The first speaker event of the semester for the Risk Management and Insurance Society, held jointly with the Actuarial Science Club, introduced us to the newest addition to the distinguished faculty of the UW-Madison Department of Risk and Insurance. Daniel Bauer is an Associate Professor and the Hickman-Larson Chair in Actuarial Science.
Professor Bauer spent most of the event informing us about machine learning and what its impact could be on the insurance industry. He explained at a basic level what machine learning is without diving into the computer science details: machine learning combines long-known statistical methods with recent advances in computing power to create algorithms that have predictive power and attempt to produce repeatable and reliable decisions.
Machine learning programs preclude the need for humans to write tedious and infinitely long algorithms because the machine writes them as it goes through its iterations of analyzing inputs and producing outputs. These programs allow more variables and inputs to be analyzed using popular statistical regression tools than humans could perform manually. Machine learning programs also build complex decision trees that consider more possibilities than a human mind could handle to determine which actions to take.
For example, technology for self-driving cars is constantly sensing its surroundings and attempting to learn how to best react. When the sensors sense an animal in front of the car, they must learn to brake. They must also repeat this action for other objects or people that come in the path of a car. Through many trials, the technology will learn how to react to that situation and many others that drivers face on a daily basis. Machine learning uses virtual repetition and pattern recognition to provide decision outcomes.
In terms of the insurance industry, machine learning has the ability to supplement traditional insurance knowledge and actuarial methods. Professor Bauer did warn, however, that utilizing machine learning simply to add as many variables as possible to an actuarial analysis would be improper, especially for traditional actuarial models that already have strong predictive power. Adding variables does not automatically translate to more accurate results. As is the case in many industries, technologies like machine learning can perform the non-value-added tasks of actuaries, insurance agents, and underwriters, and run complex calculations that are too intensive for people alone.
Humans in the industry would be best positioned to understand the new technologies and utilize them to their advantage to be more efficient and produce data-driven decisions. On the other hand, the very intangible relationship skills and innovative ideas that humans possess will be what distinguishes person from robot in a more automated insurance industry.