The 8,700 lost public sector jobs will be replaced, in part, by AI. But nobody can say how it will work or what it will cost, writes Catherine McGregor in today’s excerpt from The Bulletin.

To receive The Bulletin in full each weekday, sign up here.

The plan

The response to the government’s pre-Budget announcement has not been warm. Finance minister Nicola Willis set out plans last week to cut 8,700 public service jobs across around 40 core agencies by 2029, saving $2.4 billion – with artificial intelligence named as the mechanism to fill the gaps. Critics, economists and AI experts have spent the days since asking two blunt questions: does the government know what a pivot to AI really entails, and has anyone actually worked out what this will cost?

The announcement framed AI as a “basic expectation” across government departments. As Emily Broadmore – a former political adviser and workforce transformation strategist – writes in The Post (paywalled), the plan makes sense in principle. The problem is the execution, and whether those in charge know enough to drive the strategy in the right direction. After all, “the people ordering a transformation of our public service,” she writes, “just admitted they are using the most powerful tools ever created to… edit speeches.”

The bill

Talking to RNZ’s Phil Pennington, University of Auckland law and technology professor Alexandra Andhov foresaw significant costs that are likely to eat into that $2.4b figure. Enterprise-scale AI is not a one-off purchase: it comes with ongoing licence fees, model upgrades, and the costs of auditing and oversight. “The costs that we pay for AI today are heavily subsidised while the AI companies are trying to capture as much of the market,” she noted. “These are not the real costs that AI will cost.” In parliament, digitising government minister Paul Goldsmith said he didn’t have any figures on the estimated rollout and licensing costs, nor seen a cost-benefit analysis.

Andhov raised a second concern: almost all the AI the government is likely to use is built by American companies, and the money spent on it will leave the country.

The know-how

Even if the costs stack up, a deeper problem remains: does the government understand what it’s doing? Broadmore argues that the announcement conflates two very different things. Yes, AI will compress certain kinds of work – things like first drafts, briefings and standard comms copy – and “if this is done properly, this new leaner, sharper workforce will significantly outperform the way our public service currently operates”.

But the government doesn’t seem to have grappled with what that means for the traditional career pipeline. “The grad policy adviser learned to read political risk by drafting OIA responses,” she writes, by way of example. If that entry-level work disappears, so does the experience base that leads junior staff into senior management.

The rules (that don’t exist)

Then there is the question of how AI use in the public sector should be regulated. New Zealand’s only government guidelines, released last year, are voluntary. Asked about this by Newsroom’s Fox Meyer, Goldsmith rejected a suggestion that rules should be put in place before the AI transformation got underway.

His position is quite different to the approach in Australia, where a chief AI officer is employed by every federal department. Writing in Newsroom, Andhov notes that in NSW, every agency must complete a risk assessment before deploying AI, with high-risk systems listed on a public register. The New Zealand government has none of that.

The risk of getting it wrong is not theoretical, Andhov says, pointing to crises like the Dutch childcare benefit scandal, in which an automated fraud-detection system resulted in tens of thousands of families being falsely accused of cheating, causing people to lose their homes and children to be taken into care. “‘We’ll use more AI’ is not a budget line,” she writes. “It is a major policy decision, and right now it is being made without the basic safeguards.”