By Pita Ligaiula
Pacific media should use artificial intelligence to handle routine newsroom tasks but keep reporting, verification, context and relationships firmly in the hands of journalists, says Pasifika AI creator Kevyn Wahuzue.
Wahuzue, founder of Kanaky Tech, says AI can help Pacific newsrooms work faster, but one inaccurate story can damage trust that journalists have built with communities over years.
“Trust in Pacific journalism doesn’t come from speed. It comes from people knowing who you are, knowing you went there, and knowing you’ll be back next week to face them.
“AI can’t give you any of that. But it can cost you all of it in one bad story,” Wahuzue said in an interview with PACNEWS.
He said Pacific journalists should use AI for tasks such as transcription, initial translation drafts, document sorting and archive searches, but not for decisions that require human judgement or relationships with communities.
“My message is simple: use AI for the work behind the story, never for the relationship in front of it.
“Speaking for a community, deciding what matters, putting words in someone’s mouth: that stays with the journalist,” he said.
Wahuzue said AI-generated information should never be treated as verified fact.
“Treat everything an AI gives you as a tip, not a fact.
“A tip from a source you don’t know, who has never been to the place, and who sounds confident about everything.”,” he told PACNEWS
Wahuze said this principle was central to the development of Pasifika AI, which is designed to provide answers using a defined set of institutional sources.
“That’s how I built Pasifika AI. It only searches a published list of 66 institutional sources, and every answer shows its references so the reader can open them,” he said.
“Even then, the site says clearly that a listed source doesn’t guarantee the sentence is right.”,” he said.
Wahuzue said the same standard should apply to AI use in newsrooms.
“AI can find the documents faster. The journalist still reads them, calls the people, and explains what it means here,” he said.
Wahuzue said one of the biggest threats posed by AI to Pacific languages and cultures was not obvious nonsense, but information that sounded correct while being wrong.
“The biggest risk isn’t silence. It’s fluent mistakes,” he said.
He said he had recently tested several AI models using Pacific languages and found significant problems.
“Some went into loops, repeating the same phrase for a whole page. That’s the easy case: anyone can see it,” he said.
“The dangerous one wrote smooth Samoan that was wrong,” Wahuzue said.
Wahuzue cited an example involving the pe’a, the traditional Samoan male tatau.
“It described the pe’a, the traditional male tatau, as reaching the elbow.
“It reads well. A Samoan reader sees the mistake in a second. A student overseas won’t, and will repeat it,” he said.
He also warned about the use of Indigenous knowledge in AI systems, particularly knowledge that is held by families, clans or specific custodians.
“Some knowledge belongs to a family, a clan, specific custodians. Being able to ask a machine doesn’t give anyone permission to receive it,” Wahuzue said.
“Every archive, recording and story that goes online can be pulled into a model without anyone asking the community it came from.”
Wahuzue said another major problem was the tendency of AI systems to combine distinct Pacific peoples, languages and cultures.
“To be honest, these models don’t ‘understand’ anything. They reproduce what they were trained on, and very little of that was written by Pacific people, in Pacific languages,” he said.
He said his testing had shown how easily AI could confuse related but distinct languages.
“In my tests, when I asked for an answer in Cook Islands Māori, some models wrote New Zealand Māori instead.
“Related languages, different peoples,” Wahuzue told PACNEWS.
Wahuzue said Pacific AI systems needed to distinguish between different peoples and communities rather than treating the region as one cultural group.
“A model trained mostly outside the region treats the Pacific as one culture with one voice,” he said.
He identified three steps to address the problem by using regional sources, keeping Pacific peoples and languages distinct, and giving knowledge holders a role in correcting AI systems.
“Ground the answers in sources from the region: our institutions, academies, archives, not the open web.
“Keep each people distinct and named,” Wahuzue said,
He also said communities should have the ability to correct AI systems and seek the removal of information where appropriate.
“And let the people who hold the knowledge correct the system, with the power to have something taken down,” he said.
Wahuzue highlighted Te Hiku Media in Aotearoa New Zealand as an example of Indigenous-led technology development, noting that it had developed te reo Māori speech technology while retaining control of its data.
“The rest, the region has to build itself,” he said.
