Revealed: National's rejected AI attack ads

An unused ad generated by a user who also generated two images used on National's social media accounts. Prompt: "Scared woman clutching her purse at night."

National has acknowledged tasking an AI generator with creating political attack ads, but not all of its work appears to have made the cut.

It was revealed on Tuesday that the party had used generative AI tools to produce realistic-looking photos for social media attack ads on Labour.

Speaking to Breakfast yesterday, National Party leader Christopher Luxon said there was little difference between prompting an online service to make an image and purchasing a stock photo that used actors to portray scenarios.

One of those online AI services, Midjourney, has an accessible log of publicly generated images previously made by users. The service runs on Discord, a social messaging platform, and requires a paid subscription to use (around NZ$16).

1News found two of National's AI-generated images on the platform and several dozen others, which appear to be unused iterations generated in the past few weeks.

An account with the name "TG Creative", prompted Midjourney to generate images of Vin Diesel that were later used in a National Party meme, alongside a picture of a woman peering out her home at night — which would later be used in an attack ad on crime.

"TG Creative" on Discord.

Several of the unused images appear to relate to policy areas that National has previously criticised the Government on — including for violent crime, cost of living, health, and education.

1News asked the Opposition Leader's office about whether the account's generated images were all intended for political ads, whether the party employed external agencies that worked with AI, and who "TG Creative" was.

A spokesperson for Christopher Luxon said in response that National had "nothing further to add". The party also refused to disclose the prompts it used for two other AI-generated images of nurses and masked robbers that were posted on its social media.

In the 24 hours after 1News reached out to National for comment, TG Creative changed its screen name to "IntMJ".

"IntMJ" was approached for comment late this morning.

Unused AI generations and their prompts

Some generation prompts like camera type, resolution, and theme details have been removed from the below verbatim descriptions.

In Midjourney, users can type in a text prompt and the AI will create four images based on the description. Users can then choose to upscale one of the images to a higher resolution or use one of them as a starting point for another generation.

Where to draw the line?

Most Aucklanders approached by Breakfast yesterday said they were relatively unconcerned by National using AI to create stock images.

"Whenever you see political marketing, it tends to be not a great representation of the truth. Whether using AI or not isn't necessarily a good or a bad thing," one said.

The line around ethically using generators, like Midjourney, remains fuzzy, with a wide scope of concerns around misinformation, copyright, ownership, and regulation.

For example, image-generating AIs are trained on billions of images available on the internet, which has raised complicated questions about copyright law.

Meanwhile, researchers are concerned that the tools could be used to boost the power of disinformation and misinformation.

AI commentator Paul Duignan and Disinformation Project research fellow Sanjana Hattotuwa discussed the potential of the new technology.

AI commentator Paul Duignan said Luxon should be given some "credit" for saying he wouldn't use AI to generate videos, however that political parties needed greater consensus about where they would draw the line.

"The problem is stills of what? Can you generate stills of the other party?" he told Breakfast. "There's a bit of a slippery slope problem here.

"Once you start using it as a question of where you draw the line… If I was a political party, I'd be very careful about bobbing my toe in the water in regard to this."

Prime Minister Chris Hipkins said that Labour hadn't used any AI-generated campaign imagery, while other minor parties said they hadn't yet touched the technology.

Speaking to Stuff, an ACT spokesperson said the party hadn't used AI imagery and isn't considering using it while the Greens said they wouldn't use the technology and "especially not to depict people".

The National leader said the party was being "very responsible" in how it used the new technology.

Disinformation Project research fellow Sanjana Hattotuwa told Breakfast that ease and speed of the new tools was their key differentiator for bad actors.

"People don't need to know Photoshop or video editing, or photo manipulation to create synthetic media that looks good enough, and real enough to manipulate public opinion.

"Synthetic media has the potential to create content at such a scale and scope, that it really ramps up what we have seen in former campaigns, and makes it accessible to anybody.

"It is very worrying and concerning globally, but also domestically, because we're heading into the general election in October," he said.

Earlier this month, a US legislator proposed a bill that would force politicians to disclose when AI tools were used to generate imagery in their advertising.

It came after Republicans released an attack ad on Joe Biden that used AI-generated imagery to paint a dystopian picture of a potential second term.

SHARE ME

More Stories