A risk model lets Philadelphia supervise low-risk probationers lightly

In a 2007-08 Philadelphia trial rated Promising by CrimeSolutions, people whom a random forests model forecast as low risk were supervised in caseloads of about 400 with no rise in reoffending.

  • 1 an organization

    Two probation officers each supervised about 400 people forecast as low risk with about 45 percent fewer contacts and no measured rise in reoffending.

  • 1 a person

    People in the low-intensity caseloads reported in person once every six months, with no measured rise in rearrest.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

An independent assessment report supports the benefit.

Compared with

Standard regional supervision of people forecast as low risk. Measured.

The system’s part
Not isolated. The trial randomized supervision levels among people whom the model had already forecast as low risk. It did not compare the model's selections with another way of choosing whom to supervise lightly.
Limits
Not yet checked.
Costs and harms
Not yet checked.
Later
Not yet checked.

Download data Cite Suggest a correction

Benefit type
Improved Organizational Efficiency
System
Random forests risk forecasting model (modified from Berk et al. 2009)
Made by
Richard Berk and colleagues (Berk et al. 2009), with the Philadelphia Adult Probation and Parole Department
Used by
Philadelphia Adult Probation and Parole Department, First Judicial District of Pennsylvania
Where
Philadelphia, Pennsylvania, United States (western and northeastern regions)
Stage
In a pilot
Date
Kind
Aggregate result
Broad benefit
Judging the Qualities of Objects, Services, or People, Organizing, Planning, and Prioritizing Work
Collections
Scrutiny and findings
Last revised
31 October 2026

Sources

  1. Assessment report, by authors with no tie to the system.

    CrimeSolutions, National Institute of Justice, U.S. Department of Justice. Program Profile: Philadelphia (Pa.) Low-Intensity Community Supervision Experiment. Rated Promising. Posted 15 June 2011.

    The program is rated Promising. There were no statistically significant effects between groups on rearrest, time to rearrest, reincarceration, frequency of offending, or seriousness of offending.

Details

AI system

Workflow

The model forecast risk from each person's prior record and other data at the start of supervision. Cases were eligible only when more than half of the model's votes forecast no serious offense.

Follow-up

Conditions for replication

People return to standard supervision after a new arrest or six months without contact. Recent two-year outcomes of people under the department's supervision are available to base the forecasts on.

Entry details

Sector

Criminal justice (probation and parole).

Purpose

Richard Berk and colleagues built a random forests model to forecast homicide or attempted homicide among people on probation and parole. A modified version forecast which people would not be charged with a serious crime within two years.

Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). A risk model lets Philadelphia supervise low-risk probationers lightly (AIBD-0726). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0726.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions