Slashdot Mirror
Law Enforcement Wants To Try 'Predictive Policing'
Harperdog with this excerpt from a story about using statistics to fight crime:
"It’s great when cops catch criminals after they've done their dirty work. But what if police could stop a crime before it was even committed? Though that may sound like a fantasy straight from a Philip K. Dick novel, it's a goal police departments from Los Angeles to Memphis are actively pursuing with help from the Department of Justice and a handful of cutting-edge academics. It's called 'predictive policing.' The idea: Although no one can foresee individual crimes, it is possible to forecast patterns of where and when homes are likely to be burgled or cars stolen by analyzing truckloads of past crime reports and other data with sophisticated computer algorithms. 'We know where crime has occurred in the last month, but that doesn't mean it'll be there next month,' Los Angeles Police Department Lt. Sean Malinowski says. 'The only way for us to continue to have crime reduction is to start anticipating where crime is going to occur.'"
Are they spending a lot of money for a fancy computer system that will tell them to watch out for crime in the crime ridden part of town?
I read the internet for the articles.
Arresting someone before a crime is committed is a bad idea. Arresting someone in the process of committing a crime is also okay. What they are talking about here, it seems, is predicting crime like predicting the weather and manning the areas most likely to have precipitation.
Alternatively, if you live in a bad neighborhood, just keep a bunch of donuts on-hand. They can smell it!
first futurama and now slashdot. I think my daily dose of minority report has been satisfied
Policing the Dunkin' Donuts isn't going to prevent many crimes. Policing areas where crimes occur will prevent crimes, or at least force the criminals to expend energy going elsewhere. This is called "the police being smart and doing their jobs" and it's nothing like Minority Report.
*ALERT* Crimewatch 0.1b has detected a possible future equestrian molestation. Units have been dispatched and are authorized to use deadly force.
I'll be honest, we're throwing science against the wall to see what sticks. -Cave Johnson
The police seem to have no problem analyzing data to figure out the best places and times for speed traps. It's about time they used the same principles to stop real crimes.
Support Right To Repair Legislation.
My Criminal Justice teacher always taught this. The example that I remember from him was unmarked patrol cars.
When he was a captain in the local Sherrif Department he fought against using unmarked cars for patrol. His reasoning was that a visible patrol car detered criminal and traffic violations wherever it went. It also let the general public know that the police were in the area and there for you. And in case of an emergency a member of the public could quickly recognize a police vehicle to flag it down.
The only upside of the unmarked cars was that you could collect more ticket revenue easily. But ticket revenue was not the purpose of the department, so why should they give up ground in crime prevention for marginal gains in catching offenders unawares.
It boils down to the question, is it better to prevent a crime or catch the criminal after the fact?
Somebody mod parent up. One of the best ways to reduce crime is to reduce the inequalities between the very rich and the very poor. Look at the crime rates in countries where this gap is lower. Another way is education. So, if you want to fight crime, invest in police training, urban tanks, SWATs, fancy pre-cogn algorithms, etc. If you want to prevent crime, invest in raising the quality of life of the poor and in teachers.
2019 is going to be the year of Linux on the desktop.
Sounds like the new term for "Racial Profiling"...
So what's wrong with racial profiling if it accurately (e.g. passes statistical tests) for predicting crime rates in certain areas?
If a bunch of white males in suits drive into a neighborhood where that racial profile is uncommon, and the mortgage fraud rate goes up by a statistically significant amount, shouldn't that type of profiled activity cause increased fraud investigation in that area?
You want to look for factors that can forecast a certain type of event or events before they occur. If you find the right ones you can take action to prevent undesirable outcomes.
The problem is proving that it works. I used to do simulation of manufacturing systems for my day job about a decade ago. The problem with it was that if you build a good model which avoided a cost, only rarely could you actually prove that the money spent on the model was worthwhile. After all, if you never incur a cost (or a crime), how do you know what the ROI on the analytic model was? Very difficult to prove most of the time since you can't prove a negative. An organization like the FBI or maybe the NYPD *might* be able to justify it but most police organizations simply would not find the ROI to be acceptable.
That's not to say simulation modeling is a bad idea. It does work and can be very powerful. But it is VERY easy to misapply it even if the analytic models are correct and validated. It also tends to be extremely expensive hire the analysts and buy the software so you have to be sure the problem is of sufficient scale to justify the expense. Then of course there is the problem of actually building the model. There is a truism that "all models are wrong - some models are useful". Getting a useful model is not always an easy thing to do. A bad (very wrong) model can sometimes be worse than no model at all.
I generally tell people that if they can solve a problem without a complicated computer simulation, they should. Most uses I've seen for simulation are somewhat like duck hunting with a howitzer. For all but the most complicated and intractable problems with lots of variables and high risk of a negative outcome there is a strong chance that there are much simpler and cheaper solutions available.
This is a joke right?
This is not minority report type stuff. This stuff is more like: data shows an increase in vandalism in the vicinity of the sports stadium after a championship game. OK, most people get that because the relationship is somewhat easy to grasp. However with data mining much more subtle trends in human behavior can be discovered. This sort of stuff has been done in the past with respect to consumer behavior. For example Wal Mart discovered that when news in the gulf region warned of a possible hurricane there was a spike in the sales of pop tarts. So when the news mentions a possible hurricane Wal Mart immediately relocates pop tarts from the mid west to the gulf region before there is any apparent demand.
What will most likely occur is that data mining of law enforcement records will be used to schedule and position officers in different areas depending on various inputs: season, weather, temperature, community events, sports events, etc.
As long as the various police forces (and the legislatures that drive them) continue to misdirect a large proportion of their efforts, I'm not inclined to pay serious attention to any theory they might come up with about why and/or how crime can be reduced. So far, they seem to be quite focused on proving they don't know how to do their jobs worth a damn.
I've fallen off your lawn, and I can't get up.