Sentencing officials are trying — again — to build an algorithm to predict future crimes. Critics still have big concerns

It’s been nearly 10 years since Pennsylvania lawmakers ordered a state commission to develop an algorithm that would help judges dole out fairer sentences and, eventually, reduce prison populations.

The Pennsylvania Commission on Sentencing, which is fulfilling the mandate, hopes its newest iteration is finally close to the finish line — even though it remains deeply controversial among criminal defense attorneys and civil liberties groups.

“I feel the pressure, and I’m concerned that we haven’t done this yet,” Rep. Todd Stephens, R-Montgomery, the commission’s vice-chairman, said Wednesday. “I do feel that we need to put this to bed and move on to other issues that we’ve got.”

The commission published its latest draft of the sentencing tool in June. It was the subject of three public hearings across the state this week, and will be on the Sentencing Commission’s agenda when it convenes in State College in early September.

The proposed assessment weighs a handful of data points, including a defendant’s age, gender, and criminal history, to predict their likelihood for committing new crimes.

Based on those calculations, the algorithm will direct county court employees to conduct a more thorough review of certain defendants prior to sentencing.

Stephens, a former prosecutor, said the tool will offer judges a more comprehensive picture of the defendants who appear in their courtrooms. That will help them determine which ones to direct into specialized courts or diversionary programs, he said.

But such critics as Dean Beer, Montgomery County’s chief public defender, say its reliance on historical crime data will penalize poor people, people of color, and those who live in high-crime neighborhoods.

“I agree that judge should have as much information on an individual as possible during sentencing,” Beer told the commission in Harrisburg on Wednesday, when he was the only participant at a public hearing in the Pennsylvania Judicial Center. “But I disagree that we should box people into groups based on historical perspectives. That’s the general concern.”

‘A moving target’

Some experts estimate that as many as 20 states across the country have started incorporating risk assessment in criminal sentencing. 

Stephens and Mark Bergstrom, the commission’s executive director, say the assessments are maligned because they aren’t well understood by the public.

The tool under development in Pennsylvania, they said, won’t be used to add more time to a prison sentence.

Instead, it will identify candidates for alternative sentencing programs such as State Intermediate Punishment, a diversionary program that allows some defendants to complete part of their jail sentence under community supervision.

The Sentencing Commission has published five draft sentencing tools in the past four years. The latest one calls on county parole and probation boards to conduct “risk, need, and responsivity” reports as part of a defendant’s risk assessment during sentencing.

County probation and parole officers already conduct these reports to determine probation and parole sentences.

They consider a slew of factors — including a person’s criminal history, living situation, hobbies, and friends and family — that may influence their recidivism risk when they’re living under community supervision, according to the company that develops a widely used assessment. 

Bergstrom said these assessments help county parole and probation boards allocate their finite resources, since they can’t afford the same level of supervision to every probationer and parolee in their jurisdiction.

But Nyssa Taylor, legal counsel for the ACLU of Pennsylvania, told the commission at a Philadelphia hearing on Tuesday that the county-based assessments weren’t designed to be used in sentencing. And since countries use different assessment methods, she said, they may generate disparate sentencing outcomes across the state.

In her prepared testimony, Taylor also argued that the county assessments reflect human biases, making them an unreliable foundation for Pennsylvania’s new sentencing tool. 

Beer leveled an additional charge on Wednesday when he said that the proposed assessments lack transparency. 

He said the only information judges and attorneys receive from the proposed algorithm is a directive to conduct a “risk, needs, and responsivity” report. They won’t know which factors contributed to that recommendation, he said, or whether or not the defendant was classified as low or high risk.

“You’re asking me to place trust in something that nobody wants to be open about,” Beer said. “I can’t risk that for my clients.”

Bergstrom and Stephens said that’s by design. 

Critics rejected previous versions of the assessment tool on the grounds that it would reduce defendants to “high-risk” and “low-risk” labels. The current algorithm would flag defendants of both classifications, they said, and direct judges to learn more about them before sentencing. 

“In the interest of fairness and minimizing stigma on the defendants, you have to withhold some information,” Bergstrom said. “That’s a response to the criticism that labeling people [“high-risk”] would stay with them for life.

The change was one of the ways that the commission has responded to public feedback as it tried to fulfill a seemingly impossible task, Stephens said. 

“It’s been a moving target,” Stephens said. “I don’t think anything we do satisfies everyone… but I think it’s a disservice to everyone involved with the criminal justice system, including the defendant, for us not to do a risk assessment. We should focus our resources on those [who] need it most.”

A previous version of this story incorrectly referred to Rep. Todd Stephens as the chairman of the Commission on Sentencing. He is the vice-chairman.