A monkey and a Generative AI walk into a bar ... *

*The author works in the field of machine learning/artificial intelligence. The views expressed herein are his own and do not reflect any positions or perspectives of his employer.

A monkey and a Generative AI walk into a bar, each looking glum. After they are served, the monkey looks over and says “Why are you so down?” The Generative AI replies “I make beautiful pictures, but the Copyright Office says I can’t own them.”? Astonished, the monkey exclaims excitedly “Me too! I took a fantastic photo of myself once that went viral around the world, but the Copyright office said ‘No.’ The guy whose camera I used couldn't own or protect it. And neither could I.”? The Generative AI and the monkey clink their glasses and continue chatting. They then come up with an idea. “Perhaps we can work together,” the monkey muses. Something about that just isn’t right.”? The Generative AI, quickly putting thoughts together word by word, replies “Yes, that’s it. Since I’m superhuman and you’re subhuman, acting together we can be human. Let’s collaborate on a piece and submit it to the Copyright office.”? “Brilliant!” exclaimed the monkey. “And great to meet you. I had heard AIs hallucinate and weren’t that good at math.”        

Only parts of the account above are fictional: the bar and the plan by the monkey and the Generative AI to join forces. The U.S. Copyright office and then the U.S. federal courts have decided that a ‘selfie’ taken by a macaque is not Copyrightable by the monkey, Naruto, who took the snap because creators for purposes of the law can only be human beings. Similarly, the Copyright office ruled that images created by a Generative AI operated by Stephen Thaler cannot receive Copyright protection. Mr. Thaler is pursuing litigation challenging that ruling, but is likely to face a similar fate. The same holds for patent protection. The US Patent Office had determined in 2019 that the very same AI developed by Mr. Thaler, DAUBUS, cannot be a named inventor on a patent application, even if the AI autonomously creates an inventive design. A federal appeals court agreed, ruling against Mr. Thaler last August. In the world we now inhabit, inventions and creations that would otherwise certainly be eligible for intellectual property protection are not — because the laws, as written, assume that only humans can create or invent. What’s at stake in changing this status quo, or not, is the focus of this piece.

A paradox at first blush

The litigations mentioned above all concern the question of authorship/inventorship and whether non-humans are eligible to enjoy the protections of copyright or patent law. There is an additional set of litigations now being brought against the operators of various Generative AIs (or products leveraging such generative AIs) made available for public use. One case is concerned with the software source code that GPT3 (and products with an API to GPT3, such as Github’s Copilot) generates. Another case is concerned with the images generated by Midjourney, Stable Diffusion and DeviantArt. At issue is (1) whether Generative AIs training on copyrighted works is copyright infringement or fair use, and (2) whether the outputs from Generative AIs are infringing derivatives of the works on which the Generative AIs trained. The legal issues presented in the lawsuits are in turns interesting and complex. They deserve fuller consideration, which I will pursue in due course in a separate piece. One paradox presented by the more recent lawsuits concerning Generative AIs in light of the older suits involving Naruto and DAUBUS is the following: Can Generative AIs be understood to create illegal derivative works if they cannot legally create copyright protectible works in the first instance?? That is, if Generative AIs cannot be named creators for copyright purposes, should imaginative works they create ever be deemed infringing???

The simple answer to this question is that it is a false comparison. Generative AIs, like other digital tools, can be used by humans to create infringing works. Even if Generative AIs give a semblance of independent production in their rendering of creative works, it is their operators or users who induce them to generate such outputs. To the extent those outputs are generated from, or too closely resemble, copied training images, it is the humans responsible for developing the Generative AIs (or those using them to generate the outputs) who are responsible for copyright violations. Human agency, however attenuated from the act of creation undertaken by the Generative AI, is ultimately in play. Regarding infringement, these are critical precepts in the copyright portions of the lawsuits brought against the publicly available Generative AIs that are creating source code and art. The suits allege human agency in the acts of copyright infringement they assert. This reasoning, extended to acts of creation, might go as follows. A creative tool (e.g., camera, Generative AI) can facilitate an act of creation, just as it can an act of infringement. But we would never say that a camera was itself an infringer, and neither would we say it was a creator. Without human assistance, we similarly would not say that a Generative AI on its own is an infringer, nor would we say it is a creator.?

