Predictive Policing Algorithms: How Do They Affect Prison Sentencing and When Do They Fail?

The United States has the highest incarceration rate in the world. As the U.S. incarceration rate is rapidly increasing, the amount of resources required to maintain the criminal legal system is getting larger. Advances in artificial intelligence have led to the development of “predictive policing” algorithms (PPAs). These tools are designed to predict and prevent crime and aid in forensic detection. Although PPAs are extremely cost-effective and data-driven, their effectiveness is uncertain and several biases which target specific racial groups have been found in the underlying algorithms.

While several researchers have reported on how PPAs work, their biases, and their effects, there's still no clear policy recommendation for governments to follow. This study brings together findings from past quantitative and physical experiments, the technological infrastructure of PPAs, and scholarly research. In addition, this work considers the statistical models used in PPAs, machine learning algorithms used in forensic detection, and statistical studies regarding major current uses of PPAs. A practical policy that would make police departments more uniform in their operations and use resources more wisely is suggested.

This work finds that despite their current flaws, successes in PPA usage show great potential for future use. However, they should not be used in the present-day as they can cause innocent deaths in majority-minority cities with a current problem of racial injustice in the police system. But, it is critical for the U.S. government to allocate more funds and labor into making PPAs a reality for the future.