Art, artificial intelligence and the law

As artificial intelligence (AI) makes rapid strides and art generated by using AI becomes more and more sophisticated, the legal regimen might need to be updated sooner than was previously assessed, argues Aarushi Anand. 

—

BACK in October 2018, Christie's sold a piece of art called Portrait of Edmond Belamy for a whopping US $432,500. The portrait presents the sitter in a murky setting, clad in a standard dark coat and a contrasting white collar, with undone, almost pixelated facial features.

The painting exudes overtones of eighteenth-century vintage portraiture aesthetics combined with a touch of modernity, quite like the compositions of Glenn Brown, a post-modernist known for appropriating and subverting historical art.

Though the encompassing look evokes a sense of timelessness, what makes it one-of-a-kind is its authorship, a mathematical formula. The artwork, evocative of a human's emotional experience of artistic innovation, was the product of a 'generative adversarial network' (GAN), a form of artificial intelligence (AI) technology.

In November 2023, Christie's teamed up with Gucci to auction off artworks developed via AI and generative techniques. The primary objective was to re-conceptualise pieces of Gucci's different eras and extrapolate them into the future.

While art and fashion are two peas in a pod, the escalating convergence of mankind's inventiveness with computational accuracy to redefine creativity brings to scrutiny the tendencies and limits of AI to excel in a human-centric domain.

Understanding creativity

Creativity is a burst of generative energy that singles out ideas, alternatives or possibilities to solve complex issues and create surprising goods with value. In his Treatise on Painting, Leonardo da Vinci described how a soiled rag thrown at a blank canvas can serve as a stimulus for the subsequent step.

Simply put, the human brain is a cognitive network that runs a recognition engine. It strives to harmonise bits and pieces into a meaningful array of symmetrical shapes or patterns.

Also read: India's quest for a workable AI legislation

Inspired by the human brain, GANs are trained to identify patterns in a certain dataset and then duplicate them until a second 'discriminator' fails to recognise distinctions between the original and the processing set.

An additional breakthrough is the emergence of creative adversarial networks, a mutated version of GANs, that propels networks to produce visuals explicitly beyond the training set while theoretically retaining image coherence.

The rewiring of neural networks is best illustrated by the Portrait of Edmond Belamy, which was created via an algorithm acquiring knowledge of 15,000 portraits ranging from the fourteenth century to the present.

While artists typically go through a historical study of visual arts before affixing their own style, Immanuel Kant stringently associates artistic genius with 'exemplary originality', when an original composition inspires others, as opposed to non-exemplary non-originality, i.e., superficial replication.

AI-generated emulative and not creative art is merely an aspect of computer creativity that has yielded monetisable outcomes. While the seamless circulation of AI art tools highlights democratisation of creativity, it also contributes to the current threat landscape where technology can overtake what has traditionally been a purely human outpost: intellectual creativity and its associated right to control and profit from it.

The transition from 'interaction' to 'convergence' of technology and humans in the production of art from scraping copyrighted sources possibly dampens the future of human agency and autonomy.

Legal outlook

In January 2023, Sarah Ander­sen, Kelly McKer­nan and Karla Ortiz filed a copyright infringement lawsuit against AI image-generating businesses Sta­bil­ity AI, DeviantArt, and Mid­jour­ney.

Also read: Generative AI and the copyright conundrum

Matthew Butterick identifies the interpolation of diffused images via the software Stable Diffusion as primarily a twenty-first century collage tool that develops latent images in the style of a certain artist.

The massive diverse training dataset for Stable Diffusion was obtained from, LAION-5B, an information repository that contains 5.85 billion web addresses of visuals scraped from across the web.

While Stability AI founder Emad Mostaque publicly recognised the imperative of utilising licensed training imagery, the fact that there is virtually no impediment against the emergence of text-to-image generators has created an entirely different ballgame.

In a sea of several hundred million images, the endeavour of licensing creations ethically and in accordance with copyright is not naturally occurring; rather, it is laborious and pricey.

According to James Gurney, an American illustrator who has become a well-known reference for consumers of text-to-image computational models, artists should be empowered to opt in or out from having their output integrated into a training dataset. While he considers AI to be a creative partner, a synthetic genie or an inspirational ally to many others, it is a slow poison, replacing humans at the table.

According to critics, each image originating from Stable Diffusion is a derivative version of the training image. Derivative work is centred on or adapted from previously created pieces, such as translations or art reproductions of pre-existing sources.

Since the ability of an individual to manufacture and distribute a derivative work known as the adaptation right is provided to the copyright holder, a derivative work is prohibited from being developed without the consent of the rights holder.

Hence, an unapproved adaptation of an original work could result in an infringement of copyright. Furthermore, such works may be put up for sale on the internet, syphoning off revenue from creators.

While conditioning and interpolation can be employed to tweak an image's mathematical network of data, the reconstructed copies based on the contours of a specific style can be utilised to feature inapt visuals like sacrilegious or sexualised images, resulting in grave repercussions on an artists' personal and professional life.

Also read: When and how will the law wake up to deepfake technology?

