world wanted to create more jobs, not fire everyone who is currently working... ;)
But in my car, if it could drive itself, I could get in the back seat and sleep. I could have a tiny apartment back there and read or play video games.
I think that if people will text while they drive, risking their lives for a tiny message, they will gladly let their car drive for the payoff of getting to do whatever they want during the trip.
Side effect: if people can sleep comfortably in the back seat while their car drives, suddenly a 4-hour commute may sound reasonable. That's going to screw with city planning, pollution, etc.
As for driverless cars... It just seems folk are getting a bit ahead of themselves - can computers really handle all the edge cases that come up while driving? Like when you have to break some road rules temporarily in order to circumnavigate an obstruction. Or when the car gets stuck. Etc.
Does not surprise me. Most Europeans have not even adapted to automatic transmission in cars, and still believe that a human can switch gears better with manual transmission (which has not been true for decades now). Makes me wonder if they ever adapt to fully automated driving.
(And yes, I'm saying this as a European).
Either way, I don't think the prevalence of manuals in Europe is entirely (maybe not even mostly) about not giving up control to the machine.
Frankly, I'm a little dubious of automatic transmission. Perhaps I've just been driving the wrong cars, but I haven't found an automatic gearbox that can match a human being. For instance, if I'm descending a steep hill, I'll stick the car in a lower gear, but an automatic always chooses the higher gear.
And personally I never do for the simple reason that brakes are a lot cheaper than transmissions. And I rarely descend a hill long enough and steep enough that brake heating is a serious problem. (And when I do I usually just slow down, and stay slow.)
F1 drivers did until it got banned. In fact, if I'm not mistaken I believe our more advanced automatic transmissions came from F1.
They're an option on the more expensive Audis and a few other manufacturers have them.
A quick search for "CVT reliability" produces pages of horror stories related to the Audi (and other) CVTs.
The CVT is only available on the A4 and A6 (not the A3, A5, A7, A8, Q5, Q7, et al). Also, the CVT is only available with the smaller engine options. Finally, the CVT cars are front wheel drive only, no quattro available.
That is what I mean by "not mature".
If 1 pilot in a hundred million decides to override the autopilot, and crash the plane into a landmark building, this is a cost.
Autopilots are adjunct flight controls. They are not...the pilot in command.
The transition to driverless cars is (IMHO inevitable. At some point it will be cheap enough that the additional cost will pale in comparison to the lives that will be saved as well as the simple convenience of being able to do something else while commuting somewhere.
Likewise I see this kind of thing replacing many forms of public transportation. There will simply be a fleet of cars. You'll say where you want to go and some system will route people and cars to destinations.
But, the transition won't be quick or easy. You need look no further than the aviation industry to see why.
Basically, automation in modern aircraft is a double-edged sword. It seems to erode the ability of pilots to actually fly [1], software errors causing deaths [2] and (I can't find the link to this) I also read a study that in more automated planes, pilots are more likely to believe erroneous instruments rather than their own senses and experience.
The issue won't be how the car normally behaves because as demonstrations have shown, current systems require very little human intervention.
The issue will be extraordinary circumstances plus the huge liability problem of any errors.
Example: if someone runs a red light and causes a crash, killing someone, that person is responsible. If an automated car does the same thing, the manufacturer will be responsible.
That alone will impede adoption.
Instead I think you'll have what we already have: slowly adding automation to cars. Cars already have radars and can stop themselves from colliding, they can park themselves and so on.
But at some point the driver will need to go away and that will be a tremendously challenging leap forward for society.
[1]: http://www.tourismandaviation.com/news-4530--Pilot_Reliance_...
Note I said "interesting", not "the beginning of the inevitable victory". I say "interesting" because I think we can pretty much all agree how this plays out until that happens, but I can't see past that event.
The problem is a systemic one that stems from the belief that the gov't is the omnipresent nanny of its citizenry.
Note: Before I get downvoted, I recognize that statistically seat belts save lives, however, there are conditions in which seatbelts cause fatalities (even when used properly), and that the choice of whether to wear or not wear one should be that of informed consent.
Edit: Added "healthy".
Do you have an example?
