Comment: At the first leaders’ debate, Labour Party Leader Chris Hipkins said artificial intelligence was an existential threat, and National’s Christopher Luxon spoke of its “serious dangers and evils”. There was even an offer from Luxon to Hipkins to work together on a regulatory framework.
How much things have changed. Just over a year ago, 36 AI experts including myself wrote an open letter to all political party leaders calling for a bipartisan approach, risk-based legislation, and a national AI oversight body. Only the Green Party leaders were willing to sign on.
Today, AI is an election issue, with even the parties most wary of regulation, such as Act, acknowledging the need for action.
Even in the deeply polarised US, senators from both parties have worked together on AI safety bills. Democrat and Republican voters have very different views on President Donald Trump and former President Joe Biden but do agree on two things: they love Dolly Parton and they can’t stand Mark Zuckerberg (chief executive of Meta). Only 13 percent of Democrats and 23 percent of Republicans view him favourably.
What changed? The Overton window (the political concept that refers to what’s publicly acceptable at any point in time) has clearly shifted.
One big factor is the increased fear about the existential risk of AI development. This year’s turning point was the Hugging Face story, where OpenAI’s AI agents were able to breach the cyber defences of Hugging Face, another AI company. Since then, we’ve seen more breaches, including of Australian government websites.
But it all really came to a head when Jacob Coxon, an AI researcher, resigned from the AI company Anthropic because of fears of the tech getting out of control. Rather than dismiss his concerns, leaders of AI companies have publicly agreed that something needs to be done. Big AI says it wants to be regulated!
We could argue about whether this is a marketing stunt in the lead-up to these companies going public. Or maybe their chief execs know AI regulation is inevitable, and would rather the Trump administration introduce laws that suit them while he is pro-AI (also known as regulatory capture). That could also help the big AI companies suppress new competitors.
But whichever it is, our answer should be the same. If the big AI players believe their own warnings, they shouldn’t be left to regulate themselves. If they don’t, then they are still misleading us for their own gain. Either way, they can’t be trusted to behave.
We also don’t need to know exactly how far away superintelligent AI is. The AI we have today is already causing harm, from deepfakes and scams to biased decision making. If we put regulation and real oversight in place for these harms, we’ll be in a much stronger position for whatever comes next, too.
What would that oversight look like here? Our political parties have finally started to tell us. And their policies have a key thing in common: we should build on what’s happening overseas.
National suggests we should be “fast followers”; Labour points to Australian standards; the Greens to the European Union’s AI Act; NZ First to Unesco principles; and Opportunity to “international standards”. That makes a lot of sense. We’re late to the regulation game, so we shouldn’t start afresh. And aligning with other countries’ rules is a good strategy for getting Big Tech to actually obey our laws when the US is failing to lead.
The parties differ in their ideas for oversight, though. Labour wants an AI office in the Department of Prime Minister and Cabinet. The Greens want something similar, but with a movement towards an independent AI body over time. National would spread the work across existing agencies. The other parties haven’t been specific.
But we can’t just copy and paste regulation from overseas. New Zealand has a distinct culture and demographic makeup that needs to be accounted for. No overseas law covers our obligations under te Tiriti o Waitangi, or Māori rights over Māori data.
AI systems trained mostly on overseas data can also work poorly for Māori and Pacific peoples, who comprise about a quarter of our population. If we don’t have sufficient testing and oversight, not everyone will benefit from AI equally.
We’ve done this before with our privacy law. The Privacy Act was based on work by the OECD, but was adapted for the New Zealand context. It also gave the Privacy Commissioner powers to introduce new legally binding rules as technology changes. An “AI Commissioner” would help us keep pace with new advances in AI without going back to Parliament every few months, albeit we’d want one with more teeth than the Privacy Commissioner.
AI regulation is long overdue. Now that it’s an election issue, with broad political support, surely we will see new AI law in 2027. I hope so. But there are already some signs of backpedalling.
A few days after the first leaders’ debate, National’s technology spokesperson Penny Simmonds confirmed National was sticking with a “light touch” approach. She also warned against regulating “on hypothetical scenarios”. That’s quite a different message from her leader’s “serious dangers and evils”.
Surveys show fewer than a quarter of New Zealanders think current safeguards are enough, and nearly half say clearer government standards would make them more confident using AI. We must keep the pressure on our lawmakers. Once the election dust settles, whoever wins must be held to what they said on the campaign trail, including that offer to work together.
AI is changing our society fast. Our leaders owe it to us to make sure AI benefits New Zealanders, not just Big Tech.
This article was originally published on Newsroom.
Andrew Lensen is a senior lecturer in Artificial Intelligence in the School of Engineering and Computer Science at Te Herenga Waka—Victoria University of Wellington.