Slashdot Mirror
Crime Prediction
pipingguy writes "More than a decade of extensive crime data collection matched with new technology may soon allow police to predict to a surprising degree of accuracy the number and type of crimes that will occur in a given neighborhood one month in advance."
Mr. Smith, you will commit a murder tomorrow morning at 2:34 PM. You will murder your wife. We are taking you into our custody now.
But what about...
Mr. Smith, you will break free from our custody tomorrow and attack a security guard in the process. Thus, we are taking you into our custody now.
I like paradoxes, but not when they're amber.
I like paradoxes, even though my name's not Bamber.
--TheOrangeSquid Is it any wonder things seem so awry? We swim in a sea of confusion and don't have to think to survive
Does this sound to anyone else like the beginning of psychohistory?
It's a neat idea, but it seems like the complex and chaotic nature of the neighborhood would preclude anyone from being able to draw any substantive conclusion. I mean, if we can't get the weather right within 80% more than 12 hours in advance, we should we be able to predict the behavior of humans, even in large groups?
So, if I want to stage a robbery, now I should find out where the least likely spots are for said robbery, demographically (do they have to publicize the specific information they gather? Civil liberties advocates would probably push it)--and then commit crime where they expect it least. Such a system, if acted upon in the manner suggested, would allow an informed criminal (or gang of criminals!) to act with even less resistance than before. That's the major flaw with demographic information, of course; it only gives averages and likely outcomes. But "they" (local police forces who use this information) will have to be careful in how they use it, because an overreliance on statistics means that the outlier criminal could take away someone's life or property with little chance of being caught.
We recently had heard in the office over one of the Yellow Machine that's made by Anthology Solutions.
No, most likely what will happen is they'll get data supporting the idea that neighborhoods consisting primarily of minorities have higher violent crime rates than other neighborhoods. Then they'll be accused of racial profiling. Bad cops (and unfortunately, probably some good cops, few though they may be) will have their lives ruined.
"The evil of the world is made possible by nothing but the sanction you give it." -- Ayn Rand
They aren't going to be able to stop crimes before they happen, because that's impossible (unless you count mitigating the factors that cause crime, like poverty or mental illness). If you stop it before it happens, then it hasn't happened.
The only concern I have about this is that it will heighten the notions of where to send police in the first place. In Minneapolis we have a system called CODEFOR that is used to help police track crime and prioritize resources, but it's based (I believe) on reported crime. That's quite different than using statistical measures like arrest rates to project crime rates and send police into an area based on that. One has to be very careful, since using the results of this to direct limited police resources could influence the numbers that go into the model in the first place (feedback looping).
Luckily, the people the article talks about were using a wide variety of inputs (not just police or crime-related) to what are likely a host of regression analyses.
I do not have a signature
According to the article, they seem to be grasping at straws:
They are apparently matching coincidental data with crime statistics rather than finding causal relationships. By this I mean they can correlate such weird things as percentage of dirt versus grass in yards, average number of health club memberships, and population density to crime rates, but that doesn't mean they have a reliable means to accurately predict crimes.
Further, finding causal relationships in a system as complex as human social interaction is impossible. A simple example is the relationship that poverty causes robberies.
The first flaw is that there are numerous counter examples, both where impoverished individuals do not rob, and where non-poor people do rob (how many pens do you have from work at home, thief?) ... for the causality to be true, there should be a somewhat proportional relationship (how can it be that the richest people *cough*Enron*cough* still rob?)
Second, and more importantly, is that this is a correlation--if you can find a statistic that correlates poverty to higher crime rates, this does not show that one causes the other--is it the crime that causes the poverty or the poverty that causes the crimes? You can find more obtuse correlations ... oh, I don't know, how about that there are more crimes committed where there is street lighting. A causal relationship would say that removing the street lighting would reduce the crime, but that seems kinda silly.
Anyway, I went around the block just to say that a fair predictive system should deal only with causal relationships to ensure long-term accuracy. I'm sure an impartial computer analysis would find that high crime areas also have a lot of police patrols ... coincidence?
--- Jason Olshefsky
Karma: Poser (mostly affected by adding this line long after everyone else did)