Plain definitions of the terms used in the framework, its Q&A and its brief — written for readers who run parishes, clubs, rūnanga offices and small councils, not data centres. No prior technical knowledge assumed.
Causal-loop diagram (CLD) — A map of how things push on each other. Instead of a list, it draws arrows: more of this leads to more of that, which feeds back and changes the first thing again. The deliberation used one to make sure every team could see how its topic — electricity, jobs, data — connected to everyone else’s.
Deliberation — A structured group argument designed to produce a decision, not just talk. Participants start from evidence, disagree in the open, test each other’s positions, and vote at set points. It differs from consultation, where people are asked for views someone else then weighs in private.
Consent vote — A way of deciding where the question is not “does everyone love this?” but “can anyone not live with this?”. A proposal passes unless someone raises a reasoned objection, which must then be dealt with. It stops one enthusiast railroading a group, and it stops one perfectionist stalling it.
Ratification — The final sign-off, by everyone, of work done by a smaller group. In this deliberation each team’s position went to the full membership with a plain-language briefing and an anonymous vote. One position failed ratification, went back to its team, was rebuilt, and passed the second time.
Jurisdiction of inference — A question to ask before buying any AI system: where does the AI actually do its thinking, and whose courts and laws govern that place? A tool can have a local office and a local price in dollars while every answer it produces is generated in another country, under another country’s law.
Data residency — Where information is physically stored and processed. If your community’s records sit on a computer in another country, that country’s laws can reach them, whatever your contract says. Residency asks: does this information stay under New Zealand law?
Sovereignty (authority you can hold) — In this framework, sovereignty is not a flag or a slogan; it is authority you can actually exercise. Can you inspect the system, refuse it, switch away from it, or take it to a New Zealand court? If the answer is no, you do not hold authority over it — someone else does.
Māori data sovereignty — The principle that data by, about or from Māori is a taonga, and that authority over it — how it is collected, protected, used and shared — belongs with Māori. It is for Māori to define and exercise; the framework’s role is to make room for it in how public systems are built and bought.
Te Kāhui Raraunga — The Māori data leadership organisation whose Māori Data Governance Model, co-designed with sixteen Crown agencies in 2023, sets out eight Pou (pillars) covering how Māori data should be handled, and proposes a Māori Chief Data Steward. The framework’s te Tiriti chapter draws on this published Model and is offered back for its authors to review.
Cryptographic sealing (and the provenance record) — Sealing a record means locking it mathematically at the moment it is made, so that any later change — even one character — is detectable by anyone who checks. The provenance record is the sealed trail itself: who argued what, what was voted, what changed and why. Together they let a reader verify the story without having to trust the teller.
The GAP register — The framework’s public list of what New Zealand does not know: figures that simply do not exist, such as how many jobs AI has displaced here, or how much water each data centre draws. Where a number is missing, the framework declares the gap instead of inventing one, and names the instrument that would create the missing evidence.
Frontier model — One of the handful of most capable AI systems in the world at any moment, built by a few very large companies at enormous cost. New Zealand cannot build one and the framework does not pretend otherwise; its interest is in the terms on which such systems are used here.
Compute / FLOPS — Compute is raw processing power: the capacity of the machines that train and run AI. FLOPS (floating-point operations per second) is the standard unit for measuring it, the way litres measure volume. New Zealand’s largest research computer runs at about 1.4 petaFLOPS — thousands of times below what frontier AI training uses.
Procurement test — A pass-or-fail check applied before a public body buys something. The framework proposes two for AI: where does the system actually run and under whose law (jurisdiction of inference), and where does the data live (data residency)? A system that fails is never bought — which makes buying rules the strongest lever a small country holds.
Reversibility — The design rule that every step must be undoable until the evidence says otherwise: pilots that can be stopped, funded programmes with named stop conditions, open standards rather than one central system everyone is locked into. It means no future government inherits a machine it cannot change.
This glossary accompanies “Our Own Machines”, an AI-generated demonstration framework offered for public discussion — not the policy of any party, and not the position of any organisation.