Voyager Broadband‘s (Actrix Networks, 1st Domains / Affordable Domains / Register Direct) latest advertisement (via RNZ) shows two Māori/Pacific men, one of whom is attacking a Pākehā man by having him in a chocker hold on the floor, after their internet drops out during a UFC match on the television.
The advert was made with generative AI and lacks cultural due diligence while perpetuating false stereotypes of Māori and Pacific men are aggressive. It also takes away employment opportunities form Māori and Pacific actors, reflecting a wide spread issue in America for all actors and people involved in the film industry.
Voyager says it is reviewing the prompts used to generate the video. But this is a much bigger issue than that, it is a human rights, ethics, corporate culture, racism and bias issues.
This situation highlights the need for Human-in-the-Loop for all AI generated outputs from AI. Also, to New Zealand companies that chose to replace humans with AI generated images, that they must add a disclaimer that the advert is AI generated and then make your AI prompts public. The same applied to political parties who create AI generated adverts.
Disclaimers are good internationally recognised best practise for AI governance and reflects another best practice of using AI Cards.
Voyager Broadband’s review should start from one principle ‘the AI did it’ is not a defence.
Someone (a human) wrote the prompt(s) using their own human bias and steryotypes that now reflect the view of Voyager Broadband, reviewed the output, approved it and paid to place the advert online.
An organisation that publishes AI-generated content is fully responsible for it. Choosing a tool with known bias raises the standard of care rather than lowering it.
Image generators amplify racial and gender bias, and detailed prompts have not fixed it. These models learn from what the internet has said about us, and for Māori and Pasifika men, much of that archive is crime reporting and deficit imagery.
This is also a Māori Data Sovereignty issue. Our likenesses were taken into training data without consent, and are now handed back to us as caricature to sell broadband.
Te Tiriti o Waitangi is a threshold for organisations operating in New Zealand, not a box to tick afterwards. Our Pasifika cousins are owed the same care.
In practice this includes:
- treating AI depictions of people as higher risk
- use AI cards and public disclaimers of prompts and keep records of them
- give Māori and Pasifika reviewers real authority to stop content, not just comment on it
This will happen again because generative AI makes it fast to create content without the people who would have caught the problem. The machine did not choose the violence. People chose to publish it, and accountability has to sit with them.
All organisations using AI generated materials should have an AI Policy and or access to a human with AI Governance/Safety/Ethics perspectives.