Rembrandt vs.‘The next Rembrandt’: who owns the AI content?
For Neil Clarke, the pandemic hit differently than others. Neil, the owner of Clarkesworld Magazine (a Hugo and World Fantasy award-winning magazine), started to notice a lot of spammy submissions to his site. At first, it seemed to be a minor nuisance: plagiarized entries were typical for any literary magazine. But towards the end of 2022, these entries were back with a vengeance.
Interestingly, this phenomenon coincided with AI chatbots like ChatGPT - one that could 'generate' a story from prompts in mere minutes - gaining mainstream recognition. More and more videos were emerging encouraging users to make a quick buck as a 'side hustle' using tools like these.
Read about the top 5 AI tools that can help you enhance your content creation process.
On 15th February, Neil reported that submissions to Clarkesworld had more than doubled in the past 15 days. Permanent banning that resulted from submitting plagiarized content also reached an all-time high of 38%.
Five days later, the volume of submissions had increased so much that Clarkesworld announced they were closing submissions temporarily.
“It’s clear that business as usual won’t be sustainable,” Neil wrote in his blog post, “I worry that this path will lead to an increased number of barriers for new and international authors.”
Neil wasn’t alone who was venting out against AI-generated content. Last year, a person named Jason Allen won the first prize in the digital category at the Colorado State Fair, creating a furor online. Midjourney, an AI software that translates text prompts into images, created the artwork "Théâtre D'opéra Spatial".
To quote Matthew Butterick, a lawyer working with plaintiffs in the GitHub Copilot lawsuit, wrote in a blog post that he had "heard from people all over the world who are concerned about AI systems being trained on vast amounts of copyrighted work with no consent, no credit, and no compensation."
This provokes multiple questions - should we consider people like Allen, who use computer algorithms or programs to create content, as 'artists'? Allen says that he had spent weeks perfecting the image prompts and even editing the final piece - suggesting a high human intellectual involvement.
But does that make these creations ‘normal’? Should we consider objects created by other humans as art?
Of late, there has been an explosion of AI-powered selfie tools like Facetune, Lensa, and DrawAnyone. These can recreate your picture in several pre-trained styles by scraping the artist's work over the web - almost always without their explicit permission.
Generative AI requires datasets to analyze and create content. But is it legal to use copyrighted materials as 'sources' to train these tools? Let’s find out.
The un-Stable Diffusion: AI content and their copyright
Hollie Mengert wasn't prepared for the shock. On one fateful November 2022 weekend, she woke up to several emails, tweets, and messages.
All of them pointed to a Reddit thread where someone had posted images that mimicked her artistic style. The Redditor, MysteryInc152, fed 32 of Hollie’s artworks to Stable Diffusion, a text-to-image AI tool - and fine-tuned the outputs to recreate her style.
He then released the AI model that uses Hollie’s name as a prompt to generate images that copy her style of work - for public use under an open license.
As expected, a ruckus ensued in the comments section of that Reddit thread. Most poignantly, someone had asked, “Whether it’s legal or not, how do you think this artist feels now that thousands of people can copy her style of works almost exactly?”
When asked this, Hollie - an LA-based illustrator, told technology blogger and reporter Andy Baio, that she initially felt “invasive that my name was on this tool, I didn’t know anything about it and wasn’t asked about it.”
Would she have given her permission if the Redditor had asked? “If I had been asked if they could do this, I wouldn’t have said yes.”
But there’s a catch. Ogbogu Kalu, a Redditor, used many copyrighted images from clients such as Disney and Penguin Random House to train the AI model. “So even if he had asked me and said, can I use these?” Hollie continued, “I couldn’t have told him yes to those.”
However, Ogbogu thinks that people object because they do not understand how generative AI models work.
He remains optimistic that his and future use of copyrighted content comes under fair use. “It is transformative,” he said, like “trying to recall a vivid memory from your past.”
In this case, who owns any image created using this open-source AI model?
Organizations like OpenAI have expressed that training AI models on copyrighted data sets is fair use. Many AI startups, researchers, and organizations share the view that the fair use doctrine in the US allows the usage of copyright-protected data to promote freedom of expression.