I'm not sure that would be classed as "used properly", though.
I believe the legal point is the same as you've stated - the law is "not correctly restrained" or something.
Not that I'm this much of a bastard, but for those of us that have faith-based issues around community-provided health services.
The laws and regulations then turn towards cost reduction along with "taxes" to discourage higher health costs.
We only have the cigarette laws, seatbelt laws, and other 'endanger self only' actions that are regulated.
But agreed in principle ;)
The human brain is not setup for 3D dimensions of orientation, unless you can accurately see the horizon it's more than likely that your perception of the planes orientation and vector is wrong than the instruments. In most turbulent situations the brain is wrong and the instruments are right. This is why pilots are trained to fly on instruments only. On most planes the instrumentation system actively and accurately informs pilots that their instruments are broken when they are in such a state.
1:Parenthesis :: >1:Parentheses
With computerized drivers, it will finally be possible to fully enforce speed limits, by introducing some ceiling to the speed attainable by the car. I'm sure some "well-meaning" legislator will make it his or her priority to ensure that speed limits are never exceeded. However, at least in Massachusetts, if you go on the highway, everyone (including police) drive at ~75 MPH even though the posted speed limit is 55 or 65 MPH. Few will buy a car with this kind of handicap, were it to exist -- and I worry that it will. Many speed limits in the US were imposed decades ago, with less safe and responsive cars -- it would be a pity if potentially revolutionary technology advances were thwarted by this fact.
Legislation has already crippled or made useless many useful automotive innovations. In the US, technologies that allow for adaptive cruise control (maintaining a distance to the car in front of you) can only decelerate the car, and not accelerate the car. This forces the driver to have to constantly accelerate, greatly reducing the effectiveness of this feature. Many computer-laden vehicles with navigation systems are similarly crippled -- they automatically "lock" when the car is in motion, and some (like in Lexus vehicles) cannot even be overridden by people sitting in the passenger seat... often causing unintended risks like drivers pulling over on busy highways just to readjust their GPS target.
Jailbroken cars: mark my words, that will happen the day after automated drivers are released.
Why the hell would you want speed limits on a road with computerized cars? That's the whole point of eliminating human error and human limitations from the control loop, that 200 mph+ driving can be extremely safe.
For example, what about edge cases? Suppose the Google car does just fine in normal driving conditions, but in a blizzard w/ 26 mile per hour gusts of wind (as I drove in recently), or when a tractor trailer flips over on the road in front of you? Humans have a certain intuition that allows them to do bizarre twitches in extreme situations (even including supernormal strength) that presumably no machine intelligence will be able to approach for a long time (if ever).
Or what about the possibility of someone hacking the car? Could a worm engineered by some hostile government take millions of cars off the road -- or, worse, cause them all to steer into the median and cause mass damage and thousands of instant casualties?
It is, frankly, irresponsible not to consider edge cases like these when drafting legislation, and while I'm all for gradual introduction and more testing, the author of this article has convinced me that senators sitting on their hands not doing anything are probably acting on the interests of the people much more so than those who wish to simply hand over driving and navigation functions to machines as soon as possible.
My assumption is that this is a first step towards driverless cars, as it provides the manufacturers a way to test critical technologies in a safe environment. If the dealerships aren't downloading data from these systems, and sharing it back to the engineering groups, thats an enormous wasted opportunity.
I think another major milestone towards this will be the next generation of GPS, with the increased accuracy it will help guide the cars where they need to go.
What would happen to the current drivers? Which new jobs should they pick?
There was a time when 80% of the population used to work on farms. Now much less than 10% work on farms in the western world. Do we have 70% unemployment?
The bigger question is: Will we enter a new social contract where all people benefit from automation, our will only the property owners? Currently, we're trending heavily toward the latter. Regardless, that is a political issue, and unavoidable. Whatever we decide, even to my detriment, I'm of the opinion that we shouldn't let it guide technological progress. At least not just yet.
Making more use of telecommuting will drive down the number of accidents without the potential risks and liability issues of driverless cars.
Not that driverless cars aren't neat, but I agree it is the solution to the wrong problem.
