The NZAIM unifies technical guidance, while supporting frameworks can be re-engineered

As this paper argues, the existing legislative framework already offers the foundation to enforce best practices where necessary. The focus of the recommendations is on re-engineering supporting frameworks and guidance. Figure 5 illustrates a proposed, more streamlined guidance ecosystem. Normative commitments around the trustworthy and considerate use of any information, especially system-specific commitments in the Algorithm Charter and Ngā Tikanga Paihere, as well as partnership-specific commitments promoted by Te Kāhui Raraunga, can be refactored around the existing Data Protection and Use Policy (DPUP). A single, enhanced, and mandated DPUP+ will serve both as a blueprint for agencies to develop a high-performing, trustworthy, and respectful data culture, and as a benchmark to independently evaluate an agency’s adherence to best practice.

Operational considerations related to the development and end use of AI systems can be unified within a New Zealand Artificial Intelligence Manual and streamlined Responsible AI Guidance sections. Both should be aligned with the principles of the Public Service AI Framework to clearly demonstrate how this strategy is operationalised. An NZAIM will incorporate (rather than simply reference) existing obligations and guidance on accessibility, security, cloud risks, record management, and privacy.

Figure 5: Proposed state for a streamlined guidance ecosystem, refining the roles of the GCDS and GCDO into single frameworks at each abstraction layer: DPUP+ and PSAIF respectively.

This streamlined ecosystem clarifies the overlapping roles of the GCDO and the GCDS by defining – but functionally linking – the ethical “why” and the operational “how”. Abstracting the “why” at the most fundamental level provides a lasting foundation for any “how”, both now and in the future.