cover story

Destination Driverless: Will Vehicles — Not Drivers — Become the Center of Risk?

Last August marked the first injury-causing Google car accident. This 16th crash, amid the more than two million miles the cars have traveled, shared one important factor with the others — they were all caused by human error.

A shift in fault from drivers to vehicles is just one factor that could forever change the traditional automobile insurance industry. Other considerations, such as the potential risk of new technologies and the exchange of lower claim frequency for higher claim costs, are also likely to torque the industry’s future.

Automated vehicles are no longer the fodder of science fiction. Some technology companies and car manufacturers anticipate they will be commercially available within the next five years.

These cars will have limits, said John J. Leonard, a roboticist and associate head of research for the Massachusetts Institute of Technology’s (MIT) mechanical engineering department. “If vehicles do become truly autonomous,” he said, “they would likely require two to three decades.”

It’s not too early for insurers to study the potential risks and advantages of robotic vehicles, said Michael Stienstra, chair of the Casualty Actuarial Society’s Automated Vehicles Task Force and vice president of the actuarial department at ACE Private Risk Services.

Which is the bigger risk?
1. Vehicles.
2. Drivers.
3. Both.
4. Not enough data.


The task force’s goal is to demonstrate the value casualty actuaries can add to the public conversation about these technological marvels. To quantify their influence on auto insurers and their customers, actuaries need to begin examining the ambitious claims being made by technology and vehicle manufacturing companies and to anticipate future risks and advantages.

To Err is Human

The most important expectation of driverless cars is that they will be safer because they will reduce the potential for human error.

Headlines and predictions affirm this assumption. Auto insurance premiums could drop as much as 60 percent in 15 years as self-driving cars hit the roads, Donald Light, head of the North American property and casualty practice for the research firm Celent, recently told Bloomberg Business News.

But will driverless cars be safer? “We still don’t know yet,” said Leonard, a self-proclaimed technology optimist. “We need a lot more data to know if we can meet or exceed human performance.”

There is limited information available to anticipate the potential of driverless cars to reduce accidents. One commonly used statistic is that 93 percent of vehicular accidents is caused by human error. The logic is that the more driverless cars operate, the less opportunity there will be for mere mortals to make accident-causing decisions.

The statistic, which comes from the 2008 National Highway Transportation Safety Administration’s (NHTSA) “National Motor Vehicle Crash Causation Survey,” was even offered by many witnesses who testified about driverless cars before the U.S. Senate and House in 2013.

When the task force took a look under the hood of the NHTSA study, however, it found that the 93 percent statistic is problematic for several reasons.

According to the CAS Task Force report, “Restating the National Highway Transportation Safety Administration’s National Motor Vehicle Crash Causation Survey for Automated Vehicles,” the NHTSA study is based on old data — 6,950 auto accidents from 2005 to 2007. It was also never intended for considering the safety of driverless cars.

Source: NHTSA.

The data, however, was useful in helping the task force identify circumstances where the technology behind automated cars could be limited. As a result, the task force report reaches significant conclusions that deserve attention from policy makers, insurance companies and the general public.

According to the task force report, if automated technology could not overcome weather, vehicle errors and inoperable traffic control devices, it could only address 78 percent — not 93 percent — of accidents. That’s a difference of 830,000 accidents or $45 billion annually, according to the task force report, which was issued in 2014.

The Control Factor

Tonight Show Host Jimmy Fallon joked earlier this year that driverless cars will allow people to eat, talk on the phone and even apply makeup while driving — activities they are already doing today!

Despite the technological prowess of self-operating cars, there remain many instances where drivers will need or desire to take control of the car. These circumstances fall under two basic categories. Either the technology requires driver intervention or human action makes them inoperable.

The most important expectation of driverless cars is that they will be safer because they will reduce the potential for human error.


“Part of automated vehicles’ value comes from their predictability,” the task force report said. “However, the more involved the driver is, the less predictable the driving becomes.”

The task force study found that 49 percent of the accidents in the NHTSA report had at least one limiting factor that could disable the technology or reduce its effectiveness (see Chart A).

The bar chart represents technological and human behavior-related hurdles that might have to be overcome. None of the risks negates the technology’s potential, but instead, they indicate the relative importance of each risk, provided the assumptions are true. The behavioral issues identified indicate the importance of the driver’s relationship to the technology. Improper technology use can offset its potential safety benefits so it is important that the technology is both safe and used correctly to realize its maximum benefit.

