Motor insurance in the age of AI

The rise of autonomous vehicles (AVs) introduces many challenges to insurance companies: traditional models of fault attribution will inevitably be altered, requiring insurance vendors to examine both claims management and policy costs.

 

When issuing policies for people who own, drive or are driven by autonomous vehicles, there are several critical issues that will need to be examined.

 

Who is at fault?

 

As AVs increasingly take over from humans when cars are moving, it can be argued that there should be a shift in liability from the driver (or rather the “human in control”) to the vendor, manufacturer or provider of particular vehicle parts such as software or sensors.

 

This may mean insurers having to pursue complex product liability claims rather than simply allocating fault to one of the drivers in a collision.

 

These more complex legal processes are inevitably going to be more expensive and present insurers with additional risk, at least until some precedents have been established. This issue becomes even more complex in mixed-traffic environments. For the foreseeable future, roads will have a mix of manual and autonomous vehicles. Determining fault in incidents between a human-driven car and an AV will be difficult. Insurance companies will have to establish whether the AV system behaved as designed – and whether the design was appropriate.

 

In addition, there may be circumstances where a human driver has acted in a way that the AV designers could not have foreseen or which puts the human at fault – for instance, if they overrode automated safety equipment. This isn’t a new issue, though: drivers can already override safety features such as lane assist and speed limitations.

 

It will be essential that a record of who or what was in control of the vehicle is kept; in many cases, there will be shared control, with the vehicle in operational control and the human driver in the role of overseer. If the principle remains that the human is ultimately in control of the vehicle, allocating fault will be simple: if you fall asleep at the wheel when your car is on autopilot, you are still liable if there is an accident.

 

However, as AV systems become more sophisticated and (as is likely) are proven safer than human operators, such a simple assumption will be hard to maintain. In this case, humans may be liable for an accident if they decide to override automatic systems.

 

At the moment, there are accepted levels of autonomy vs driver responsibility that help to indicate who may be liable. For example, the Society of Automotive Engineers has defined six levels of driving automation, from level 0, where the human driver is fully in control with no automation, to level 5, where the vehicle is fully autonomous and the driver is essentially just another passenger.

 

Level	Description	Control	Monitoring	Liability
1	Driver assistance	Human	Human	Driver responsible
2	Partial automation	Shared	Human	Driver responsible
3	Conditional automation	System; fallback to human; complex handoff scenarios	Human and system	Shared or alternate liability
4	High automation	System (no human needed in defined conditions)	System	Manufacturer or system liable (in defined conditions); need for cyber and software insurance
5	Full automation	System (no human needed in all conditions)	System	Manufacturer or system liable; need for cyber and software insurance

 

These standards are certainly helpful, although they perhaps may need revisiting when new types of AV are developed, such as flying or submarine cars. However, they assume that the AV system will be flawless and that, with levels 4 and 5, there is no role for a driver. Given the glitches seen in some AVs, this may be optimistic. And for level 3, where control shifts between the driver and the AV system, there will need to be clear definitions of driver responsibilities.

 

With AVs, there will be considerably more data available to the insurance company in its attempts to allocate fault: AV data logs will include sensor information as well as all the data held today in a black box tachograph.

 

Level	Additional data available
1	Potential for enhancements in claims data collection from sensors.
2	Potential for tachograph and telematics data to record driver behaviour and driving conditions. Evidence of driver inattention or misuse of automation could be provided. Data could be used to personalise policy costs.
3	Will require clear records of who was in operational control at the time of an incident. Claims complexity will increase, with inevitable disputes about handover timing or system faults.
4	Role of traditional insurance companies may move from insuring drivers to insuring manufacturers. A way of regulating the degree to which drivers can instruct vehicles to operate outside designed-for conditions will be necessary. Drivers may still need some form of travel insurance.
5	Individual driver liability insurance could become obsolete, although drivers may still need some form of travel insurance.

 

More data may at first mean longer (and more expensive) analysis, although in the medium term, artificial intelligence (AI) systems should be capable of automatic data analysis and fault allocation.

 

Note that if AI systems are used for fault allocation, a human will be required to review and sign off on a decision made by the AI for ethical and probably regulatory reasons. In addition, there will be issues around data ownership and access, including many privacy concerns, although current practices around the use of data in tachographs should help here.

 

Effects of AV on driver insurance

 

At the moment, there is considerable regulatory uncertainty as the laws around AVs are still evolving, and liability frameworks vary greatly by jurisdiction. These changes will necessarily influence the way policies are structured. However, while the effects of this evolution are hard to foresee, there are some changes that are more predictable.

 

Policy wording is likely to change with terms such as “driver” and “operator” being redefined and new terms including “control system”, “automated system” and “handoff” being added. Data consent clauses will be included, with policies containing agreements to share sensor and AV system data. It is also likely that policies will have to cover liability from AV software vendors or infrastructure providers (where infrastructure communicates with AVs).

 

The effects on policy costs are harder to calculate. It is likely that the way premiums are calculated will change, with less attention being paid to traditional risk factors such as age, occupation and postcode and more to details such as vehicle model, software version, the reliability of sensors and the manufacturer’s track record. 

 

Additionally, insurers will be able to collect real-time data and use this either for micro-insurance products (e.g. insurance per journey) or simply to generate premiums that reflect an individual driver’s behaviour more accurately.

 

Some developments might indicate that costs for drivers will fall. For example, it is likely that there will be a reduction in claims frequency as AV systems take over decision-making from humans. The vast majority of vehicle collisions are due to human error: AVs can help to eliminate dangerous behaviours such as speeding and tailgating while alerting drivers to dangers such as distracted driving and fatigue. In addition, advanced data processing capabilities can use sensor data to process claims quickly and automatically and to reduce fraud, saving costs and putting downward pressure on premiums.

 

On the other hand, there are reasons why premiums may rise. For a start, AVs are likely to be more expensive than today’s motor vehicles. This means that when accidents occur, they may involve costly repairs. AVs are packed with expensive sensors, cameras and software: even a minor bump could require costly replacements or recalibration.

 

Another consideration is that the owners of AVs will face new risks, including cybersecurity risks, which allow data theft or unauthorised control, and the risk of software malfunctions causing an accident. Furthermore, the overall complexity of claims, especially where a handover of control is involved, will lead to higher costs for insurers due to dispute resolution and litigation, which may be passed on to customers.

 

 

An uncertain future

 

The insurance industry is currently facing enormous changes regarding motor insurance. Legislation has yet to be crystallised, the technology is evolving rapidly and the effects of automation on driver behaviour are still being researched.

 

What is clear, though, is that great change is coming: the picture we see today will be very different from the situation in three years, with further major changes arriving over the next decade. Insurers that are alert to the opportunities that technology brings, while mitigating the parallel dangers, will have an opportunity to thrive. Success in this unpredictable and volatile environment will take considerable attentiveness, imagination and a willingness to take many calculated risks.