Monday, July 9, 2012

Is Predictive Policing A Crime Analyst In A Box?

There has been quite a lot of interest lately in technology that's been called "Predictive Policing". In fact, I've covered quite a few stories about this here on The Crime Analyst's Blog. Just last week there were these stories on the topic:
At least one of those stories is an AP News piece so it's probably popped up in quite a few other publications that subscribe to AP stories.

These stories sparked a discussion over at the International Association of Crime Analysts mailing list.This list is a members only list that has quite a number of participants from academia to practitioners and all ranges in between. On thread on the topic I thought was interesting centered around the discussion of whether this type of technology could replace a crime analyst either now or in the future.

Years ago, in what sometimes seems like another life, I was a hard charging street cop in a mid sized agency of around 200 officers. I worked hard and got on our department's SWAT team and was a tactical officer for about seven years. Our agency's team regularly participated in the Texas Tactical Police Officers Association. We'd usually send several officers to the TTPOA conference every year. Like any good conference, there was always a vendor expo with vendors hawking all kinds of kit they though every tactical operator needed.

Many vendors had great gear. However, there were some vendors that weren't so great. In fact, it became kind of an inside joke among operators that for these less stellar vendors, they would take an ordinary product, paint it black, stick some velcro on it, and call it "tactical".

Often times vendors who sell things to government entities will hear a buzzword like "predictive policing" and then decide to jump on the buzzword bandwagon. In the field of crime analysis, the buzzword du jour seems to be "predictive policing". As news stories about predictive policing proliferate, we'll likely see an increase in vendors hawking their latest product that does predictive policing.

The problem with this is that predictive policing is a definition that's a broad as a barn door. Is predictive policing only valid if it's something like this: Self-Exciting Point Process Modeling of Crime? Or is predictive policing as simple as saying the guy that has robbed five pizza joints in your sleepy little burg almost always does it on Wednesdays and Thursdays between 9PM and midnight so I'll make a prediction that "I think he's going to hit again on Wednesday or Thursday from 9PM to midnight"?

Even more problematic is the idea that someone could get the idea that you can replace crime analysis with the latest whiz bang software. You might make your analyst's life easier with this kind of tool but I doubt that you're going to be able to replace an analyst with software, especially since the definition of what makes up predictive policing is not settled.

Just because you paint something black and stick velcro on it doesn't make it tactical. Neither does sticking a "predictive policing" label on the box make a crime analysis tool worthwhile.This is an emerging technology and I think it's going to be a while before it's a "crime analyst in a box".


  1. Reminds me of the Star Trek (TOS) episode "The Ultimate Computer." Let's just hope that PredPol doesn't start raining down missles in these boxes in order to destroy the criminals before they can commmit crime!

    We actually have been in contact with PredPol about their product, and it sounds promising for more efficiently deploying resources. It's important to keep in mind two things. First, this is a for-profit company. They are selling something that they want us to buy. Second, even if it comes close to working as advertised, it's an algorithim that can't read a report, can't analyze MO, can't talk to police officers, can't produce crime bulletins, and finally, can't think. Until it becomes sentient and proclaims "Cogito ergo sum," we'll still need the human crime analyst.

    BTW. PrePol's literature quotes some crime analyst by the name of Scott Dickson.

  2. Tim, Thanks for the comment. I am not saying that predictive policing software doesn't hold promise. I think that it really does. Anything that makes policing more efficient and effective is probably a good thing. I can't speak specifically to that product as I haven't used it.

    I'll have to look at their literature to find out how I was quoted. I get to talk to lots of people about crime analysis. Not because I claim to be an expert but probably because I am pretty visible.

  3. Tim is right -- computer algorithms simply can't do most of what crime analysts do so well. I think at this point any discussion of replacing a human crime analyst with a computer is naive. Humans do have limits though, and it's much easier for our algorithms to pour through years of data and flag patterns than it is for a human.

    Having been a part of the Predpol team since before we decided to start a company using the technology, I can tell you a little bit about how it started. After George Mohler and the others listed on the self exciting point process modeling document you linked to published their paper, the question arose: How well could such a model be used to predict and stop crime before it happens? The LAPD volunteered to try it out, and the answer turned out to be "pretty well".

    At first, Mohler thought the best way to bring the software to others was to just sell the algorithm to another corporation. We eventually decided the most effective thing to do would just be to bring it to market ourselves. It was a great move because the same academics who started the research are now running the company, and we've been able to focus on creating something truly great rather than focusing on ripping off municipal governments around the world. As news of our successes in Los Angeles and other places hit the media, other people started using predictive policing to describe their existing software that just puts pins on the map. That's pure marketing -- as of right now I think we're the only ones doing "real" predictive policing.

    Scott, I don't know how your name showed up in our literature but I'll find out how that happened. My apologies -- obviously we don't want to make you look like you're endorsing a piece of software you've ever used.

    If you're interested though, feel free to contact me, and I can show you the software and answer any questions you might have about it.

  4. Omar, Thanks for your comments and your offer. I have spoken generally about the potential benefits of predictive policing and crime analysis. Based on what articles I have seen, I do see the potential for good to come of it. Anything that makes law enforcement more effective / efficient at crime suppression is a good thing. If you don't mind, send me an email with just what I was quoted as saying. It may not need a retraction if I was speaking in generalities and not specifically about PredPol.

    That being said I would like to learn more about what you guys are doing, how it works, etc.


I reserve the right to remove defamatory, libelous, inappropriate or otherwise stupid comments. If you are a spammer or are link baiting in the comments, a pox be upon you. The same goes for people trying to sell stuff. Your comment will be deleted without mercy.