Responsible Use of AI in Electoral Roll Verification

Abstract

Electoral rolls constitute a foundational administrative component of democratic elections because erroneous registration, duplication, outdated information, or wrongful exclusion can affect both electoral integrity and citizens' ability to exercise voting rights. Artificial intelligence offers potentially valuable tools for identifying duplicate registrations, reconciling inconsistent records, detecting anomalous data patterns, normalizing multilingual names, and prioritizing cases for administrative review. However, automated electoral-roll verification also creates significant risks involving false matches, algorithmic discrimination, opaque decision-making, privacy intrusion, cybersecurity, and inappropriate delegation of legally consequential decisions to computational systems.

This study develops a rights-preserving governance framework for AI-assisted electoral-roll verification and evaluates its implications through a reproducible synthetic audit of 2,400 electoral-record cases. Four record conditions were modeled: standardized digital records, transliteration and name variations, recent address changes, and OCR-degraded legacy records. Duplicate-detection recall ranged from 85.6% to 98.3%, while potential false-positive rates increased substantially as record quality deteriorated. A simulated governance intervention requiring authorized human review reduced the aggregate potential erroneous-exclusion rate by approximately 86%. These results demonstrate that strong average model performance can conceal condition-specific risks capable of disproportionately affecting voters whose records contain transliteration differences, mobility-related inconsistencies, or legacy-data errors. 

The proposed Responsible Electoral Roll AI Governance Framework therefore restricts AI to evidence generation and case prioritization rather than autonomous registration cancellation. Responsible deployment requires lawful purpose limitation, data-quality assurance, independent validation, subgroup auditing, explainability, cybersecurity, voter notification, correction opportunities, human decision authority, appeal mechanisms, and continuous public accountability.