Confronting Label Indeterminacy in Automated Bail Decisions DOI: XXXXXXX.XXXXXXX Conference: 21st International Conference on Artificial Intelligence and Law; June 08–12, 2026; Singapore ISBN: 978-1-4503-XXXX-X/2018/06 TBD

Cor Steging, Tadeusz Zbiegień · 2026

Read Confronting Label Indeterminacy in Automated Bail Decisions DOI: XXXXXXX.XXXXXXX Conference: 21st International Conference on Artificial Intelligence and Law; June 08–12, 2026; Singapore ISBN: 978-1-4503-XXXX-X/2018/06 TBD by Cor Steging, Tadeusz Zbiegień (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-04073.txt.

Description

Cor Steging and Tadeusz Zbiegień confront a structural problem in using machine learning to support bail decisions: label indeterminacy. When bail is denied, whether the defendant would have appeared in court is never observed. So historical bail data only partially reveal the outcomes that future models are trained to predict. Building systems on such data risks bias and self-reinforcing feedback loops.

In a case study of bail decisions from Pennsylvania's Unified Judicial System, the authors evaluate five approaches to handling indeterminacy. They test them across three machine learning models. One approach is a novel label imputation method motivated by bail dynamics. Every method rests on unverifiable assumptions. Yet the choice among them influenced predictive behavior sometimes more than the choice of model. Explainable AI analysis showed the effects reach the models' internal decision processes.

Catalog record

CREATORCor Steging, Tadeusz Zbiegień
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.04073

Browse the full library · About the Tower of Babel