It sounds like a scene from the popular Tom Cruise movie, 'Minority Report,' where an artificially intelligent machine identifies a person who was released from prison will likely commit a crime again. However, this is real life and an MIT professor was the one who developed it.
According to the latest crime statistics released by the Federal Bureau of Investigation (FBI), crime in the United States has decreased but it doesn't mean that security will slack off. According to a study, a majority of these crimes are committed by almost the same people. Data has shown that two-thirds of prisoners who have been released commit a crime three years after their release.
In order to address this issue, Professor Cynthia Rudin of the MIT Sloan School of Management developed a simple prediction model of recidivism using machine learning methods.
Her model holds the criminal records of more than 33,000 individuals and has the ability to predict what type of crime will be committed by these individuals. It can also be customized by each jurisdiction to predict the time and place these crimes will be committed.
To make it more accurate and identifiable, she assigned points for various factors. If the total of these points reached the threshold that has been set, then that person is more likely to commit a crime three years after their release. All of these are determined and calculated by the machine learning algorithm used in the collected data.
Despite the promise that it holds, Rudin also cautions that this also has a potential to be abused. For example, some authorities might use it for discriminatory punishment where a person's sentencing will depend on their skin color.
Taking this possibility in mind, Rubin and her team decided to leave race and socio-economic factors in the algorithm because they deemed it not helpful at all.
The study titled 'Interpretable Classification Models for Recidivism Prediction,' is published in the Journal of the Royal Statistical Society Series A.
