SoylentNews is people
SoylentNews
from the don't-tell-them-about-motorcycles dept.
Like record companies at the dawn of online music file sharing, Allstate, Geico, State Farm, and others are grappling with innovations that could put a huge dent in their revenue. As carmakers automate more aspects of driving, accidents will likely plunge and car owners will need less coverage. Premiums consumers pay could drop as much as 60 percent in 15 years as self-driving cars hit the roads, says Donald Light, head of the North America property and casualty practice for Celent, a research firm. His message for insurers: "You have to be prepared to see that part of your business shrink, probably considerably."
Auto insurance has long been a lucrative business. The industry collected about $195 billion in premiums last year from U.S. drivers. New customers are the source of so much profit that Geico alone spends more than $1 billion a year on ads to pitch its policies with a talking lizard and other characters. Yet even Warren Buffett, whose company, Berkshire Hathaway, owns Geico, is talking about the long-term risks to the business model. "If you could come up with anything involved in driving that cut accidents by 30 percent, 40 percent, 50 percent, that would be wonderful," he said at a conference in March. "But we would not be holding a party at our insurance company."
The loss of revenue for the insurance industry gives me a sad.
(Score: 2) by tempest on Friday July 31 2015, @04:51PM
If AI can't react to each other properly, then there's no way they can judge human drivers properly and that's going to be a prerequisite for a while. Eventually I imagine we'll have something like a Vehicle Communication Protocol, where two vehicles aware of each other communicate (via wireless or whatever), indicating status and intent, which would boost accuracy considerably compared to vague reactionary systems required now.
(Score: 2) by Tork on Friday July 31 2015, @05:05PM
If AI can't react to each other properly, then there's no way they can judge human drivers properly and that's going to be a prerequisite for a while.
Sorry to be nitpicky, but this is not true. The software being designed to deal with human drivers. If automated cars drive differently than humans then it has to be tested against the new behaviour. Either these companies will need to work with each other to iron this out before the production phase, or the rules of how to handle certain events will have to be regulated by a governing body.
🏳️🌈 Proud Ally 🏳️🌈
(Score: 1) by timbojones on Friday July 31 2015, @05:48PM
To the extent that automated cars will drive differently than humans, they will be more predictable with fewer outliers. Automated cars will not present entirely new driving strategies -- their strategies will be a subset of human driving strategies.
Automated driving software is being designed to deal not just with human drivers, but with human cyclists, pedestrians, streetcars, school buses, construction detours, falling trees, balls rolling into the street, birds swooping across the windshield, a woman in an electric wheelchair chasing a duck in circles in the middle of the street. They are being designed to deal with any moving or stationary road hazard of any size.
Other automated cars will present zero challenge and zero surprise.
(Score: 2) by Tork on Friday July 31 2015, @06:08PM
To the extent that automated cars will drive differently than humans, they will be more predictable with fewer outliers.
I don't think that's a safe assumption, especially when you factor in that robotic reaction time is different from human reaction time. That alone will deviate the behaviour enough to require extra rounds of testing. It'll likely mean more software updates during the life of each car as well.
Other automated cars will present zero challenge and zero surprise.
I hope you end up being correct, but I've heard similar arguments against sanitizing inputs on web-facing code.
🏳️🌈 Proud Ally 🏳️🌈
(Score: 2, Insightful) by timbojones on Friday July 31 2015, @07:15PM
True, but irrelevant. The car is still a car with a car's momentum and maneuverability. An automated driver needs to respond to a car swerving regardless of whether the human driver is reacting to something that happened a second ago, or another automated driver is reacting to something that happened a tenth of a second ago, or the car itself blew a tire just now.
(Score: 2) by tempest on Friday July 31 2015, @07:09PM
While I sort of agree that AI might drive differently than an average person, I don't think we'll really know until AI has a track record. But I think AI cars will be far more predictable, even if different than humans, because it's nearly impossible to group humans together as driving one way. Teenage punks, girls texting, drunks, old folks, and "normal drivers" all drive completely differently and inconsistently. Not accounting for random other factors, like the time my girlfriend smashed into a curb because there was a bee in her car. Computers are generally kept fairly simple, as in keeping it on the street between the lines obeying traffic rules and not hitting things. Unfortunately humans are often not concerned with these things like they should be and can't be expected to even act in their own safety.
I think you're very right about new protocols will be needed first. I'd guess that will eventually fall to a regulated body of some sort. Especially when these cars will have to decide how to weight human life and damage in scenarios where there is no "safe" option. Thus far I can't recall anyone stepping up to the plate to handle this yet, so I'm guessing we'll have to wait for a disaster to make news headlines.
(Score: 0) by Anonymous Coward on Friday July 31 2015, @09:29PM
I think that a communication protocol would not be necessary for safety but it would be beneficial for improving efficiency given the same degree of safety. Without such communication a car would have to be much more conservative assuming and planning for the worst case scenarios and taking the (less efficient) path that would be safe under multiple different possible scenarios because it doesn't know the intent, status, location, and orientation of surrounding cars. With such communication then cars can collectively coordinate the most efficient scenario and no longer have to assume the worst case scenarios (or take into account multiple possible scenarios) since they can plan for and hence know the scenario ahead of time.
