The New Zealand government's enthusiasm for replacing public servants with artificial intelligence is running headfirst into reality: AI isn't a magic solution, and the costs might be higher than anticipated.
Critics are warning that Wellington's plans to use AI to deliver government services with fewer staff could come with significant hidden costs, according to RNZ. These include system failures, loss of institutional knowledge, potential discrimination in automated decision-making, and the cost of fixing mistakes when algorithms get it wrong.
The government sees AI as a way to maintain or improve service delivery while cutting staff numbers - essentially having your cake and eating it too. It's an appealing vision: government services that are faster, cheaper, and available 24/7, all powered by artificial intelligence.
But as anyone who has actually worked with AI systems knows, the reality is more complicated. AI is brilliant at certain tasks: processing routine applications, answering common questions, identifying patterns in data. It's terrible at others: handling exceptions, exercising judgment, understanding context, or showing empathy.
Mate, this is tech-utopian thinking meeting the messy reality of government services. The people who interact with government agencies aren't simple data points. They're humans with complex situations that often don't fit neatly into algorithmic categories.
Consider someone applying for a benefit. An AI might be able to process straightforward cases quickly. But what about someone with a complicated employment history, medical conditions that affect their work capacity, or circumstances that require discretionary judgment? Do you really want an algorithm making those decisions?
The accountability question is even thornier. When a human public servant makes a bad decision, there's someone to appeal to, someone who can be held responsible. When an AI system denies your claim or approves something incorrectly, who do you complain to? The algorithm? The programmer? The minister who approved its use?
Then there's the discrimination risk. AI systems trained on historical data tend to replicate the biases in that data. If New Zealand's welfare system has historically disadvantaged Māori and Pacific peoples - which it has - an AI trained on that data will likely continue that discrimination, just more efficiently.
The government's AI push also ignores the value of institutional knowledge. Public servants don't just process applications - they understand how systems work, why policies exist, what the edge cases are. When you fire experienced staff and replace them with algorithms, you lose that knowledge. And you won't realize you've lost it until something goes wrong.
There are also the practical costs. AI systems require significant upfront investment, ongoing maintenance, regular updating as policies change, and technical expertise to manage. Are those costs factored into the government's savings calculations, or are they hidden in future budgets?
Cybersecurity is another concern. Government AI systems would be handling sensitive personal data about citizens. They'd also be attractive targets for hackers. The cost of securing these systems and responding to breaches could be substantial.
None of this means AI has no role in government services. Used properly - to augment human workers rather than replace them, to handle routine tasks while humans focus on complex cases - AI could genuinely improve service delivery.
But that's not what the government is proposing. It's proposing AI as a replacement for public servants, driven by the goal of cutting staff numbers rather than improving services.
The motivation isn't technological improvement - it's ideological commitment to a smaller public service. AI just provides a convenient excuse and a veneer of modernity to justify job cuts.
Opposition parties and public sector unions are right to demand detailed analysis of the costs and risks. Before New Zealand embarks on a mass replacement of human workers with algorithms, the government needs to prove that it will actually work and that the costs - including the hidden ones - are worth it.
The track record isn't encouraging. Governments around the world have tried similar initiatives, with mixed results at best. Australia's robodebt scandal showed what happens when you let algorithms make welfare decisions without adequate human oversight: thousands of people wrongly pursued for debts they didn't owe.
New Zealand should learn from those failures rather than repeating them. AI can be a powerful tool, but it's not a substitute for adequate staffing, proper funding, and human judgment.
You can't algorithm your way out of the responsibility to deliver quality public services. The government needs to drop the tech fantasy and focus on the unglamorous work of making sure public agencies have the resources - human and technological - to do their jobs properly.
