AI chatbots could get you sued for copyright infringement, an expert warns. Illustration / Andrew Louis
As singer-songwriter Ed Sheeran discovered, copyright lawsuits can be pesky. He and other music superstars – including Taylor Swift and Justin Bieber – have been forced to defend allegations of copyright infringements, vastly expensive legal challenges that are unlikely to disappear any time soon.
As they say in the industry, where there’s a hit there’s a writ.
Now there’s a new kid in town, in the form of “generative” artificial intelligence (AI). Enter the chatbot in all its wonder, able to produce just about anything that writers, composers, artists and playwrights have laboured over for months, and sometimes years.
With a few simple commands, a chatbot like ChatGPT, launched on the world late last year, can do the same in a moment. But wait another moment, say copyright experts: who “owns” parts of the essay, song lyrics, poems, speeches, blogs or other features that the chatbot spurts out?
Copyright expert Ken Moon, of intellectual property (IP) lawyers AJ Park, thinks the chatbot technology arrived so quickly that users have not had time to think through the implications.
As an IP lawyer, Moon has been involved with AI on and off for 20 years, but purely on a science and engineering level. In the past, it was technology that drove production lines in a factory, managed orchards or made automatic vacuum cleaners and lawnmowers do their jobs.
But suddenly, at startling speed, AI upped its game. Here was something that could “think” and write like a human, and create visual and musical content that looked and sounded like the real thing. Magic.
But Moon’s point is that somewhere along the way, those words, sentence sequences, images and sounds are likely to have been input by a human. And under New Zealand law, that content is automatically protected by copyright for the life of the author or creator, and 50 years beyond that person’s death.
In the US, one of the most litigious countries in the world, the courts are already filling with AI-related copyright cases, and Moon predicts such cases could happen in New Zealand, where copyright protection is in some ways even stricter.
In the US, an author needs to register a work before it is protected. In New Zealand, literary works - scripts, plays, books - art, images and music are automatically protected by the much-tweaked Copyright Act 1994 from the moment they are produced.
Another difference is that in the US, only works produced by a human can be registered. But in New Zealand, all computer-generated work, including “new” content created by a chatbot like ChatGPT, is protected under copyright for 50 years.
But apart from ChatGPT outputs, there is a copyright risk for AI users who “educate” or “train” their systems to generate material, Moon says. That includes AI systems like GitHub Copilot, used to help write computer programs, which are protected under literary works.
Five years ago the Ministry of Business, Innovation and Employment (MBIE) called for submissions with a view to updating the Copyright Act, but no progress has been made since submissions closed in 2019. An MBIE spokesperson blamed “competing work priorities” in the Commerce and Consumer Affairs portfolio for the lack of progress.
However, those priorities are likely to change given new copyright-related obligations under a Free Trade Agreement (FTA) signed with the UK last year and another FTA with the US this year. New Zealand has until 2028 to extend its “human” copyright term from 50 years to 70 years beyond the death of the creator to fall into line with other countries.
The lifespan of copyright protection has become increasingly relevant to the offspring and other family members of those who created the work, wanting to safeguard their heritage - and their royalty rights.
Little did the late Marvin Gaye know that 50 years after he and Ed Townsend wrote the 1973 hit song Let’s Get It On, the next generation would be fighting it out with Sheeran in a New York courtroom, with millions of dollars at stake, over a few musical chords.
This year, the ginger-haired musician batted away two claims that he had copied chords in his hit song Thinking Out Loud. He won a case brought by Townsend’s long-lost daughter Kathryn Townsend Griffin, and a second $168.2 million lawsuit brought by a company owned by investment banker David Pullman, which owns copyright interests in Gaye and Townsend’s song.
In 2015 the Gaye family took Pharrell Williams and Robin Thicke to court over their 2013 hit Blurred Lines, claiming it copied Gaye’s 1977 song Got to Give It Up. The family won an $8.9m judgment and half of the song’s future publishing royalties.
Enter AI on to the music scene and the copyright issues become even more complex. To show just how easy it is to make a fake, French DJ and music producer David Guetta created a song in an hour using ChatGPT for lyrics, and voice bot Uberduck for rap vocals which sounded exactly like Eminem.
And CNN described AI as “a new Goliath” for music companies to contend with after Universal Music Group (UMG) this year took exception to an AI-generated fake duet supposedly featuring Canadian rapper Drake and singer/songwriter Abel Tesfaye, aka The Weeknd. Both will perform in Auckland this year, Drake in November and The Weeknd in December.
TikTok user @ghostwriter977, clad in a white sheet and dark glasses, taunted the superstars by uploading the song, Heart on my Sleeve, which mimicked the artists’ voices and musical styles.
By the time UMG, which represents stars such as Ariana Grande and Nicki Minaj, got vocal about copyright and the song was taken down, millions of people had viewed it on TikTok and listened to it on Spotify.
The music company, which vilifies AI music, describing it as “deep fakes” and “fraud”, urged streaming platforms like Spotify and Apple Music to stop platforms from allowing AI “training” using music and lyrics that are protected by copyright.
Stock photo company Getty Images is currently suing UK-based Stability AI in the US, alleging that the AI art generator “scraped” more than 12 million Getty images from its database to “train” its Stable Diffusion AI image-generation system.
Under New Zealand copyright law there are two aspects to be aware of when using AI, Moon says.
Literary works, images and music accessed by AI might have copyright protection, but the end user may be unaware of the violation. He warns that users don’t have to copy a whole image or piece of text to have infringed copyright, just a “substantial part”, which isn’t measured in quantity, but in quality. It could be part of someone’s face, a striking part of an artwork, or an important piece of text from a book, he says.
And because computer-generated material is protected for 50 years in New Zealand, people may not realise they can’t copy or use something created by a chatbot without permission.
An MBIE spokesperson said no decision had been made on whether the timespan for computer-generated copyright protection would be extended from 50 to 70 years. Moon predicts there could well be a debate over whether New Zealand should drop its copyright protection for computer-generated works altogether. He thinks it’s a debate worth having, given that few other countries offer that protection.
Jane Phare is a senior Auckland-based features and investigations journalist, former assistant editor of NZ Herald and former editor of Viva and the Weekend Herald.
"If you’re loyal to them, they will be loyal to you."