Predicting Crime Through Data

Monday, August 26, 2013

Tom Cruise in "Minority Report."

Throughout history, America’s law and order branches have profiled citizens, both unofficially and under the auspices of policies like “stop and frisk.” Most of this profiling has been based on race, gender and neighborhood.

But what if those identifying factors were combined with other information, like how many tattoos we have, and bits and pieces collected of data by the National Security Agency (NSA)?

Would the powers that be start predicting which among us might commit felonies? Would we find ourselves in a real-life version of the movie “Minority Report"?

Jim Adler knows from experience that these questions aren’t just the stuff of science fiction. He recently created a program that makes predictions about criminal behavior based on identity. And as he sees it, programs like his aren’t good for society, but they can raise questions that are.

Adler is an entrepreneur and thought leader on big data, privacy, security, and voting systems. He’s testified before the U.S. Congress, Federal Trade Commission, and National Institute of Standards and Technology. He joins The Takeaway to discuss his crime prediction program and the way surveillance data can be used by more than just the NSA.

Guests:

Jim Adler

Produced by:

Kristen Meinzer

Editors:

T.J. Raphael

Comments [5]

At one point the host proffered gun ownership as a predictive factor for crime. The guest neither introduced gun ownership nor commented on it. Nevertheless, I believe the host's comment left listeners with exactly the wrong impression.
The facts quickly show a strong INVERSE relationship. With +/- 300 million guns in the US, recent crime data show about 11,000 murders with firearms each year. Only 1 of every 27,200 guns is used in a murder each year, leaving a 99.996% annual chance that a gun will NOT be used in a murder. When adjusted to number of gun owners rather than number of guns, an overwhelming 9,999 out of 10,000 gun owners do not murder anyone in a year.
Offenders used firearms to commit 8% of violent crime incidents in 2009. http://www.bjs.gov/index.cfm?ty=tp&tid=43 making one 11 times more likely to be victimized by a criminal who presumably does not own a gun than by a criminal "gun owner".
Conclusion: Gun ownership is extremely correlated with non-violent behavior and highly predictive of a continued ABSENCE of felonies.

Aug. 27 2013 11:08 AM
Page Schorer from El Cerrito, CA 94530

What is actually being predicted here? In the real world it can not be actual criminal activity because we just don't know. I suspect it is being con viced of a felony. I also suspect that this differs radically from actual criminal activity. Most of those convicted of a crime plead guilty to a “lesser charge”. Why? At the individual level it makes sense to plea to a “lesser charge” rather than risk trial and conviction of what prosecutors load the victim up with in order to get those plea bargains. I don't know but strongly suspect that many innocent people plead guilty because they are so powerless relative to the power of the state.

So what we would actually being predicting in the likelihood of being convicted of a felony not the likelihood of actually committing a felony. I have no doubt that Bayesian statistics and big data are totally capable of this kind of prediction, but it would only lead to increased circularity in our already racist criminal justice system. It is not a system, it is not justice but it is criminal.

Aug. 26 2013 04:39 PM
sbechtel@zoomtown.com from Cincinnati

In an odd strategy, we used similar though small scale technique to scale down the number of guests we could invite and accomodate for our wedding. There were people who absolutely could not come, such as the uncle in Alaska. We applied a zero probability to that guy, but the co-worker with 4 kids, we applied a positive of 6 to. It worked really well with my engineer husband and his spreadsheet. We invited a number of people but predicted the precise number who would actually show up.

Aug. 26 2013 03:58 PM
RT from Santa Clara

The tools described are accurately called formulae. They employ mathematical expressions

I guess that these days, journalists just like to say and write the word "algorithm".

Just as they like to say and write "amygdala".

Aug. 26 2013 03:55 PM
Ed from Larchmont

Statistics are still digital, finite, and life is analog, infinite, so it's only possibly a prediction, and sometimes not a very good one.

Aug. 26 2013 09:47 AM

Leave a Comment

Register for your own account so you can vote on comments, save your favorites, and more. Learn more.
Please stay on topic, be civil, and be brief.
Email addresses are never displayed, but they are required to confirm your comments. Names are displayed with all comments. We reserve the right to edit any comments posted on this site. Please read the Comment Guidelines before posting. By leaving a comment, you agree to New York Public Radio's Privacy Policy and Terms Of Use.