NZ’s new AI strategy is long on ‘economic opportunity’ but short on managing ethical and social risk

The government’s newly unveiled National AI Strategy^[1] is all about what its title says: “Investing with Confidence”. It tells businesses that Aotearoa New Zealand is open for AI use, and that our “light touch” approach won’t get in their way.

The question now is whether the claims made for AI by Minister of Science, Innovation and Technology Shane Reti – that it will help boost productivity and enable the economy to grow by billions of dollars – can be justified.

Generative AI – the kind powering ChatGPT, CoPilot and Google’s video generator Veo 3 – is certainly earning money. In its latest funding round in April, OpenAI was valued at US$300 billion^[2].

Nvidia, which makes the hardware that powers AI technology, just became the first publicly traded company to surpass a $4 trillion market valuation^[3]. It’d be great if New Zealand could get a slice of that pie.

New Zealand doesn’t have the capacity to build new generative AI systems, however. That takes tens of thousands^[4] of NVIDIA’s chips, costing many millions of dollars that only big tech companies or large nation states can afford.

What New Zealand can do is build new systems and services around these models, either by fine-tuning them, or using them as part of a bigger software system or service.

The government isn’t offering any new money to help companies do this. Its AI strategy is about reducing barriers, providing regulatory guidance, building capacity and ensuring adaption happens responsibly.

But there aren’t many barriers to begin with. The regulatory guidance contained in the strategy essentially says “we won’t regulate”. Existing laws are said to be “technology-neutral” and therefore sufficient.

As for building capacity, the country’s tertiary sector is more under-funded than ever^[5], with universities cutting courses and staff. Humanities research into AI ethics is also ineligible for government funding as it doesn’t contribute to economic growth^[6].

A relaxed regulatory regime

The issue of responsible adoption is perhaps of most concern. The 42-page “Responsible AI Guidance for Businesses” document, released alongside the strategy, contains useful material on issues such as detecting bias, measuring model accuracy, and human oversight. But it is just that – guidance – and entirely voluntary.

This puts New Zealand among the most relaxed nations when it comes to AI regulation, along with Japan^[7] and Singapore^[8]. At the other end is the European Union, which enacted its comprehensive AI Act in 2024, and has stood fast against lobbying to delay legislative rollout^[9].

The relaxed approach is interesting in light of New Zealand being ranked third-to-last out of 47 countries in a recent survey of trust in AI^[10]. In another survey from last year, 66% of New Zealanders reported being nervous about the impacts of AI^[11].

Some of the nervousness can be explained by AI being a new technology with well documented examples of inappropriate use, intentional or not. Deepfakes^[12] as a form of cyberbullying have become a major concern. Even the ACT Party, not generally in favour of more regulation, wants to criminalise^[13] the creation and sharing of non-consensual, sexually explicit deepfakes.

Generative image, video and music creation^[14] is reducing the demand for creative workers – even though it is their very work that was used to train the AI models.

But there are other, more subtle issues, too. AI systems learn from data. If that data is biased, then those systems will learn to be biased, too.

New Zealanders are right to be anxious about the prospect of private sector companies denying them jobs, entry to supermarkets^[15] or a bank loan because of something in their pasts. Because modern deep learning models are so complex and impenetrable, it can be impossible to determine how an AI system made a decision.

And what of the potential for AI to be used online to mislead voters and discredit the democratic process, as the New York Times has reported may have occurred already^[16] in at least 50 cases?

Managing risk the European way

The strategy is essentially silent on all of these issues. It also doesn’t mention Te Tiriti o Waitangi/Treaty of Waitangi. Even Google’s AI summary tells me this is the nation’s founding document, laying the groundwork for Māori and the Crown to coexist.

AI, like any data-driven system, has the potential to disproportionately disadvantage^[17] Māori if it involves systems from overseas designed (and trained) for other populations.

Allowing these systems to be imported and deployed in Aotearoa New Zealand in sensitive applications – healthcare or justice, for example – without any regulation or oversight risks worsening inequalities even further.

What’s the alternative? The EU offers some useful answers. It has taken the approach of categorising AI uses based on risk^[18]:

• “Unacceptable risk” – the likes of social scoring (where individuals’ daily activities are monitored and scored for their societal benefit) and AI hacking – is outright banned.

• High-risk systems, such as uses for employment or transportation infrastructure, require strict obligations, including risk assessments and human oversight.

• Limited and minimal risk applications – the biggest category by far – imposes very little red tape on companies.

This feels like a mature approach New Zealand might emulate. It wouldn’t stymie productivity much – unless companies were doing something risky. In which case, the 66% of New Zealanders who are nervous about AI might well agree it’s worth slowing down and getting it right.

References

1. ^{^} National AI Strategy (www.mbie.govt.nz)
2. ^{^} US$300 billion (www.nzherald.co.nz)
3. ^{^} surpass a $4 trillion market valuation (apnews.com)
4. ^{^} tens of thousands (www.ornl.gov)
5. ^{^} more under-funded than ever (www.nzherald.co.nz)
6. ^{^} contribute to economic growth (www.beehive.govt.nz)
7. ^{^} Japan (www.csis.org)
8. ^{^} Singapore (www.whitecase.com)
9. ^{^} delay legislative rollout (www.reuters.com)
10. ^{^} trust in AI (kpmg.com)
11. ^{^} nervous about the impacts of AI (www.ipsos.com)
12. ^{^} Deepfakes (theconversation.com)
13. ^{^} wants to criminalise (www.nzherald.co.nz)
14. ^{^} music creation (www.thepost.co.nz)
15. ^{^} entry to supermarkets (www.nzherald.co.nz)
16. ^{^} may have occurred already (www.nytimes.com)
17. ^{^} potential to disproportionately disadvantage (theconversation.com)
18. ^{^} categorising AI uses based on risk (digital-strategy.ec.europa.eu)