Year: 2026
Type: article
Abstract: Traditional correctional risk‑assessment tools depend on static indicators alongside subjective human judgment, and they often fail to reflect real‑world conditions in both prison and community correction contexts. This work presents a different approach. Building on recent progress in large language models, we outline an alternative paradig centered on four capacities missing from existing tools: contextual understanding, causal reasoning, unstructured data processing, and dynamic learning. We further map these capacities to a practical system design, including three principles, a four‑layer architecture, and a closed‑loop workflow covering both prison and community correction. The resulting framework can serve as both a foundation for further research and a practical reference for system developers.
DOI: 10.54097/mzdke887
Author: Yintao Wang
Institution: Zhejiang Police Vocational Academy
Language: English
Subfield: Law
Field: Social Sciences
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W7236853868
