AI art has muddied intellectual property rights

Beyond the digital space, an entire industry is getting upended. Visual art alone is supported by artists, curators, art conservators, critics, art collectors, auctioneers and so forth. All of them need to rethink their future role

The battle between creators and generative-AI companies has only just begun. Getty Images, the stock images company, has sued AI art generator Stability AI for allegedly illegally copying the former’s library of images for commercial interests. Generative-AI companies have been using publicly available image libraries to train their AI model. Many of the images generated by Stability AI’s text-to-image model has Getty’s watermark.

In January 2023, three artists filed a class-action lawsuit against Midjourney, Stability AI, and DeviantArt for scraping billions of images from the Internet without the consent of the creators. Copyright infringement is one of the many problems with the messy business of AI art, whether it is visual art, music, or literature.

But we are still not sure if AI art is art after all. ChatGPT, the clever AI chatbot at the epicenter of the rising generative-AI movement, reckons that “the question of whether AI-generated art can truly be considered ‘art’ is a matter of debate’. Yet, in 2018, Christie’s sold an AI-generated portrait picture rendered in an 18th-century oil-on-canvas style for US$432,500.

For centuries, we have been trying to work out what amounts to a work of art. In Eugene Veron’s 1873 book, Aesthetics, art is defined as the manifestation of an emotion. When I laid eyes on the Water Lily Pond by Monet at London’s National Gallery, the feeling was visceral. A great work of art emotionally connects us with the artist’s expression.

In 1898, American sculptor and art critic, Frederick Ruckstuhl had testified on an architect’s case involving an imported church altar that had attracted customs duty. American customs officials did not consider it a work of art. Ruckstuhl defended the case by asserting that a work of “art is human work made …with the purpose of expressing, or stirring, human emotion”. The architect went on to win the case and the US government accepted Ruckstuhl’s definition of art. AI art will eventually appeal to us at an emotional level. ChatGPT agrees that art is about human creativity and imagination.

Art was never meant to be neatly demarcated. Pop art, minimalism, and stop motion animation have different effects on our sensibilities. Perhaps art will keep changing, just as cave art had its brush with time. History proves that there will always be a market for the more enduring art.

Digital tools have extended human creativity and embodied our imagination. With each new online hack, art gets diversified and commoditised, thus giving rise to millions of creators. Many creators feel that their livelihood is threatened by AI art, especially when their work is used for AI training.

Beyond the digital space, an entire industry is getting upended. Visual art alone is supported by artists, curators, art conservators, critics, art collectors, auctioneers and so forth. All of them need to rethink their future role.

There are concerns about our inability to distinguish between art produced by humans and machines, just as scientists are finding it difficult to determine if a scientific paper’s abstract was written by a human or AI. But do we really need to make that distinction if AI art appeals to us? Yes, especially when authorship is linked to commercial interests.

Intellectual property rights (IPR) of creators need to be protected in this new AI paradigm. Can copyrighted material be used to train an algorithm? We do not have a clear answer yet. Regulators are only starting to wade through the copyright laws, muddied by generative-AI. We also need a consensus on the copyrights of AI art. Many courts have ruled that AI-generated art cannot be copyrighted because the artwork was produced autonomously by an algorithm, without human intervention. The machine made its own decisions while creating the artwork.

Yet, no machine has absolute autonomy, yet. So, they do require human intervention such as the choice of training data for the AI model, which helped generate an artwork. Even with unsupervised training, some amount of human oversight is needed. Software development and the training data used to produce a specific artwork could be deemed as human intervention. Then again, the training data even if synthetic, needs some copyright material from other creators. Therefore, the doctrine of human authorship in the case of generative-AI is unclear. The European Union (EU) makes a distinction between AI-assisted and AI-generated outputs. The former qualifies as ‘work’ and hence can be protected under the EU copyright law.

Contractual agreements between all parties involved, for authorship or co-authorship of AI art may help resolve future legal wrangles among the parties. But this would not cover third-party disputes. Creators, artists, and their lawyers are still learning about AI art. The wording of the artists’ class action lawsuit is technically flawed, giving AI companies room to defend themselves. Creators need to fully understand the AI training process before taking any action. There will be many more lawsuits while the IPR regime in different countries are reevaluated in the light of generative-AI.

(Shalini Verma is a serial entrepreneur. She tweets @shaliniverma1)

• Opinion