HOW ARTIFICIAL INTELLIGENCE CREATES JOBS, PROTECTS THE INSURED, AND SAVES INSURANCE AGENCIES MONEY – EPISODE 074
In this episode of The Digital Broker, Ryan Deeds talks to Ron Glozman, CEO and founder of Chisel AI, about artificial intelligence and its applications in the field of insurance. By listening to this episode, you will learn:
- How artificial intelligence subdivides into specific AI’s, what each of these specific AI’s does, and how they differ from a general AI
- How specific AI’s are already helping people in other industries, and what they can do in the insurance industry in particular
- How specific AI’s create employment opportunities by relieving human beings of soul-sucking work
Artificial intelligence. You have heard of it, all right—probably at an Insurtech conference. AI this, AI that. But how well do you understand it? Most Insurtech conferences are geared primarily toward carriers (although that’s changing), and artificial intelligence has consequently become viewed as a “carrier thing,” leaving insurance agents and brokers in the dark about what, if anything, it can do for them.
It’s time to demystify artificial intelligence for insurance agencies, and we’re going to be the ones to do it. It would not be the first time that we’ve taken on a controversial topic: our episode about Millennials comes to mind. But whereas Millennial habits simply puzzle some employees, artificial intelligence has gotten everyone on guard. What, exactly, can artificial intelligence do? How reliably can it do it? Will it take away my job? Will I wake up one morning and find out that my AI-powered personal assistant has transformed into the Terminator and turned against me?
We’ll attend to every one of those questions—yes, including the Terminator one—in this episode of The Digital Broker with Ron Glozman. There is perhaps no better person to weigh in on the intersection between insurance and artificial intelligence than Ron. He is the founder and CEO of Chisel AI, a company that teaches computers how to read insurance policies. According to Ron, the term “artificial intelligence” alone is too vague and imprecise to form the basis of any serious conversation on the topic. To get a clearer view of what artificial intelligence can do and is doing, Ron would rather that we break artificial intelligence down into its seven subsections, which are:
- machine learning, “the purest form of artificial intelligence” that everything else is based on;
- natural language processing (NLP), teaching computers how to read and understand text;
- expert systems, long decision trees (“If this happens, do that.”) and one of the most diffuse and longest-running types of AI;
- machine vision, the ability of a machine to read and recognize images;
- speech, the ability of a machine either to understand human speech and convert it to text (speech-to-text) or to transform text into a human-sounding voice (text-to-speech);
- planning systems, like logistics systems popular in manufacturing; and
- robotics—humanoid machines capable of touching and interacting with the physical world.
These are all specific AI’s. A Terminator-like machine would be a general AI, comprising several specific AI’s at once: robotics, speech, vision, natural language processing, and machine learning. We are quite a long way away from that. As far as we know, there is no real-life Skynet, no company that’s making serious strides in assembling a general AI. And why would there be? Specific AI’s are immensely challenging enough on their own—plus, they’re actually getting things done in the real world. Amazon uses expert systems to recommend other products you might enjoy. Facebook uses machine vision to recognize and tag you in photos. These are all impressive and sophisticated uses of AI, but they are mundane in comparison to how some specific AI’s are keeping people safer and saving lives. The Transportation Security Administration (TSA) is studying ways in which AI can help screen passengers more effectively. And deep learning medical imaging is helping doctors and radiologists extract a lot more information about patient conditions.
Other industries, in short, are way ahead on how to leverage specific AI’s to solve problems and make progress. Insurance agencies have to do the same, and some have already started. We’ve talked about policy checking in our previous episode with Ron Glozman. Currently, most policy checking is performed by human beings, but this has its drawbacks. Humans are imperfect readers, and they can only handle so big a workload. At some insurance agencies, the amount of policy checking to do is so voluminous that a lot of policies simply go unchecked, making for significant E&O exposure. Ron’s company, Chisel AI, employs natural language processing to teach computers how to read and understand insurance policies in way less time than a team of humans would and at a higher accuracy rate. This is amazing, but it’s also controversial. What happens to the old policy checkers after a machine takes over their job? Do they get fired? Not necessarily. Policy checking is rarely a full-time profession. Generally, it’s just one part—a big, soul-sucking part—of an employee’s total responsibilities. Cutting down on it doesn’t get those employees fired—it frees them up to do other things. Furthermore, if policy checking becomes easier and cheaper to do thanks to AI solutions like Chisel, more agencies can afford to do it, thereby delivering better value to customers and increasing profit margins—so that, if anything, the agency can now afford to hire more people.
Yeah, we’re Terminators, all right—of soul-sucking tasks. Killing soul-sucking work creates jobs.
Okay, but can you trust a machine with something as sensitive as policy checking? Can an AI ever be 100% accurate? Probably not—then again, neither can human beings, and nobody’s forbidding them to do policy checking. A machine, on the other hand, can be supremely more reliable than a human being. Ron explains: “Suppose a human being can check 2,000 policies in a year, over the course of a 25-year career. That’s 50,000 policies. Our machine has seen over half a million by that point. The scale at which we’re able to train it is unfathomable—we’re talking hundreds of years worth of training. I don’t think AI will ever get to 100% accuracy—at some point, we will hit a sort of asymptote, whereby adding more information doesn’t get us much further. But it’s still a lot more computationally competent than a human.”
We’re not saying that artificial intelligence can’t be used in unhealthy ways, but it is wrong and unfair to reduce all of artificial intelligence to just those. In the insurance space alone, we’re seeing plenty of specific AI’s keeping people safer, making their lives easier, and helping agencies and carriers save money. Let’s look at just two examples:
Auto insurance. Would-be fraudsters resubmit the same pictures, or they photoshop others to make it look like they were in an accident when they weren’t. There’s at least one company that’s reducing this type of fraud by up to 90% by employing machine vision to uniquely fingerprint every photo. As before, the benefits of this trickle down. The insurance company saves money by paying out fewer fraudulent claims, and this drives down prices for everybody since auto insurance premiums are partly based on total payouts in a specific area.
Property insurance. Smart water meters identify pipe damage before it leads to flooding. Smart fire alarms extinguish a fire before it devastates a room or building. Some of the most prominent investors in these technologies are insurance companies because the best thing that can happen to an agency, a carrier, and a customer is for a claim never to be filed. When disaster is averted, the customer is safer; the carrier doesn’t have to disburse; and the agency, by way of intermediation, enjoys the credit of having brokered such a good arrangement.
If you want to keep having a debate about the merits of artificial intelligence, you aren’t lacking for perspectives on either side of it. We will limit ourselves to reminding you that there are options out there today. If you can clearly define your problem, you should be able to find an AI solution for it. And if you can’t, or there isn’t one, keep articulating your problems anyway. You might eventually meet an entrepreneur like Ron who can develop a specific AI solution for you.
Or, you might just meet Ron. Are you coming to InsureTech Connect this year? Ron will be speaking there, and so will Ryan Deeds and Indio CEO Michael Furlong. We’re all excited about this year’s inaugural Agency Connect, a brand-new one-day event dedicated entirely to insurance agencies and their issues. Come say hello to Indio at Table #K11 on the trade show floor. Join the Digital Broker LinkedIn group to tell us what you thought of this episode, what you think about AI so far, and whether we can expect to see you at InsureTech Connect later this month.