Technology-related disabling factors were present in 21 percent of NHTSA study accidents, according to the task force report. Such instances included weather, non-working traffic control devices (which interpret the environment to prevent the accident) and vehicle condition/error (which was used as a proxy in the task force report for vehicle failures).

Meanwhile, the report indicates that another 30 percent of the NHTSA accidents involved technology-undermining human activity. These actions include operating a vehicle under the influence of drugs and alcohol, sleeping, distraction or physical impairment such as a heart attack or low blood sugar.

Becoming More Human

While artificial intelligence-guided vehicles offer great promise, they still are being tested under generally ideal driving conditions. These futuristic wonders generate a great deal of media attention, but the fact remains that they have a long way to go before being able to handle the multiplicity of real-world traffic situations that drivers are able to handle today.

“People,” said Leonard, “do not like to talk about the technology not being perfect in many ways. As we deploy the system, we will be building the statistical evidence that they are safer.”

The irony is that for robotic vehicles to be safer than people-driven vehicles, they need to acquire more human-like decision-making capabilities. “The question of how people pay attention when monitoring a highly autonomous driving system remains a research question because humans are often not good at taking the ‘handoff’ from an autonomous system in a difficult driving situation,” he said.

Source: HLDI.

Automated cars will need to better reflect human judgment in several ways. The driverless car currently “lacks the courage” that humans have to handle currently difficult scenarios such as making left turns across high-speed traffic at junctions without traffic lights, Leonard said.

To improve judgment, artificial intelligence software adjusts itself through parameters, he explained. There are several parameters to perfect so that driverless cars can detect and correctly respond to other vehicles, people and cyclists. “Another research challenge is responding to gestures from police officers or crossing guards directing traffic,” he said.

Driverless cars also use more conservative judgment than humans because they are programmed to operate according to safe driving practices, such as following traffic rules. “It’s possible that an autonomous vehicle that strictly obeys speed limits might frustrate human drivers who would typically drive faster than the speed limit,” he said.

And there are specific technologies that still require expansion and fine-tuning. Driverless cars depend on Global Positioning Systems (GPS) via the Internet, but for the cars to work anywhere, the entire world would need to be precisely mapped and constantly updated. Current GPS technology, such as the Google Maps app, is a boon to drivers, but human judgment is still necessary.

Insurer Adjustments

While some experts anticipate that driverless cars will reduce overall insurance premiums by preventing accidents, Robert Hartwig, president of the Insurance Information Institute, does not believe they will have a significant impact due to concurrent premium-raising trends. “I think the tale of the death of the auto insurance industry has been greatly exaggerated,” he said.

“The auto insurance industry will continue to grow,” Hartwig said. “As the number of vehicles continues to increase, the number of drivers continues to increase and the average value of the vehicles on the road continues to increase.”

Driverless cars will likely reduce overall premiums in the long term, he believes, but some of these reductions are already taking place due to safety features being introduced in conventional vehicles.

New safety features being introduced in cars are not only showing reductions in crashes, but they are also the building blocks of driverless car technology, said Russ Radar, senior communications director for Highway Loss Data Institute (HLDI).

Electronic stability control, for example, has been shown to reduce fatal single-vehicle crash risk by 49 percent and fatal multiple-vehicle crash risk by 20 percent for cars and SUVs. Other features are also making cars safer (see Chart B).

According to HLDI, if all passenger vehicles were equipped with forward collision warning, lane departure warning, blind spot detection and adaptive headlights, and all of them worked as intended, about one in three fatal crashes and one in five injury crashes could be prevented or mitigated.

The irony is that for robotic vehicles to be safer than people-driven vehicles, they need to acquire more human-like decision-making capabilities.


Among the newest features, forward collision warning and automatic braking systems are proving beneficial for vehicles now on the road. HLDI has found that insurance claims for injuries in front-to-rear crashes dropped by as much as 35 percent for some vehicles equipped with automatic braking. And there are more safety features coming soon.

Even though safety features are preventing accidents overall, insurers are not offering discounts on particular ones, said Roosevelt Mosley, a principal with Pinnacle Actuarial Resources. The reason is practical because it would take agents too much time to determine the existence of features for each insured car, he added.

