WASHINGTON — Valentine’s Day this year was less than sweet for Google. For the first time since it began testing autonomous cars in 2009, one of Google’s cars caused an accident.

It was the 17th accident for a Google Autonomous Vehicle in California, but the first in which the driverless vehicle was at fault.

The Google car was trying to merge from the right lane to the left lane of a road in downtown Mountain View, California, when it was hit by a bus. The test driver, who is required by state law to be in the car but doesn’t drive, reportedly thought the bus was going to slow down to allow the Google AV into the lane, and apparently so did the car.

No one was injured in the accident. The Google AV sustained damage to its front fender, wheel and one driver’s-side sensor.

The accident highlighted an important question facing insurance companies: Who’s going to cover accident claims for autonomous vehicles?

As autonomous vehicles continue to develop, insurance policies and regulations are going to have to change to keep up with the technology. Google, which declined comment on the insurance issue, is road-testing its autonomous vehicles in California, Washington and Texas. Michigan, Florida, Nevada and Washington, D.C. allow testing as well. Tesla, Honda and others are also road-testing, and General Motors is one of many other groups still in the development phase.

“I can tell you that they are certainly paying attention to this because it is really going to change the business model,” Michael Barry of the Insurance Information Institute said “It is going to take a main criterion, namely the driver’s ability to drive, out of the equation.”

There will always be accidents because of weather, vandalism and other random acts. For driverless cars, though, the issues for insurers are how well the cars can drive, how to handle the dangerous combination of human drivers and autonomous vehicles on the road at the same time and which policy-holder pays – whether states move to no-fault insurance systems or decide to hold manufacturers responsible.

An added wrinkle is that because insurance is regulated by states, experts say it’s hard to generalize about how the insurance industry will change.

Richard Wallace, director of transportation systems analysis at the Center for Automotive Research, an independent research group, said creating good AV “drivers” is the hardest part .

“Replacing your eyes and ears with sensors and even vehicle-to-vehicle communication isn’t that hard,” Wallace said. “The hard part is replacing your brain and mine and every other driver’s.”

He said manufacturers are solving the problem using deep learning, which is a way to program machines so they can imitate complex thought. It’s like giving technology the ability to learn from mistakes the way humans do. In fact, some of the research behind deep learning comes from neuroscience and copies communication patterns that happen in a human being’s nervous system.

“It’s repetition, example, the same as kids learn (through) object repetition and classification,” Wallace said, such as seeing a ball and hearing it called a ball again and again.

Perfecting those systems is going to take time. Wallace said he expects to see the people who are most willing to take the risk in self-driving cars by 2020 and the rest following suit by 2025.

Barry estimates autonomous vehicles won’t be commonplace for at least two decades.

“One of the key challenges is you may have just bought a car and you’re going to keep that six, eight years,” Wallace said. “The average length is 11.5 years.”

Self-driving vehicles already co-exist on the road with human-operated cars during the testing phase. But a quick shift to autonomous technology parked in most Americans’ garages – or getting us around town — is unlikely.

Instead, AVs and cars with drivers will be on the road together for quite a while – and that phase of mixed technology is going to be one of the more dangerous parts of the switch to autonomous cars, according to Wallace. Human drivers are unpredictable, which could be a problem for driverless cars.

“The two don’t mix,” he said. “You have adamantly rule-obeying technology and you have selectively rule-obeying technology like you and me.”

Insurance providers need to consider how those two types of “drivers” are covered by their policies.

“Insurance isn’t going away,” Wallace said. “First of all there’s all that mixed traffic for a long time to come. There’s weather-related things. … There are crash scenarios where it doesn’t matter that a computer can react 10 times faster than you or I can react.”

Even with driverless cars, accidents will still happen, but the idea of fault is going to change, Barry said. A survey of insurance industry leaders by KPMG in June 2015 indicated that insurance groups are going to wait to see how laws are changed before they make their own changes.

He said one good possibility is a shift to no-fault policies, where the policy-holder files the claim with his or her own insurance provider, which keeps claims from clogging up the country’s court systems. Barry also said he would not be surprised to see fault shift to manufacturers.

The same survey by KPMG indicates that providers expect a new player to enter the industry: Google. Nearly 40 percent of those surveyed expect technology firms like Google and Intel to start providing their own insurance, and 81 percent anticipate that those firms will own the data related to driverless cars that will be used to make industry decisions.

The good news is that insurance costs for consumers are likely to go down, Barry said, because autonomous cars should reduce the number of accidents, meaning insurance will become more about weather-related damage or vandalism and less about covering injuries.

“It costs more to repair a human being than it does a vehicle,” Barry said. “Anything that cuts down on the amount paid to take care of a person will decrease the price of insurance.”