The question of who should be paid when an artificial intelligence system learns from a song, a novel or a painting has become one of the most contested in the creative economy, and a group of Australian researchers has now put forward a framework they believe could settle it. As reported by Limelight magazine, the proposal sets out a way to compensate creators whose work is used to train generative models, rather than leaving that work to be absorbed for free.
It lands at a fraught moment. For the past two years the global technology industry has built its most valuable products on vast datasets scraped from the open internet, much of it copyrighted material that was never licensed for the purpose. Songwriters, authors, illustrators and composers argue that their life’s work has effectively become raw material for machines that may one day compete with them. The companies building those machines counter that training is a transformative use, closer to a person reading widely than to copying, and that the sheer scale of modern datasets makes individual licensing impractical.
What the researchers are proposing
The thrust of the Australian framework is to accept that AI developers will keep using large bodies of creative work, while insisting that a payment attaches to that use. Instead of trying to block training outright, which many legal scholars consider unrealistic given how the technology already operates, the model treats access to creative material as something that must be licensed and remunerated, much as radio stations and streaming platforms already pay to play recorded music.
That framing matters because it shifts the conversation away from an all-or-nothing fight over whether AI firms may use copyrighted material at all, and towards the more tractable questions of how much they should pay, who collects the money, and how it is distributed back to the people who made the work. Collective licensing bodies have run exactly this kind of plumbing for decades in music and publishing, which gives the proposal a template to build on rather than a system to invent from scratch.
The appeal of a pooled, pay-per-use approach is that it sidesteps the impossible task of tracking which individual poem or melody shaped a given AI output. Rather than demanding forensic attribution, a levy-style scheme spreads payment across the pool of rights holders whose material fed the system, in proportion to its use. It is a pragmatic answer to a problem that pure copyright litigation has struggled to resolve, because courts in several jurisdictions are still working out whether training even counts as infringement.
Two sides of a hardening debate
Creators and the organisations that represent them have generally welcomed any mechanism that puts money back in their pockets, after watching their work vacuumed up without consent or payment. Their argument is straightforward: if a company profits from a model that could not exist without human creativity, some of that value should flow to the humans involved. A workable payment framework, on this view, is long overdue and preferable to the current situation, in which the default is zero.
The technology industry’s position is more guarded. Developers worry that a mandatory payment regime, if set too high or structured too clumsily, could make it uneconomic to build AI models in a small market like Australia, pushing research and investment offshore to countries with looser rules. They also point to practical headaches around valuation, given that no one has a reliable method for pricing the contribution of any single work to a model trained on billions of items. There is a genuine risk that a scheme designed to protect creators ends up benefiting large collecting societies and well-catalogued rights holders while leaving independent and emerging artists with little.
A third perspective, often voiced by legal academics, is that frameworks of this kind only work if they are backed by law. Voluntary codes tend to be ignored by the companies with the most to lose, so any Australian model would likely need legislative force and an enforcement mechanism to have teeth. Without that, a well-designed scheme remains a paper exercise.
The Australian stakes
Australia has a particular reason to care about getting this right. The country’s creative industries, from its music sector to its screen and publishing businesses, contribute billions to the economy and employ tens of thousands of people, yet they operate in a market small enough that a handful of policy decisions can tip the balance. The copyright debate here has already been heated. When the Productivity Commission floated the idea of a text and data mining exception that would make it easier for AI firms to use copyrighted material, the backlash from musicians, authors and rights groups was immediate, with the music sector in particular warning that such an exception would legalise the mass appropriation of local work.
That context is what makes a payments framework interesting to Australian policymakers. It offers a potential off-ramp from a standoff in which creators demand protection and technology companies demand access, by reframing the argument as one about price rather than permission. For a government that has signalled it wants both a thriving AI sector and a protected cultural one, a model that lets both coexist has obvious political appeal.
There are also sovereignty questions. If Australian creative work is going to be used to train AI systems regardless, many in the sector would rather that use be governed by Australian rules and Australian payments than by the terms of overseas platforms. A homegrown framework, developed by local researchers with local conditions in mind, speaks directly to that concern.
What happens next
A research proposal is not policy, and the distance between the two can be considerable. For the framework to matter it will need to be tested against the practical realities of valuation and distribution, scrutinised by the creative organisations it aims to help, and ultimately picked up by lawmakers willing to translate it into enforceable rules. The federal government‘s broader work on AI regulation, together with ongoing reviews of copyright in the digital age, gives it potential openings.
For now the significance of the proposal is that it moves the conversation forward. After two years in which the debate has too often collapsed into a binary between banning AI training and allowing it for free, a serious attempt to design the machinery of payment is a welcome addition. Whether it becomes the basis of a new deal between Australia’s creators and the companies building on their work, or simply another contribution to a crowded field, will depend on what happens once it leaves the page.
Sources: Limelight magazine.
