Artificial intelligence is already becoming part of the machinery of government. Used well, it could help public agencies work faster, reduce backlogs, and improve services. Used poorly, it could quietly shift real decision-making power away from people and into systems the public cannot properly see, question, or hold to account.

In Algorithmic Government: Of the people, for the people, by AI?, Senior Fellow Dr Paul Henderson and Researcher Thomas Scrimgeour examine how governments are using artificial intelligence, what New Zealand can learn from comparable countries, and how public-sector AI should be governed.

The paper argues that New Zealand should pursue the benefits of AI for public-sector capability—but only with clear limits. Productivity matters, but it is not the only measure of success. AI use in government must remain bounded by public accountability, constitutional norms, and human judgement.

The risk of post-democratic drift

AI in government is often discussed as a technical question. This paper shows why it is also a political and constitutional one.

If AI systems are used only to support public servants, the risks may be manageable. But as systems begin to triage, prioritise, recommend, and shape decisions at scale, the line between “government with algorithms” and “government by algorithms” becomes harder to maintain.

The question is not whether AI can make government more efficient; it’s whether the public can still know who made a decision, why it was made, and who is responsible if something goes wrong.

What the paper argues

The report analyses public-sector AI through three frames:

• DEI-oriented governance: focused on redress, participation, and fairness
• Competitive national utility: focused on productivity, responsiveness, and state capability
• Conservative governance: focused on ordered liberty, institutional continuity, and rule-of-law accountability

Its central argument is that New Zealand should pursue a strategy of competitive national utility disciplined by conservative governance. In practice, this means using AI where it can improve internal operations and service delivery, while placing strong guardrails around rights-affecting decisions, public accountability, procurement, data use, and constitutional norms.

Key findings

The paper finds that the near-term value of AI lies mainly in internal operations and service delivery—not autonomous policymaking or governing.

It also warns that Western governments are moving quickly from isolated AI experiments to system-level AI infrastructure. That shift creates real opportunities for New Zealand, but also serious risks if public agencies become dependent on opaque, vendor-controlled systems.

Without clear governance, AI could create strategic fragility, weaken public accountability, and make it harder for people to understand or challenge decisions that affect them.

Recommendations for New Zealand

The paper recommends practical steps to keep AI useful, accountable, and bounded:

• Create a public-sector AI register so New Zealanders can see where AI is being used, what it is used for, how much autonomy it has, and what accountability structures govern it.
• Use cognitive forcing functions in important decision systems, requiring public servants to record their own judgement before seeing an AI recommendation.
• Treat procurement as governance by requiring model portability, inspection rights, and safeguards against vendor lock-in.
• Build workforce capability that goes beyond tool training and supports critical vigilance, independent judgement, and meaningful human oversight.