Does Rembrandt own ‘The Next Rembrandt?”
Kris Kastanova wants to build a safe AI community. So, in 2022, she filed a copyright claim to the US Copyright Office (USCO) for her comic book. Images in "Zarya of the Dawn", an 18-page story, were exclusively created by feeding text prompts to Midjourney, an AI image generator tool.
In September 2022, the USCO granted copyright registration to Kris. At the time, people hailed this as a landmark for copyrighting AI-generated content, at least in the US.
But this was not to be. Earlier this year, the USCO partially rescinded its original copyright grant, stating that the images “are not the product of human authorship.”
Kris, however, retained full rights to the texts, as she “is the author of the Work’s text as well as the selection, coordination, and arrangement of the Work’s written and visual elements.”
The USCO refused copyright to an artwork by Stephen Thaler, named 'A Recent Entrance to Paradise' a second time in February 2022.
Thaler, who first filed for copyright back in 2018, claimed a ‘creativity machine’ created this artwork. He was “seeking to register this computer-generated work as a work-for-hire to the owner of the Creativity Machine.”
USCO, in its first denial, cited that Stephen’s claim “lacks the human authorship necessary to support a copyright claim.”
USCO has cited previous cases (including the infamous PETA monkey case), and the case law (e.g. Feist Publications v Rural Telephone Service Company, Inc. 499 U.S. 340 (1991)) that specifies only “the fruits of intellectual labor” that “are founded in the creative powers of the mind.” are protected by the copyright law.
This is in line with their 2017 announcement that they will only “register an original work of authorship, provided that the work was created by a human being.”
Who should own AI-generated content isn’t 100% clear at the moment, but we seem to be heading towards a situation where lawmakers will have to either define a new law or amend the existing copyright laws to address this.
Van Lindberg, Kris’ lawyer, agrees. “AI-assisted art is going to need to be treated like photography. It is just a matter of time,” he says.
There seem to be two ways of dealing with the copyright question. Countries like the US can continue to deny copyrights to any AI-generated content. But that has a butterfly effect. It dissuades people from using the tool and ends up hampering research and development of AI-based tools in general.
Considering the extraordinary potential that tools like ChatGPT and DreamBooth have, this would be a coup de grâce.
Alternatively, people could treat AI-generated content the same way as human-generated content in the future. Copyright could then be either owned by the content creator or shared between the content creator and the owner of the images used to train the AI tool.
Some, like Dr. Andrés Guadamuz, Senior Lecturer in Intellectual Property Law, University of Sussex, suggest that there may be another way out: by granting authorship of AI-generated content to the program’s creator.
As technology evolves, our perspective on laws surrounding those evolutions need to be progressive.
Art imitates Life. AI imitates Art. Where do we go from here?
The fight for copyrighting AI-generated work stems from two contentious areas. One is whether people can consider what a computer program creates as 'art'. The second is whether someone trained an AI model on the copyrighted data set(s).
Should the person or company who owns that base training data, own the content generated by that model?
Answering these questions may be simple. AI tools have made it easy to create content, but that does not mean it is easy to perfect the output. Often, to generate the desired output, the creators spend hundreds of hours, feeding the AI model thousands, (and in some cases, millions) units of data.
As a result, the content generated by these AI models carries subtle differences from their 'source materials'. Whether people write material using ChatGPT or create images using Stable Diffusion, one can argue that these AI tools generate uniquely personalized content due to their differences in input, both prompts and data, by the user or creator.
Hopefully, with more research and development poured into these AI tools, these user-specific outputs will distinguish artists using the same AI models from each other.
Generative AIs harness unprecedented power. People are already using AI tools to synthesize drugs, design chips, analyze proteins, and more. But, as with all emerging technology with extraordinary potential, there will always be bad players with the intent of causing inconceivable harm. That shouldn't distract us from reality.
We are on the verge of something incredible.
Beyond this, Rembrandt's self-portrait and The Next Rembrandt can coexist side by side, and be admired by everyone.