In class, a teacher recommended the Propublica article on machine bias. I already knew how algorithms are used in the criminal justice system, but I was still interested in the article’s conclusions. Although it is a bit outdated, written in 2016, I believe the issues it discussed are still relevant today.

The article revealed some interesting insights on how risk assessment algorithms, like COMPAS, are used in the criminal justice system.  These tools, designed to predict the likelihood of reoffending, showed an alarming rate of bias against black defendants, falsely labeling them as high risk nearly twice as often as their white counterparts. The accuracy of these predictions is also concerning as they are around 20 percent higher than random guessing, which would seem high at first until you realize that actual humans could be affected by the wrong choice. 

Considering this was published around 8 years ago, I’m sure these algorithms would’ve been modified and updated with new data. The problem with this is that if this data is also inherently biased in sentencing because of things like race or over-policing, it would be tough to incorporate data into training the AI.  

I think these computer algorithms are not advanced enough to warrant inclusion in sentencing. The original goal of the problem was not to help in sentencing but rather to analyze these people’s situations to get them the help they need. I would love to know your thoughts on this, so feel free to send me an email or leave a comment below.