Recently, it appears that the discourse surrounding the intersection of art and artificial intelligence (AI) is increasingly permeating scholarship and public spheres. This includes the emergence of public programs, institutional symposiums, and art exhibitions dedicated to these themes. This growing interest seems to signify a potential recognition of the significance of this intersection. However, while this dissemination of knowledge may facilitate a deeper understanding of the subject, it also raises concerns about the reliability of information available on the relationship between art and AI. This prompts a reevaluation of this rapport, aiming to comprehend how our perception of it has evolved over time. This reassessment is particularly pertinent given the rise in exhibitions featuring artworks generated through artificial intelligence algorithms.
Historical recognition of art generated by the use of Artificial Intelligence
In this regard, it is pertinent to note that the nexus between art and AI traces back to the 1960s and 1970s, notably marked by Harold Cohen (1928-2016) developing the pioneering artificial intelligence program for art creation at University of California. Dubbed AARON the program was conceived as an exploration of artistic creations as a means of divine communication. Drawing inspiration from the biblical figure, Moses’ brother, AARON was envisioned as the mouthpiece of artistic expression. Cohen programmed it to generate firstly artworks by simulating human decision-making processes, adhering to rules and directives set forth by its creator. This program emerged as a distinct digital persona, prompting the artist to establish a collaborative relationship with it. This dynamic allowed for an exploration of creativity within the dialogue between program and programmer, underscoring the nuanced interplay between artificial intelligence and artistic expression. The collaboration between Cohen and AARON has evolved, maturated, and refined over the decades, culminating in its current manifestation where the software stands as the primary creative force behind the generated artworks. This development enables a profound examination of concepts surrounding creativity, authorship, and collaboration within the realm of AI and allows the beginning of an artistic revolution.
The advent of deep learning and neural networks in the 2010s marked a significant turning point in AI-generated art. These advanced techniques enabled AI systems to glean insights from extensive datasets, resulting in the production of increasingly intricate artworks. One noteworthy development was Google’s Deep Dream, introduced in 2015, that operates as a conventional neural network, mirroring the workings of the human brain to process images with an ethereal and surreal aesthetic. Thoughts its algorithm, Deep Dream delves into images, seeking out recognizable patterns akin the phenomenon of pareidolia. This subconscious tendency, akin to Walter Benjamin’s notion of mimetic faculty, allows humans to perceive and reproduce familiar forms within seemingly random shapes, such as spotting distinct shapes in clouds. Deep Dream’s reinterpretation of images echoes this innate human ability to discern and recreate semblances within the abstract.
During the same period Alexander Mordvintsev, a Google researcher, found himself delving into the realm of convolutional neural networks. He began to experiment artistically with Deep Dream, eventually leading to the viral spread of images. Google then recognized profound innovative potential of Deep Dream, leading to the establishment of the Ami collective (Artists and Machine Intelligence) within Google Seattle’s division. In 2016 the collective curated the exhibition titled DeepDream: The Art of Neural Networks, where a total of 29 artworks were showcased and sold.
In 2018 the prestigious auction house Christie’s sold the first artwork created through AI tools in New York. This piece was crafted by Obvious, a collective deeply engaged in exploring the societal and philosophical dimensions of artificial intelligence. Comprised of artists and scholars, including Hugo Caselles-Dupré, Pierre Fautrel, and Gauthier Vernier, the group ventured into the realm of Generative Adversarial Networks (GANs), a type of unsupervised machine learning primarily employed for image generation.
These networks were conceived by the prominent computer scientist Ian Goodfellow, renowned for his contributions to the fields of deep learning and AI and essentially, they are tasked with producing original works based on a dataset of real-world examples, aiming for a such realism that distinguishing between computer-generated and human-made becomes challenging. The collective Obvious has curated a collection of over 15000 portraits spanning from the 14th and the 20th century. This extensive archive served as the foundation of the creation of the artwork later auctioned by Christie’s. Specifically, the artwork titled Portrait of Edmond Belamy (2018) bears resemblances to a Frenchman, possibly a clergyman, yet its indistinct contours evoke the portraiture styles typical of European Renaissance and Baroque art. It is worth noting that this piece was crafted by an algorithm, resulting in a distinct lack of intricate facial features, which nonetheless renders the works easily identifiable. The artwork also features a signature at its base, resembling a mathematical formula. The decision to use such a restricted dataset inevitably led to a foreseeable outcome. Nonetheless, the venture spearheaded by the renowned auction house sparked a significant focus on the burgeoning phenomenon.
