You could get compensated for what you’ve posted online – as long as it’s been used to train AI.
At least that’s according to Dr Margaret Mitchel, chief ethics scientist at US open source AI company Hugging Face, who’s calling for AI companies to use their technology to trace generative AI content back to the original creators.
“There’s a lot of content that’s being taken from creators – including artists and authors, but also general people with what we write online – who aren't being compensated for that work,” she said, speaking at AI Everything in Cairo, Egypt.
“I envision a future where we’re able to actually attribute and identify who was in the input space that’s made it possible [for a model] to create its outputs, and compensate them.”
Generative AI has relied more heavily on some creators than others, and there’s often a clear, direct link between input and output – with distinct writing styles or the signature of an artist identifiable in some AI-generated works.
Recently, Japanese animator and filmmaker Hayao Miyazaki levelled criticism at this process, which pumped out images copying his signature Studio Ghibli style.
But it’s not just musicians and artists whose work gets used – after all, large language models (LLMs) like ChatGPT and Google Gemini have been trained on a vast bank of online data
“We’re all creators,” Mitchell told BBC Science Focus, saying a compensation model should therefore take into account contributions from any online user. Your experiments with poetry, for example, or that photo of a sunset from your holiday five years ago.
Fortunately, she added, there are potential models that can trace output to input, and compensate those creators based on the amount of input they have provided.
Yet no system like this currently exists, and the funding to create one is not possible within current AI business models, Mitchell said, though noting that some AI companies have started to investigate how it would work.
For example, in 2021 documents from AI company Anthropic (which were recently unsealed with a court order), CEO Dario Amodei proposed his “crazy idea” for such a model, theorising that the payment distribution system could work similarly to monetisation platform Patreon’s.
Existing technological approaches could be adapted to make this possible – even for LLMs that have already been built, Mitchell said. For example, clustering algorithms (a machine learning technique that groups data points) can help trace similarities and work out authorship attribution.
To ensure privacy, the model would require consent from online users, so that you could opt in to being associated with your data (and then be attributed and compensated if used), or choose to stay anonymous.
“You have to open the door for this kind of research in order for there to be innovation,” Mitchell said. “It's not really open now.”
Read more:
