Computer Vision: From Diagnosing Cancer to Transforming the Collision Claims Process
From the day Watson beat Ken Jennings at Jeopardy, artificial intelligence (AI) has captured our collective imagination. Just six years later, AI has gone from game show gimmick to one of Gartner’s 2017 Top Ten Strategic Technology Trends. Every day brings more news about the latest advancements, and verticals of all types are turning to it to help meet evolving customer needs and business demands. One reason AI is of particular relevance to our industry is because of the role it plays in computer vision—and one of the most relevant applications for computer vision is self-driving cars. Computer vision basically seeks to enable computers to ‘see’ images and extract information from them, in much the same way a human does. It goes beyond sensors that simply capture information. It layers in deep learning—the ability to actually perceive, interpret and respond to what’s happening in the environment. And this ability is essential for vehicles to be truly autonomous.
Healthcare Is at the Forefront of Computer Vision InnovationComputer vision also has many applications beyond self-driving cars. And as is often the case, the healthcare industry is at the forefront of innovating key trends. Take, for example, a team of researchers at Stanford who used computer vision, deep learning algorithms, and a database of more than 120,000 images to train a computer to identify and diagnose skin cancer—with the same accuracy as 21 dermatologists. Now, that same team of researchers is looking to connect consumers to that process via their mobile devices. Ultimately, they envision a solution in which people, especially those without easy access to medical care, can use their mobile phone to connect with physicians, who then view and digitally capture skin abnormalities and use computer vision-based software to remotely diagnose malignancies.
Potential Applications for P&C and Collision Repair IndustriesSo what might the P&C and collision repair industries learn from this? Here’s one scenario: just as the computer was taught to recognize and diagnose various types of skin cancer, it could be taught to recognize vehicle make, model and year, type of damage, and extent of that damage. Based on experience, it may then be able to extrapolate appropriate repair procedures. Now, like the researchers at Stanford, imagine connecting that process to a consumer with a mobile device—like a claimant who uses their phone to take photos or video of an accident. Those images, when submitted as part of the first notice of loss, could provide an enormous amount of information for a given claim, perhaps even determining total loss without the expense of towing a vehicle to the shop.
Just as the computer was taught to recognize and diagnose various types of skin cancer, it could be taught to recognize vehicle make, model and year, type of damage, and extent of that damage.And looking even further down the road, what if that recommendation could then automatically trigger a series of activities that would increase the efficiency of the repair process—identify a qualified repair shop for that vehicle, create a preliminary estimate, place an order for replacement parts, and even schedule the work to be done? Similarly, could that information be used to predict the types of injury a claimant is likely to sustain and set the wheels in motion for treatment? Sure, it’s a leap, but you get where I’m going with this.
Mitchell Innovation and Computer VisionAt Mitchell, these are just some of the questions we’re asking ourselves—and looking to answer. Photos have always been a critical component of the claims cycle. To that end, we have launched a compelling innovation project incorporating computer vision and artificial intelligence. In October 2016, we formally announced that “Mitchell Partners with Tractable to Bring Artificial Intelligence to Insurance Claims for the First Time”.
The ‘Mitchell Review’ project utilizes millions of photos from damaged vehicles to ‘train’ computers to recognize vehicle damage across all makes and models of cars and trucks.The ‘Mitchell Review’ project utilizes millions of photos from damaged vehicles to ‘train’ computers to recognize vehicle damage across all makes and models of cars and trucks. Then, once an estimate has been written, Mitchell Review will use computer vision to double-check a repair vs. replace decision— based completely on photos. This means that appraisers and experts could potentially be sent into the field more selectively, which may save time and boost productivity. And finally, given the growing skill shortage in the industry, this is a clear case of machines aiding humans. We expect to launch the solution in late summer 2017 and look forward to more innovation using new technologies.