One of my first significant impressions of computer vision came from The Terminator when the robot was able to see and identify objects as they moved. That computer vision of the future has started to emerge, and as computers start to see things, there are far reaching implications for the property and casualty insurance industry.
How computers see
Like humans, a computer's vision implies two distinct things: identification and categorization. Humans do this unconsciously as we view the world around us; however, computers perform each task separately and quite 'consciously.' A simple example is how Facebook recognizes faces and categorizes them as your friends. A more advanced example is where a computer recognizes objects within running movies (or live video feeds) and categorizes them as "safe" or "dangerous." I recently saw computer vision used to recognize people and objects in near real-time within a video, including people running and a bag lying on the ground. Fascinating technology.
In insurance, there are virtually countless use cases where humans need to recognize and categorize object, e.g., "Ford Focus — total loss." In that example, the object is both recognized "Ford Focus" and categorized "total loss." This kind of computer vision is already is starting to take hold and has the ability to dramatically speed claims and other insurance processing.
I mentioned that sedentary bag in the video example because computer vision is being used to look through closed-circuit television to spot suspicious objects. While the video feeds are not yet universally sharp enough to see "everything," this is quickly changing and the security implications are pretty amazing.
Hyperbole aside, humans still have a big edge on how quickly we read and recognize pictures (movies, or moments in our lives, are seen as a massive number of individual pictures) compared to computers, but not for much longer.
Humans also have a big edge because there are virtually countless exceptions to a single object type because of the nuances that exist in objects. Taking the totaled Ford Focus example again — maybe it is not really a complete loss because the image was somehow skewed. In a non-total loss example, there may be frame damage unseen in the picture, making it a total loss. Point being that humans are still required.
Computer vision is doing even more than helping identify and adjust claims; collision avoidance is helping prevent claims from ever happening. (Photo: iStock)
Insurance use cases make computer vision tangible
In the world of claim inspections, computer vision has the potential to significantly speed up the process, reduce errors, and lower fraud. The "identification" aspect is the object type (vehicle, house, etc.) like the Ford Focus example. Like all computer vision use cases, the real issue is getting enough samples so the machine can accomplish its identification task. Until recently, the number of permutations and combinations was simply impossible for computers to process, which is exactly what is changing now to make computer-assisted inspections possible.
As classification gets more sophisticated, insurance agents can use computer vision to adjust some claims. A good example is in situations where relatively simple geometry is used to estimate a loss. Consider damage to manufactured homes. There are relatively few types of building materials, and many manufactured homes have straightforward geometry. Now that computers can interpret distance (see the Tape Measure app on the iPhone X), a computer can identify not only the object but how big it is, and then how much it may cost to replace.
Taking both inspection and adjusting to the next level, some insurance companies are using drones to not only perform identification and classification tasks, but also provided the added value of reducing the risk of harm to adjusters. For example, instead of climbing a ladder to inspect roof damage, a drone can just fly there. Companies also are using drones to obtain high level views of catastrophes. Doing so not only provides rough order-of-magnitude loss views, but also keeps adjusters out of harm’s way.
Computer vision is doing even more than helping identify and adjust claims; collision avoidance is helping prevent claims from ever happening. Collision avoidance requires extremely fast identification speeds, and have some relatively simple applications (e.g., drones avoiding objects) along with complex applications (e.g., autonomous vehicles and proximity-based collision avoidance). It is also not outside the realm of comprehension to imagine alarms that can help employees avoid workplace accidents. Collision avoidance is risk management taken to an entirely new level of sophistication.
Making computer vision accessible
Unlike some big ideas of the past, it seems that computer vision is happening openly, through application programming interfaces (APIs) that anyone can use. This is very good news for the insurance innovators of the world, since they can focus on the use of the technology versus the technology itself. There are a number of vendors supplying computer vision APIs, which is a very good thing for innovation, and likely why the technology is progressing so rapidly.
What we can expect in the near future
Computer vision has come a long way, there is still much further to go. As the speed of identification and categorization increases, so does the applicability for insurance. It is not beyond comprehension to have computers rapidly creating estimates for entire property damage claims, across a wide range of objects including vehicles, buildings, and interior objects. When handwriting recognition progresses far enough, the next step is significantly improved transcription services as the work moves towards validation and evaluation and away from transcription itself.
Regardless of what comes to pass, one truth still remains: People plus computers are still more powerful than computers alone. The range of reality far exceeds the computational power of computers. This means that people should focus energy on learning how to interact and help computers maximize our collective value.