I suppose we could get self driving motorbikes but that would be missing the point of riding.
They did try in the 19th century though, at least in the UK, with the Locomotive Acts[1]. The way those laws went out of their way to protect the status quo (i.e. horse-powered transport) is an interesting parallell to today's possible transition from human-controlled to computer-controlled transport.
(1) Off-highway driving happens. That means that the algorithms have to manage an uncontrolled environment where 'anything' can happen.
(2) It is very expensive to create bug-free software.
(3) You can't iterate by failing fast on life-critical systems after it is released. Failure means killing someone.
(4) Legal liabilities. It's not going to work to say something like, "This car's driver software is not warranted free from defects".
(5) Humans can manage situations utterly outside the norm; algorithms can not see beyond the vision of the designer.
I work in an industry which operates below the levels of software assurance that the medical/flight industries work at, and it is incredibly painstaking as it is. A fully automated car will be very expensive to build.
I am not a paranoiac regarding software. I am a paranoiac regarding software bugs and the limits of the software designers.
While I have no way to prove it, I'd bet my right hand that Google's PR people made this story happen. I'd bet they also made the first NYT piece blowing up the Chauffeur project happen and they made it look serendipitous for authenticity. I think "The Suit is Back," and I think it's going to come back again and again.
Prediction: Driverless cars will be portrayed in a very positive way in a major motion picture within the next year.
> imagine that the cars would save many lives over all, but lead to some bad accidents when a car malfunctions. The evening news might show a “Terminator” car spinning out of control and killing a child. There could be demands to shut down the cars until just about every problem is solved. The lives saved by the cars would not be as visible as the lives lost, and therefore the law might thwart or delay what could be a very beneficial innovation.
It's otiose to point out that the premise of personal motor vehicles is not called into question every time a human driver spins out of control and kills someone.
DDOS and MITM attacks take a whole-new meaning if the networked entities are 3-ton objects moving at 65 mph.
Such systems would place limits on how fast you could accelerate, turn the wheel or apply the brakes. And they would also brake for you when other vehicles, pedestrians and animals appear to be on a collision course.
Such systems will have to be proven in the real world. And we are starting to see them. The newest Mercedes have such features. I predict that full blown systems will dramatically lower accidents for older people, teenagers and those who drive under the influence. Eventually all new cars will have these systems.
And by then it will be a lot easier to trust the machines.
As a matter of policy, the FDA will not approve any device that makes a medical diagnosis without human intervention. [1]
Recently there have been numerous studies and much punditry accusing the American medical regulatory environment of stifling innovation and driving up costs. [2][3][4] The arguments are strikingly similar to Cowen's argument re autonomous cars.
[1]: http://www.technologyreview.com/biomedicine/37596/ [2]: For a quick summary, see http://reason.com/blog/2011/05/25/will-fda-regulation-kill-t... [3]: A much discussed recent report [PDF] is at http://www.nvca.org/index.php?option=com_docman&task=doc... [4]: An article in JAMA opined similarly [PAYWALL], http://jama.ama-assn.org/content/305/15/1523.extract
http://arstechnica.com/old/content/2008/09/future-of-driving...
Can we stop calling all motor vehicle crashes accidents. If the driver was drunk or purposely distracted it is NO more an accident than if I was randomly firing a gun hoping no one got hit.
Having lived in Oregon, Arizona, and California, I have never had anything other than emissions routinely inspected. Demonstrate a car smart enough to monitor its own brake pad wear, alert on burnt out bulbs, and provide a clear readout of all detected issues (i.e. not a coded blinking service light, or plug interface) before you start trying to make it drive itself.
(I do love the idea of an automated train of cars, and driving my drunk self home, though)
I don't know about other cars, but all model Porsche's will do exactly this. Granted, it doesn't have as many sensors as I would like--leading to it only giving an 'engine warning' light for innumerable problems, but when it comes to breaks, oils, lights, etc. it has a pretty fine grained detection pattern.
"alert on burnt out bulbs"
I'm not sure I've seen a car from the current or past 2 decades that doesn't do this...
We also have mandatory annual safety inspections where I live, which should catch things like worn suspension, rust-damaged parts, etc.