Replacing public servants with AI could come with hidden costs, critics warn

by · RNZ
Replacing public servants with artificial intelligence could come with more costs, critics say.Photo: 123RF

Replacing public servants with artificial intelligence could come with more costs that will eat into any savings, say critics.

The government says the plan to cut another 8700 people sector jobs in about 40 core agencies will save $2.4 billion over four years.

The coalition government's basic recipe for cutting the public sector size and wage bill is to reduce the head-count and increase use of artificial intelligence (AI) technology.

When Roger May, a retired forestry consultant of Motueka heard this, he emailed Finance Minister Nicola Willis to say that "is going to take time and money".

He then told RNZ: "8700 knowledgeable bureaucrats are about to be axed. She expects that AI will replace a lot of these people.

"But there's not been any explanation of the costs and time involved, and I wouldn't mind betting that's going to eat into their $2.4 billion."

He got back an acknowledgement from the minister's office.

Finance Minister Nicola Willis.Photo: RNZ / Marika Khabazi

'I'm not aware of a current local AI provider'

Labour tried to get some details on Wednesday afternoon in Parliament, asking what the rollout and licensing cost of AI would be.

Digitising Government Minister Paul Goldsmith responded: "I don't have that exact figure at the moment, but of course it varies.

"And what we inherited, of course, was a wide variety of arrangements in place across many government departments, and that is precisely why we're putting together a more coherent and centrally guided system."

Goldsmith was then asked if it would use local or overseas AI technology.

"Mister Speaker, I'm not aware of a current local AI provider in the scale of Claude or Copilot, but what I would say is that we'll be making use of the best technology available."

Digitising Government Minister Paul Goldsmith.Photo: RNZ / Mark Papalii

Claude and Copilot are leading AI models from US companies Anthropic and Microsoft.

ACT's David Seymour then asked if the public sector would build its own silicon chips or import them.

Goldsmith replied, "My suggestion is that we focus on the things that we do well and sell them to the world and then we purchase the things that other people in the world do better than us and hopefully if we have made enough money in our economy we can afford to have the best ones."

Professor Alexandra Andhov, chair of law and technology at the University of Auckland, foresaw huge challenges to the government's savings goal, quite apart from the lack of clarity at how it settled on the $2.4b figure.

"The published material doesn't really show the cost side of the AI," she said on Wednesday.

Enterprise-scale AI was not a one-off buy but had ongoing costs in licence fees, model upgrades and responding to companies dictating when and how models were replaced or integrated, plus audit and oversight of its operations, Andhov added.

"And even I think more importantly, it's to recognise that the costs that we pay for AI today are heavily subsidised while the AI companies are trying to capture as much of the market, these are not the real costs that AI will cost."

Real-world pricing, and cyber security demands "will push up the cost of AI and generally any kind of digital infrastructure to such an extent that I think that we are not yet even in a position to foresee", Andhov said.

The security landscape is in flux after Anthropic released its new model Mythos recently to a select few organisations to test how good it was for hacking. In response, one of the world's largest cybersecurity firms Palo Alto Networks put out 26 security advisories at once last week - when it normally issues five a month - because Mythos had found so many more vulnerabilities than usual.

The security landscape is in flux after Anthropic released its new model Mythos recently to a select few organisations to test how good it was for hacking.Photo: Picture-Alliance via AFP

"We intend to fix every vulnerability we find before advanced AI capabilities become widely available to adversaries," the US firm said.

Here, Treasury told public agencies last year they must prepare for the encryption-breaking threat of quantum computing due in 2030.

Andhov said there was still more cause for pause.

"But wait a second, who we are actually using, who are these AI providers?" she asked.

"To my understanding, the majority of the providers that government is considering are not New Zealand companies, not the companies that are governed by New Zealand law, but they are US-based companies that only need to comply with the US law, given the fact that also New Zealand doesn't want to regulate AI."

(The government has chosen a lighthanded approach to regulation.)

"If the New Zealand government ultimately uses, let's say, Microsoft's AI ... then they're paying OpenAI, which is based in California, which doesn't pay any taxes [here].

"All of this amount is taken to the US and actually brings nothing back to New Zealand ... and it has involved loss of jobs here."

Professor Alexandra AndhovPhoto: Chris Loufte/University of Auckland

Australia is undergoing its own AI moves across its public sector AI, and it also relied on US Big Tech, said Jeannie Paterson, Professor of Law and director of the Centre of AI and Digital Ethics at the University of Melbourne.

But it had some things New Zealand did not have, like a new AI Safety Institute and a new AI Employment and Workplaces Forum that just had its first meeting with unions and businesses.

Australia also has a new central registry the public can use to check how agencies are using AI, [www.gov.au/articles/new-central-register-ai-transparency-statements-commonwealth-entities stating], "Bringing this information together in one place provides a clearer, more complete picture of AI use across government."

Jeannie Paterson, Professor of Law and director of the Centre of AI and Digital Ethics at the University of Melbourne.Photo: Supplied / University of Melbourne

"The first question to ask is, what's the framework for deploying AI?" said Paterson.

"Because we know that AI is not the genie ... and that there's a number of risks. AI only works well with expert humans around it.

"So unless there's an investment in training and resources from humans and a pretty clear democratic commitment to what role AI should be playing in society, then it's sort of a rush to lowest, to the bottom."

The New Zealand government has issued a Digital Government Target State for a centralised approach to improve tech while saving billions.

RNZ approached ministers Willis and Goldsmith for comment but they did not respond by deadline and they declined to do a recorded interview.

Sign up for Ngā Pitopito Kōrero, a daily newsletter curated by our editors and delivered straight to your inbox every weekday.