The Justice (or lack thereof) of Math
There is something I was completely unaware of. Mathematical models are becoming a ubiquitous tool in police departments for improving the efficiency of units on patrol. In cities with historically bad crime rates, police departments were under pressure to alleviate the emerging problem. Unfortunate for department heads, there just isn’t enough to funding to have cops posted on every street corner, so they had to find a way to more efficiently cover their jurisdiction. Enter: mathematical modelling.
Many jurisdictions have adopted mathematical models to best predict where crimes will occur. By feeding models historical records they can determine crime hotspots and predict criminal patterns. In theory, this is a great idea. Efficiency in the police force is not inherently bad, but the consequences of the system is what turned these models into WMDs.
These types of models, that use historical data to make decisions, have a self fulfilling feedback loop. By having more cops posted at these crime hotspots, more criminals are arrested and added to the historical records, further reinforcing the fact that this is a crime hotspot. For what O’Neil calls “Phase 1” crimes, which are typically the most egregious and violent in nature, this is not a bad thing; however, when so called “Phase 2” crimes are included, which are comprised of petty crimes like public drunkenness and jaywalking, things can get out of whack. This is exactly what has happened.
In certain cases outlined by O’Neil, cities that are heavily segregated, which is the case in a lot of American cities, geographic information is racially sensitive. Poorer, economically disadvantaged areas are notorious for having high crime rates, thus the maps outputted by these mathematical models were effectively maps of the economic distribution and ethnic makeup in many cities.
To hinder this type of bias in these models, it is important in what data is being fed to these models. Big Data is a growing trend among mathematical models, but having the most data is not always the answer. By curtailing the data it is fed, selectively choosing the most statistically significant data points will give the most accurate models. Similarly, inclusion of petty crimes that are not violent in nature is a mistake on the part of the developers. These types of crimes are prolific across entire cities and are not necessarily important to heavily police. My final point of critique, is that feeding data aggregated after the model has been implemented creates a self perpetuating feedback loop that exponentially affects the bias in the models. The data itself is as important as the functions and graphs extracted from the data.
This chapter provided a lot of interesting insight into the inner workings of modern day police departments. The use of mathematical models to increase the efficiency of police forces is a novel idea, but its implementation is extremely important. Applying mathematical models to extremely important governmental functionality, such as policing and court rulings, is a dangerous road that must be travelled very cautiously.
