High Use, Low Trust: What a Trust-Builder Makes of New Zealand's AI Problem

Last month the AI Forum released the third edition of its AI Blueprint for Aotearoa. It's a big document. One phrase did most of the work for me.

High-use, low-trust.

The Forum reviewed 28 New Zealand adoption studies and found uptake sitting somewhere between 40 and 80 percent, depending how you count, with people reporting real productivity gains. And yet trust hasn't moved. A global study by KPMG and the University of Melbourne puts us near the bottom of the table for believing AI's benefits outweigh the risks, around four in ten. We're some of the most enthusiastic users and some of the least convinced people in the world, at the same time.

The Forum is clear-eyed about what that means. The next phase, they write, "won't be determined by access to technology alone. It will be shaped by how well organisations, government, and society address trust, skills, and governance alongside continued use." In other words, the hard part from here isn't the tech. It's us.

The easy read is that low trust is simply the problem to fix. Make people more comfortable, the thinking goes, and adoption deepens. I'm not sure that's right. So I took the finding into a conversation with someone who has spent her whole career on exactly this question.

Sanel Tomlinson is a chartered accountant with close to three decades in corporate reporting and assurance. She was a partner at KPMG in New Zealand, London and Hong Kong, and as interim project director at the External Reporting Board she led the early work on this country's climate-related disclosure standards, the first mandatory regime of its kind anywhere in the world. Her job, for thirty years, has been getting people to believe in numbers and claims they can't check for themselves. If anyone knows how trust actually gets built, it's her.

Low trust might be the smart part

Here's the first thing Sanel said that made me rethink the Blueprint's framing:

"People are sceptical, and sometimes I think that's a good thing. We need to be sceptical, because then we challenge and we think things through. And when you try and it doesn't meet your expectations, you know how to feed back and what you need to change."

That stopped me. We keep treating New Zealand's caution as a deficit, a number to drag upwards. Sanel sees it as a feature. Scepticism is how you avoid sleepwalking into a tool that doesn't do what it promised.

The catch is that scepticism is a starting point, not a destination. Healthy doubt gets you asking the right questions. It doesn't, on its own, get you to trust. Something still has to earn that.

Trust is earned, not declared

This is where Sanel's career stops being a side note and becomes the point. You don't make people trust a set of accounts by telling them to. You build the machinery, the standards, the disclosure, the independent sign-off, and the trust follows.

A lot of organisations are trying to skip that step with AI. "Just get everyone using it" is the strategy I hear most often. Sanel has been watching the same thing:

"Nobody's gone out with, here is our strategy, here is what you can and cannot do... Until some of those pieces are in place... it's very difficult to put a person in place and say, you shall now be accountable, because what are they accountable for?"

That's the gap in one sentence. In every other part of a business, someone's name is on the line. With AI, we've handed the tool to everyone and named no one. The same KPMG and Melbourne study found people would trust AI far more if they simply knew who was accountable when it went wrong. Accountability isn't a brake on adoption. It's the thing that lets adoption go faster.

There's a second piece, and it's one I run into constantly in workshops:

"People will never trust something, and they will continue to struggle, if they get an answer but they don't know why they got the answer they got. They don't understand what went into the decision, and they don't understand why that led to the outcome."

A lot of the time AI is a black box. You put a question in, an answer comes back, and you've no real idea how. You can narrow that gap with better context and more rigour around your prompts, but you can't pretend it isn't there. Trust needs explainability, and explainability is still hard.

We've seen this movie before

The useful thing about talking to Sanel is that she's lived through a version of this. Sustainability reporting went from voluntary and nice-to-have to mandatory standards. Greenwashing forced the issue. Now we're watching AI-washing arrive right on cue, everything suddenly "powered by AI" whether it is or not.

So does AI need to make the same journey, from guidance to rules?

"My view is yes. We just don't live in a world where people do things if they're not told to, if they're not required to. I don't say take a sledgehammer to it... things are moving so fast, the standard-setting process can't keep up. By the time we've got the standards and guidance for AI in place, there's going to be the next thing."

That's the bind we're in. The government has gone light-touch and principles-based, with no AI-specific law, leaning on the rules we already have. The Blueprint is comfortable with that. Sanel's instinct, and mine, is that some guardrails are coming, and the trick is to move fast enough to be useful without pretending you can freeze a target moving this quickly.

Show people the value

If you want trust to follow use, Sanel's prescription is blunt. Stop announcing AI and start showing what it's worth.

"It's been in place now for a number of years, but it hasn't yet proven its benefit for everyone... I think that just means work harder on showing where's the value, where's the benefit for the person."

She was talking about climate reporting, but it lands just as hard on AI. People don't trust what they can't see the point of. And here the Blueprint is sobering. Only 2.7 percent of our workforce are genuine AI practitioners with the tool embedded in how they work, and 79 percent of employers say they don't know how to train their people. We've handed everyone a tool and skipped the part where they learn to get value from it.

That isn't a technology problem. I think of it as people, process and technology, and the technology is the easy corner. The organisations getting real value have their house in order first. The ones struggling have brilliant people, processes living in those people's heads, and a shiny tool they can't quite plug in.

The one thing AI can't do

I close every episode of The Cairn with the same question. In the age of AI, what's the one trait that defines human leadership? Sanel didn't hesitate.

"For me, it's honesty. You say what you mean, you mean what you say, and you do what you say. From a human perspective, AI can't do that. It can't say one thing and then show you what it does. People hear what you say and they see what you do, and those two things need to match. That, to me, is honesty."

That's the whole thing, really. Honesty isn't soft. It's the match between what you say and what you do, held over time, and a model can't offer it. Ask it the same question tomorrow and you're talking to a different thing. There's no self there to be held to its word. Which is exactly why, for all the talk of autonomy, the human stays in the room.

New Zealand doesn't have an AI technology problem. We have a trust problem, and trust has only ever been built one way: by someone willing to stand behind what they said. AI can say anything. Being held to it is still our job.

Listen to the full conversation with Sanel Tomlinson on The Cairn.