A research group from the University of Chicago has developed a machine learning tool that can predict crime up to a week before it happens with 90 percent accuracy. The tool has limitations tied to biases in the programmers, as well as legal ramifications depending on how it used.