In chapter 8 of the book Weapons of Math Destruction, the author highlights the impact of socioeconomic scores on people’s ability to secure loans and jobs. The author mentions the FICO credit scores and escores. At the same time, she compares and contrasts between these two different types of scores. FICO scores carry reports of a person’s financial history. FICO is used by the major credit unions like Experian, TransUnion and Equifax to determine whether a person is financially responsible. Credit reports are highly regulated by the government. So, these reports are transparent and can be challenged anytime by anyone. Due to rigorous regulation and the inclusion of personally identifiable information, these credit reports are prohibited for marketing usage. FICO represents an ideal and useful WMD system with an appropriate feedback system.
The prohibitive regulations around FICO motivated marketeers to devise alternatives like escores that are built on top of proxy data. Escores use an individual’s ethnicity, resident zip code, personal web history etc. to profile consumers. This practice is highly debatable since juding someone by race, location or web history leads to stereotypes and a highly biased distribution. To make this profiling process look more sound, some data brokers go as far as to include a person’s credit report in the profile. Eventually, these escores are used as a proxy to determine someone’s reliability and trustworthiness in paying back a loan and in carrying out professional responsibilities. And, as with any WMD based on proxy data, this system creates a vicious poverty cycle. Because of poor credit score and adversity, people from the lower echelon of economy remain unemployed, are unable to secure a loan to start a business or are unable to pay bills because of not getting a payday loan. Unemployment and missed payments eventually creep into their credit reports and start the negative WMD feedback loop.
To solve this problem, some people can try to game the system by rigorously following and optimizing the attributes that are used to profile them. However, this is not an ideal solution. Sometimes, there are some people who go beyond the scope of WMDs and escores to properly evaluate an applicant. However, such people are rare and WMDs are designed particularly to get such people out of the way. Rigorous regulation can be a solution to ensure fairness. But, sometimes systemic bias creeps into the regulation itself. So, even before feeding data to a WMD, the data itself might reflect stereotypical human judgement. On the other hand, peer to peer lending systems sometimes help but they have the potential to become even more biased due to the absolute absence of regulation and the peer to peer scope. Therefore, to rise above the bias in WMDs, a concerted effort of better human judgement, critical regulation and unbiased data are required to create fair systems.
