Velvet Sundown: The new band that doesn't exist

‘Dust on the Wind’ has been listened to more than 1.2 million times. But the band that plays it, Velvet Sundown, doesn’t exist,

Spotify has now changed the band's bio to confirm the music has been created using artificial intelligence while, “guided by human creative direction”.

It's not a trick, the bio states, it's a “mirror”.

Related stories:

But what original work was ‘Dust on the Wind’ trained on? It remains legally contended, Dr Joshua Uvaraj explained to RNZ’s Nine to Noon.

Uvaraj is a senior lecturer in the Faculty of Law at the University of Auckland and co-director of the New Zealand Centre for Intellectual Property.

The song was likely made with an AI generator called Suno, Uvaraj says.

“That works, as I understand it, in similar ways to other chatbots like ChatGPT or other programs in that you would put in an instruction and generate some sort of song or musical pattern, and the AI generator would generate that.

“Now, the question then becomes, to what extent was there some sort of complexity in the instruction? And to what extent did whoever's involved edit the output? Did they add more manual instruments? Did they slow it down or speed it? Did they record the vocals themselves? So, it's very opaque, what has happened here, but at least at some level and some important level, AI has been involved to generate the music involved.”

There are legal challenges going on throughout the world to AI generated music, he says.

“There is a big dispute around the world at the moment as to whether or not an AI company can legitimately train its models on vast quantities of information, the vast majority of which they probably have not appropriately licensed, although they are starting to license content now from, say, the New York Times and other publishers.”

Artists have been suing AI companies in the US, and recently in the UK, arguing that training AI on their work is copyright infringement, but very few legal determinations have so far been been made, he says.

In the US AI companies are using ‘fair use’ conventions to access copyrighted material, he says.

“If you want to reproduce a work for your class or something like that, or a whole bunch of other things where common sense would say that's really not copyright infringement. The AI companies have been arguing that what they're doing is fair use.

“Two lower district courts in California, I believe, have indicated some tentative agreement with that, but we have a while to go before we have a firm legal position on that.”

The UK, Australia and New Zealand don't have a broad fair use exception, he says.

“We have a very narrow exception for things like reporting the news or doing criticism or review.

“We were on the verge of getting that guidance from the UK, but then the company that was suing an AI, Getty Images, they dropped their copyright claim during the trial.”

Artists in New Zealand are waiting on the Government to indicate whether or not they will amend existing the intellectual property framework to address AI training, he says

“Also questions like, if an AI produces a copyright infringing work, well then who's liable for that? Who can you sue? Who can you seek compensation from?

“So those are some of the questions that copyright law really needs to address, and we're waiting for the government to see what they will do.”