"Now we can forcast criminal activity just like the weather." http://www.wired.com/wired/archive/11.09/view.html?pg=1 Most of the time, crime seems utterly random: A stray bullet kills an innocent bystander. The apartment next door is burglarized. A high school kid steals a purse. These acts occur with little or no warning. But what if we could predict the unpredictable? What if a crime wave could be forecast with the same confidence as a heat wave? Armed with this information, police departments could deploy extra patrols to hot spots before crime happens, not after someone gets hurt. Sounds like a Philip K. Dick novel, doesn't it? In fact, a crime mapping and forecasting system is already in the alpha-test stage in two American cities. With funding from the Justice Department, computer scientist Andreas Olligschlaeger, criminologist Jacqueline Cohen, and I amassed individual reports from police departments in Pittsburgh and in Rochester, New York. We assembled every electronic record ever keyed in at their police stations, starting with the earliest Cobol entries. After standardizing the data, we plotted the offenses on a digital map of each city. Then we laid the outlines of the precincts and the patrol car routes over the map. Finally, we crunched the numbers, running the data through trend-spotting programs originally designed for big business. Our model turned out to be on target down to the beat level: We were able to predict monthly criminal activity before it happened, with 80 percent accuracy. In the process, we found that the way to predict lawlessness is to identify and track leading indicators. Companies look at consumer spending data; meteorologists keep tabs on barometric pressure. In our case, we studied soft crimes such as mischief, disorderly conduct, and trespassing. An increase often precedes a rise in hard crimes like burglary and assault. If this sounds like a no-brainer, it is. Businesses have been coupling common sense with information technology for years to predict consumer demand. This is just crime "demand" by precinct "sales territory." To a computer, there is no essential difference between shoplifting and shopping. Take, for example, suddenly succumbing to a sales pitch for a miracle ion bracelet or an infomercial for an electric bagel slicer. It may be an impulse buy, but to a major retailer these purchases are anything but random. Billions are spent trying to predict our collective whim and adjust supply levels accordingly. The question now isn't whether a spike will occur - it's how big the spike will be. To that end, companies churn out tens of thousands of forecasts per month, and electric bagel slicers are manufactured and distributed accordingly. So when's the criminologist going to show up on your local newscast to give tomorrow's felony count? It may be a while - there are still a few kinks to work out. For starters, even the latest crime-prediction models are accurate only on a month-to-month basis, not day to day. Plus, they're relatively low-resolution, being correct only within a 100-square-block area. Finally, presaging crime is actually a lot harder than mapping rain showers, because crime forecasters - unlike meteorologists - must account for the effects of their own reports. Think about it: A forecast is made, action is taken, and wrongdoing is averted, along with the very trends and statistical indicators needed to make the predictions in the first place. No one ever does anything about the weather, except complain. Currently, law enforcement is a little bit like a giant game of Whack-A-Mole. Hit one crack house or chop shop, and another pops up to take its place - usually in the high-risk neighborhood nearby. But with an intelligent, self-adapting forecasting model, that won't happen. We'll use the predictions to post patrols in emerging danger zones before the criminals have a chance to relocate. We'll give cops the precog tools they need to spot the bad guys 10 moves in advance. The future isn't another Philip K. Dick dystopia. It's a crime-free city. Don't worry. We're working on it. Won't be long now, until they learn how to "attach" your name and SSN to their Prediction Models. .