But is the false comparison account quite right when it comes to Generative AIs?? Is an AI merely a tool for creation like a camera? And thus never quite worthy of being deemed a creator?? My prediction is that, in the months and years ahead, we’ll keep coming back to the question of copyright authorship for works created by Generative AIs. Generative AIs, acting on their training data with the slightest of prompts, are able to generate remarkably creative, novel works. Unlike the monkey selfie situation, acts of AI creation won’t be serendipitous one-offs but plentiful and diverse. The issue of legal capacity to create copyrightable work won’t be directly litigated in the pending infringement suits involving the publicly available Generative AIs mentioned above. But it bears on the bigger question of economic displacement of artists and coders, and how the global marketplace for creative works more broadly should be structured in view of the powerful creative modes being brought to society by Generative AIs. We can expect many players to seek amendments to copyright law that permit works created by AI’s to be protection eligible. Some may simply seek clarification of what amount of human contribution is required to deem a work produced by a Generative AI copyrightable. For reasons that I discuss below, the current legal status quo serves no one well.

Monkeys & cameras – exposing the Grand Illusion?

Naruto showed that an animal, perhaps accidentally, can use a modern human tool to produce something creative. The photograph is in focus and is well framed. The captivating thing about Naruto's photo is the sense conveyed that she posed for the camera – that somehow she was aware that she was about to photograph herself and knew that smiling was an appropriate facial posture for doing so. This was the impression left by the viral selfie. But was it anything more than accidental???

On the one hand, the infinite monkey theorem may explain Naruto's situation. Her winning selfie came about purely by chance. On the other, why is Naruto's snap any less worthy than a random, blurry point and shoot photo by a human, which would be copyright protection eligible?? The Supreme Court has ruled that to be original and deserving of copyright protection, a work requires only “independent creation plus a modicum of creativity.” Feist Publications, Inc. v. Rural Telephone Service Co., Inc., 499 U.S. 340, 346 (1991). If copyright requires only a modicum of creativity, why wouldn’t a work embodying more than a modicum of creativity be eligible for protection? Naruto is not alone. There are other known instances of highly creative art production from a range of non-human mammals.

The animal rights organization PETA certainly sympathized with Naruto, bringing suit against David Slater to prevent him from getting copyright in any images captured by Naruto (or other members of her troop). They argued that non-human animals can and should have the right to express themselves creatively and receive authorship protection under the law. The lawsuit was settled between Mr. Slater and PETA in 2017. Instead of dismissing the case, the Ninth Circuit Federal Court of Appeals took the opportunity in 2018 to affirm that animals have no rights under the Copyright Act (or any other statute) unless such rights are explicitly stated in the law. They also ruled that PETA has no standing to sue on behalf of Naruto.

But the question of Naruto’s authorship was not the only issue presented. The photographer, David Slater, whose camera was used had actually asserted copyright in his own name. He took issue with Wikimedia and certain web publishers posting the monkey selfies recorded in his camera without his permission. He explained (here, in his own words) that he worked over a multi day period to earn the trust of the macaque troop that included Naruto, staged the setting for the photo, arranged the camera and camera settings just so, and used a cable release to facilitate the monkeys being able to take photos of themselves. Still, the US Copyright Office proactively took the position that, where a monkey actually triggers a photo taking mechanism, the resulting photo cannot be copyright protected. The Copyright Office had long held that works of authorship must be created by humans to be copyright eligible. In 2014, with knowledge of Naruto’s and Slater’s exploits, it updated its official compendium of operating rules (Sec. 313.2) to state that works not eligible for copyright include “a photograph taken by a monkey.”? Though not specifically litigated, Wikimedia resisted Slater’s calls to take the photos down on this basis. Human agency in arranging a work of art in which an animal played a critical role is not enough to secure copyright protection. To what extent, then, must human agency in the operation of Generative AIs be present to deem their outputs copyright eligible?

For a long while, many have expressed skepticism about photography as a form of creative expression that should merit copyright protection. This goes back to the 19th century, both before and after the US Congress amended the Copyright law in 1865 to include protection for photography. As litigated in the 1884 U.S. Supreme Court case of Burrows-Giles Lithographic Co. v. Sarony (finding that photographs can be copyright eligible), the concern is that most photography non-creatively captures the world as it is. While photographers can and do bring creative framing, exposures, light settings, post-production, etc. to a work of photography, it is questionable whether most photos reflect the “modicum of creativity” or de minimis originality that is otherwise meant to be the threshold for copyright protection. The Copyright Office embraces this fact to an extent, in that it refuses to evaluate the artistic merit of a particular work. Its Compendium (Sec. 310.2) states: “[T]he Office will not look for any particular style of creative expression. Likewise, the Office will not consider whether a work is visually appealing or written in elegant prose.”? However, for purposes of evaluating sufficient originality, the Office allows that:

"The creativity in a photograph may include the photographer’s artistic choices in creating the image, such as the selection of the subject matter, the lighting, any positioning of subjects, the selection of camera lens, the placement of the camera, the angle of the image, and the timing of the image." compendium (Sec. 909.1)?

