Results may show a biased output.
By Sreekanth A. Nair
Soon, the US may be able to crack down on criminals much before they commit a crime. Richard Berk, a statistician, and a professor at the University of Pennsylvania, is working on an algorithm that can predict the likelihood of a person committing a crime.
Berk’s tools have been used by different government departments in different states. He is currently working on an algorithm to predict at the time of birth, whether a person will commit a crime when he turns 18. Finding data to feed into the system is the only limitation in using the algorithm according to Berk, reported Bloomberg.
Berk started working on crime cases from the 1960s when violence broke out after the assassination of Martin Luther King Jr. he started focusing on machine learning in mid-1990s and has helped Philadelphia’s Adult Probation and Parole Department, and Maryland’s and Pennsylvania’s statewide parole systems.
Berk said that it is not necessary to know why a person commits a crime in machine learning. But, even though the program is a neutral one, it may generate a biased output. Studies have found that sometimes, the arrest records show the names of more blacks than white. This is mainly because of the years-long racial discrimination in the legal system. So, when the input goes wrong, the output also turns wrong.
The statistician claims that his algorithm is free of racial disparity and shows risk-scores of people regardless of race.
Admitting that his own work has proved wrong in 29 to 38 percent in predicting whether a person is low-risk, he said that it is the government to draw a line to decide on the issue. “The policy position that is taken is that it’s much more dangerous to release Darth Vader than it is to incarcerate Luke Skywalker,” Berk told Bloomberg.
However, eliminating sensitive factors will reduce the predictive power of the algorithm said, Berk. “If you want me to do a totally race-neutral forecast, you’ve got to tell me what variables you’re going to allow me to use, and nobody can because everything is confounded with race and gender,” he said.
Berk wants to predict whether a child commits crime at 18 at the time of his/her birth based on the environment and history of the parents. But he is not sure if he can succeed in that, as a person’s biological information is in the hands of many agencies.