Later, in March 2019, the renowned artist Mario Klingemann, recognized as one of the leading figures in the field of artificial intelligence art, auctioned Memories of Passersby I (2018) at Sotheby’s. This series comprises an almost limitless array of portraits depicting fictious individuals, all generated in real-time by an AI-powered machine. In contrast to the initial iterations of generative art, this installation does not utilize a database. Instead, it relies on an “artificial intelligence brain” trained by Klingemann using thousands of portraits spanning from the 17th to the 19th centuries. This AI software is programmed to generate entirely new images. The artist developed an application akin Tinder to hasten the learning process. Notably, this artwork marks a departure as the AI itself becomes an autonomous creative entity, with the sequence of images influenced by the AI’s output.
The sale of the Portrait of Edmond Belamy shed light to the pivotal issues within the discourse on AI Art, which differs from computer/digital art in the fact that a portion of the creative process escapes the artist’s direct control and is instead delegated to the software. These include the discussion around machine creativity’s legitimacy, the blurred lines of the artist’s identity, and the aesthetic merit of the outcomes.
The question of artificial creativity between art and technology
Margaret Boden, a cognitive science professor at University of Sussex, proposed a theory of creativity in 1994, defining creativity as «the ability to come up with ideas or artefacts that are new, surprising, and valuable. Ideas, here, includes concepts, poems, musical compositions, scientific theories, cooking recipes, choreography, jokes […] and so on, and on. Artefacts include paintings, sculpture, steam-engines, vacuum cleaners, pottery, origami, penny-whistles […] and you can name many more» (Boden, 2009). Boden categorizes creativity into combinatorial, exploratory and transformative forms. Combinatorial creativity involves combining known ideas to create new patterns; exploratory creativity entails exploring the conceptual space to discover new areas or pathways; transformative creativity builds upon exploratory creativity by introducing modifications, thereby generating entirely new concepts. For AI, exploratory creativity is accessible, as it can be taught the conceptual space of a particular field, exemplified by Obvious’ use of European portraiture. However, combinatorial and transformative creativity pose challenges for AI, as they rely on human cognitive process that AI cannot access.
Boden’s work lays the groundwork for exploring the various perspectives that have evolved over time concerning this issue. On one hand, art historian Mariam Mazzone posits that AI can indeed function as a creative entity capable of producing artworks. For example, GANs generates images that are similar yet not identical to the data used for their training. Mazzone, a member of Rutgers Art & Artificial Intelligence Lab, has been involved in modifying GAN process to create AICAN (AI-Creative Adversarial Networks), drawing inspiration from the theories of psychology professor Colin Martinadale. He theorized that the model for artistic creativity can be predictable, suggesting the potential to “map” it and transform it into an algorithm. On the other hand, artists such as Anna Ridler delve into the nuances of the training set to showcase how even works created with AI possess a creative essence. (Barale, 2020). Ridler avoids biased datasets and limitations in experimentation or critical thinking by utilizing programs that requires smaller training sets. This approach enables her to create her dataset and meticulously examine each image. Similarly, Mario Klingemann’s work in Neural Glitch (2018) involves the use of a more refined and creatively potent code. In this project, Klingemann randomly alters the references used for the GAN’s training, resulting in the creation of novel textures and semantic information.
The modification or alteration of networks stands out as a crucial element for artists in assessing the originality and creativity of their work. However, as Francesco D’Isa said in his book La rivoluzione algoritmica delle immagini «[…] there are still few software that allows the customization of the data set by the user (fine-tuning)» (D’Isa, 2024). Continuing the discourse, as D’Isa aptly notes, individuals creating artworks with Text-To-Image (TTI) software showcase their creativity through the dialogue they engage in with the machine. This dialogue represents a wholly novel interaction, characterized by its unique nature, emerging from the necessity to communicate with and, in a sense, for the machine.