These are exactly the kinds of contributions made by David Slater to the photo that Naruto snapped. However, his contributions wouldn't count because his "joint author," Naruto, is not human. For creative expression, one’s human status is more important for protection under copyright law than the actual creative merit of a work. This is therefore the crux of the grand illusion. In a world of digital ease, the step of capturing a compelling photographic image is relatively easy. Point and shoot digital cameras have automatic focus, light exposure settings and filters. Photos taken on a timer, where no one is actually holding or manipulating the camera, execute no differently than when a monkey tugs a cable release. Except, in that instance, there is just the camera acting on its own!? Creativity as something uniquely human is encoded in the law, but our lived reality steadily belies this precept as well grounded. Humans may create works having minimal artistic merit and little to no aesthetic appeal and nonetheless obtain the protection of the law. But works of high artistic merit embodying substantial creativity cannot be protected if created by a non-human animal. In cases where non-humans are involved in the production of a work, contributions by humans can be disregarded, leaving the work without any legal protection at all.?

This leads us back to considering the implications for Generative AIs. To flirt further with the merits of having a low bar for originality of a work, but a rigid eligibility bar for who (or what) can count as an author, we’ll spend some more time below with a selfie-creating robotic AI named Ai-Da. The law (as is frequently the case) embodies a status quo that is reflective of bygone eras of technology. US Copyright law last underwent a substantial revision in 1976. Without the same sophisticated means of creative production we have today, it was never imagined that non-human mammals could produce copyrightable works having even de minimis creativity. There were no digital cameras, no public internet and computing was substantially less powerful. Almost fifty years later, we occupy a world far beyond the imagination of that time.

The deeper paradox – the case of Ai-Da and AI selfies

As explained in a prior piece, Generative AIs don't truly understand the material they train on or the content they generate. Rather, they use vast computing resources and advanced analytical frameworks to form mathematical associations between elements of the data on which they are trained. In response to user prompts, Generative AIs leverage such associations to generate new or recontextualized content. The training and dynamic production of responsive content is done at a speed and scale that no single human could match. That content may not have the same inspiration or motivation to create as more traditional human-created work. But AI-generated art can be quite creatively stunning, good enough to win art contests.?

Ai-Da the robot is a special instance of artificial intelligence. Developed in 2019, Ai-Da is named after Ada Lovelace, the 19th Century English mathematician and writer who is widely considered the first computer programmer. Ai-Da the robot was designed with cameras for eyes and an ability to adaptively translate what she “sees” into works of art. As an AI that occupies a mechanical body and physical space, Ai-Da is able to perform, give speeches and create tangible works of art. Ai-da is best known for her self-portraits, but is able to create portraits of anyone who poses for her, creatively generating new works of art. Ai-Da has been outfitted with different mechanical arms that permit her to work in pencil, paint or sculptural materials. Her works are extraordinary and have been displayed at some of the world’s most elite museums. No two are alike, even if rendering the same subject using the same media. Ai-Da created a work of art called “Algorithm Queen” that was displayed at the Platinum Jubilee held in 2022 for the late Queen Elizabeth II. She is outfitted with an embedded AI language model, an actuating mouth that coordinates with internal speakers, and is able to generate and then recite poetry in the combined style of several well-known writers. Ai-Da has given her own Ted Talk. In short, Ai-Da is quite a versatile creator of works of artistic expression and has gained more exposure and fame than most artists can ever aspire to.

As the creators of Ai-Da have aptly put it:

“Extraordinarily complex, our online worlds are pushed and pulled by forces and personalities that are sometimes apparent but largely oblique. Ai-Da, the machine with AI capacities, highlights those tensions: is she an artist in her own right? Is she an artist’s alter ego? Is she an avatar, or a constructed character? All these options bring powerfully to the forefront the complexity of our interacting digital and physical worlds and the masked identities we can assume in them.”

