Archive for January, 2013

Minority-Report-2002-movie-props Anyone remembers the movie Minority Reports (BTW, probably the only Spielberg movie I’ve really enjoyed since the third Indiana Jones)? The premise of the movie was a special crime unit called Precog where, using the psych juice of three guys, they could predict the future and see who was going to murder someone before it actually happened. Indeed, Tom Cruise’s job in the movie was to figure out from the scattered perception of the psychs where things would happen and stop them in time. I’m not commenting the movie any further in case you have not seen it (I really recommend it if you like scifi). Instead, we can focus on the idea of crime prevention based on action prediction. Scifi? Until now.

A new crime-prediction software is under use in Maryland and Pennsylvania at the moment and they promise it will reduce the homicide rate by predicting which prison parolees are likely to commit murder and therefore receive more stringent supervision. It is not new that there are many algorithms and methods for prediction in signal processing, Artificial Intelligence, etc, like Markov Models, Artificial Neural Networks, Case Based Reasoning, ARMA models … They have been widely used to predict how stock change will evolve, how weather will change, the demands of electricity during a hot sport event (or of flush water during the breaks in those events :D) … but it is probably the first time these tools are used to predict if someone is likely to offend again any time soon.


The idea is actually simple: you give the software the full criminal record of a given person and it calculates the probability of a new crime when he is released on parolee, instead of letting the judgments parole officers make the decision. How does the software decide? Basically, by extrapolation. It has been designed from a dataset of more than 60,000 crimes, including homicides. Each criminal was categorized according to about two dozen variables, including criminal record and geographic location. Apparently, the type of crime and the age at which it was committed turned out to be two of the most predictive variables: “People assume that if someone murdered then they will murder in the future,” the creator told the news outlet. “But what really matters is what that person did as a young individual. If they committed armed robbery at age 14 that’s a good predictor. If they committed the same crime at age 30, that doesn’t predict very much.”

After clustering criminals according to these variables, each time a new person is evaluated, the software checks which groups he belongs to and checks how many people in the group committed murder when paroled or on probation. If your group seems fishy, you are out of luck! In fact, the main concern with respect to this software is: how reliable is it against false positives? or maybe how more reliable is human judgment to this respect? For further consideration, go watch the movie, it’s worth your time!