It is worth noting that the term “creativity” might not be entirely fitting in art generated through AI’s software, as these systems lack the conceptual tools to comprehend what they create. While humans make choices and associations both visually and conceptually, AI operates primarily at a visual level due to its absence of conceptual understanding and experimental knowledge. A fundamental disparity between images crafted by humans and those generated by AI lies in the origin of the visual stimuli. For humans’ inspiration often stems from nature, while for machine, it is rooted in numerical data related to specific shapes, colors and lines.
In the realm of HumaniTies and Artificial Intelligence Inke Arns discusses of artists regarding the nature of AI, emphasizing its lack of autonomous thought despite its seemingly misleading title. German artist Hito Steyerl goes as far as to dub it “artificial stupidity”, highlighting the notion that AI operates based on pattern recognition within vast dasets, often shrouded in mystery due to the unknown origins and classifications of these initial datasets. (Grunert, Craglia, Gómez, Thielen-del Pozo, 2022)
Despite this perceived autonomy, the role of the human artist remains pivotal and influential at various stages of AI art process, such as the selection of datasets or the curation of the most relevant outputs. Ridler, for instance, suggests that the discussion around the creative potential of AI is somewhat misplaced, as true creativity emerges from the collaborative interplay between human and machine. Even in Klingemann’s view, where the machine is seen as a tool under the artist’s indirect control, the resulting artworks maintain a sense of autonomy and novelty. Through meticulous data selection and image curation, artists steer the AI process towards the realization of their creative vision, yielding artworks that resonate with audiences on emotional and intellectual levels. Central to this emerging art form is not merely the artwork as a static object, but rather the dynamic process itself, that is an intricate relationship between human agency and machine intelligence.
Rise and identification: AI-generated works as legitimate art in the exhibition The Rights from Future Generations – A Perspective on (A)rt and (I)nnovation
In the evolving landscape of art and technology, the designation of AI-generated works as legitimate artworks serves not only to acknowledge the creative agency of artists but also to expand understanding of art itself. These pieces, born from the fusion of human imagination and computational algorithms, challenge conventional notions of authorship, creativity, and artistic expression. As such, they rightfully claim their place within the broader canon of art history, inviting contemplation, interpretation and appreciation alongside traditional works of art.
The discourse surrounding the classification AI-generated images as art is challenged when the art world embraces the exhibition of these reproductions within gallery spaces. This action, initially seeming to resolve the quandary of the relationship between art and artificial intelligence, still warrants through investigation today.
Ongoing concerns regarding the viability of utilizing such programs in the artistic realm suggest, to many with the creative community, a certain oversight of this burgeoning phenomenon. To some, image-generating prompting programs are perceived as autonomous entities, potentially supplanting human creativity entirely, complicating the acknowledgment of the artist’s inherent authorship in defining an artwork. However, this presupposition, as we have already demonstrated in this essay, encounters a dissonance with reality.
It is good to remember at this point that already Walter Benjamin in 1930s and later with John Berger in the 1970s discussed about the artistic essence, that could transcend the ritualistic and artisanal aspects of manual production. Hence, the utilization of a tool does not definitively determine the aesthetic value of a work.
The excessive appraisal accorded to AI programs by artists must give way to their redefinition as artistic instruments. While not innately groundbreaking, these tools fundamentally shift our relationship with imagery. The potential for this transformation is realized through art exhibition showcasing AI-generated works.
One such exhibit, The Rights from Future Generations – A Perspective on (A)rt and (I)nnovation, held at Reggia di Monza in September 2023 and that I have curated, presents the works of three artists who have integrated AI software into their creative process: Francesco D’Isa, Roberto Fassone and Andrea Meregalli.
