How Big Data Asses Insurance
In chapter 9 of “Weapons of Math Destruction (WMD),” Cathy O’Neil addresses how erroneous data is used to evaluate an individual’s insurance value. She focuses on the car insurance industry because every person who drives is required to have one. As mentioned in previous chapters, the prevalent characteristic embedded in the WMD that evaluates insurance is its opacity. The WMD sorts individuals into subgroups that is dependent on specific data inputs. The problem with the data inputs is that it is not based on the person’s ability to drive. Instead, it is based on personal data such as credit scores and academic standing. We have seen in previous chapters that this widens the wealth gap because WMDs are making life difficult for those who are already struggling. In the end, profit is all that matters; thus, the creators of the WMD deems their model to be a success because it brings profit to the company.
To mitigate the use of faulty data that is not related to driving, some insurance companies have offered discounts to their customers if they agree to install a data analyzer in their vehicle. Like a black box, it allows insurance companies to gather data on their customer’s driving habits and activities. While it is a step in the right direction, there are still subtle ways that faulty data is still collected. For example, if a person drives on the street filled with bars often, then their insurance company might mislabel them as an alcoholic.
Big data is more prevalent than ever. No matter what we do, patterns will emerge from big data, and models will feed on it and divide us into groups without being aware of why we are assigned to a specific group. This procedure is very much like an unsupervised machine learning model. Models will eventually create their definition on how they perceive data and its patterns with little human input. Thus, it will make it difficult for humans to decipher or understand how these models work. Therefore, it is important for data scientists to be aware of a model has the potential to be a WMD so that it can be prevented from becoming so.
