What would it take to stop a crime before it occurred? It's a challenge Richard Berk, Professor of Criminology and Statistics, is tackling with software he's designing to help increase public safety and squeeze more crime prevention from the constrained budgets of criminal justice agencies.
"Common sense says if you want to prevent crime you need to anticipate it. A range of criminal justice officials routinely consider 'future dangerousness' when they make their decisions about charging, sentencing, prison security, release on parole or probation, and supervision during probation or parole. The idea isn't new; crime forecasting has been around since at least the 1920s. But thanks to very recent developments in statistics and computer science, our ability to accurately anticipate future crime has improved dramatically."
Parole boards, a key example of both past and present forecasting, use a variety of background information to make case-by-case judgments about whom to release from prison. One of the most important factors is whether an individual poses a significant threat. With the power of modern computers it's now possible to develop forecasting algorithms from information on tens of thousands of cases, in which hundreds of potential risk factors can be considered simultaneously. "These algorithms will generally make far more accurate forecasts of risk than projections relying on human judgment, even when those judgments are informed by insights from criminology or psychology."
"Common sense says if you want to prevent crime you need to anticipate it. A range of criminal justice officials routinely consider 'future dangerousness' when they make their decisions about charging, sentencing, prison security, release on parole or probation, and supervision during probation or parole." – Richard Berk
Berk harnessed this form of data mining to create the software that is being used in several jurisdictions so that criminal justice resources can be allocated based on priority. The Philadelphia Adult Department of Probation and Parole uses his software to determine the kinds and intensity of supervision provided to offenders. The state of Maryland uses a different version to forecast whom among those on probation or parole is most likely to be the perpetrator or victim of a homicide. These individuals can then be provided with special services and more frequent contacts with supervising officers. Professor Berk is also working with the Pennsylvania Board of Probation and Parole, Philadelphia's District Attorney's Office, and the California Department of Corrections. "The details of the forecasting procedures will vary setting to setting, but the same basic approach is applied: The computer uses a technique often referred to as 'machine learning' to gather valuable statistics from very large databases."
With colleagues in Penn's School of Law, Professor Berk is turning his forecasting procedures to "crime in the suites." Software will track the probability of health and safety violations in the workplace in an attempt to guide the oversight activities of the Occupational Safety & Health Administration. It will then provide a recommendation as to which employers or facilities require inspection.
Berk's software is mindful of privacy, for example, concerns about profiling by race or gender. Neither the race of individuals nor their ethnicity is included in his forecasting algorithms. "The way the forecasts are made not only explicitly precludes the use of race or ethnicity, but insofar as the forecasts determine key decisions, it can prevent the use of illegitimate factors by others, such as law enforcement. Gender is a different matter. We use gender as a predictive factor because of the well-known fact that most violent crimes are committed by men. In this instance, public safety trumps everything else."
The inevitable question remains: How can you punish someone for a crime they haven't committed? "All of the criminal justice forecasting I do is thoroughly vetted by attorneys representing a range of interests, and concerns about 'future dangerousness' seem to have been accepted as legitimate for many kinds of criminal justice decisions."
"This isn't Minority Report," Berk jokes, referring to the Philip K. Dick short story and Tom Cruise film by the same name, in which oracles predict crimes before they happen. "It's not a perfect system—there will always be errors when you're trying to forecast crime. But modern forecasting and its historical counterparts have long been a part of criminal justice decision-making. The intent is to increase accuracy even further, while maintaining a high standard of transparency."
