The ethical considerations inherent in trying to predict future events — such as criminal offending — are not new. Indeed, as the NIJ-funded researchers who worked on the Philadelphia risk assessment tool point out, one of the reasons some offenders are sentenced to longer prison terms is to prevent crimes that they might commit if they were not incarcerated.
Geoffrey Barnes and Jordan Hyatt, from the University of Pennsylvania, believe that random forest modeling offers a different — and potentially more accurate — approach for building a prediction tool. Nonetheless, they recognize the ethical crux that lies at the heart of building such a tool: deciding which "predictors," or fact variables, are acceptable to use.
In their final report, for example, they ask, "Would it ever be permissible … to include an offender's racial background as a predictor variable in one of these models? If not, what about the use of predictors such as residential location or familial circumstances, which could indirectly communicate the offender's racial identity into the forecasting model?"
Would it be permissible to use controversial predictors in "lower-stakes" forecasting models — to control admission into a treatment program or govern supervision decisions, for example — but prohibit their use in "higher-stakes" decision-making such as sentencing?
Furthermore, some note, aren't the age of criminal-behavior onset, possession of a juvenile record or the neighborhood a person resides in (factors that could be used as prediction variables) all "extrajudicial" factors? As such, should they be considered in an individual criminal justice decision?
Considering potential "collateral consequences" of decision-making based on a forecasting tool is also an important part of the process. As mentioned in the main article, for example, Philadelphia's Adult Probation and Parole Department used the random forest prediction tool to identify offenders who were at a high risk of committing a serious crime in the two years following return to the community — and these people were supervised more closely, under more stringent parole terms and conditions. This could increase the likelihood that technical violations of their parole would be more likely to be detected and punished, including imposing additional custodial sanctions.
There are no easy answers to these questions, but they will have to be addressed head-on as increasingly technologically advanced forecasting methods become available for use in our nation's criminal justice system.
About This Article
This article appeared in NIJ Journal Issue 271, February 2013, as a sidebar to the article Predicting Recidivism Risk: New Tool in Philadelphia Shows Great Promise by Nancy Ritter.