(Score: 1) by tftp on Friday July 31 2015, @05:15PM
I can already imagine a new class of "pranks" - to hack into that protocol, stand on an overpass with a pocket transmitter, press the button, and watch the chaos down below as cars are advised of "intents" to cut each other off, suddenly brake, etc.
One would think that the protocol would have to be very robust if the vehicles are to trust each other. Not only the transmission must be cryptographically signed; the key must be in a tamper-proof storage that cannot be desoldered and used elsewhere. The vehicles probably also want to accompany their intents with something else that they know of but that is pretty hard for a prankster to collect - such as visible locations of nearby cars, for example.
One possibility is to use vehicles' cameras for purposes of optical communication. But currently the frame rate is abysmally low. There is also blocking of light by obstacles and rain/fog. Short range radio (in tens of GHz) will work within a hundred meters, and you don't need more than that anyway. But ultimately this is how it will be. Today drivers are just have to read subtle hints to predict intentions of other drivers. An autonomous car will just know ahead of time.
None of that will entirely eliminate accidents, though - a pedestrian can always jump into traffic.
(Score: 2) by tempest on Friday July 31 2015, @06:46PM
I don't think cars would communicate directly, although in an ideal non hostile world that would be easiest. Instead I'd guess we'd have something like the browser CA system, where cars register via the mothership authority and pass signed messages to the third party where they are signed and authenticated. Pulling the message computer out I think would be the biggest issue to worry about, but this seems like something that would be keyed to the engine, like transponder keys currently do with anti-theft systems.
It makes me cringe thinking about the privacy implications, but it seems like everyone can't throw that away fast enough these days, especially if they offer it via Twitter integration so people can see real time status updates like "I just started my car."
Assuming we're not all chipped by then too :-/
(Score: 1) by tftp on Friday July 31 2015, @07:10PM
I'd guess we'd have something like the browser CA system, where cars register via the mothership authority and pass signed messages to the third party where they are signed and authenticated.
I don't think this will work at all. We do not have enough of wireless bandwidth. Usable frequency range is only somewhere from 1 to 3 GHz, and you need some small latency - which implies high data rate, and you need guaranteed service, and you do not want handoffs, and you don't want towers every 100 meters along highways - it's just too expensive and unreliable. On every morning there are 100 million cars on the road - what kind of a third party can receive that, let alone to forward to those who "subscribed" - and those subscriptions constantly change, as cars are moving and their areas of interest vary as well.
Compared to that, peer to peer networks are naturally self-organizing, as if a car hears another car it has a business to hear it. The latency is defined only by the packet length. Privacy is not affected because if you can hear a car you can also see it with your own eyes; the cars do not broadcast more than what will be obvious a few seconds later (like "I'm about to take this exit.")
The mesh also solves the problem of infrastructure. There is no way to cover the whole country with a network of towers. Most of the roads do not have any power whatsoever [google.com]. Cars on such a road will be able to communicate even better than in a city.
(Score: 2) by tempest on Friday July 31 2015, @07:25PM
I think you're right about the latency, but as for bandwidth I think the very first thing people will demand when having cars drive themselves will be internet access in their car. If we don't have the bandwidth now, it will likely come with growing demand. That assumes we use existing infrastructure, so maybe the cars will talk to receivers built into the roads themselves. Something like a smart highway that kept track of temperature, road conditions, traffic congestion as well as allowed nearby cars to talk to each other. It could be that once a car has authenticated with a section of road, it is allowed to message other vehicles directly. Like kerberos tickets. Obviously due to many factors not all roads would be covered, but cities and major highways could be. At other times the AI would default to the reactive mode used today.
(Score: 1) by tftp on Friday July 31 2015, @08:23PM
as for bandwidth I think the very first thing people will demand when having cars drive themselves will be internet access in their car
They can demand all they want, but Shannon's Law [wikipedia.org] will be a problem. It's like demanding that one could jump up and reach the Moon. Good luck with that.
Furthermore, there is NFW that the consumer's Internet access and the real time driving information would be sent over the same link.
maybe the cars will talk to receivers built into the roads themselves
You can go that way, but it will take time and money - and I don't think anyone has enough money to upgrade all the roads. I'd be happy if they just add a lane or two, forget the radios.
It could be that once a car has authenticated with a section of road, it is allowed to message other vehicles directly. Like kerberos tickets.
Authentication with the road is kinda pointless here, as the road does not do much, and it's unclear why it would be a trusted authority. It's much easier to give OEMs (Ford, Toyota, etc.) the CA keys, and tell them to sign keys of the vehicles that they make. With key revocation over the air, once per day, this should be both secure and sufficient. Then the cars can talk to each other on all roads, not only on upgraded ones. And outside of roads as well - plenty of construction and agricultural machinery operates there, and automatic functions there are even more important because the job is so repetitive, like plowing the field from here and until the Sun sets. Collisions with farm equipment are not all that rare either.
(Score: 2) by Francis on Friday July 31 2015, @06:07PM
A lot of it is surprisingly straightforward, the car keeps at least 2 seconds behind the next car as long as it can do so at a safe speed. The car needs to watch out for lane intrusions and slow or speed as appropriate. Which is a pretty limited number of reactions necessary to get along with other cars.
Obviously, the actual execution of it is difficult, but realistically, the things human drivers do that cause crashes are generally a lot more easily identified than one might imagine. The motorcyclists, bicyclists and pedestrians are a much bigger problem though.