In 2022, an AI-generated artwork wins a Colorado State Fair art competition. The artist, Jason Allen, used the Midjourney system – a generative AI system trained on art from Internet – to create the widget. The process was not fully automated: Allen went through about 900 iterations over 80 hours to create and refine his message.
However, his use of artificial intelligence to win the art competition sparked a backlash on the Internet, with a Twitter user promptingWe are watching the death of art unfold before our eyes.
As generative AI technical tools like Midjourney and Stable Diffusion are highlighted, so do you have questions about ownership and authorship.
The generative ability of these tools is the result of their training on dozens of previous artworks, through which the AI learns how to create an artistic output.
Should artists who scrape their art be compensated to train models? Who owns the images produced by AI systems? The process of fine-tuning the claims of generative AI is a form of Authentic creative expression?
from direction Rave techies To work like Allen. But on the other hand, many working artists consider using their art to train artificial intelligence Exploitative.
We are part of a team of 14 experts from various disciplines who have just published a paper on generative AI in the journal Science. In it, we explore how to advance artificial intelligence It will influence creative work, aesthetics, and the media. One of the main questions that has arisen is related US copyright lawsand whether they can adequately deal with the unique challenges of generative AI.
Copyright laws are designed to promote the arts and creative thinking. But the advent of generative AI has complicated current notions of authorship.
Photography serves as a useful lens
Generative AI may seem unprecedented, but history can serve as a guide.
Take The emergence of photography in the nineteenth century. Before its invention, artists could only attempt to depict the world through drawing, painting, or sculpture. Suddenly, reality can be captured in a flash using a camera and chemicals.
As with generative AI, many have argued that photography lacks artistic merit. In 1884, A.P The US Supreme Court considered this case And I found that cameras were tools that the artist could use to give the idea of visual form; The court ruled that the “masterminds” behind the cameras must own the images they create.
From then on, photography developed into its own art form and even took off New abstract art movements.
The AI cannot own the outputs
Unlike inanimate cameras, AI has capabilities — such as the ability to turn basic instructions into exquisite works of art — that make it capable of making this happen. susceptible to anthropomorphism. Even the term “artificial intelligence” encourages people to believe that these systems have human intentions or even self-awareness.
This has led some people to wonder if AI systems can be “owners”. But the US Copyright Office has stated this unequivocally Only humans can own copyright.
So who can claim ownership of AI-generated images? Are they the artists whose images were used to train the systems? Users who write requests to create images? Or the people who build AI systems?
Infringement or fair use?
While artists indirectly rely on past works that have taught and inspired them to create, generative AI relies on training data to produce output.
This Training Data consists of prior artwork, much of which is protected by copyright law and which was collected without the artists’ knowledge or consent. Using art in this way may violate copyright law even before the AI generates a new work.
Was that a form of trespass? Or is it a new form offair useA legal doctrine that permits the unauthorized use of a protected work if it has been sufficiently transformed into something new?
While AI systems do not have verbatim copies of the training data, they do He sometimes manages to recreate works of training data, which complicates this legal analysis.
Does contemporary copyright law favor end users and companies over artists whose content is in training data?
To alleviate this concern, some scholars are proposing new regulations to protect and compensate artists whose work is used for training. These proposals include the right of artists to Opt out of using their data For generative artificial intelligence or a method Compensate artists automatically When their work is used to train artificial intelligence.
However, training data is only part of the process. Artists using generative AI tools frequently go through many rounds of revision to improve their claims, which indicates a degree of originality.
Answering the question of who should own the output requires consideration of the contributions of all participants in the generative AI supply chain.
Legal analysis is easier when the output is different from the business in the training data. In this case, the person who asked the AI to produce the output appears to be the default owner.
However, copyright law requires meaningful creative input – a standard that is met by clicking the shutter button on the camera. It remains unclear how the courts will decide what this means for the use of generative AI. Is authoring and proofreading enough?
Things are more complicated when the outputs are similar to the business in the training data. If the similarity is based solely on general style or content, it is less likely to infringe copyrights, because style is not protected by copyright.
Illustrator Holly Mengert faced this problem firsthand when her unique style was imitated by generative AI engines in a way that didn’t capture what was in her eyes, making her work unique. Meanwhile, singer Grimes embraced the technology, “opening up” her voice and encouraging fans to create songs. In its style using generative artificial intelligence.
If a director includes key elements of a work in the training data, he or she may be infringing the copyright of that work. Recently, the Supreme Court ruled that Andy Warhol painted a portrait Not allowed by fair use. This means that simply using AI to change the style of work—say, from a photo to an illustration—is not enough to claim ownership of the modified output.
While copyright law tends to favor an all-or-nothing approach, scholars at Harvard Law School have proposed new models for joint ownership Allows artists to obtain some rights to outputs that are similar to their work.
In many ways, generative AI is another creative tool that allows a new group of people access to image making, just like cameras and paintbrushes or Adobe Photoshop. The main difference, however, is that this new set of tools is based explicitly on training data, and thus creative contributions cannot easily be traced back to a single artist.
The ways in which existing laws are interpreted or reformed — and whether generative AI is appropriately treated as a tool — will have real consequences for the future of creative expression.
