I am quoted several times in this article (in Danish) by Peter Svarre on Artificial Intelligence and copyrights. Who owns the AI images?
Here are some of the quotes in my translation from English to English:
And that brings us back to the question of who actually owns these AI-created images. Or actually, there are three questions:
- First, one might ask whether it’s even morally or legally okay to harvest billions of images from the internet, use them to train an artificial intelligence, and then make money selling access to that artificial intelligence.
- Second, many of those who have tried the systems are probably wondering whether they actually own the copyright to the images they have created.
- And finally, some companies are probably concerned that they may be infringing the copyright of others if they use images created by an artificial intelligence.
WHERE DO THE IMAGES COME FROM?
Our new creative helpers have become creative because they’ve been trained on billions of images from the internet. And it doesn’t matter which images we’re talking about. Some images are protected by copyright and can only be used to train an AI if the owner of the image has given explicit permission. Other images, however, are in the public domain or covered by Creative Commons licences, which allow a web crawler to “scrape” them to train an algorithm.
“There’s no human’s consciousness or creativity in these machines,” says Martin von Haller Grønbæk, a lawyer and digital copyright expert. “The artificial intelligences are trained on images, and if there were no images at the beginning that ordinary people had made, then nothing came out the other side. And so it’s debatable whether there’s copyright infringement when Dall-E and the others process these images. And there would be, for example, if they hacked a database and copied all the images. But if they are based on publicly available databases, then they have permission to use the images to train artificial intelligence, and then there is no copyright infringement.”
It is therefore also interesting that artists around the internet have started to criticise these systems because they can see how their own (copyrighted) images have very likely been used to train the artificial intelligences.
There are examples of images where the artist’s signature has been preserved and is being recreated when creating images in that artist’s style. And stories have been told of people who can recognise almost 1:1 a photograph they have taken. It’s not exactly identical, but so close to the original that it seems unlikely that the original hasn’t been used to train artificial intelligence.
The problem, of course, is that the original images are no longer present inside the artificial intelligences, so it can be incredibly difficult to prove outright copyright infringement. But if, down the road, cases emerge where it is quite clear that the artificial intelligences have been trained on copyrighted material (for example from Disney, Heinz or others with a strong interest in preserving their copyright), Martin von Haller Grønbæk could well imagine that legal action would be taken against the companies that developed the artificial intelligences.
DO YOU OWN THE IMAGES YOU’VE CREATED WITH AN ARTIFICIAL INTELLIGENCE?
I’ve had a paid subscription to Midjourney for a little over a month and have generated a couple of hundred images, and all these images feel like my images. I may have only provided a few words while the artificial intelligence did all the rough work, but they still feel like my unique creations. And I’m confirmed if I go and read Midjourney’s Terms of Service, which says: “Subject to the above license, you own all Assets you create with the Services.”
But that promised ownership is bullshit, say both von Haller and Jonathan Løw. You can’t own something created by a machine. Indeed, copyright can only be granted to humans, and a machine cannot create a work and then assign copyright to a human.
In the US, there have been several cases where artists have created generative art using artificial intelligence, and in almost all cases the US Copyright Office refuses to grant the artists copyright in the work. A few years ago there was even a case where a monkey had taken a selfie and this photograph could not be copyrighted either because it was not taken by a human.
WHO OWNS THE IMAGE?
Martin von Haller has heard the question many times before when it comes to AI-created images: ‘Who owns the image?’ he asks, continuing: ‘I guess you’d have to say that nobody does, because it’s generated by an AI, and so under copyright law nobody can get the copyright. But if the person using the AI service, or the service itself, can be said to have a verifiable creative part in the creation of the individual AI-generated work, then you can start talking about independent copyright protection. But this requires a thorough understanding of the AI service in question, the underlying algorithm and the user’s interaction with it, for example by establishing criteria and parameters.”
In other words, it might get a little grey in the grey, but most lawyers can agree that you don’t own the copyright to the images you create in Midjourney or Dall-E because your creative efforts are simply too small. Therefore, if you choose to use your images commercially, other people are free to use the images for other commercial purposes because they are not protected by copyright. This is worth bearing in mind before you commission Midjourney to design your new logo or corporate visual identity.
The problem, of course, is that it quickly becomes a bit unclear when something is created by a machine and when it is created by a human. You can already see examples of designers using AI-created images for inspiration, which they then work on to design a shoe, a handbag or a piece of art. Here, artificial intelligence is used only as a working tool, and humans provide the essential creativity required for copyright.
In the future, we will see more and more creative work tools such as Photoshop, Figma and CAD programs that have built-in artificial intelligences that can generate designs as an integral part of the human creative process, and when is the design created by a human or a machine?
There is much to suggest that copyright law will need to be updated or at least clarified in a world where humans and intelligent machines are increasingly merging.
Here, however, von Haller is more sceptical. He believes it will be really difficult for rights holders to prove that images created by an artificial intelligence are a direct copy of another work. It has to be really close to the original image to be considered a copyright infringement. But it certainly can’t be ruled out, he says. For example, if you find that a specific AI-generated image is so close to a copyrighted image that you think the former is a derivative work of the latter. However, he believes that companies in particular should be aware of what he calls reputational risk, the risk of being dragged through a shitstorm on social media because a big bad company created a million-dollar business on an AI-modified design of, say, a vase or a coffee pot. Just think of Christian Bitz or Kenneth Plummer.
In other words, you can probably use your AI-created images on social media, in articles and blog posts, but as soon as you start using your images commercially, you should think twice and really take the same considerations you would in any other copyright matter.
SCALE IS EVERYTHING
From a legal standpoint, there’s really not much new about creative artificial intelligences. Humans have always used machines of various kinds to create creative works (passports, rulers, Photoshop, CAD programs and so on). Many of the doubts discussed in this article could therefore be dealt with by existing legislation and case law. As expressed in an article in Wired, only directly man-made things can confer rights. Machines CANNOT be copyrighted and cannot seriously in a legal sense be creative, until now. The big problem is that generative AI art questions what creativity is and where the line is for human-created copyrighted creativity. What about non-human creative geniuses like Midjourney and Dall-E?
“THE PURPOSE OF copyright, per the Constitution, is to promote the progress of art. The dominant justification for intellectual property among US legal scholars today is the incentive theory, the idea that in granting authors a limited monopoly over their work, we stimulate artistic production by dangling a monetary reward. Without the promise of a payday, the theory runs, creative people would stop creating. But US copyright law does a much better job of accommodating the old model of the lonely genius toiling in solitude than it does collaboration, even collaboration between human authors. When a collaboration extends to include nonhuman authors, the law recoils.”
However, there is every indication that the new creative artificial intelligences will become ubiquitous in our everyday and working lives. In the future, we may create our own personalised family videos, feature films or 3D meta-verses using artificial intelligence. And design companies, advertising agencies and architects will have to work in entirely new ways. As these tools become bigger, better and far more widespread, we will see industries challenged and new industries emerge, and therefore we will certainly see court cases challenging the use and creations of creative artificial intelligences. It will be Dall-E versus man, and the verdict will require a new creative legal understanding of human creativity.