Transition Trials

Mosley sees several challenges to the insurance industry during the transition period from driverless cars becoming available for purchase to when 200 million vehicles will be replaced with driverless cars.

“What happens when 25 percent of the vehicles are driverless and 75 percent are not?” he asks. “As the percentage of driverless cars begins to increase there will be unique risk issues for non-driverless cars,” he added.

To offer coverage for driverless cars, Mosley said, “Insurers will need to adapt while continuing to offer insurance for traditional manually operated vehicles that are operating in a very different environment.”

Driverless cars will rely on Internet connectivity, raising real concerns about cyberattacks or thieves remotely taking control of the cars.


Hartwig believes that the transition period will span until the 2040s because the automobile manufacturing cycle takes five to six years and cars are being better built. “Fifty percent of all vehicles purchased 11 years ago are still on the road, so vehicles are lasting longer,” he said. “People will not rush out and buy a fully autonomous vehicle if they have a fully functioning traditional vehicle.”

Anticipating the effect of driverless cars on insurers and their customers remains challenging because there is little available autonomous vehicle data. “To create a pricing model of the most predictive 20 to 40 variables, actuaries have to start with a dataset that includes hundreds to thousands of variables,” Stienstra said.

Therefore, actuaries will not only have to wait to observe the technology’s impact in their data, they may also have to create new models to accurately capture the new variables’ impacts.

More Considerations

There are other driverless car-related factors that can affect overall premiums during the transition period as well. During the gradual transition from conventional to driverless, Mosley said, there should be a decline in accident frequency that will reduce overall premium.

At the same time, claim severity is expected to rise for decades from the expense of repairing driverless cars until the technology costs decline and driverless vehicles become more common, Mosley said. Priced at about $320,000, Google’s Prius, if available today, would clearly cost more to repair than a traditional car.

It could take decades until technology costs decline enough to make up for the cost of claim severity, Mosley said. “But the common thinking is, at some point, the reduction in claims will more than offset severity costs.”

Google is quick to point out that human drivers, not technology, are at fault when its cars experience accidents, but laws, regulations and court decisions ultimately will guide such determinations.

How much will auto accident claims be shifted to product liability? Looking at current modern safety technology, such as electronic stability control, provides some indication, said Richard L. Fox, vice president and chief actuary for the West Bend Mutual Insurance Company. “We are already relying on more systems to take control in certain circumstances where the machine can respond faster and better than people,” he added.

While he sees the potential for liability to shift from auto insurance to product liability, the evidence, based on safety features so far, show that the impact will likely be minimal.

In fact, he believes that car manufacturers could become so confident in the safety of automated cars that they might compete directly against car insurance companies by offering free or discounted insurance as a selling point — whether the fault lies with the vehicle or the driver. “We have already seen manufacturers start to dabble in auto insurance,” he said. BMW, for example, sells auto insurance to owners through its agency.

Driverless cars could also significantly reduce the substantial costs of third-party damage, which would mean substantial premium reduction, Fox said.

There are also the new risks to consider, Stienstra said. For example, it was thought in 2000 that the GPS would reduce accidents from drivers getting lost. When GPS became available via cell phone, however, new and unforeseen risks did emerge, he added. Other car features that rely on technology, such as the gas-saving hybrid vehicles, can automatically shut off on impact, disabling the car in traffic.

Determining future risks means looking at today’s hazards and using some scenario imagination. Driverless cars will rely on Internet connectivity, raising real concerns about cyberattacks or thieves remotely taking control of the cars. “Cyberrisk is a huge potential concern,” Mosley said.

For example, when security researchers hacked a Jeep via the Internet, taking over dashboard functions, steering, transmission and brakes, Chrysler recalled 1.4 million cars within days, according to The automaker sent customers USB drives with software updates to correct the problem.

The Road to Destination

The autonomous vehicle is quickly approaching its spectacular advent. Paved with technological innovation, the road to destination driverless offers exciting possibilities and the potential of new risks. By offering their unique perspectives and insights, actuaries — the seers of the insurance industry — can also help manufacturers, developers and public policy makers better prepare for the future.

Annmarie Geddes Baribeau has been covering actuarial topics for more than 25 years. Her blog can be found at