Multi-stakeholder transparency evaluation and dynamic accountability mechanisms for AI-assisted criminal sentencing

Review

Li and Zhu treat transparency in AI-assisted criminal sentencing as a stakeholder-specific governance problem rather than a single disclosure obligation. Their framework maps differentiated information needs for judges, defendants, developers, and the public; evaluates transparency across technical, process, outcome, and interaction dimensions using AHP-weighted fuzzy assessment; and connects those measurements to ex ante, in-process, and ex post accountability mechanisms. Empirically, the study combines COMPAS data with a structured survey of 300 respondents. The reported system rankings remain stable in 97.3% of sensitivity trials, while the dynamic mechanism reduces accountability-response latency by 70.5% against a static audit baseline. Structural equation modelling further associates accountability completeness more strongly with stakeholder satisfaction than transparency alone, with accountability mediating much of transparency's observed effect.

The governance contribution is the move from visibility to consequence. It makes explicit that different actors need different evidence and that disclosure has limited institutional value unless someone can act on what is disclosed. That distinction matters in sentencing, where an explanation that cannot trigger review, correction, liability, or remedy may improve observability without improving a defendant's practical position.

The framework nevertheless leaves the decisive allocation of public authority underspecified. Blockchain anchoring can strengthen provenance and tamper evidence, but it does not establish who is legally entitled to determine that an accountability threshold has been breached, compel disclosure, suspend a system, reopen a sentencing decision, or provide remedy. Stakeholder satisfaction is also an incomplete proxy for legitimacy or rights protection in a coercive setting. A system can be understood, accepted, and rapidly audited while still operating under defective substantive rules or inaccessible appeal conditions. The empirical results therefore support the paper's measurement and workflow claims more directly than they establish the adequacy of the proposed governance arrangement across jurisdictions.

For implementation, the framework would become substantially more governable if each accountability phase were bound to named institutional owners, mandatory escalation triggers, preserved evidentiary records, appeal and correction procedures, liability allocation, and measurable redress outcomes. Its durable contribution is not blockchain traceability itself, but the proposition that transparency becomes governance only when evidence is connected to institutions capable of imposing consequences and correcting decisions.

Key Insight

Traceability can preserve evidence about an AI-assisted sentencing decision, but accountability exists only when institutions attach that evidence to authority, contestation, remedy, and enforceable consequence.