The Potential Challenges of Rating Autonomous Vehicles
By Jared Smollik, Actuarial Director of Personal Auto Product Development
In 1929, Paul Dorweiler, an influential actuary and a notable contributor to the advancement of actuarial techniques, suggested that in auto insurance rating, mileage is superior to the car-year as an exposure base because it more directly relates to the usage of the car, but the devices and effort necessary to use mileage render it impractical. He also said that factors such as road conditions and traffic density, weather and time of day, and speed influence auto risk. It’s clear that people have seen the value of this information for a long time.
Now, almost 90 years after those sentiments were published, the modern technologies giving rise to telematics and usage-based insurance have only recently started to unlock the predictive power of those and similar risk factors for auto insurance rating. Ironically, just when we thought we had the tools to understand the entirety of the vehicle risk, the same sorts of technological progress are redefining the auto exposure as we know it. Of course, we’re referring to autonomous vehicles (AVs). And just as Dorweiler tried to address questions about rating manually operated vehicles, we must now think about what considerations are important to the rating of AVs.
As we mentioned in our previous discussion of insurance coverage and autonomous vehicles, about 90 percent of all auto accidents are estimated to result from driver error. Even if autonomous vehicles only perform half as well as expected, there will still likely be a significant reduction in auto claims frequency and insurance loss severity. The nature of the auto risk can be fundamentally different, and it may no longer be appropriate to rate based on the variables used today, which typically attempt to discern how the vehicle will be operated based on driver characteristics.
The problem with partial autonomy
In the nearer term, it may be reasonably expected that the first autonomous vehicles introduced to the market won’t be fully autonomous in all situations. Two examples that come to mind are when the vehicle encounters extremely adverse weather conditions (a situation where the vehicle may be programmed to cede control to the driver) and when the driver actually wants to drive, since many people enjoy driving. In the latter case, there may be no question that traditional insurance coverage, and rating, will apply. And here we may find that telematics offers the best solution to a hybrid manual/autonomous rating scheme. Even if telematics and driving behavior data aren’t used to their fullest, at the very least, the vehicle should be able to accurately track and report how often it’s being operated manually versus autonomously, allowing insurers to determine how much of the exposure should be rated traditionally.
But a problem arises during the transition from autonomous to manual operation. When the vehicle cedes control to the driver in such emergency situations, rating and coverage during the transition are complicated because the vehicle is not completely autonomous. Distracted driving is already a leading cause of auto accidents in the United States, even though people are supposed to be paying full attention when driving. Imagine how many driver-passengers will read books, catch up on work, and take naps when their cars are in control! How many of those people will be instantly ready to assume control of the vehicle in adverse conditions? Some auto manufacturers have already pledged to assume liability if their autonomous vehicles get into an accident, but who is responsible in such a transitional situation? It could be argued that the vehicle was programmed to cede control during unsafe conditions, and it was doing exactly what it should, but the driver wasn’t paying attention and that resulted in an accident. How this transition will affect claims and losses remains to be seen.
Rating autonomous systems
Regarding the rating of autonomous driving, the natural analog to the human operator for autonomous vehicles is the vehicle itself, and especially the software. When vehicles drive themselves, the rating characteristic of the age of a driver may need to be replaced by another factor, such as the version of the software. The vehicle manufacturer and software programmer may also need to be considered. When autonomous vehicles become mainstream, we should monitor how different implementations perform over time, just as we monitor the experience of different makes and models of cars today. We may need to develop a system of rating factors for self-driving software as we currently do for other physical characteristics of the vehicle, such as vehicle series rating (VSR).
One outstanding question that deserves more consideration is how software updates will be treated, especially if they’re automatic and over-the-air. Will manufacturers and/or insureds need to notify insurers of software revisions and new features and bug fixes? If the newest software is “the best,” will insurers need to audit fleets to ensure the most current software is being used? Or instead, will the latest and greatest software even be compatible, available, and supported by the vehicle manufacturer for older cars, or will it possibly be withheld to encourage the purchase of a new vehicle?
There are other interesting rating changes to consider. For example, routes that autonomous vehicles take may be more or less safe than others, with more differentiation based on the time of day. Insurers will want to leverage telematics to determine where and when autonomous vehicles are driving and evaluate pricing based on the riskiness of each mile of road, just as they would for human drivers.
Potential loss impact of autonomous cars
Hopefully in the future, the damage resulting from accidents with autonomous vehicles will be largely isolated to the vehicles themselves, because even in situations where a collision is unavoidable, the superior reaction time and precise maneuvering of autonomous systems may be able to reduce overall severity of claims, especially to vehicle occupants and bystanders. Even so, the vehicles themselves will experience some damage, and the sophisticated sensors and other electronics in the vehicles will be sure to complicate repairs and increase such costs, as they can today. As mentioned above, vehicle rating should remain relevant because property damage and collision losses may increase for some time, even as bodily injury and medical payments may diminish.
Effects on theft and weather damage
We’ve already considered how fewer accidents can affect liability and collision coverage, but what about other than collision (also known as comprehensive) losses? Thieves, for instance, will need more than a crowbar and a basic knowledge of car electronics to steal an autonomous vehicle. They’ll likely need to know how to hack into the vehicle’s computer systems. Does that necessarily mean there will be fewer thefts of autonomous vehicles? At the same time, will autonomous vehicles be able to protect themselves by reviewing weather forecasts and driving out of an area in advance of an expected hailstorm or flood? What if the insured declined to use such a feature?
The auto exposure is changing in ways never before imagined, and rating must adapt. It’s exciting to think about how we will address these challenges.