How the autonomous car will upend auto insurance
04 Feb 2015
Autonomous and semi-autonomous cars will significantly reduce the frequency and severity of car accidents as we know them. However, they will also come with new risks. The insurance industry will have to demonstrate agility and flexibility to successfully accommodate what could be a huge societal shift.
In the United States, car insurance is a USD 200 billion industry, though the estimated cost of accidents at over USD 800 billion includes the loss of life, pain and suffering that are often not covered by insurance. Insurance exists to manage and distribute risk, but in automobiles it does far more.
The car accident is by far the most common major tort in society; 30,000 happen every day in the USA. The people who cause the accidents do not pay for them; the costs are pooled through insurance. The vast majority of those 30,000 accidents do not go to court - they are resolved much more simply through the insurance system.
Taking every car accident through the courts would be hugely expensive. As insurance companies handle all the costs, they are strongly motivated to keep those costs low. In fact, in many places, insurance firms have lobbied for "no fault" accident systems where blame is not assigned at all. Insurance companies realise that fighting expensively over who is to blame in one accident is pointless to them, they will just be on the other side next time and a lot of lawyers will get richer. Individuals, on the other hand, are strongly motivated to spend money to sway things their way in an individual case.
This is important as one of the first consequences that might be expected by low-accident robocars is the decline of the size of the auto insurance industry. The path is not entirely simple. Makers of robocars will not release them until they have demonstrated that they are safer than human drivers, reducing accident frequency and severity; demonstrated to themselves, to their lawyers, to their insurers, to the public and in some cases to the government. This means that accidents go down, and one presumes claims go down, and that means premiums go down. They will fall slowly at first, but eventually and ideally to very small numbers.
One might liken it to other automatic vehicles, like automatic trains or even elevators. There is no significant business in insuring against elevator accidents today.
At the same time, it changes who takes the risk. Passengers in eventual full robocars are not in control of anything but their choice of vehicle, and it will be hard to get them to take the risk upon themselves, other than simply as part of the price. It makes tremendous sense for robocar products and services to have insurance bundled in as part of the price.
This means the risk is borne by either car vendors, or robotic taxi fleet operators. For large entities like this, there are incentives to self-insure. Self-insurance is almost always the cost effective path if you are large enough. This is particularly true for the makers of the robocars. They are not just large-enough to self-insure, they are also the ones who best know how to measure the risk.
Designing and testing the software for a robocar is also about managing risk. The vendors of robocars will design them with safety as the first priority, and they will work extensively to measure their safety record. They take their cars out on the road for millions of miles, and observe how frequently they have problems. They will work out those problems, but they will also measure the ones they cannot solve. They will simultaneously be the designers of the product and perform the actuarial tasks of calculating its failure rates. They will be the ones best equipped to calculate that total cost, though they may borrow a little expertise from existing insurers.
They should get a good handle on how often they will have accidents, though they cannot quite predict who will be hurt, and that might vary the cost. Insurance companies know that a little better, and it will make sense to offer reinsurance to the big firms that decide to self-insure. However, in general, if they can tell you what their accident rate will be, they probably do not need primary insurance; and if they cannot tell what it will be, the insurer will not know how to underwrite the risk.
All human-caused accidents are unanticipated, but the patterns of accidents are well understood by analysts and actuaries. This is less true for robocar accidents. As I outline in Robocar Accidents, every accident will be quite different, because each accident will quickly cause all the cars to get new software that assures the same type of accident never happens again. The human race as a whole learns almost nothing from any given accident, but all robocars will learn a great deal.
Unfortunately, this unpredictability is frightening to both vendors and insurers. People are correctly scared of risks they cannot calculate. The vendors make a decision to enter into this field for the huge potential upsides and take the risk. For insurers there is not the same upside.
The law (and mixed driving)
In most places, the law requires those who drive motor vehicles to have insurance. New law likely needs to be written for cars that operate unmanned, as Google plans to do, though existing law may suffice under 40 km/h. In California, due to a ballot proposition that is almost impossible to reverse by statute, car insurance must be priced first by experience, driving record and number of miles driven before the safety features of the car can be weighted in. This will keep premiums there high until a work-around is found.
