Experts cautiously optimistic about expanding AI in education, health sectors

This story has been updated to clarify Mahsa McCauley's job title.

Experts are cautiously optimistic about the government's interest in expanding artificial intelligence, but want to see a deliberate and well-scrutinised approach.

Minister for Digitising Government Judith Collins wants to start with the health and education sectors, saying the emerging technology could be used to assess mammogram results and tutor children.

New Zealand Initiative senior fellow Dr Michael Johnston said many in the education sector hoped the technology would level the playing field, with studies showing students who received one-on-one tutoring performed better (two standard deviations above) those who did not.

"Two standard deviations takes you from the 50th percentile to the 95th percentile, so that's massive. The trouble is, of course, the problem is that we can't afford a one-on-one tutor for every child in school, so if AI can become some proxy for that it could make a substantial difference to education equality."

However, Johnston warned the technology should not be used in place of learning basic skills.

"The use of any technology, certainly including AI, in education is that it mustn't undermine children, young people learning knowledge and skills that they'll later need to rely on for further learning," he said.

"A good example is in early mathematics, learning arithmetic. If we allow children to use calculators to do addition and subtraction before they know how to do those things themselves, before they've mastered those skills, then anything that comes later that needs to build on those skills will have shaky foundations."

Increased AI usage also raised questions about whether young people would spend even more time on screens, he said.

"You are putting them at risk, not only of not developing social skills but actually a lot of children now are starting school with insufficient oral language skills, not only for learning to read and write but to access what's going on in the classroom, to understand what's going on. That is a real concern."

AI could lighten teachers' administrative workload - academic

A leading artificial intelligence academic says using AI in education could bring significant benefits for teachers and students.

Senior lecturer at Auckland University of Technology and an executive council member of the AI Forum, Mahsa McCauley, told Checkpoint the technology could free teachers' administration workload.

AI could take care of the repetitive, mundane tasks, leaving teachers to focus on things like more one-on-one time with students, she said.

There were still a lot that was unknown about the adoption of AI in education, she said.

A potential benefit of AI could be in creating a personalised learning experience for students, she said.

"So every student they can have their own tailored curriculum that is really based on their needs and their pace and the other one it is really having access to their intelligent tutoring system, that they are available for a student, [an] assessment system available to support a student 24/7."

It would allow every teacher to have their own virtual assistant, she said.

But she said AI also came with its risks and it was important to address them, while being open to "the transformative potential of this technology".

McCauley said parents, teachers and children needed to be educated about how to use AI before the technology was rolled out.

"We need to advocate for a responsible and equitable use of AI in the classroom."

AI could be big 'game-changer' in health sector - general surgeon

On the health side of things, Collins wanted AI to take on tasks like processing mammogram results. General surgeon and Auckland University associate professor Dr Matthew Clark said it would be the "biggest game-changer" in the health sector for generations.

"One of the big challenges at the moment is different languages that people use in Aotearoa. Perhaps using chatbots in a person's native language to improve health communication, information transfer, that's going to be an early win."

However, he agreed the technology would still need human oversight.

"It still needs to be approached with caution because you know we just need to make sure the technology's safe and people aren't being poorly diagnosed or wrongly diagnosed at these very high stakes levels."

AI was currently used in healthcare on a small, isolated scale, Clark said, so it was difficult to know how it was already being applied.

"One of the things I would hope for in the future is a clearing house of technology and AI so that everyone knows what is being done and those that are interested can use the wins that we generate and rapidly diffuse them around the motu."

Government-level AI guidelines needed, Allyn Robins says

AI lead of public interest think tank Brainbox, Allyn Robins, said there were reasons to be enthusiastic, but warned people should not get carried away by what he described as "the peak of a hype-cycle about AI".

For example, Robins said the technology was not yet capable of things like tutoring.

"These AI systems, certainly the ones [Collins] seems to be talking about, they are not thinking in the way that the term 'artificial intelligence' often implies," he said.

"They're all essentially doing the same thing, which is putting words together how words have been put together before, and that means that fundamentally they are not built to provide access to facts, and they are not built to give coherent explanations of things like math problems.

"It still might not get even the most commonplace facts correct, and for more obscure things you are really just rolling the dice and you have no way of verifying in the system, because it will tend to double-down and ... insist that it is correct."

Scanning mammograms was a different type of AI that bordered on a more traditional machine-learning, Robins said, which was now relatively well understood.

However, a broader concern was how personal information would be used.

"Data powers these AI systems, it's what allows them to continue to improve, it's what lets you build them in the first place. So any and all data that these AI producers can get, if they can they will likely want to feed it back into their systems," he said.

"There are ways of using these systems that are relatively safe in terms of the data - you can run local versions of the AI, you can make sure that the data is not being used for fine-tuning in various ways. So it's not an insurmountable problem, but it is a problem that has to be confronted directly and accounted for when you're designing how you're going to use these things."

The European Union has regulated AI usage in high-risk sectors, including education and health, but New Zealand is yet to follow suit. Robins said some type of guidelines were needed.

"Whether it is legislation or direct regulation, whether it's just a set of expectations at a relatively high level for the government to publish, but I do think some form of guidance is good to have at this time, and I imagine some form of regulation is going to be an inevitability."