Many people will be familiar with Agent Based Modelling (ABM), even if they don't recognise the term, as it's the principle underlying games like The SIMs™ and SIMCity™, along with most computer games where you battle against the machine. Prior to ABM, most computer representations of people ('models') were done by lumping them together and treating them with mathematical equations or statistics. ABM allows computers to represent individuals as individuals.
The big advantage with ABMs is that you can build more realistic behaviour for a criminal or victim into the model. This allows us to make a model that has computer-generated criminals in it, in which the criminals interact with their environment and search out opportunities to commit crimes. With enough detail, such models can become predictive, allowing us to say, in general terms, where crimes will be committed.
However, these models have their issues: they aren't accurate enough to predict specific crimes at specific times, and there is considerable potential for decisions based on them to be problematic if real data about real individuals is used in the system to attempt to predict their specific behaviour – not least because such systems are unlikely to ever be accurate enough to say anything about individuals (even if we wanted to!). As a group, we look at how we can make these systems more useful while also warning of the potential misuses.
For more information, see Nick Malleson's blog, and the presentations on Andy Evans' webpage.