The DOI matches its Crossref record.
Reprogramming Recidivism: The First Step Act and Algorithmic Prediction of Risk
Abstract
The First Step Act, a seemingly miraculous bipartisan criminal justice reform bill, was signed into law in late 2018. The Act directed the Attorney General to develop a risk and needs assessment tool that would effectively determine who would be eligible for early release based on an algorithmic prediction of recidivism. The resulting tool—PATTERN—was released in the summer of 2019 and quickly updated in January of 2020. It was immediately put to use in an unexpected manner, helping to determine who was eligible for early release during the COVID-19 pandemic. It is now the latest in a growing list of algorithmic recidivism prediction tools, tools that first came to mainstream notice with critical reporting about the COMPAS sentencing algorithm. This Article evaluates PATTERN, both in its development as well as its still-evolving implementation. In some ways, the PATTERN algorithm represents tentative steps in the right direction on issues like transparency, public input, and use of dynamic factors. But PATTERN, like many algorithmic decision-making tools, will have a disproportionate impact on Black inmates; it provides fewer opportunities for inmates to reduce their risk score than it claims and is still shrouded in some secrecy due to the government’s decision to dismiss repeated calls to release more information about it. Perhaps most perplexing, it is unclear whether the tool actually advances accuracy with its predictions. This Article concludes that PATTERN is a decent…
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
While a belief exists in the literature that deep learning will be applied to risk assessment in the future, the risks, lack of interpretability, and need for large training data sets currently outweigh the potential benefits.
“While a belief does exist in the literature that deep learning will be applied to risk assessment in the future [22], the risks, lack of interpretability [23], and need for large training data sets [24] currently outweigh the potential benefits.”
From Efficiency vs. Equity: A Structured Interdisciplinary Review of AI in Criminal Justice Risk Assessments (2026) · cites this paper as [22]
The specific biases identified by the authors of the sources are race, gender, age, and economic status.
“The specific biases identified by the authors of our sources are race [14,22,33,46], gender [6,14,47], age [48,49], and economic status [17,[50][51][52].”
From Efficiency vs. Equity: A Structured Interdisciplinary Review of AI in Criminal Justice Risk Assessments (2026) · cites this paper as [22]
ARA tools have existed at least since the 1970s, and many of the early tools were simple actuarial tools rather than AI or ML.
“While ARA tools have existed at least since the 1970s [22], many of the early tools were simple actuarial tools rather than AI or ML.”
From Efficiency vs. Equity: A Structured Interdisciplinary Review of AI in Criminal Justice Risk Assessments (2026) · cites this paper as [22]
Citations
Cites
No references could be resolved yet.
Research works printed but not linked: 9
- See supra notes 157, 160 and 161 and accompanying text
- Updated See, at 13; see also supra note 96. 253 "In crafting the System, the Department worked to make the benefits of the FSA as widely available as possible without compromising predictive reliability, Attorney General). 254 UPDATED REPORT, 55
- Oversight Hearing, supra note 32 (letter attached to statement of David E. Patton, Executive Director
- Id, 1765, Patton goes on to note that "[s]electing the right tradeoff between these sometimes competing goals are more rightly within the power of policymakers and stakeholders
- Mayson, 2017, An algorithm can be designed to achieve any one of the above metrics of output equality, but not all of them together. That is, an algorithm cannot be designed to achieve 'total fairness.'"); see also Richard Berk et al., Fairness in Criminal Justice Risk Assessments: The State of the Art 1
- E G See, Oversight Hearing, supra note 32, at 4 (statement of Prof. Melissa Hamilton) ("[T]he AUC has serious limitations and thus cannot present a holistic portrait of a tool's abilities
- Eaglin, at 91; see also Hopkinson, supra note 128, at 742 ("[I]n statistics, there is no consensus as to which AUC scores represent small, moderate, or large effect sizes
- SETON HALL LAW REVIEW, 51, 10.5840/chesterton2015411/223 · DOI
- SETON HALL LAW REVIEW, 51, 10.5840/chesterton2015411/223 · DOI
Cited by
Sorted by citations
Atkinson, G., & Casagrande, K. (2026) · Information · doi:10.3390/info17060574
Metadata
- Authors
- Cyphert, A.
- Venue
- SSRN Electronic Journal
- Publisher
- Elsevier BV
- Published
- 2020
- ISSN
- 1556-5068
- Cited by
- 0
- References
- 0
- Language
- en
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
