Artificial Piracy: Digital Art, AI, and Ownership

The recent popularity of Lensa AI, an app that creates digital portraits from selfies, has reinvigorated debates about plagiarism, creative rights, and the nature of AI systems. In this article I seek to contextualize the Lensa AI debate within the wider structure of intellectual property law, using it as a framework to answer questions of ownership, and examine the possible role of copyright protection as a solution to art theft. 

Lensa AI is a for-profit company that makes portraits for consumers using the Stable Diffusion program (owned by Stability AI, a nonprofit organization), which in turn makes AI-generated art by learning from pieces in the LOIRON5-B database. LOIRON5-B is an open source, public domain database of artworks pulled from the internet. In their terms of use, Lensa AI does not claim to own the portraits, but they do retain some permissions to use the works at their discretion. The fundamental problem is that individual artists’ works are pulled and used to teach the AI program— whose product is then sold for profit — without the artist’s knowledge, consent, credit, or compensation. (Read more about the Lensa AI issue from The Washington Post and The San Francisco Standard.) 

Although Lensa AI does not claim copyright over their products, looking at the structure of copyright in the U.S. offers some clarity on the meaning of ownership and the roles of AI programs. Under the status quo, creators can request a copyright for their work, which ensures their full ownership and exclusive right to profit from that work, from the U.S. Copyright Office. U.S. copyright law specifies that the author (or creator) of a work must be human, and that AI does not count as human since it cannot be held responsible in court. The code of an AI program can be copyrighted by its human author, but AI-generated products are different: they can only be copyrighted if they were influenced by a human. Otherwise, the products are simply in the public domain. Human influence is not clearly defined and remains open to interpretation. For instance, between the author of an AI program and the artist whose work was used to teach said program, who would win the most influence over the result? 

Copyright law reveals some of the wider legal ambiguity surrounding artificial intelligence. To summarize one definition, AI is intelligence synthesized through computer software and hardware (where intelligence is the ability to reason, achieve goals, synthesize information, and produce creative works). Legally, the status of AI is extremely vague. Examples from relevant copyright literature regard AI as a tool used by the codewriter, and anything it produces is a product of the codewriter. This perspective follows an 1884 precedent regarding photography that established the basis of US copyright law — in Burrow-Giles Lithographic Co. v. Sarony, the camera was ruled an instrument of the photographer and the photograph belonged to the photographer. That being said, the tool or instrument perspective disregards the inherent intelligence of AI as well as the sources and creators of the information that is fed to it.

Applying this status quo to companies like Lensa AI raises two interesting questions of human influence and profiting from public domain images. If a company selling AI-generated art sought to copyright its images, would they qualify as having human influence? And if it could be proven, who would have more influence on the final product: the codewriter or the artist whose work the AI learned from? These points are largely unanswerable because of the continuing lack of legal clarity. Given that Lensa AI is using the Stable Diffusion program as a foundation, though, I would argue that the most direct proportion of human influence on the final product — if present — would go to the artist. On the other hand, suppose AI generated images are legally proven to not have human influence. By the status quo of copyright law, they would be in the public domain, so what right would a company then have to profit off of them? 

In summary, analyzing the influence and ownership over Lensa AI products through current copyright law does not provide a clear right or wrong answer. Furthermore, it is worth examining whether copyright theory is even a fitting approach to these questions of artificial intelligence, creativity, and ownership of art. Copyright is designed to provide an economic incentive for people to create art. Thus, copyright protections are ascribed value because they control who gains profit from a given work. Allowing copyrights for AI programs is designed to encourage industry growth; copyrights for art are supposedly designed to protect artists from theft as well as encourage the industry. 

In theory, copyright could be a solution for artists affected by art theft. Copyrighting one’s art would prevent a separate company from profiting from it, and there would be no possibility for online posts to be interpreted as implicit consent for image reproduction. In practice, though, there are two arguments for the inadequacy of copyrights in art. Firstly, there is the issue of accessibility. Small-scale digital artists, like those whose work have been copied by AI, cannot necessarily afford to pursue or enforce a copyright. If an artist did pursue copyright protection or an argument of theft by Lensa AI, they would almost certainly be met with a strong force of corporate lawyers. Secondly, an argument made by New York University Professor of Law Amy Adler asserts that copyright has no place in art at all. Art pirates cannot profit off of their theft because of the norm of authenticity in the art market. But Adler’s argument only applies to the large-scale, high-reward art market, and not to the dynamic of corporation versus artist that is at play in the Lensa AI case. 

Before circling back to solutions, the copyright system as a product of capitalism should be contextualized further. Why is ownership such a contested right? Why does creativity need to be incentivized and monetized? Copyright is the product of a profit-protecting legal system and a framework that indubitably prioritizes money over creativity, but under current structures, a total lack of intellectual property protections could exacerbate the corporate theft already occurring. Examining the systems of art and copyright also raises salient questions of what qualifies as human, what is influence, and what is the purpose of art; these are philosophically relevant and important to consider but beyond the scope of the law and of this article.

Setting aside the issue of human influence, and whose, Lensa AI is in some fashion exploiting artists’ labor. The app produces images with uncanny resemblance to the  creations of others who did not consent to having their work reproduced and were not compensated by the company. The legal system provides little clarity on the intersection of artificial intelligence, copyright, data mining, and for-profit versus nonprofit use.  Due to this murkiness, it is difficult to imagine the system finding a solution that is fair to plagiarized artists and does not unduly benefit corporations already profiting from questionable practices. 

Leo Worthington is a first-year staff writer concentrating in history and can be contacted at leo_worthington@brown.edu.