3. Involvement of the public in a deliberation process to define acceptable risk and use of AI in government systems
3. Involvement of the public in a deliberation process to define acceptable risk and use of AI in government systems
If AI is going to radically transform the way that public services are delivered, we should have a say in how that is done.
We want the government to fund ongoing public discussions so that all New Zealanders can help decide how AI should be used in our government. This should also include targeted consultation with important groups, like minority communities, rural communities, iwi and Māori, and those most likely to be effected by use of AI in government.
For example, if a government agency wants to use an AI system to review social housing applications, the public should weigh in on whether or not that is appropriate. We might collectively decide that it's okay for a machine to help sort the paperwork. We might also decide that a human official must always make the final decision to ensure that empathy and fairness aren't lost.
There are lots of areas where AI use requires some trade-offs. For example, using AI could make the delivery of public services more efficient and therefore less costly for taxpayers... but increase the potential error rate and make it harder for people to correct those errors. Is that a trade-off we are willing to make as a country? Is that trade-off OK in some scenarios, but not others?
We might also want to collectively define some "unacceptable risk" areas that the government should just not do. For example, we might collectively decide that we don't want our military to use lethal autonomous weapons. Or maybe we're okay with it! We don't really know, that's why we need to have the conversation.
We, as co-petitioners, don't have the answers. We want the people of New Zealand to be given the time and space to figure out our collective answer, as a country.
The public should help draw the line between 'acceptable use' and 'overreaching automation'.
Right now, decisions about when to use AI are being made behind closed doors, generally without public consultation or engagement. The government has even passed some AI-enabling legislation under urgency, removing the opportunity for the public to provide input through the Select Committee process.
But AI is not just an IT upgrade; it is a technology that shifts how public power is exercised.
The impacts of AI systems will be felt by all of us, and therefore we all deserve the opportunity to have a say in how AI systems will be used. People have a variety of concerns, ranging from immediate issues such as algorithmic bias and loss of privacy, to longer-term issues such as environmental damage, the future of work, cultural loss, and value lock-in. The government needs to stop making assumptions about what the people of New Zealander are okay with... and just ask them. Our perspectives deserve to be heard.
We are calling for funded public deliberation processes - such as citizens assemblies and structured community workshops - to enable all of us to participate in a national conversation about government use of AI. What constitutes "high-risk" needs to be at least informed by the public's views and risk tolerances, and it needs to be grounded in our unique societal context. We cannot rely on AI systems imported from overseas to automatically understand Te Tiriti o Waitangi obligations, or reflect the distinct values of our diverse communities. True public engagement gives people the time, space, and unbiased information to grapple with these complexities.
Seeking this input from the public, and being transparent about the process and its outcomes, is how we build genuine social licence. It moves us past broad anxieties about AI and allows us to be more specific. When the government is led by the public, it earns public trust, and therefore public support, to use AI to improve our public services.
2020: The Digital Council for Aotearoa New Zealand ran a series of workshops to understand what is needed for public trust in automated decision-making. They gave participants six scenarios to discuss how comfortable they were with that use, and why.
A citizens’ assembly in Switzerland was convened in 2025 to give guidance on national AI policy.
A report from the University of Edinburgh recommends the establishment of a ‘standing citizens’ assembly on AI and society’ (see e.g. p4).
The Ada Lovelace Institute in the UK ran a Citizens' Biometric Council to understand the public's view on biometric technologies like facial recognition.
Deliberative democracy processes such as Citizens Assemblies have been run successfully (on non-AI issues) in Auckland (Watercare), Tairāwhiti (Te Weu), Whanganui (District Council), and Wellington (City Council).
See also this article calling for Representative AI Governance, from Kris Rose at Georgetown McCourt School of Public Policy.
Existing frameworks like Te Hiku Media's Kaitiakitanga License, their approach to developing a reo māori speech tool while abiding by Māori tikanga, and Luke Munn's adaption of Tā Hirini Moko Mead's five Tests give examples of how AI can be designed and built while adhering to values that are not built into the default AI development process.
2019/2020: Several submitters on the government's consultation on the Algorithm Charter pointed out that a top-down code would not work, and that creating trustworthy systems requires processes for the public, affected communities, and Māori to shape how these tools are developed and deployed.
2020: The Digital Council for Aotearoa New Zealand's report was clear: everyday New Zealanders want to talk about automated decision making, but that they need structured, neutral spaces to do so.
2025: The government's Responsible AI Guidance for the Public Services suggests agencies should be completely transparent with citizens and maintain human supervision over AI outputs. However, this is unenforceable guidance, rather than formal process. It is also inconsistent with some approaches already taken by government agencies.
2025: The Government Chief Digital Officer's Public Service AI Framework explicitly includes "social licence" as one of its core foundational pillars, asking agencies to understand how to build public trust and worker engagement in government use of AI. It does not yet provide information on how thorough this should be.
Remember when we spent $22 million on the flag referendum and then ended up making no change? About $7 million of that was spent on public consultation and marketing as part of a "national conversation" on what the public wanted to have reflected in our flag. We're not saying that we have to spend that much, but some funding should be allocated to enable a robust public deliberation process that tells us all what we want from AI use in government.
An example of this could be to hold an AI Citizens' Assembly, an independently run process where a randomly selected but demographically representative group of New Zealanders comes together to discuss government use of AI. Participants would be compensated for their time, with childcare/travel costs covered so that no one is priced out of participating. The Citizens' Assembly would seek to understand what issues people are worried about, and define the risk thresholds for when an AI system needs greater oversight and when it is low risk. Experts can brief the participants on how current government AI tools work here and overseas, and where the risks lie.
This can be combined with other sources of public input, such as focus groups, surveys, and submission processes to enable all people of New Zealander to make their voice heard. These various sources of input can be combined to identify acceptable use-cases and allow government to understand the boundaries between uses that have public support and uses that are likely to lead to backlash.
If this process works well, then it could be repeated every one to two years to ensure that the government continues to align with our public's attitude towards AI. This could be legislatively mandated as a centrally-funded and repeated activity to keep the government's understanding of the public's views up-to-date.
The lack of a clear public mandate is slowing things down already! Government agencies have to guess what the public and other stakeholders might feel about the deployment of AI. In some cases, they guessing incorrectly. This creates an environment of anxious hesitation, or legal challenge and PR nightmares.
Getting public feedback and building a map will make sure that the government is heading in the right direction. Once the government has that direction, it can adopt AI at pace. But first, we collectively need to decide the destination, pave the road, and paint the road markings. The cost of getting it right allows us to move faster overall, even if it feels slow to start. Further, the results of this consultation will help businesses know what their customers want, and therefore allow them to move forward with confidence, too.
When people are uncertain and AI is new, it's easy to fear the unknown. However, if people are given unbiased, plain-language information about how a technology works, the conversation naturally shifts away from broad sci-fi panic (or misinformation from overseas) towards constructive, pragmatic discussion. This doesn't slow down adoption—it ensures that when AI is deployed, it is aligned to the New Zealand public.
The public aren't being asked to sit in a room and review code or audit algorithms; they are being asked to help write the high-level principles and rules that will allow the government to navigate high-risk AI systems safely. The public should be involved in deciding which values must be protected, what safeguards are non-negotiable, and where human discretion must not be removed.