RICHARD PREBBLE: AI Could Make Big Government Even Bigger

On a flight to Wellington the passenger beside me introduced himself.

“I’m from IBM. We are developing for Railways a world-leading wages management system.”

All my alarm bells went off.

“World-leading”, “computer system”, and “government department” are three phrases that usually mean cost overruns and eventual disappointment.

I was Minister of Railways. I had never heard that Railways was buying a “world-leading” wages system.

When I reached my office, I summoned Railways management.

They explained that Railways had unique requirements developed over a century: wet money, dirt money, overtime rates, shift allowances and dozens of special categories.

Armies of wages clerks struggled to keep track of it all.

No standard payroll system could supposedly cope. Railways needed a bespoke system costing millions.

I rang the rail unions. They agreed the system was unmanageable. Workers were often paid incorrectly. Workers had a low base pay made up by allowances. As their superannuation was based on the low base wage it meant a low income in retirement.

We reached agreement to simplify the entire structure. Staff shifted to salaries instead of complicated allowances and overtime calculations. Retirement incomes improved. 

Even with extra cost of superannuation payments the savings to railways was significant.

The cheapest computer reform was not building a better computer system. It was simplifying the system itself.

I had a similar experience as Minister of Police.

Police Headquarters wanted me to approve a bespoke computer system called INCIS — the Integrated National Crime Information System. Officials said it would be “world leading” and revolutionise policing by monitoring and coordinating virtually every aspect of frontline police work.

I thought they were asking the wrong question.

Frontline police are not motivated by helping headquarters monitor them more efficiently. If anything, they will quietly ensure such systems fail.

I asked a simpler question: what actually wastes police time?

Headquarters did not know.

A survey found frontline officers were spending huge amounts of time two figure typing repetitive reports.

I explained that I had seen many large government computer systems that had failed and predicted that the INCIS system would also fail. I suggested a far simpler reform: modern word-processing equipment and touch-typing competency for recruits.

Headquarters preferred the grander vision.

The Police waited until I left the portfolio and persuaded subsequent ministers to green-light the INCIS project. It consumed around $100 million, took years to build and was ultimately abandoned as unworkable. It became one of New Zealand’s worst public-sector IT failures.

That history is worth remembering when ministers now announce that artificial intelligence will dramatically reduce civil service numbers.

It is far more likely that bespoke AI systems will result in huge cost blow outs.  The AI systems may even require even more civil servants.

AI certainly can improve productivity.

Anyone who regularly uses tools like Copilot or ChatGPT knows they can draft reports, summarise documents, prepare briefings, search regulations and answer routine queries remarkably quickly.

Simply giving many public servants access to competent AI tools would probably lift productivity noticeably.

But governments consistently misunderstand where technology savings come from.

The biggest gains rarely come from giant bespoke systems designed by consultants and officials. They come from simplifying processes, eliminating unnecessary rules and asking frontline staff what wastes their time.

New Zealand governments have an especially poor record buying large custom-built computer systems. Departments always insist their requirements are unique.

Complexity expands. Consultants multiply. Costs blow out.

AI could easily become the next version of this cycle.

Every department will want its own carefully tailored “AI solution” designed around thousands of pages of existing processes and reporting requirements.

The risk is that we automate bureaucracy instead of reducing it.

Technology by itself does not usually shrink government. Often it merely allows government to generate more forms, more reporting and more compliance.

National risks repeating the same mistake that is the reason the cost of government keeps increasing.

Nicola Willis and Christopher Luxon have not properly analysed why spending and borrowing keeps rising.

The Government is still trying to carry out almost every activity created by the previous Labour Government — only more efficiently.

That may improve administration at the margin. It will not significantly shrink the state.

There is only one consistently proven way to reduce the size of government: stop doing things.

New Zealand now has 28 ministers holding about 80 portfolios across more than 40 departments. Every new activity creates another layer of officials, reporting, coordination and compliance.

Governments do not become smaller because they buy better computers.  They just produce more paperwork faster.

A smaller state requires ministers willing to stop doing some things altogether.

That is a political decision, not a technological one.

The Honourable Richard Prebble CBE is a former member of the New Zealand Parliament. Initially a member of the Labour Party, he joined the newly formed ACT New Zealand party under Roger Douglas in 1996, becoming its leader from 1996 to 2004.