Interest.co.nz is unpacking how AI will change your everyday life - the risks, the opportunities, and what to actually expect. Our new series brings you the policymakers, experts and industry leaders from New Zealand and overseas.
By Anna Whyte
Three weeks ago, Anthropic's most advanced AI models were suspended by US export controls over national security concerns and access disabled for foreign nationals, including foreign national Anthropic employees, and restored only after a fortnight of negotiation.
As the New Zealand Government embarks on a technological overhaul of the public service, with AI expected to transform the way it works, Green Party public service spokesperson Francisco Hernandez says the Anthropic example is exactly the scenario New Zealand should be worried about.
Hernandez's view on a successful AI rollout in the public service is that control is needed ahead of deployment.
"AI can potentially be a powerful tool but guidelines and safeguards need to be developed first in consultation with workers and the wider public," he tells Interest.co.nz.
"If the guidelines and safeguards are robust enough and supported by workers and the public, it can be deployed in certain areas to enhance what the public service does.
"But we need to make sure that we actually have control over the models - the Trump administration’s unilateral withdrawal of Fable from Anthropic shows how vulnerable NZ could be if we relied on overseas models without adequate safeguards."
A scathing review of the Government's digital delivery agency raised potential data sovereignty issues, using an example of a large agency forced to rely on AI tools based in Australian data centres rather than in NZ.
"The hard issues around data privacy, Māori data sovereignty and the ability of the public service to actually own the models they are using and the data associated with it have to be resolved in a way that protects the public, workers, tangata whenua and people who might be disproportionately negatively impacted by it," Hernandez said.
He said AI use should be optional, and there should be auditability and human responsibility "at the heart of it and an option to always request a human."
"The Green party co-leaders signed onto a call on parliament to regulate AI and we agree with the letter’s authors on the need to ensure that it serves the public good."
For Hernandez, sovereignty over the technology is only half the risk, the other is who gets left behind.
"We should also ensure that AI does not widen the digital exclusion gap for retirees, low income and rural communities that have issues with connectivity or accessing digital tools."
On the importance of upskilling the workforce, Hernandez said the priority had to be enhancing public services.
Hernandez says he agrees with the position paper the Public Service Association released, which says; 'the use of AI should be aimed at enhancing public and community services and improving the quality of jobs, not aimed at job displacement."
Earlier this year, the Government’s digital delivery agency functions were moved from the Department of Internal Affairs to the Public Service Commission, and with Public Service Commissioner Brian Roche now at the helm, he is eyeing a major overhaul with new veto powers to stop agencies spending frivolously.
Hernandez says the Green Party would support transitioning the leadership of the Government's Digital Delivery Agency (GDDA) to; "an expert because the impression I got from the Minister’s answers [in a select committee hearing] and his answers to several media interviews, is that he doesn’t have a single clue about AI and barely understands it."
That was in reference to Public Service and Digitising Government Minister Paul Goldsmith.
"It is deeply dangerous that at the same time the Government is undertaking a comprehensive transition into AI that the Minister actively doesn’t understand the subject which is why having an expert leading the [G]DDA is so critical," Hernandez said.
In response, Goldsmith said the comments did not seem to be following "the so-called Greens kaupapa of not being abusive and making personal attacks."
"The Government has absolutely got a clear sense of the opportunities that AI will bring to deal more effectively with data and information and help provide better services," Goldsmith said.
8 Comments
"an expert because the impression I got from the Minister’s answers [in a select committee hearing] and his answers to several media interviews, is that he doesn’t have a single clue about AI and barely understands it."
I get the feeling Francisco Hernandez isn't that much wiser than Goldsmith. All he has done is take some of the issues and weave them in to an ideological narrative.
Unless you are using Frontier models and understand how they work, and where they process the data you give them, You probably do not understand the risk and governance issues.
Only 6 months ago we could not do any of this, I suggest that legal, commercial and procurement departments do not under stand the risks well themselves yet.
Giving everyone models risks possible end user computing nightmares and no governance with our data. and possible non deterministic outcomes.
Imagine if a resource consent could be granted or denied depending who, inside council, ran it through which model on a day, or even the same LLM different day different outcome...
but it is probably great for looking up cricket scores.
The loony left no mere fertilizer party do not deserve to be in Gov, but AI does indeed need rules around it, Humans make dumb calls, like the fertilizer policy, we need to make sure AI does not make random dumb calls as well
"Imagine if a resource consent could be granted or denied depending who, inside council, ran it through..."
IIRC that's happened b4 without any AI involvement
"Unless you are using Frontier models and understand how they work, and where they process the data you give them, You probably do not understand the risk and governance issue"
Isnt the problem (or one of them) that even those that are developing these models dont themselves know how they work?
Maybe AI will at least be consistently inscrutable?
.
I agree. The risks are real. We need to listen to the warnings given by the Godfather of AI Geoffrey Hinton and take action.
It's now way too late for this discussion as AI technology has been, and is, evolving so fast any guidelines that come out of the usual elephantine consultation processes will be obsolescent before they get written down. We're not that special that special rules are required.
A better option would be to do a cut-and-paste from the Scandinavians about data privacy, and go like hell to try and make the vast pools of government data at least a bit breach-resistant while making it uniform and useful.
We welcome your comments below. If you are not already registered, please register to comment
Remember we welcome robust, respectful and insightful debate. We don't welcome abusive or defamatory comments and will de-register those repeatedly making such comments. Our current comment policy is here.