Regulations could provide a defence against an allegation of copyright infringement. 'Fair use' is an assortment of exceptions that permit copyrighted material to be reproduced despite a lack of authorisation from the original inventor.

Appropriation of elements and style, which artists deemed as art theft, is perceived as an entirely novel fair use of art by AI businesses. Within the precept of the fair use doctrine, the US law provides a clause of exemption for works that are transformative in nature.

These are creations that modify the intended purpose of an original piece by incorporating a fresh perspective or revelation and hence do not disqualify as a copyright infringement.

Attributing creativity to GANs (or other digital entities) raises concerns about the status of a company. From such a standpoint confusion lingers whether or not AI-generated transmogrified content qualifies for protection under copyright law.

In August 2023, computer scientist Stephen Thaler was denied copyright registration for an AI-generated artwork called A Recent Entrance to Paradise. The court as well as the copyright office reaffirm the prohibition against awarding patents for creations made by an AI system in the absence of a traditional human inventor.

The fundamental principle of copyright demands originality. AI-generated outputs are the culmination of algorithms, instructions and templates, which fails to represent the creator's distinct personality.

Eran Kahana, a fellow at Stanford Law School, reasons that intellectual-property regulations function so that IP owners can reap profit from it while barring third parties from illicitly capitalising on their work. Conversely, an AI does not have any of those needs but is only a tool in the process of work creation and patent invention.

Societal response

Understanding creativity or the element of 'art-ness' in a work is not feasible without public engagement. In December 2023, American rock band Kiss unveiled their computer-generated avatars to be able to live on eternally.

Also read: Does artificial intelligence need a constitution of its own?

The usage of technology to memorialise and immortalise personalities for posterity has been a real threat to the entertainment industry. The Wundt curve, which reflects the arousal induced by an artwork, assigns a negative hedonic value to a piece that is too novel or challenging to grasp and a positive hedonic value to a work that is familiar, yet captivating.

Hence, the discriminator's purpose is to constantly spur the generator to yield novel projects that get positioned at the peak of the Wundt curve. Today a human's limited consciousness obstructs the imagination of a cube in four dimensions, while Midjourney's Deep learning model can build out a mathematical space with way more than three dimensions.

At the same time, interlinking creativity and devices powering the information economy could have an unsettling prospect. In the backdrop of an imminent 'infocalypse', a period in which civilisation is swarmed by disinformation, a searchlight on the dark corners of AI models reflects the usage of risky unsupervised learning techniques.

Biases in training data in terms of race, ethnicity or gender have resulted in algorithms making partial selections when presented with a text prompt such as categorising black persons as gorillas, CEOs as white elderly men and nurses as women.

When art begins to imitate life, inspiration and imitation turn out to be essential elements of the artistic endeavour. In such a case, how does one identify the fine line between an original and a copied work?

In blind surveys, participants found it difficult to distinguish between AI-generated and human-created art. The issue at stake is whether machine intelligence deserves to be scrutinised in the same way humans are.

Also read: The possibilities and pitfalls of ChatGPT

While laws are still being worked out, the urgency of the issue reveals impactful consequences on the way people understand and reflect on their own civilisation. That will only safeguard a cherished anthropocentrism to a certain degree.

The Indian context

A closer look at home reveals that the distinctive flavours of India are gently simmering within the bubbling saucepan of AI art. With each new prompt, the output gets more elaborate and imaginative, ranging from Van Gogh-styled packaging for Maggi and Parle G to political figures as rock stars and met gala attendees.

While AI tools have successfully adapted to the standards of Indian aesthetics, the question lingers: how does Indian legislation keep a tab on the creation of AI technology?

The prevailing Copyright Act of 1957 does not explicitly regulate AI-generated art. Under Section 2(d) of the Act, a human or legal person who "causes the work to be created" is referred to as the "author".

The term 'computer-generated work' has no specific meaning in the Act and can be construed at face value. It is, therefore, arguable that AI-generated works are excluded from the ambit of a 'natural' person, thereby excluding them from claiming authorship.

The Indian legislation continues to remain uncertain about the application of this definition. A copyright submission that identified an AI painting app (RAGHAV) as the exclusive creator of a 2-D art named Suryast was dismissed by the Indian copyright office in 2020.

Subsequently, the copyright office temporarily cleared an application in which the AI tool and the app's proprietor, Ankit Sahni, were listed as co-authors. Soon afterward, the office sent out a withdrawal letter. In the latest decision issued in December 2023, the registration request stands denied.

Even though Al has made enormous strides in producing work, tasks like data entry, template filtering and style management still have to be performed by a human programmer.

Henceforth, the causality dilemma of law, that is, whether to recognise the creator of the programme or its user as the individual who 'causes the work to be created' persists.

In addition, the inseparable red flag embossed on any AI art in the form of copyright infringement stands tall.

Section 52 of the Indian Copyright Act recognises fair dealing in the following contexts: (i) private or personal use; (ii) criticism or review; and (iii) reporting on current events.

Accordingly, using licensed material for purposes of instruction might theoretically qualify for legal protection. What time and again remains unrecognised is what constitutes 'fair' in the fair use of scrapping and profiting from inspired creations. Since technology is still at its embryonic stage, legal uncertainty persists.