Ineligible for Service
In the eighth chapter of Cathy O’Neil’s novel, she accounts the use of WMD’s to streamline the hiring process. Specifically, the story starts with a college student taking time off from his studies at Vanderbilt applying for a low level job at the local Kroger. On the reference from a friend, he was made aware of the opening in the chain grocer, and subsequently applied. During his application he followed the normal procedure and was made to take a personality test. After hearing back that he had not been offered the position, he questioned his friend as to what the issue could be. He was a strong student, who was just attending a prestigious candidate, perhaps he might be over qualified for minimum wage positions; however, this was not the case. His friend who had referred him said he had been red-lighted due to the results of his personality test. The personality test had been the deciding factor in whether he would be hired or not, and this raised a lot of questions into the hiring practices of major corporations.
Human resources, or HR, is an incredibly daunting field of work that requires a lot of personnel to actively manage a corporation’s workforce. Hiring proceedings fall within the reach of HR and requires a lot of man power to process the number of applicants a lot of these positions receive. Similarly, having a large amount of people making personnel decisions leads to a lot of inconsistency in the type of workforce received from this process. In order to remedy this situation, and in the case of all WMDs, the goal was to institute fairness and bias-less decision making using mathematical models.
To begin their feat, corporations started pulling information from their huge swaths of employee information to train their models. Corporations hoped to reduce churn, the turnover rate of their positions, and increase efficiency by hiring better prospects by using less work to find them. Pulling from their historical data, they found many trends that were implemented into their models that seemed to insinuate a candidate is worthy of the cause.
Unfortunately, this is where the story becomes much less pleasant for the hard working job seeker. Many of these tests used to determine aptitude and work ethic were inherently biased or at worst, ambiguous. In the original case, a personality test was used to determine whether the applicant was a hard worker, intelligent, and wanted to stay with the company for a while. Normally, these types of tests are easy to game, and thus were amped up to try and catch only the top nth percent of applicants. Additionally, there has been a lot of legislation about using these types of tests to evaluate applicants, most of which has fallen under the order of the Americans with Disabilities Act. In order to do increase efficiency and skirt the long arm of the law, tests became a lot more gray than they were intended to be. This caused a lot of applicants to be unsure about what the companies were looking for. Although that is somewhat the intent of the corporation to produce a test that is not easily gamified, this resulted in questions that had no “correct” answer in terms of the hiring process. Applicants would have questions that both answers would result in ambiguous responses that depended entirely on the will of the model. Similarly, other models not specifically mentioned in the original case, utilized application scanning tools that whittled down piles of resumes before they were even seen by human eyes. Unfortunately, since these cases are WMDs, many of the inputs to their models unfairly categorized lower socioeconomic and ethnic groups into the “unsuitable for service” category.
A way that could have at least helped in building efficient models to ascertain employment eligibility would be to institute a more concise way to implement feedback from their models. If an applicant at Walgreens was denied then went on to work at CVS and performed spectacularly, the hiring model Walgreens used would not be tweaked to see what went wrong in this specific situation. There is no feedback back to these systems, outside of crazy circumstances that require a deep search into what systemic issue caused, for example, a series of kleptomaniac employees.
My take away from this chapter is that WMDs are essentially ubiquitous. Their use in every facet of business is inevitable, but there is a lot that can be done to make sure that these mathematical models are not biased and sufficiently fulfill their purpose. The most important part of building a mathematical model, especially that fulfills a purpose as important as hiring, is understanding why the results are what they are, and analyzing them for important biases that could unfairly affect certain populations.
