The question today concerning autonomous vehicles is not whether they will become a standard technology, but how long until we have to reckon with the effects of their proliferation.
A widely covered consequence will be the loss of jobs for the 3.5 million truck drivers in the US and mass retraining of 5.2 million people employed in other capacities related to trucking. A less acute but more far-reaching change will be the shift in liability for car insurance. Late last year, when the Michigan legislature approved the testing of autonomous vehicles without human occupants, the tangle of legal – not to mention philosophical – issues that such a world entails also crossed the threshold from theoretical to practical.
Who is responsible for a self-driving car?
The good news is that autonomous vehicles are much, much safer than those driven by humans, but they will still damage property (including themselves) and kill people. Until now, manufacturers like Tesla have been careful to avoid liability through explicit driver opt-ins, “beta” labels and terms of service agreements. That has proven to be a wise choice, as people relying entirely on their driving software have died as a result.
When the Michigan law goes into effect, however, we will witness the pilot program for what will likely become a tidal shift in liability from drivers to vehicle manufacturers.
The evaporation of personal car insurance will be good news for people – and it will happen because fewer objects will get hit by cars – but catastrophic for insurers who do not stay ahead of the curve.
KPMG predicts that the loss in business for insurers will reach $75 billion, a massive 40% for the $125 billion industry. There are 260 million registered vehicles in the US that need to be insured, and currently no paradigm exists for how that will be done. For the sake of individuals damaged by autonomous vehicles, manufacturers who will find themselves liable sooner rather than later, and a multi-billion dollar industry facing collapse, now is the time to plan for technology insurance.
Autonomous vehicles are better and will continue to get better
Whether autonomous vehicles are safer than those piloted by human operators is increasingly a question beyond rational disagreement. Even the current generation has a fatality rate per mile that is half that of human-operated vehicles. That difference should only increase – we are barely beyond the proof of concept for autonomous vehicles while humans reached maximum proficiency in driving decades ago.
The only improvements in human performance have been due to improvements in car technology, like the introduction of seat belts and air bags. Driving is simply the kind of precise, repetitive, and concentration-demanding labor for which machines are better suited.
The fact that insurers have aligned their projections to a rise of the machines and a decrease in fatalities is the most telling sign of all. Insurers only make money by accurately predicting how often bad things will happen; they have one job and they do it as well as our knowledge of math and the physical world will allow. Just as the introduction of climate change into insurance models signaled the end of any apolitical opposition to the science of global warming, KPMG’s study of insurer predictions should be taken as the highest vote of confidence that autonomous vehicles will increase public safety.
Safer vehicles are bad news for the auto insurance market
Not only will safer cars reduce the overall risk pool for auto insurers, but also their autonomous nature will shift liability away from the individuals who purchase personal auto insurance and to the manufacturers.
At first, manufacturers will resist the unplanned liability of providing coverage for their customers, but ultimately it will be a competitive advantage for those with the best driving algorithms.
Manufacturers of autonomous vehicles will compete on the quality of their driving algorithm (like search engines and mapping applications did in the days of Ask Jeeves and MapQuest) or license them from other vendors (like search engines and maps do now in the age of Google Maps). The company with the best driving algorithm will eventually be happy to insure drivers of vehicles because of the competitive advantage granted by such a massive benefit.
And when a manufacturer has a strong enough market position to insure its own customers, they will have a host of new problems. Cybersecurity is on a trajectory to keep getting worse before it gets better, and new types of connected entities will have their own learning curves.
The cost of a self-driving car accident could be enormous since the technology needed for autonomous operation will itself be more expensive. That trade-off is one of the key findings of the KPMG study: while number of incidents will decrease, the cost per incident could double due to more expensive components in advanced vehicles.
Liability and growth opportunities galore
That concentration of risk makes attributing liability all the more important, and here, parallel developments in technology can both improve the fairness of that system and provide some hope for reclaiming the $75 billion shortfall.
Root cause analysis can be painfully complex for information systems, but it can also be unequivocally precise. Take for example the Tesla driver who tried to claim his car crashed instead but was dismissed after the vehicle’s logs showed he slammed on the gas pedal. Monitoring and systems of record for autonomous vehicles will be indispensable for establishing liability, and if insurers take the lead, they can capture the growth in that market.
As such disputes become more complex, the traceability of machine systems becomes more important. A common method for challenging a speeding ticket today is to request documentation of the radar’s last calibration. If you can’t prove the radar gun was correctly configured, the evidence may not be admissible. Similarly, if computer logs are submitted as evidence, then the natural challenge will be confirm the configuration of that machine’s logging system. Computers can be set up incorrectly, and when they are, their output cannot be trusted. For that very reason, the Center for Internet Security already publishes recommendations on how to configure logging, among other subsystems.
Autonomous cars could conceivably – and reasonably – be subject to process validation like pharmaceuticals. Because they operate as black boxes, manufacturing conditions must be tightly controlled to ensure they perform as predicted by test models. Demonstrating that is a massive data collection and analysis challenge, another area where insurers can partner with technology companies to grow alongside the market for autonomous vehicles.
Benefit from the rising tide rather than get lost in it
For all their improvements, autonomous vehicles will still cause damage and take lives, and we will require new technology to treat those injuries justly. Innovative insurers can make up the shortfall from the drop in total coverage by leading the way with the types of solutions needed to assess machine drivers.
Not only will actuarial models need to be entirely rebuilt, but also the data gathering infrastructure to feed them. There is an immense amount of work to do in order for autonomous vehicles to be insured with the same confidence as the death traps we use today. Whoever succeeds, we’ll all benefit from the rising tide of autonomous vehicles rather than be lost in it.
And soon, at least in Michigan, this will be a question we start to answer.