AI and the Work Force

In the seventh chapter of Cathy O’Neil’s book, “Weapons of Math Destruction,” she delves into the use of mathematical models in optimizing the work place. Her first examples describe the optimization of work schedules to reduce lost revenue caused by overstaffing during under performing periods of time. For the corporation, this is a no brainer, slow Tuesday mornings don’t return much in terms of sales and the bored employee has little work outside of watching an empty store. Corporations had major incentives to reduce these types of situations since non-revenue producing periods were still staffed by hourly employees.

To solve this problem, as is becoming the theme with this novel, corporations turned to mathematical modelling to more effectively utilize their payroll pools. These models attempted to identify the lowest revenue-producing portions of the day and reduce staffing during those periods of time. On the surface, this seems like a pretty novel idea that works well for everyone. Corporations are not spending as much money on slow periods of the day, and employees aren’t bored out of their minds watching an empty store; however, as is later illustrated in the chapter, this is not the case. The models were heavily relied on by corporations and resulted in erratic work schedules for low-income employees. Clopenings, as they were called by Cathy, were where an employee would close the store and then come back a few hours later to open it again. These types of work schedules are unsustainable and resulted in erratic work hours that were infeasible to keep. Schedules were being optimized by the week, so employees would only receive their staffing schedules at the latest point possible, resulting in a scheduling hell that required upending their lives just to schedule around work. After backlash, the corporations said they would set their schedules at the very least a week in advance, but even then the problems were not solved. These models, although powerful in their revenue protection power, were extremely inhumane to the workers. Instead, I would have proposed a system that optimizes the schedules quarterly and then set static weekly schedules that evaluate the consumer trends over this period using historical data. This would result in a much more simplified system that would also keep from having to change staffing needs by the hour.

These mathematical models went on to extend beyond optimization of working hours, and went on to rate the efficacy of employees. Employees would have their work evaluated by the models and be given a store that gives their relative performance compared to their peers. Again, the intention of this model was pure in that it wanted to determine who the best employees are. These models became an issue though when they garnered a negative feedback loop due to a lack of feedback. Models were making hiring and firing decisions, an employee that was let go due to poor performance in the model’s eyes could have gone on to work at another company and perform spectacularly; however, the corporations did not feed this information back into the model to reanalyze why their results were inconsistent with their performance.

In another example, teachers were being evaluated after some legislation changes in the late 2000s that waived low-income and low-performance schools from funding hits due to the No Child Left Behind program. One teacher, Tim Clifford, received his yearly evaluation that showed he scored a measly 6 out of 100 possible. Amazed by this, but with no insight as what to change, he kept teaching the way he had been. The following year, he received his evaluation showing he scored 96 out of 100 possible with little to no changes in the scope of his teaching. This came from a mathematical model that attempted to nullify racial and income disparities in evaluation but in doing so, based the results off a statistical baseline set by the performance of preceding students. This resulted in scores that were essentially random due to the random nature of the quality of students in a teacher’s classroom.

The teacher value-added model was inherently flawed from the beginning. Teachers being compared based on the performance of their students is difficult nigh impossible to compare. Some schools receive the wealthiest students with the most opportunities due to extrinsic and intrinsic attributes, whereas some schools receive the poorest students with little to no opportunity outside the immediate vicinity of their lives. These types of biases are inherent in the performance of students, and basing a model off of these factors results in an unequal system. Similarly, basing them off only the students before and after them results in too much noise in the data to develop meaningful results. One year, a teacher may be handed a classroom with many strong students, and the next a classroom of weak students who do not perform to the same level. No amount of teaching prowess will result in a classroom of geniuses without the right students.

Similarly, universities are now making standardized test scores optional for admissions. Brought on by the 2020 pandemic, schools are foregoing the need for SAT and ACT scores for admission. While I believe that there is some merit to this idea, I’m not sure that this is the solution to the problem. Standardized tests are equally flawed in terms of their inherent bias; however, the alternatives are not much better, especially when looking at prestigious schools like the Ivys. When all students there are presented with perfect GPAs and stellar extracurriculars, what more data can be used? Prestigious schools with low acceptance rates have started looking at low-income individuals, legacies with ties to the schools, and racially diverse students as their best options. I do not think that this is fair either, as being chosen by race or income (either low or high) has little to do with their academic ability. Unfortunately, I’m not sure what the solution to this problem is. Standardized tests are built to be exploited by the wealthy who can afford sophisticated prep and educations, but there is not much else that can be done. At least in these cases, the majority of students are equitable. A strong student at a low-income school has just as much opportunity of achieving a 36 on their ACT as a strong student at an exclusive preparatory school. I’m sure a perfect solution to this problem doesn’t exist, but I’m unsure that axing the entire idea is better than the alternative.