Wednesday 7 October 2026

What New Zealand should do about AI

People often ask me, “We have all these concerns about AI, but what can we actually do about it?” There are multiple parts to that question — what can we do as individuals? For my Christian friends, what can we do in our church community? and what can we do at the (New Zealand) national level? These questions are often rooted in a feeling of powerlessness, but I think there’s actually a lot we can do at both levels. I want to record a snapshot of my thoughts, and in this post I’ll tackle the national level.

First we have to acknowledge the limitations inherent in trying to answer these questions. AI is changing fast (too fast), and predicting the future is getting more and more difficult. Planning several years out is largely futile, although in some cases we still have to try. Whatever national strategy we choose will have to be flexible and respond to unexpected events with potentially drastic changes. Building up that flexibility is actually a recurring theme in what I suggest.

We need to assume that AI progress will not slow down, and that we will not discover fundamental limits on AI/robotics capabilities. It will be great for humans if either of those assumptions are false — we should definitely try to slow down, or ideally pause, the frontier — but we can’t bet on that happening. We need to plan for a future where AI and robots can do most of the work humans are currently doing. New kinds of work will appear, but we should be prepared in case (unlike in the past) this does not create jobs for humans, because we are able to adapt machines to do that work faster than we can retrain humans. So we need to prepare for a long-term future where there is very little economically productive work where it makes sense to employ humans rather than machines. This won’t happen all at once, and we don’t know what the timelines will be for different kinds of work — it’s so difficult to predict the schedule on which AI capabilities emerge and how they interact with the rest of the world. We need both a vision for the long term, and a plan to get us there that we can adapt as those timelines unfold.

The economy

Here are some of my guesses about the long term. For now I will assume we avoid the worst catastrophes (major loss of control, wholesale alignment failure, a superintelligent AI despotism, etc), mainly because they make whatever we do in NZ irrelevant, but see some discussion of that below.

• The cost of knowledge and services work will fall to approximately zero. (AI will be doing all this very cheaply.)
• The moats around information-work platforms will become much weaker. (AI can port your systems from one platform to another very easily, economies of scale won’t matter as much.)
• So there won’t be lucrative monopolies in these areas, in fact not much lucrative business at all.
• The cost of manual labour will drop precipitously on a longer timeline as robotics matures.
• The cost of manufacturing will drop sharply due to robotics, but there will still be moats in these areas because economies of scale still apply.
• However, standing up your own supply chains by leveraging robots will be a lot easier, if you don’t need the cheapest option.
• Sustainable natural resources — energy, land, water, minerals, beauty — will still be subject to scarcity. These will become the main economic bottleneck. Since NZ has a pretty good stock of natural resources per capita, that puts us in a relatively good position.
• If we maintain democratic control of our government, then as unemployment rises to unprecedented levels (especially among the upper middle class), we will dramatically expand the welfare state. Politicians who resist this will simply be swept aside.

If those guesses are reasonably likely, then here are some things I think we could and should do at the national level:

• Maintain our sovereignty over our natural resources. Don’t sell them off or ruin them. They’re the most important assets we have.
• Invest in projects like renewable energy to increase our sustainable natural resource harvest.
• Monitor the job market very carefully so we can see what impact AI is having month to month. For example, it took only six months this year for AI to overturn the field of mathematics, if that happens to, say, accounting next year, we will need hard data fast.
• Start taxing AI automation to slow down AI adoption a bit (giving us more time to adapt) and raise funds to pay for transition costs. Maybe this will be an interim measure, maybe permanent.
• Have politicians describe their contingency plans for quickly and massively expanding our welfare state if and when mass unemployment happens, and how to pay for it. Keep in mind that personal income taxes are not a long-term option.
• Protect our democratic institutions at all costs.

Safety

We can’t influence what the big AI companies do, but we have some control over how they operate in New Zealand. We can promote AI safety by imposing minimum requirements on the products they offer here. For example:

• Require certain minimum safety evals to have been performed and publicly reported, and keep updating that set of evals frequently. Preferably impose some minimum standards that must be reached on those evals.
• Mandatory standards for reporting use of models in dangerous ways.
• Mandatory standards for reporting security incidents, loss-of-control incidents, etc.
• Mandatory standards for truthfulness. Models that lie or conceal the truth in too-obvious ways (e.g. about the Tiananmen Square massacre) should be banned. (Avoid whataboutism on this; there’s lots of room for debate about where to draw the line, but there are some easy minimum lines to be drawn.)
• Eliminate accountability voids. If someone deploys an AI model that causes harm, either the deployer or the model developer or both must be accountable, depending on the situation. If an AI company deploys its own model and that model causes harm, the company must be held fully responsible. “AI did it” should never absolve anyone of anything.
• Limit the use of AI models by children.

AI companies may react by simply not offering their products in New Zealand. To some extent that’s fine — we don’t want the less responsible companies operating here. But I think people also overestimate this as a problem. Big tech companies are quite reluctant to exit even a small market, because it creates an environment where potential competitors can grow up and threaten them elsewhere. Another aspect is that when one country pushes sensible regulation, other countries are likely to copy it, so we will probably not stand alone.

Those sorts of regulations are harder to apply to open-weight models running on local hardware, but we need rules around those models as well, even though enforcement is difficult. It’s a similar situation to 3D-printed guns, which are hard to regulate but we can’t just stop trying.

Catastrophe planning

There is a certain class of AI-related catastrophes that NZ could survive and that deserve some contingency planning. Kinetic wars (even nuclear wars) and cybersecurity catastrophies could shut down our trading networks and force us to become far more self-reliant. Cyberattacks could cripple our IT infrastructure. This suggests some national-level efforts that I think are worth investing in:

• We should identify layers of essential services that we want to continue to work if NZ is disconnected from the outside world’s infrastructure. (Either because infrastructure is destroyed, or we disconnect as a defensive measure.) For example this should include the electrical grid, supermarkets, freight transport, and a lot of agriculture.
• Identify the problematic dependencies (overseas cloud providers, GPS, embedded systems phoning home, essential medicines, fossil fuel imports, etc) and fix them, or at least have a plan for addressing them if the need arises. (Personally I’m particularly interested in studying the cybersecurity of the embedded systems powering our essential infrastructure, and making sure it is resilient in these scenarios. AI helps here.)
• Some strategic stockpiling of equipment or knowlege is probably worth investing in. General-purpose robotics may dramatically lower the cost of maintaing small-scale emergency supply-chain stand-ins.
• Work out contingency plans that describe the conditions which would trigger us to disconnect ourselves from the outside world (to the extent we can).