Keeping track of all the evidence needed to bring down a dangerous criminal requires a mind like a steel trap— or perhaps like a silicon chip. Lars Kangas and his colleagues at Pacific Northwest National Laboratory in Richland, Washington, have created a computer program that searches through huge police databases to spot correlations human detectives might take years to find.
Computer Aided Tracking and Characterization of Homicides, or CATCH, uses an artificial neural network to sort crime files into a giant grid. The more details two cases have in common, the closer the program places them. A police officer can study the resulting clusters and ask the program to highlight common factors— the type of murder weapon, for instance. Or he can ask CATCH to display the crimes on a map to determine whether they were committed near each other. The program can even generate rough evaluations of personality and behavior. "You look at solved crimes that were committed in a similar way and see what you know about these offenders," Kangas says. "Did they drive old cars or new ones? What was their level of education? Were they neat or sloppy people?"
Washington State police are now evaluating the system. In a test run, the program correctly identified the similarities in a high percentage of serial murders known to have been committed by the same offender. And CATCH has already led to one notable but grisly success. It suggested that two heaps of body parts, recorded as separate murders, might literally be connected. Comparing DNA tests proved they were in fact pieces of the same victim.