EECS 649 Blog 7: Justice in the Age Big Data

Author: Megan Rajagopal

In Cathy O’Neil’s book Weapons of Math Destruction, O’Neil describes how AI is used to create Artificial Intelligence models that predict crime. These models will process historical crime data, and calculated (hour by hour), which crimes were most likely to occur. These models would be used to help position cops in areas that are more likely to have crimes occur. While these models can be helpful, there are some flaws in them. These crime models create a negative feedback loop. These models are trained and developed using petty crimes, and are acting under the assumption that catching people who commit these smaller and usually unreported crimes will prevent larger crimes from occurring. However, this is not the case. Instead, the use of these models will result in more people getting arrested from impoverished/lower class neighborhoods. In general, policing has a focus on poorer areas that consist of black and Hispanic people compared to the wealthier people who commit both small and large crimes. With that said, this shows how these crime models are flawed in some ways. A way to fix this is to make sure that crime data from all socioeconomic people and neighborhoods are included in the model.

O’Neil also talks about New York’s stop and frisk initiative This is only a partial weapon of math destruction because it take person choices into consideration. The initiative ended up targeting racial minorities, which means that people were faced with selecting fairness or efficacy for the models. A solution to make the models fair is to remove purposefully remove data from the model, this would set up the crime prediction models to find smaller crimes being committed by all people, not just minorities. It is important to create a fair crime predicting model to help ensure the model is accurate and not skewed.

I thought this chapter was very interesting. It was interesting to see how AI models were used with the intention to help prevent and predict crime. Given everything that has been going on in our world, it was very comforting to read O’Neil’s perspective on policing. She believes that the police should work to build trust with a community instead of focusing on arrests. I also think that these prediction models should not be the only tool used when it comes to preventing crimes. While these tools are helpful, police should make sure they are making arrests based on evidence, not a prediction from an AI model. All in all, if these models can be developed to be fair and predict crimes everywhere, they have potential to help society.