This couple launches an organization to protect artists in the AI ​​era

Want a surefire way to get AI image generators like DALL-E to create captivating images? Ask the system to imitate an artist.

It is something that the internet has quickly taken up, knowing that the technique gives the best possible benefit. By doing so, it is possible to travel in time to see how e.g. Donald Trump would be portrayed by Picasso or Van Gogh or even a prehistoric cave painter.

Such “X in the style of Y” prompts work so well in large part because the volumes of data used to train AIs like DALL-E, Midjourney, and WOMBO are pulled from the Internet, which is often populated by copyright protected images. The legality of using this training data is disputed in the European Union; in the United States, much of it is widely believed to be permissible under the fair use doctrine.

What is legal and what is moral are different questions – this is where artists Mat Dryhurst, an academic, and Holly Herndon, a renowned musician, come in. The Berlin-based couple are AI veterans: Around 2016, they started training neural networks, later building such networks themselves. One project, Holly+, launched last year, allows anyone to upload a polyphonic track that can then be “sung” by a deeply faked version of Herndon’s voice created using AI generative tools.

The pair are very careful about what training data they use. “We ended up making the decision that we wanted to train our machine learning systems on data that only came from us or people who gave consent,” Dryhurst says.

In keeping with this thinking, Dryhurst and Herndon are developing a standard they call Source+, which is designed as a way to allow artists to opt in and out — or opt out — of allowing their work to be used as training data for AI. (The standard will cover not only visual artists, but also musicians and writers.) They hope that developers of AI generators will recognize and respect the wishes of artists whose work could be used to train such generative tools.

Source+ (now in beta) is a product of the organization Spawning – a partnership between the couple and Jordan Meyer and Patrick Hoepner, founders of the WolfBear software studio. Officially launching today, Spawning also developed Have I Been Trained, a site that lets artists see if their work is among the 5.8 billion images in the Laion-5b dataset used to train Stable Diffusion and Midjourney The AI ​​generators. The team plans to add more training datasets to go through in the future.

The entire project is designed to empower artists at no cost to them. “The advantage of working with us is that we can serve or withdraw your data for all services on request, as opposed to hunting down individual organisations,” says Dryhurst.

Dryhurst says Spawning has been in contact with the developers of some of the most popular AI tools and received sympathetic responses. “I’m very optimistic that if we can establish a verified database of opt-in, opt-out wishes from artists, then we can honor those wishes,” he says. “That’s the basic foundation on which a lot of good can come from these tools.”

Project goal

Dryhurst and Herndon chose now to launch their project — and start the discussion about ownership of data that trains AI — because of the huge public response to generative tools in recent months. “Incredible imaging systems like DALL-E have been instrumental in bringing this conversation mainstream, so now is a good time to intervene,” Herndon says via email.

Herndon adds, “It’s also important to present a different narrative of how spawning is different from the 20th-century practice of sampling, so as not to derail a productive discussion about how we can deal with this new terrain fairly and with excitement.”

Spawning and sampling, the artists argue, are two entirely different things. “Growing is a more reproductive process,” says Dryhurst. Rather than taking an element and remixing it to create new art, spawning is creating new artwork from a training data corpus.

The project does not aim to prevent people from setting e.g. “A McDonalds restaurant in the style of Rembrandt” into DALL-E and look at the wonder produced. “Rembrandt is dead,” says Dryhurst, “and Rembrandt, one might argue, is so canonized that his work has crossed the line of extreme consequence by creating in their image.” He is more concerned about AI image generators affecting the rights of living mid-career artists who have developed their own distinctive style.

What Dryhurst does not want to do is create a third-party police force. “We’re not looking to build tools for DMCA takedowns and copyright hell,” he says. “That’s not what we’re going for, and I don’t even think it would work.”

He also believes — contrary to what AI companies may fear — that artists will be more willing to accept their work being used than you might think. “I believe that ultimately more people will opt out than opt out, but first we need to establish a mutual respect,” says Dryhurst. “A lot of good will come from getting everyone on the same page.”