By Ran Harpaz
Artificial intelligence has been at the peak of its hype cycle over the past few years, with everything from chatbots to machine learning and facial recognition to natural language processing all getting high attention and technology spend.
Subscribe to the Crunchbase Daily
Since the term AI is so broad, industry experts are starting to differentiate purpose-built applications of machine intelligence from general-purpose AI, or the kind that large companies like Google, Amazon and Microsoft offer. The foundation of general-purpose AI technology is necessary to get started, but not enough to advance beyond industry-average outcomes.
Keeping in mind that winning applications of machine intelligence are uniquely built-for-purpose solutions, one must apply specific domain expertise and a particular body of knowledge of a specific industry to achieve step-function changes. As a result, unique built-for-purpose solutions applied to the domain requirements of a specific industry are the next frontier to explore.
Using the insurance industry as a case study example, we see a revolutionary change as we combine general-purpose AI with purpose-built human-in-the-loop AI to completely rewrite the way we approach risk management.
Achieving scaleRan Harpaz is CTO at Hippo.
In the field of insurance, specifically home insurance and proactive home protection, our industry has accumulated an enormous depth of knowledge over the past century. Industry-specific domain expertise relating to actuarial, underwriting, claims, coverages, fraud and more has been developed and improved beyond any layman’s ability to decipher.
The component that has been lacking, however, is scale: The ability to apply human expertise repeatedly, consistently and accurately a vast amount of times without losing quality. This is where expert systems and augmented intelligence make a difference.
Some experts define an expert system as programs that mimic the thinking of the human experts who would otherwise have to perform the analysis, design, or monitoring. More recently, robotic process automation has been introduced to reproduce tasks ordinarily performed by humans—like common underwriting decisions—but with the consistency, speed and scale of computers.
“Augmented people intelligence is a human-centered partnership model of people and artificial intelligence working together to enhance cognitive performance, including learning, decision making, and new experiences,” according to the research firm Gartner. “The goal of AI should be to empower humans to be better … It assists machines to do their best and people to be their best.”
Specifically for home insurance and home protection, in order to mimic human intelligence when using machines, companies have to accurately model elements like preventative maintenance, the type of roof, water proximity, fraud likelihood and inspection recency. Those elements don’t show up in the training set of Google Images, Siri, Alexa or any of the general-purpose AI databases.
With a purpose-built system, the system is trained to understand and discern the specific data or imagery that relates to the problem at hand. That’s why purpose-built AI machines need to be built, and why companies need to build them for their specific industries.
Ultimately, making a difference results from choosing an ambitious purpose to pursue together with acting on the developments that make it possible. The better outcomes we set our sights on are ultimately happier customers, safer homes, fewer claims and lower costs.
At Hippo, we employ these technologies to drastically scale and achieve our ambitious purpose: to make homeowners worry-free. At the outset, we have the inhouse domain experts and the depth of data required to make the right decision at the right time for each and every person/property combination.
Then, to emulate the decision-making of human experts at scale, our system employs a constantly expanding knowledge base alongside dynamic rules, addressing the ever-changing factors of proactive risk management. We accumulate knowledge from multiple domains and data sources, going deeply and broadly, processing far more information and considering huge numbers of factors, in a much more sophisticated and finely attuned way than any human can do repeatedly.
While we are still a long way from independently learning computers with the level of versatility that human experts possess, we know how to develop expert systems with human-level, or better, competency in a set of particular fields. Specifically, predictive modeling, mass personalization and underwriting are showing impressive results, relying on the right mix of data and computing power to drive machine learning processes.
We appreciate how IBM was early to describe augmented intelligence this way: “systems that enhance human capabilities, rather than systems that aspire to replicate the full scope of human intelligence.”
The insurance industry has been built on paying out claims for losses, and it has gotten to an equilibrium around this commercial structure where customers pay for the aggregated cost of claims.
At Hippo we see the world differently, so our risk management goal is drastically different. If most home damage is man-made, and we can reduce this preventable damage to zero, it is drastically better for customers. Life would be better if you didn’t have to fix it, pay a contractor and go to a hotel while the damage is fixed. That’s why our focus is on protecting homeowners and their homes. Technology and, specifically, augmented intelligence and expert systems allows us to do that.
Ran Harpaz is chief technology officer of insurance startup Hippo.
Illustration: Dom Guzman