Richelle Khor and Robbie Morris
Intertwining art and technology: the future of human creativity
Updated: Jan 17
Jason Allen’s work, “Théâtre D’opéra Spatial”
The recent win of AI-generated art in the Colorado State Fair's fine arts competition has sparked debate over the authenticity and definition of art. The use of AI in art foreshadows the potential replacement of human artists. The absence of expressive elements and human techniques in AI-generated art makes its position difficult to justify. However, with the continuous improvement of AI image generation, it can no longer be ignored. We investigate the debate on whether AI-generated art should be considered independent and expressive, or simply advanced plagiarism.
After submission to the Colorado State Fair's fine arts competition, Midjourney became the first AI to win a human art contest, obtaining a gold medal in the digital art category. This competition sets a new precedent for artists but raises some concerns. Was Midjourney the winner, or was it a tool used by Jason Allen, the 'artist' who entered one sentence into software that generated the winning piece? Where do we draw the line between using tools and cheating with them? How does AI art fit into human artistic expression, and should we be concerned about the replacement of real artists? AI continues to improve image generation; frankly, the art is so good it can no longer be ignored.
Most people believe art is unique because it contains expressive qualities that are not quantifiable or tangible. This defining feature of art is currently being challenged by technology's most recent developments in AI, sparking discussion over what truly defines the subject. The absence of expressive elements, alongside the omission of human techniques, makes it difficult to justify the position of machine learning models' creations in the art world. Over recent years, we have seen the development of various machine learning models, notably Midjourney, DALLE-2, and Stable Diffusion, that create art from text prompts, each better than the preceding one, effortlessly emulating famous painter's styles.
Stable Diffusion has been improving its capability of doing these things faster than others. As we write this, Stable Diffusion easily tops the trending charts of GitHub repositories by a wide margin. Its first version was a variant of the diffusion model, meaning that it consisted of Markov chains (sequences of probabilistic events) trained using some complex integral approximations. Stable Diffusion 2.0 was released just this November and has implemented more features on top of its original tech stack, including upscaling and partial image replacement. These are features that no other current big-name AI art software has. The feature sets of software, such as Stable Diffusion, are impressive, but their training process still needs to be improved due to massive data requirements. Choosing training data that doesn't teach models to replicate (or, as some may say, steal) images created by other people is an intricate balancing act. The art community still needs to decide whether these machine-learning models make independent expressive pieces or plagiarise them from others.
The idea that art is only used to express passion and culture was contested by Elea Zhong, a sophomore at USC's Hanman Academy, who believes there is a second category of "corporate art" that may not demonstrate those qualities. In reality, most people do not appreciate the process of creation and regard visual pleasure as the sole factor in distinguishing art. The modernisation of art dominates the bulk of the debate, inviting us to consider the practicality of insisting on a criterion that is no longer as highly regarded as before. Think about it this way - generative AI models are the first pass at prototypes in the creative industries. This may be a blindspot in many analyses of AI art.
A rococo painting of a Persian woman covered in peacock feathers standing before a red mosaic wall
The question is one of the conventional men versus machines conundrums; evolution on the one hand and originality on the other. Affirmed by TEDxMileHigh's research on imagination, humans' cognitive association plays a central part in materialising their inspiration into artistic expressions. Will generators that merely merge pre-existing images into a particular art form drive artists out of the industry? Mats Borges drew a parallel to the time when photography was first invented, and similar concerns about portraiture were raised. Photography, a widely accepted art form today, shows that art can expand. While his point silences the clamour about the job security of artists and has its merits in history precedence, a distinction must be made: the processes of creating photography and portraits both involve establishing a link between what the creator knows and what the creator wishes to express, whereas, in the case of AI art, the process is replaced by a reference to pre-existing images. Thus, it is possible to reconcile that AI art is real art, but the fundamental differences call for further distinctions to be made.
The concerns do not end here - there are further legal implications when we seek to commercialise AI art. Dubbed by Leigh McGowran as the legal minefield, opening up AI art may lead to intellectual property litigation floodgates. AI generators are trained using a massive database of images to ensure they can interpret various concepts, attributes, and styles. As such, copyrighted materials may have been included in the database. Although OpenAI attempted to dispense the allegations by claiming that the training database of the generator (DALL-E 2) comprises publicly available and licensed sources, the claim is nonetheless undermined by the lack of transparency in disallowing public access to the database. In fact, an analysis of the database training Stable Diffusion reveals that some of the images may be copyright protected, making AI art vulnerable to infringement claims.
Furthermore, users have no legal protection in terms of their proprietary rights. Stable Diffusion requires users to forfeit their intellectual property rights, and DALL-E and Midjourney provide no redress to users embroiled in ownership disputes. The burden is on the user to carry out the necessary due diligence. Given the lack of precedent, there is an urgency to impose sensible regulations and responsible practices on AI developers. A sudden leap in using AI art will be imprudent without a sound legal framework to navigate the implications.
Tokenization of Global Illiquid Assets Opportunity by 2030
Artprice100 Index vs. S&P 500 (2000-2022)
We are also concerned with monetising AI art as a financial asset. In 2018, a portrait painted by an AI, Portrait of Edmond Belamy, made its debut on the world auction stage and was sold for $432,500. The latest research from Deloitte and ArtTactic shows that 85% of wealth managers believe art and collectables should be part of their client offerings. Consider this - blue-chip art has outperformed the S&P 500 since 2000, according to the Artprice100 index, and there's evidence to highlight that alternatives are and will continue to be the fastest-growing asset class. But what happens when we combine art, tech, and finance? There's no correct answer, but tokenised art stands out to us. A recent BCG and ADDX report forecasts that the total size of illiquid asset tokenisation globally could be $16 trillion by 2030. For example, Artory / Winston, a joint venture between Artory and Winston Art Group, has created a $25MM tokenised art fund. Artory captures the artworks' (physical art, digital art, NFTs, and collectables) due diligence data on the blockchain and provides investors with digital certification. The fund contains 68 artworks — 3 x Blue Chip ($2MM average value), 15 x Mid-Career ($600K average value), and 50 x Emerging ($200K average value). Or take Pixura, which lets you tokenise art and launch a blockchain art marketplace on Ethereum, or artists' equity management platform Lobus, which manages $9 billion in assets for clients such as the Rothko family and Louis Vuitton. The opportunities are endless for digital artists. Beeple is the best example of this - his piece, "Everydays: The First 5000 Days," sold for $69.3MM at a Christie's auction last March.
We drifted away from AI art for a moment, but AI still needs to work on comparing with human artists in certain areas. While it effortlessly draws landscapes, anything organic or intricately detailed falls apart. We are still at the stage where you can tell human-made and machine-made images apart (although we're not sure now after Stable Diffusion 2.0). However, AI has cemented itself as an unresolved point of contention among artists. We must carefully tread the line between creating art with a new tool and losing sight of what art is altogether. While machine learning fundamentally removes the spirit of the artist, it also comes with benefits like speed and ease of use. With AI, anyone can be an artist, for better or worse.