Digitising government Minister Judith Collins told RNZ AI would deliver efficiency, speed and savings in government services. Photo: 123RF
RNZ can reveal artificial intelligence is taking on diverse roles across New Zealand's public service - from surgical risk assessments at Health NZ, to fishing vessel camera monitoring for MPI.
Some academics want greater transparency in the government's use of AI, saying Kiwis have the right to know when decisions about healthcare and social services aren't being made by humans.
According to data on the Department of Internal Affairs' digital.nz website, 50 public agencies were surveyed in July last year and 37 were using at least one type of AI, from chatbots to transcription software to image analysis.
DIA briefings prepared for digitising government Minister Judith Collins around the same time - now released proactively - revealed more details.
The documents showed Health NZ clinicians used AI to help work out the risk of a patient dying after surgery, while Auckland Transport had deployed AI to help speed up its response to certain official information requests.
Since September 2018, the Accident Compensation Corporation had been using AI to automatically process and accept straight-forward claims.
The Ministry for Primary Industries used AI to sift through fishing vessel recordings, isolate the footage of people and store only that footage in a high-quality format.
It was also using AI-powered '3D scanning machines' to look for biosecurity risks in luggage and mail.
Biosecurity New Zealand commissioner North Mike Ingles said it was about giving staff the best tools possible.
"We've got one installed at Auckland Airport, and that's been in situ for a number of months. It's better viewing for our staff, in respect of the algorithm development. The standard of the image is much better for staff, rather than your traditional 2D machines," he said.
MPI said a second machine was being installed at New Zealand Post's new processing centre in Auckland.
It was due to start scanning inbound international mail shortly.
At the Department of Conservation, staff were using AI tools to process trail camera footage, including a program called Alita.
"It just gives you file name, date, time and what animal it thinks it saw," DOC science advisor Joris Tinnemans explained.
He said the technology had a way to go and Alita could only be used on certain GPU computers.
"We're still at a stage where AI tools are not miracles... but they're helpful and they save time because first of all, they weed out all the empty images. Depending on your set-up of the trail cameras in the field, if it's an especially in a grasslands environment, you might get up to 80, 90 percent of the images being empty, so you don't have to look at them all," he said.
Last week the government announced a new Public Service AI Framework, setting out guidelines for the safe use of AI.
Digitising government Minister Judith Collins told RNZ AI would deliver efficiency, speed and savings in government services.
She said it was important agencies protected people's data, as set out in the Privacy and Human Rights Acts.
University of Otago public health professor Michael Baker said it was great to see the government setting out AI guidelines, which reflected the wider OECD AI principals.
However, Baker believed it could go a step further.
In a Public Health Communication Centre briefing last year, he and four other academics urged the government to publish a register of all the AI systems in use within public agencies.
"By having a register of AI systems, that offers huge benefits for the public," he said.
"It means there's a degree of transparency and scrutiny about the extent of these systems, and it does mean they can potentially be asked to demonstrate they meet standards that we would regard as reasonable."
Baker said AI algorithms needed to recognise New Zealand's unique population.
"We know that equity is a huge challenge in New Zealand. We've got so many examples of how Māori in particular miss out on healthcare equity and because there are conscious and unconscious biases built into systems. That will be one area where we might have to adapt these principles for our local needs and our local populations," he said.
Artificial Intelligence professor at Victoria University of Wellington, Alistair Knott, also contributed to the Public Health Communication Centre briefing.
He said it was important to distinguish between simple predictive AI models trained to do one task, and newer 'generative' AI models which may need more careful oversight.
"That's the ChatGPT type of system, which can basically do any task that you can talk about in natural language. It's rather different in the sense that you don't train it to do a particular task. You train it just on language in general, and then you use it to do a task after it's been trained.
"And that means it's kind of a double edged thing - it can do almost anything that you can talk about, but you don't have an evaluation out-of-the-box on its ability to do any one task," he said.
Knott said building an AI register might not be straight-forward, but it would definitely be a "good idea".
"There are many, many discussions, many conversations about AI happening in many parts of government. It's not necessarily one government department that's going to come up with this. I think it'll be a discussion between several departments," he said.
"If a government service or government department is using an AI system, then the public should know about it, and should know as much as possible about how it works, and how it's used... we think that is going to be helpful in creating public trust."
Sign up for Ngā Pitopito Kōrero, a daily newsletter curated by our editors and delivered straight to your inbox every weekday.