One work-around, particularly for vehicles which have steering wheels and can be operated on their own or with a driver, is to have two types of kilometers with two types of pay-as-you-drive insurance, billed by the kilometer. A regular policy would cover kilometers driven by a person, and a different policy (or insurance from the vendor of the car) would cover when the computer is in charge. Over time, it can be expected that software-driven kilometers would dominate, though many people enjoy driving for fun and will take the wheel in places that are fun to drive.
A particular concern of much debate is how you do the hand-over between computer driving and human driving. When the human wants to take over, there are concerns about if they will have a clear understanding of when they are in charge, and what they might do if inexperienced. More troubling is the idea of humans taking over when there is a problem, either because they want to, or the computer actually insists. In aviation, the handoff has been shown to be the cause of many accidents, and people feel in cars, with untrained drivers, it could be worse.
That feeling is so strong that General Motors pulled back on its planned 2015 release of "super cruise," a product that required constant human supervision, and all similar products are hunting for countermeasures to assure people keep their attention on the road.
More dramatically, Google decided the handoff is a great concern, and has moved to developing cars with no steering wheel or other controls. The humans never drive and there is no hand-over. This is a much harder problem (almost as hard as going fully unmanned) but Google feels it is the right path. In 2012 and 2013, Google tested cars with a hand-over using their own (non-car-project) employees, and found that after a week of use, the operators did things that made the team "uncomfortable" in terms of poor supervision and risky hand-overs. This put them on the path away from this. Google is far ahead of other developers. Volvo plans to do Google's 2013 test in 2017 and hopes, one assumes, for a better result.
The expensive accident
As detailed in Robocar Accidents there is the risk that any robocar accidents might be dramatically more costly than if a negligent human driver were at fault. We understand the risks of the negligent human driver very well, but we do not understand the negligent robot builder very well at all, or even the non-negligent but still involved robot builder or operator.
In all early robocar accidents, all parties will get sued. These accidents will be "famous" with global attention, and many lawyers will want to be involved and set precedents. Some will also come seeking big awards from deep pocketed defendants. They will be unlike ordinary car accidents. In spite of the fact that detailed 3-D recordings of the accident will make ordinary fault immediately clear, the cases could be very expensive to try, especially with all the technological examination needed to be done by experts.
It can only be hoped that over time, the courts become almost as used to such accidents as they are to ones with human drivers at fault, and the cost becomes more manageable. It would be a great shame if accidents were reduced tenfold but cost per accident went up twenty-fold, creating a net loss for society in dollars to counter and delay a huge net win in injuries, death and vehicle damage.
Initially, this suggests a shift from accidents being a question of personal liability to them being a matter of product liability. Companies buy insurance for product liability, but it is rare for them to make products that are known in advance to be likely to injure and even kill people. Only in the medical arena is this a common practice.
Automated Driver Assistance Systems (ADAS)
It should be understood that even without robocars, ADAS systems are already producing reductions in accident volume and severity -- double digit reductions. This is only going to get better. There will be humans behind the wheels of such cars, but the cars will become harder and harder to crash, and more and more of the common human mistakes will be prevented by the system. Once it is possible to build a fairly functional robocar, it will also possible to build an almost "crash proof" car that humans will drive. Even if you take your hands off the wheel, or fall asleep, it will handle it 99.9% of the time -- and that means huge drops in accidents, and eventual drops in premiums.
One plus for insurers will be the detailed records a robocar makes of any accident it is involved with. This will mean that deciding who was at fault in the conventional sense will be very simple, and there will be little potential for fraud by either party. The debate will be instead about why the car caused an accident, if it did, and if negligence was involved which might increase the damages over the usual damages we see in accidents with human drivers.
Liability and collision claims should go down over time, but comprehensive insurance may not change a great deal. Stones will still chip and crack windshields. Cars will still be vandalised and stolen (though it may be harder to steal a car that will not operate without approval from cloud servers). There will still be fires and hits in the parking lot.
There will also be the accidents which happen today without human error, due to other factors like weather, mechanical failures and just plain strange circumstances.
People will not stop driving immediately. There is no real need, or call for the banning of human driving. If that is to happen will be many decades away. Most human drivers have very good driving records - a significant number never have an accident in their lives. As a group, auto accidents cause immense damage, but any single person is reasonably safe, particularly once they establish a good record. There is some speculation that over time, humans will drive less and lose their skills when they do drive.
As noted above, for vehicles that allow manual driving, coverage will still be needed for that manual driving, though probably paid for by the kilometer.
