Artificial intelligence: creativity or counterfeit

06 October 2025

Arts and culture, Auckland Law School, Staff news, AI, The Challenge, Future of work, Faculty of Arts and Education, Business School

When University of Auckland computer scientist Professor Michael Witbrock's dad took him to see 2001: A Space Odyssey as a kid, the robot HAL did not scare him off. It hooked him.

Following a grim 1970s summer selling boyswear in a Dunedin department store ("Don't ever do that: People do not leave things on the shelves well arranged!"), young Witbrock had earned enough to buy an early computer, an Ohio Scientific Superboard 2, and start tinkering.

Then, while doing a PhD at Carnegie Mellon University in the United States, he and a bunch of friends started playing around with something they called 'genetic art'.

Uncover the full story in the podcast edition! Season two of Ingenious, the University of Auckland's podcast celebrating bold ideas and groundbreaking research shaping our future, is out now. Episode one: Artificial intelligence: creativity or counterfeit?

"An image is just a bunch of pixels, so if you work out given the x and y coordinates, what colour and intensity those pixels should be, you can produce an image.

"Some of the pictures were quite striking; swooping pastel forms with crystalline structures in them. In the range of things you might think were artistic."

Fast forward three decades and what Witbrock and his friends were doing with AI art in the 1980s isn't so very different from what happens with present-day artificial intelligence image production systems, he says. Where those early students used tidy mathematical formulae, AI now uses huge neural networks trained on vast quantities of data.

But is it art?

Still, the core questions have barely shifted: Is the output art? Can it be original? And who owns it?

Alex Sims, a professor in the University's Commercial Law department specialising in intellectual property, has little time for 'AI slop', the term for low quality AI generated images.

But her research highlights a real tension - not all AI-generated work is slop.

Sims points to a graphic novel called Zarya of the Dawn, the brainchild of New York artist Kristina Kashtanova, who used image generation tool MidJourney to
generate the pictures.

"They spent literally days on each image. Initially they said 'Create me whatever', and then used a series of prompts to finally get to something creative."

The US Copyright Office initially granted protection for the graphic novel when it didn't know the images were AI-generated. When it found out, it reversed the decision; US law states images have to have a human author to be protected.

That's not the same in New Zealand, where computer-generated work can be protected. Sims says deciding where the line is between 'creative' and 'non-creative' isn't black and white. Photographs weren't protected by copyright law for decades, she says - people argued photographers hadn't created anything, they were just recording nature.

And creative people have always - knowingly or unknowingly - drawn on influences from the past.

"All of us, artists, everybody, are copying from what we know."

There are a number of big legal cases about companies training their AI on original creative work wending their way slowly through the court system - Disney and Universal versus MidJourney, Getty Images versus Stability, and more. And they may eventually provide more certainty.

The paradox is AI needs authentic original human works to be trained on.

Dr Joshua Yuvarej Law School, University of Auckland

In the meantime, Sims' view is there should be some protection for AI-generated work, "but the protection should be a lot lower".

"So they are not protected for as long, and you don't get what we call 'moral rights' - the right to object to use of your work in certain situations."

She also says artists should be forced to be up-front about their work being AI-generated, including stating that in the metadata for the image.

The paradox of AI creativity

Here's another conundrum: you might think generative AI makes humans redundant, but the truth is the robots can't do it alone.

It's sometimes called the paradox of AI creativity, or the paradox of digital decay, and it's something University of Auckland senior law lecturer Dr Joshua Yuvarej examines in his research.

"The way AI works is you put in a prompt: 'Give me a picture of a black cat', and the AI has been trained on millions, billions, trillions of data points, images, words to statistically predict what it thinks you mean when you say 'black cat'. And it'll come up with a picture of black cat.

"The paradox is AI needs authentic original human works to be trained on. If it is trained on AI-generated works, eventually it leads to what computer scientists call 'model collapse'."

Imagine the AI-generated black cat. "If I re-feed that image into the AI and train it on that and we keep doing that, eventually the image of a black cat that's produced is so distorted it looks nothing like a black cat. But the AI thinks that's what a black cat is."

If we don't have human output to train AI, then AI will become redundant, Yuvarej says.

"So if you look on Seek and other job sites, you have jobs being advertised for people who create stuff to train AI.
It's actually becoming an industry."

So much (human) labour

Here's another irony: it takes an astonishing amount of human hard work to get artificial intelligence to create an original piece of music.

At least, if the experience of Dr David Chisholm is anything to go by.

Chisholm is a composer and head of the University of Auckland's School of Music, and last year started a project with some expert colleagues in Chile to train generative AI on his own work. The aim: to create a new piece of music - in the style of David Chisholm.

It surely couldn't be that hard, he thought. Isn't one of the biggest beefs of creative artists about AI - the subject of legal battles - that algorithms are making work that appears to be an artist's original?

Chisholm couldn't have been more wrong. Making just five minutes of original music using AI was long and laborious. And after all that, the first iteration didn't sound like a Chisholm creation.

To get a performable piece, the composer had to sit down and do much of the work himself. It took months.

"I thought it would be so easy - I'd just transcribe it [from the AI]. Instead it was so much work, so incredibly laborious."

Media contact

Nikki Mandow | research communications M: 021 174 3142 E: [email protected]

More Ingenious podcasts