How safe is a car without a driver? That’s a decade-old question with no answer in sight...
How safe is a car without a driver? That’s a decade-old question the world keeps asking automotive OEMs, but the people who come up with the answer first might very well be auto insurance agents.
The lack of an answer thus far has created a gulf between the technology/automotive companies pushing to make autonomous vehicles (AVs) a commercial reality and consumers skeptical of the safety of autonomous vehicles (without a human driver).
The crying needs to produce safe AVs are better software, better hardware, and better systems integration, together producing a trail of evidence demonstrating safe operation. Instrumental to achieving these advances are safety standards such as ISO 26262, SOTIF, and ANSI/UL 4600. In particular, the UL 4600 standard published earlier this year is designed to bring standardization to assessing the safety case. UL 4600’s goal is to help AV developers who must present — before going commercial with self-driving cars — structured arguments to stakeholders explaining why they think their vehicle is safe.
Stakeholders include federal regulators, municipal authorities, insurance companies, industry peers and consumers. Among these, insurers might just hold the key to building consumer confidence in AV safety. Consumers might not like insurance companies, but they tend to trust their coldblooded risk assessments.
The case in point is a survey done by J.D. Power in fall, 2018. Asked who they trust most to perform reliable safety testing to fully automated vehicles, 500 auto insurance customers named the Insurance Institute for Highway Safety or its subsidiary, the Highway Loss Data Institute, as groups they would trust most (34%). Auto manufacturers (12%), the federal government (9%), and state governments (4%) lagged far behind.
Who is the driver?
Today, the insurance industry already knows quite a bit about the average driver. But with self-driving cars, a new twist enters the picture: who is the driver? In a self-driving context, the “driver”insurers must assess and rate is the software deployed in the robo-car, explained Michael Wagner, co-founder and CEO of Edge Case Research, in a recent interview with EE Times.
In his blog posted on Medium, Wagner revealed that Edge Case is adding “Risk Management Services” to its business. His company has partnered with Liberty Mutual for development efforts.
Given that AVs must outgrow their current status as a science project, insurance companies need to know enough about them that they can feel confident about underwriting the associated risks.
For regular vehicles today, insurers assess an applicant’s risk by incorporating personal information and internal claims data — which they have a ton of — into weighted algorithms. To underwrite the risks, insurers look at rating factors to predict the likelihood of a claim. The rating assigns a price based on the projected cost to the insurer of assuming financial responsibility.
But self-driving cars will depend on artificial intelligence to drive safely. Where is that wealth of data that insurers can apply as they assess the risks of a non-human cybernetic AI “driver”?
To date, very little such data exists.
Some insurance companies see their first step as participation in the standard development groups such as UL 4600.
Rather than waiting for the necessary data to come their way, they prefer to get ahead of the pack. Liberty Mutual, AXA and Munich Re America are companies already active in the Standard Technical Panel (STP) of UL 4600.
Benjamin Lewis, alliance director on the automotive and mobility strategic partnerships team at Liberty Mutual Insurance, remarked last month at a virtual panel of Automated Vehicles Symposium (AVS 2020): “As a member of the STP, we have taken part in establishing UL 4600 as the first standard to directly address the safety of automated driving systems for us.” He called the move “an important effort to materially move automotive safety innovation forward.”
Basically, Liberty Mutual needs knowledge that enables risk assessment and collaboration with partners on risk management and mitigation.
The role of hologram
The collaboration between Liberty Mutual and Edge Case Research goes several steps beyond UL 4600 activities, however.
Wagner explained that teaming with Liberty Mutual will help Edge Case add “Risk Management Services” to current offerings that include software, tools and consulting.
Edge Case Research is not ready to fully disclose details of its risk management plans. But one thing is clear. Effectively managing the risks associated with autonomous vehicles is the insurance industry’s top priority. At stake is a $308.8 billion auto insurance market, as measured by revenue in 2020 and projected by IBISWorld.
The Liberty Mutual-Edge Case Research partnership began in June 2019 when Liberty Mutual Strategic Ventures, the insurer’s venture capital arm, participated in seed financing of Edge Case Research.
At the time of the announcement, Liberty Mutual singled out “a pilot of Edge Case Research’s Hologram product” as a promising technology they will invest in so that the product can achieve scale. The press release described the hologram as a tool that “tests the software and hardware used by autonomous vehicles to perceive their environment.”
Edge Case Research isn’t underestimating the role that simulations play. But Wagner noted that it could take months or years for fleets of self-driving cars to build up the billions of miles that AV companies will need to demonstrate their systems’ safety. So, simulations are critical, said Wagner, but it’s also true that “simulation won’t be able to tell us all the unknown unknowns.”
That’s where Hologram comes in. It’s a tool to analyze how neural networks react to real-world data. Of course, the same analysis would be possible by comparing what neural networks saw with labels assigned to each object. But that would be hard to scale, given the many objects to be labeled and compared.
Hologram, on the other hand, is like a “vision test,” said Wagner. It offers a platform that “intelligently tests perception software against adversarial examples.” AV companies can use their own sensor data, enabling Hologram to identify risks. Hologram gives developers the information they need to retrain their algorithms to operate more reliably.
The impact of AVs on insurance companies
The dynamics of the relationship between AV suppliers and insurance companies was discussed in a Harvard Business Review article published in December 2017.
The article famously cautioned:
Since insuring privately owned vehicles is what the auto insurance industry has been all about, insurers have every reason to be concerned about their future growth and profitability. With fewer individual owners, there will be lower overall premiums. And since as many as 94% of accidents are attributed to human error, the number and severity of accidents and insurance claims will drop, also leading to lower premiums as insurers learn to price in accordance with real risk.
The report also identified three areas that might compensate insurers for declining revenues. They include: cyber security (as much as $12 billion in annual premiums), massive product liability (risks for manufacturers caused by the potential failure in software bugs, memory overflow and algorithm defects — as much as $2.5 billion a year) and infrastructure insurance (cloud server systems, signals, to other safeguards that will be put in place to protect riders and drivers — potentially worth $500 million in premiums).
How autonomy pans out for insurance companies remains fuzzy. Insurers will gain clarity only if they can develop: 1) the basic expertise they need to do “solid risk management,” and 2) the actuarial information that data scientists in insurance companies can apply to mathematical and statistical methods to assess risk.
Edge Case Research hopes to deliver both to Liberty Mutual. The bottom line is that for self-driving cars to become a reality, “a business deal needs to be worked out,” pointed out Wagner. Edge Case Research thinks it can devise the tools necessary for insurers to get that risk assessment right.