Ai-Da’s considerable accomplishments notwithstanding, the law as it stands (in the US and most of the world) would not afford any of her works copyright protection. The Copyright Office’s Compendium (Sec. 313.2) states:

“[T]he Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author. The crucial question is whether the ‘work’ is basically one of human authorship, with the computer [or other device] merely being an assisting instrument, or whether the traditional elements of authorship in the work (literary, artistic, or musical expression or elements of selection, arrangement, etc.) were actually conceived and executed not by man but by a machine.” (internal quotations omitted)

A conception of Generative AIs as mere tools may support having a human "contributor" being able to claim copyright in the AI's outputs. Perhaps that is the approach used by the Next Rembrandt project, which used an AI model carefully trained on a selection of 346 of Rembrandt’s paintings and directed through a step-by-step process to generate a remarkable new painting in the style of the late master. Any conception of Generative AIs as creators, and not merely tools, would doom its outputs to fall within the public domain. The choice of developers, operators or users of Generative AIs is a stark one: adopt the fiction that you had some meaningful role to play in the creation of an original, generative output or else forgo any ability to control the outputted work. According to the plain language of the Copyright Officer's guidance, the "crucial question" cannot be sustained without a fiction when we are dealing with AIs that employ the "traditional elements of authorship" to generate new works.

This doesn’t have to be the case. The UK Copyright Law of 1988 contains provisions for protecting computer-generated works. Protection is afforded under the law for 50 years, a shorter duration than the life-of-the-author plus 70 years afforded for human-created works. (UK 1988 Act Secs. 12(2) and 12(7).)? The law still requires that the listed author be human. For a creative work “which is computer generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.” (UK 1988 Act Sec. 9(3).) This is a pragmatic approach, in that while AI-generated works are eligible for protection, the notion is still that a human must be the author of record as facilitator of the work’s creation. In turn, the creating AI is not expected to handle the disposition of the work commercially or its legal protection; the facilitating human is responsible for those things. Without granting AIs human status, the UK law does embrace that AIs can produce creative works and enables a rational, economic structure for the disposition of such works.

Unlocking the paradox – disentangling the questions presented

For purposes of this piece, a larger question is certainly presented:? What qualifies as art and who or what can be deemed an artist, legally or otherwise?? We are at an uneasy boundary where our intuitions about what qualifies as distinctly human forms of expression or creation are being challenged. The ultimate, metaphysical questions raised are whether art and creativity are uniquely human undertakings, and, if so, whether the creative works Generative AIs are capable of producing call into question what it means to be human at all. These questions themselves are not new. It is not possible to study the field of AI (as many have been for a long while now) without invariably running into such questions. What is new is the pervasiveness with which Generative AIs and their creative outputs will be making their way into our lives. The lawsuits bear witness that the questions are no longer just matters for philosophers to reckon with in thought experiments. The public availability of Generative AIs have forced these matters into the real world, for everyday people to contend with. The aggrieved artists and coders bringing the current litigations are concerned about the impact of Generative AIs on their livelihoods. Whether this makes them neo-Luddites or rightfully concerned citizens is very much in the eye of the beholder.?

However, we should not allow such questions to cloud our judgment about matters of immediate practical import. Existential concerns shouldn’t cause us to freeze on important law or policy questions that can be addressed in a principled way in the meantime. It would be too easy, in view of the discomfort posed by Generative AIs, to resist any attempt to treat their outputs in a way that can benefit people. Because artists, coders, other creators (and even lawyers) are feeling threatened by Generative AIs, it may be too easy to think that affording any legal protection to their outputs is the same as giving the Generative AIs themselves legal protections at the expense of people. (The subject of whether Generative AIs themselves should have protection under the law — in the manner that corporations do or otherwise — is a rich one, that I will return to in a future piece.)? What’s immediately at stake, however, is the fact that AI-generated works would fall into the public domain, benefitting no one at all.?

Creating an avenue for the outputs of Generative AIs to obtain copy protection should carefully consider the balance of rights and obligations as between the developers of the Generative AIs, their operators, and their users, each of which may have some role to play in how the Generative AIs function to produce creative works and what instances of work they actually produce. The English approach may or may not achieve that balance in the right way; that is beyond the scope of this piece to tackle. It does represent an effort to balance the realities of computer-generated creative works with the critical interest of the copyright system in creating market structures and economic ordering around works having commercial value. It lowers incentives for humans to abide by fictions in order to obtain protection for creative works. Without any such approach, the massive quantities of creative content predicted to be created by Generative AIs going forward will flow into a disorganized void. The impact on human artists and creators may be the same — some displacement from the outputs of Generative AIs in their respective fields. Arguably, however, the public domain commoditization of creative work produced by Generative AIs in large volumes harms the marketplace for all creative work, however and by whomever produced. Even while litigation proceeds on the questions of infringement and fair use, policy makers and all stakeholders in those disputes have an aligned interest in considering extending qualified copy protection rights to computer-generated works.?

Copyright ? 2023 Duane R. Valz. Published here under a Creative Commons Attribution-NonCommercial 4.0 International License

*The author works in the field of ML/AI. The views expressed herein are his own and do not reflect any positions or perspectives of his current or former employers.