An open letter from the Authors Guild, endorsed by literary heavyweights like Margaret Atwood and Jodi Picoult, has demanded that AI firms refrain from utilising their creations without adequate compensation or permission.
Generative AI has begun its ascent into mainstream tech, but the revenue model for artists remains shrouded in ambiguity.
With tech giants diving into monetising generative AI, the original creators, upon whose content these AI models are trained, are clamouring for a piece of the pie.
However, there is a discernible lack of consensus on how much, if any, remuneration is due to the artists whose work makes the technology possible.
An open letter from the Authors Guild, endorsed by literary heavyweights like Margaret Atwood and Jodi Picoult, has demanded that AI firms refrain from utilising their creations without adequate compensation or permission.
This stance is further solidified by numerous legal actions initiated by artists against prominent AI players, including Microsoft and Stability AI, highlighting issues of copyright infringement.
In a bid to placate the disgruntled creators, some AI vendors have expressed intentions to set up “creators’ funds” or similar mechanisms to pay artists whose content they utilise.
Several have set those plans in motion, proclaiming this as a stride towards a more equitable generative AI business model. The glaring question remains: What can artists realistically expect from these funds?
The foundational premise of generative AI rests on training the model using vast quantities of examples, typically sourced online.
These examples, which range from photographs to texts, often come with copyright or usage licence tags, which vendors occasionally overlook.
While some firms advocate their actions under the “fair use” doctrine prevalent in the US, it's an area of contention that awaits legal clarity.
Despite the legal labyrinth, the winds of public sentiment seem to be favouring artists.
A handful of tech behemoths, including Adobe and Getty Images, have rolled out or pledged profit-sharing mechanisms with creators. But the opacity regarding the actual earnings remains.
Case in point: Adobe’s generative AI model, "Firefly," is trained on Adobe Stock images. The company proposes an annual "bonus" for contributors.
The bonus structure is contingent on the number of accepted images and the number of licences these images generate over a year.
However, the specifics, such as the exact value of each image or licence, are yet to be disclosed. Getty Images, on a similar note, has remained tight-lipped about its remuneration specifics for its generative AI tool.
Shutterstock, a rival of Getty Images, has adopted a different approach. Through its "Contributors Fund", Shutterstock ensures periodic payments to creators, based on their contribution and if their content finds itself in the output of Shutterstock’s AI tools.
But again, the specifics are obscure. An independent survey by stock photographer Robert Kneschke estimates that an average payout from this fund might be around $0.0078 per image.
Stability AI’s venture, "Stable Audio", a sound-generating model, promises musicians a slice of its profits. However, the exact distribution model remains under discussion.
YouTube, which recently partnered with Universal Music Group for a generative AI initiative, has hinted at compensation structures for music rights holders but remains in the nascent stages of fleshing out these business models.
The crux of the matter is the conspicuous absence of concrete figures on potential earnings for creators. While some vendors cite the novelty of the technology as a reason, others contend that the revenue range would be too diverse to pinpoint.
Such arguments, however, might not resonate with creators, many of whom rely on consistent income.
Notably, some startups are aiming for a more transparent approach.
For instance, Braia, which trains its AI on licensed images, offers a clear-cut revenue sharing model that allows artists to determine prices for each AI training run. Yet, the overarching sentiment suggests that most vendors are yet to make a convincing case to artists about the financial viability of participating in generative AI model training.
The allure of future rewards remains nebulous at best, offering little solace to artists striving to make ends meet.
*James Browning is a freelance tech writer and music journalist.
BUSINESS REPORT