The insurance industry is one that is just beginning to tap into the potential of artificial intelligence.
If you’ve been monitoring the ImageNet challenge over the years, you know that AI’s image classification surpassed human accuracy about 18 months ago, indicating the technology is reaching a stable and mature state. Once a technology is customer-ready, it’s important that it’s also customer-centric — in that it solves an inherent problem. For insurance customers, that problem might center around a fender bender, or roof damage from a hail storm.
Using AI technology to automate a visual task, such as inspecting damage to a car, is a nearly instant way to provide insurance customers with crucial information about the extent of the damage.
For example, the customer could snap pictures of all sides of the car with their smartphone and upload the pictures into a new experimental app we’re creating that detects auto damage. Image classification AI within the app compares the customer’s photos with thousands of other anonymized crash photos to generate a cost estimate for their repair. Not only does this save time for customers, particularly in the case of a minor accident, but it also reduces uncertainty and worry during a stressful time.
After a machine vision algorithm assesses the auto damage, the customer can decide whether to get a repair done immediately or wait. For those who need or want to move forward with repair, AI could assist them through to the time they pick their car up from the auto body shop. Down the road, with the help of another machine learning algorithm, the driver could potentially receive a list of nearby repair shops that may be particularly experienced, for example, in servicing a specific type of vehicle and that have positive online reviews.
Having AI capabilities to triage after an accident is one useful application, but insurance companies can also implement AI to help prevent accidents altogether. At Solaria Labs, the innovation incubator for Liberty Mutual Insurance, a team is developing machine learning tasks to enable safe routing and parking for drivers. Imagine a trip into the city that allows you to avoid the most dangerous intersections.
For a suburban-dwelling family, this technology may be what enables a family road trip with less worry, as we’ve found that driving in the city is perceived as stressful and is a deterrent for many. For urban dwellers, perhaps having your side-view mirror knocked off is enough to keep you from street parking your vehicle, but with the help of an app powered by machine learning you could be routed to the statistically safest place to park.
With all the interest around AI, it’s natural for companies to explore the options, but those looking to implement the technology have to be realistic about its current capabilities and limitations. AI is reaching maturity in some areas, like recognizing objects in images, but work still needs to be done for other tasks. Researchers continue to evolve neural net architectures, but in some cases the algorithm is not business-ready.
For example, more work needs to be done when it comes to skills like natural language understanding (NLU). Right now, if you were to ask a virtual assistant “Call Brenna, no, call Nora,” most virtual assistants will call Brenna first, or ask you to repeat your request. Proper NLU would derive your intent over the command that was spoken, and this is where it has struggled to date.
Also, keep in mind that some AI methods are better suited for text, others for audio or numeric, and others for pictures. Looking across the entire insurance ecosystem, we have data in all these forms that can be leveraged to ultimately reduce customer effort, which should be the goal for any company using AI. The more natural and seamless the customer’s interaction with the technology (think of interacting with a virtual assistant), the less effort it takes them to accomplish a task, and the more likely they are to accomplish that task. It may take a few minutes and several key strokes to request an insurance policy online, but doing so through Amazon’s Alexa takes mere seconds. Since both interactions can take place in the customer’s living room, it’s no surprise which is more desirable.
