I know screenwriters, who, with a few prompts, ask Chat-GPT to create a bunch of storylines. They say: ”Sometimes AI gives good ideas for the scripts I’m working on.” I’ve had arts students, who make sketches for their oil paintings with the help of AI. I know poets, electronic musicians and classical composers, who use AI to generate stuff from which they can pick inspiring elements for their work.
I’ve used it in my work too: we took photos of a middle-aged opera singer and asked AI to use the photos to create a 5-minute video in which the singer ages from early childhood to old age. It looked very authentic — the singer said so herself. I used it in an opera I directed.
In this sense AI is merely a tool.
Many artists believe that’s all it is, just another harmless gadget, like a sophisticated piece of audio, visual and textual software. They are wrong.
AI is also a thief.
Generative AI — the kind of AI that can create new texts, sounds and images — learns to mimic human creativity by training itself with huge amounts of human data: texts, pictures, video, sounds and music. The core of ChatGPT and other generative AI systems is called foundation model. There are many of them now tirelessly trawling (or crawling, as the term now seems to be) internet to learn more from us and of us.
The corporations who own the foundation models, claim that there is no harm done and that no human should get paid for this, since the AI is usually not downloading anything, it is ”merely browsing the internet” and when it generates something, it’s not making illegal copies, but creating something new.
Artists should not accept that view. Instead, we, the artists, must fight for our rights. We must demand that we can opt out: if we don’t want our works to be used as raw material — so-called training material — for generative AI, we should be able to say no. And that should be enough to keep the AI trawler at bay.
Some foundation models already offer the opt out option. But can it actually work? If I opt out and AI then keeps away from my website, what happens when the audience takes pictures of my works in exhibitions and posts those online?
Anyway, if we don’t want to opt out or can’t opt out, we need to get paid. That’s perhaps the most important demand: without us, the human artists, generative AI could not make anything, because at the moment it’s still very dumb. It doesn’t think in the sense we think and it has no consciousness. It’s able to produce convincing texts, but it doesn’t know the meaning of a single word. It’s just a hugely powerful mechanical parrot, which can create an endless amount of versions and mixtures — in a split second — of all the things it has read, heard and seen.
We feed the generative AI with our artworks. How could we get paid for this?
Our opponents say that we can’t, that it’s not possible to interpret the copyright laws in such a way that we would be entitled for any financial compensation for feeding the AI.
The AI Act that EU is preparing says that for the sake of transparency the foundation model owners must ”publish summaries of copyrighted data used for training”. There’s no mention of paying for the use of copyrighted material. However, we can still raise our voice and demand more. It’s great that Initiative Urheberrecht (Authors’ Rights Initiative) is putting pressure on EU through their: Authors and Performers Call for Safeguards Around Generative AI in the European AI Act.
Public Lending Rights as a model offer an encouraging example of how this could play out. People used to think that writers should be just grateful for the public library system and not ask for any money for the free loans of their books. However, in 1919 the Nordic Authors’ Association passed a resolution calling on governments to compensate authors for library lending of their books. EU said yes to it in 1992 and it has now become a significant source of income for many writers. In some but not all countries PLR payments are based on copyright law. What counts is how people and their governments changed their minds about what’s fair. The fact that this kind of turnaround was possible in how public libraries function in regard to writers and other ”content creators” proves that real and positive upheavals are positive in also other matters related to artists’ income rights.
How could the AI-based payments to artists be collected and paid? There are at least two basic models: pay-per-use model and generative AI tax model.
For the pay-per-use model we would need a sophisticated system that would keep track of what artworks and how much the generative AI actually uses. Artists would then be paid accordingly, most popular ones getting a lot and others less. Of course, it would be difficult to create a system that could keep track on how much the foundation models use individual artworks, but surely, it’s possible. If just a couple of programmers were able to create the Have I been Trained? service, with which you can check whether your works have been used or not by one of the biggest foundation models, it must be possible to create a system that measures also how much your works have been used. Artists’ organizations could then collectively bargain with the foundation model owners what would be the fair prices for AI’s use of human artworks.
The generative AI tax model means that a simple tax could be imposed on the use of generative AI and the funds collected this way would be distributed to human artists. Of course, many forms of generative AI, such as ChatGPT, are now seemingly free for users, but a lot of money is constantly invested in the maintenance and development of the foundation models. These investments, the money spent on generative AI, could be taxed.
What would the state do with the funds collected by generative AI tax?
There are many options. The money could be just shared evenly so that each professional artist would get the same amount. Or each artist would be paid according to how much the generative AI has used the artist’s works (a version of the pay-per-use model). Or the money could be collected into a fund that could give grants to artists. Or the state could get the money and bolster its cultural budget with it. Or the money could be channeled into artists’ social security system.
Our need to get paid for the AI use of our works is not just a matter of moral principle. It’s also a matter of practice. It’s already common for book publishers to use generative AI instead of human graphic designer to create book covers. It won’t stop there. Wouldn’t it be convenient for the music streaming giants if they could offer AI-generated, tailor-made music for each user — without any input from human music makers? Wouldn’t it be convenient for the gallerists if instead of human artists they had a row of AI-generated ”authors” in their stable?
All the above is relevant when we try to deal with the current, dumb generative AI.
But truly intelligent, Artificial General Intelligence may be just around the corner too. It will be more than a tool and more than a thief. It will be our colleague. It won’t just randomly mimic the poetry I write. It’ll have its own views and aspirations and it’ll write its own poems. I’ll be happy to read them. And talk with it about its rights. But right now, my main task is to fight so that I’m not exploited by its mindless precursor.
Helsinki, 24.10.2023,
TEEMU MÄKI
Artist / Director / Writer / Researcher, (Doctor of Fine Arts), Chairperson of The Artists’ Association of Finland, President of IAA Europe.
Links
Mäki, Teemu: ”AI Is Coming — Who Is Ready?” Taiteilija magazine, 22.5.2023.
www.artists.fi/fi/ai-coming-who-ready
https://teemumaki.com/theater-posthuman.html
EU AI Act: first regulation on artificial intelligence, www.europarl.europa.eu/news
Ruf nach Schutz vor generativer KI, https://urheber.info/about-us
