‘Forgive me if I’m not doing cartwheels’: Scepticism over AI payment breakthrough

We’re sorry, this feature is currently unavailable. We’re working to restore it. Please try again later.

Advertisement

‘Forgive me if I’m not doing cartwheels’: Scepticism over AI payment breakthrough

By Garry Maddox

Simon Kennedy is an anonymous voice in commercials for McDonald’s, Subaru, Domino’s Pizza and Big Bash League cricket.

As one of Australia’s most successful voice artists, he also works on video games, radio promos, animated series, audiobooks and corporate education. But in the last year, Kennedy estimates that more than 20 per cent of his income has vanished with the rise of voices created by artificial intelligence.

“I’m cautiously encouraged by what appears to be a turnaround on copyright,” says voice artist Simon Kennedy.

“I’m cautiously encouraged by what appears to be a turnaround on copyright,” says voice artist Simon Kennedy. Credit: James Brickwood

He used to be able to tell when an AI voice was used, but that is changing.

“There are some AI-generated voices that are pretty average, pretty robotic,” Kennedy, the president of the Australian Association of Voice Actors, said. “Those ones, even regular people can probably tell.

“However, the technology has evolved so rapidly, even in just the last year, [that] even experienced voice actors aren’t 100 per cent sure.”

Loading

If a voice artist can’t tell the difference, it’s no wonder that radio listeners can’t either.

In April, it emerged that Sydney radio station CADA had been using a host named Thy for a four-hour shift on weekdays for six months. Her voice was generated by a text-to-speech AI program with her image on the station’s website cloned from an employee in Australian Radio Network’s finance team.

As the use of AI advances in the creative industries, there was both cautious optimism and scepticism from the sector about the latest development - a breakthrough on the vexed issue of compensating musicians, writers, artists and possibly media outlets for the content that feeds AI tools such as ChatGPT at the economic roundtable on Thursday.

Advertisement

After talks over lunch between two key players – Australian Council of Trade Unions assistant secretary Joseph Mitchell and Atlassian co-founder and Tech Council of Australia chair Scott Farquhar – there was agreement that unions and tech giants would work on a model that would pay creators for their content being used.

“There was discussion with the Tech Council and the ACTU about wanting to address the issue of properly paying creatives, journalists and academics,” ACTU secretary Sally McManus said. “We’re going to give this a real good go at coming up with a model that makes sure that people are actually paid for what they’ve produced.”

“I’m cautiously encouraged by what appears to be a turnaround on copyright.”

Australian Association of Voice Actors president Simon Kennedy

Kennedy said it seemed like “a welcome shift” by Farquhar, who has argued that the country should reform copyright law to let AI companies mine data for large language models without payment to boost productivity.

“I’m cautiously encouraged by what appears to be a turnaround on copyright,” he said.

But Kennedy held there still needed to be protection against AI cloning of an individual’s image and voice, including mandatory labelling and watermarking of all AI-generated content, to protect against voice cloning and deep fakes.

Peter Mattessi, the showrunner for the TV series Return To Paradise who is president of the Australian Writers Guild, was sceptical about the breakthrough.

“Forgive me if I’m not doing cartwheels,” he said. “I’m glad they’re all getting along, but I’m very wary of people who stand to make billions of dollars in what is an over-hyped [industry].

Return to Paradise: showrunner Peter Mattessi says he is not turning cartwheels about a breakthrough in AI discussions.

Return to Paradise: showrunner Peter Mattessi says he is not turning cartwheels about a breakthrough in AI discussions.Credit: ABC

“Until I see what’s being offered and what’s being acquired, I’m pretty happy with the current copyright law.”

Why were Australian writers, musicians and other artists so upset?

It’s still less than three years since OpenAI released the first version of ChatGPT and, as the fastest growing tech product in history, it started to reshape work, industry, education, social media and leisure.

Loading

International tech companies are at the stage of training large language models such as ChatGPT and building data centres.

Earlier this month, the Productivity Commission released an interim report that proposed giving these companies exemptions from the Australian Copyright Act so they can mine copyrighted work to train these large language models.

Artists, writers, musicians, actors, voice artists and entertainment industry associations and unions were outraged.

Midnight Oil frontman and former Labor Minister Peter Garrett savaged the proposal to allow text and data mining of songs without compensating songwriters. “The rampant opportunism of big tech aiming to pillage other people’s work for their own profit is galling and shameful,” he said.

Among other opponents of the proposal were the Australian Society of Authors, Writing Australia and the Media Entertainment and Arts Alliance, which said the creative industries contributed more than $60 billion to the economy and employed more than 280,000 people.

But Scott Farquhar argued that the country should reform copyright law to let AI companies mine data without payment to boost productivity.

‘We already have a ‘model’ for ensuring people get paid for what they’ve produced’: ARIA chief executive Annabelle Herd.

‘We already have a ‘model’ for ensuring people get paid for what they’ve produced’: ARIA chief executive Annabelle Herd.Credit: Peter Rae

“At the moment, all AI usage of mining or searching or going across data is probably illegal under Australian law,” he said. “I think that hurts a lot of investment of these companies in Australia.”

Farquhar said AI was a chance for Australia to make “megabucks” through regional data centres.

“When I look at the opportunities it provides us as a nation, as a culture, it’s a new tool that we get to use and it gets to make us a lot more productive,” he said. “It’s going to make our jobs a lot more fun.”

Industry reacts to the breakthrough

If Mattessi was not doing cartwheels, industry associations were insisting that existing copyright laws should apply to AI data mining.

“We already have a ‘model’ for ensuring people get paid for what they’ve produced,” the chief executive of the Australian Recording Industry Association, Annabelle Herd, posted on social media. “It’s called the Copyright Act.

Loading

“It already ensures writers, musicians, journalists, researchers and filmmakers get paid when their work is used by others (including digital platforms). It already facilitates the kind of licensing at scale needed for AI.”

In a joint statement, the Australian Society of Authors and the Copyright Agency said “responsible AI development” should operate within copyright law.

“Innovation should be supported through licensed access – not through the unauthorised access of copyright content for the training of large language models,” they said.

What happens next?

The Productivity Commission is seeking submissions on its interim report by September 15 to report to the federal government in December.

But it remains to be seen whether the breakthrough in Canberra results in payment for creators for their data-mined work.

The chief executive of the Tech Council, Damian Kassabgi, was more cautious than McManus when he insisted after the roundtable that no details had been agreed on a copyright model.

“We are hopeful we can find a path forward on copyright that allows AI training to take place in Australia while also including appropriate protections for creators that make a living from their work,” he said.

On ABC radio on Friday morning, Kassabgi said it was “too early to say” whether tech companies would provide compensation for material used to train AI.

Find out the next TV, streaming series and movies to add to your must-sees. Get The Watchlist delivered every Thursday.

Most Viewed in Culture

Loading