In spite of Google's bold goals, it should be noted that all the car companies plan vehicles that run automated some of the time and are driven manually the rest of the time. In fact, their early plans are all for vehicles which let you take your hands off the wheel but insist you keep paying attention to the road -- like a cruise control. The driver will still be responsible for any incidents and need insurance. The vehicles will actually monitor the driver to make sure she is paying attention, as otherwise people will quickly over-trust the system and look away. For these systems, highway driving is one of the first goals, and fairly simple highway driving to boot. The more risky situations will be left to the humans -- and they need to be insured.
While accident rates will go down, and severity of injury to people should go down, these vehicles will be more expensive than ordinary cars, especially at first. This means higher claims for repairing them, which might mean more business. On the other hand, vehicles operated as taxis will wear out quickly -- wearing out by the kilometer rather than by the year, and so the old will quickly be replaced and may not be worth repairing, especially since the expensive technological parts keep getting cheaper and better every year.
Car companies have, for a century, sold people cars. Generally once they sell the car, they don't plan to have to deal very much with the customer, though dealerships want to see them come back for regular service.
New thinking in robocars calls for them to operate as taxi fleets. This means the vendor or fleet operator will be selling transportation as a service, not vehicles as a product. This is a natural trend for those who come from the computer industry. There are many reasons why it is compelling. Software companies love having steady revenue streams instead of one-off product sales, but that is only one justification.
The service model is particularly appealing because it retains a lot of control. With a vendor operated taxi service, the vehicles are coming home to the vendor every night. The vendor can check them out, improve them and carefully monitor how things are going. Maintenance is under their control, which helps assure safety. This is very desirable with radically new products unlike anything the world has ever seen. It's a brave company that will sell these cars to a customer and not have another chance to interact with the vehicle. Computer companies are used to constantly improving and upgrading existing tools, while car companies are not. Tesla has surprised the industry by regularly sending out updates that change how the car drives. Only now are car companies starting to do this, and even so, Mercedes has said that new software features from later-model self-driving cars will not go into older ones unless there is a safety issue.
This is a mixed bag for insurers. If the vendor is owning and operating a fleet, they are likely to self-insure, though they will want reinsurance. If they do seek outside help with managing the risk, it probably means selling insurance in bulk, rather than having to do retail sales.
Robocars present some new risks in addition to the complexity of their accidents. Computer intrusion is a concern for everybody. If attackers can get into a car over the networks and command it to do something dangerous, the consequences can be dire. Vendors will want protection from that risk, but it is a very hard risk to calculate. In theory, an entire fleet of 100,000 cars might be compromised as easily as a single vehicle. The result could be so severe that no company or insurer could handle it -- it would be a catastrophe more akin to an earthquake hitting an entire country at once. Insurers are used to dealing with catastrophe, but this is something that might have no geographic limitations.
Another risk arises when problems are found in software. Vendors will debate doing a "recall" but they will also have the option of disabling their system remotely for a limited time. This might be ruinous to their business, and more than just annoying to their customers. If the vehicles cannot be human driven, it could shut down an entire service unless there is a suitable contingency plan. People who have come to depend on a robot taxi service will be very upset if it does not come one morning because somebody found a bug which might harm people. Switching to competitors or human driven taxis would be both expensive and possibly create demand exceeding the capacity of such systems.
In today's recall system, companies notify customers of a safety issue and request they come into the dealer for a fix. Once they do that, the customer is responsible if they continue to drive without getting that fix. With robocars, the fix will come over the radio, but it may not be ready for some time. Will customers be responsible if they say, "I don't care, send me my taxi anyway?"
The future is uncertain for insurers. Whether due to robocars or just hard-to-crash ADAS equipped cars, the number of accidents and their severity is expected to decline seriously in the next 15 years, and reach a volume just a fraction of what we see today in the developed world. In the early days there may not be a decline as the world works out the complex legal regimes around liability for these accidents, because they will be different from those caused by human error. Insurers should expect an interesting ride in the coming years, but also plan for a decline in the premiums for auto liability and collision. At the same time, they should look to shift some of that business into product liability and insuring other risks. In the end, removing the huge risk of auto accidents from the world is a good thing, and a world with less risk is better off, even if there is less insurance to sell.
Download full article (PDF 185 KB)