New Zealand Police Use Two AI Translation Models to Reduce Errors

New Zealand Police have adopted a dual-model approach to AI translation, using two separate models to cross-check each other’s output. The goal is to catch and reduce errors that a single model might miss, particularly in public-facing communications where accuracy carries serious weight.

This practice points to a broader principle for high-stakes public services: relying on model diversity can strengthen reliability. When two independent systems process the same text, discrepancies can flag potential mistakes before the translation reaches the public. The method does not eliminate the need for human oversight, however. Human review and clear audit trails remain essential parts of the workflow, ensuring accountability and catching nuances that automated checks may overlook.