The exhibition pathway is introduced by Francesco D’Isa, philosopher and digital artist, who presents a curated selection of Errori (2023). These images acknowledge the unexpected element generated through the software, Midjourney, reveling an imagery partially unfamiliar to the artist’s usual compositions. Engaged in the contemporary discourse surrounding the revolution of algorithms in the art world, D’Isa has always regarded machine learning-based software as a «revolution for visual art» (D’Isa, 2022) and it is demonstrated nowadays with publication of his book, the title of which is a declaration of his thought. The artist reminds us that these software programs «are not anthropomorphic androids with their own intelligence and personality, but rather algorithmic models base on vast amounts of data created by humans» (D’Isa, 2022). While one facet exposes the limitation of a preselected training, on the other hand, there remains an ongoing inquiry about whether the resulting output can truly be defined as an artwork. Responding to this question over two years ago, Francesco D’Isa stated: «Recognizing and incorporating the machine’s styles and errors in its response will lead the only intelligence at play, namely our own, to invent something new» (D’Isa, 2022). This declaration is translated in this exhibition in a tangible representation through the series, where the error, often considered a potential discard, becomes the ultimate outcome, illustrating that «the human creative act» – as D’Isa himself remind us – «is fundamentally a choice, and therein lies the sole necessary condition for making art» (D’Isa, 2022).
Adjacent to D’Isa display, And we thought V by Roberto Fassone curated by Sineglossa, that is part of the broader research endeavor known as Food Digestion. Fassone, known for his exploration of the peculiarities of human existence, collaborated with the transdisciplinary cultural organization Sineglossa to create Ai Lai, an artificial intelligence fueled by accounts of journeys taken by individuals under the influence of hallucinogenic mushrooms. Initiated in 2021, the project now in its fifth iteration features a single-channel video and ten posters. Within the exhibition, And we thought V seeks to delve into the interactions between humans and machines in non-ordinary circumstances, revealing the increasingly noticeable parallels between the two. Much like a human, Ai Lai has learned to «compile its own psychedelic reports» (Sineglossa, 2023). Fassone, at a pivotal juncture, draws from Ai Lai’s narratives, using them as the foundation for the creation of posters, film and music. This innovative approach seems to navigate against the conventional current, offering a gateway to new realms of imagination through Fassone’s distinct artistic methods. This multifaced and distinctive proposition aims to take on an enlightening role within the exhibition’s context.
In a subsequent gallery space, Untiled (2023) by Andrea Meregalli is on display, showcasing a meticulous and sometimes schizoid process that characterizes his practice invites reflection on the infinite combinations shaping the contemporary aesthetic imagination. The displayed works reveal the integration of artificial intelligence software with images crafted manually by the artist. Through numerous stages of blending and prompting, aiming to achieve results from the software that are as unexpected and unpredictable as possible, Meregalli seeks to disrupt the very principle of image generation through AI. This typically necessitates highly precise textual input to generate the desired image. Meregalli positions himself as a mediator of a hybrid, undefined process, prompting the creation of a distinct category to address the current needs of defining an artist based on the tools they utilize. The notion that the device serves as an instrument for him appears quite intuitive here. Referenced within his works as a technique of realization, it does not seem to hold a primary role. Yet, the awareness inherent in this manipulation of reality becomes evident in how the advent of these devices has allowed his practice to evolve beyond even the envisioned parameters. The deliberate short-circuit desired and sought by the software thus admits the possibility of creating dystopian images that effectively translate the ‘monster of his psyche’ – abstract entities bearing hidden truths (Mascellaro, 2023).
Andrea Meregalli, Untitled, 2023, immagine realizzata con Midjourney, courtesy Andrea Meregalli
Shifting perspectives
The intersection of art and AI presents a compelling narrative of innovation, creativity, and the evolving nature of artistic expression. Through the examination of various artists, who use AI tools in order to generate artworks, we have navigated a landscape that challenges conventional notions of authorship, creativity and the role of technology in the artistic process. D’Isa’s exploration with synthetic images in Errori series reflects a deliberate disruption of traditional artistic processes, embracing the unexpected outcomes generated by AI algorithms. Fassone’s collaboration with Ai Lai delves into the realm of psychedelic experiences, using AI as a conduit to narrate journeys and explore the boundaries of human-machine interaction. Meregalli’s fusion of AI software with his manual creations unveils a new dimension of aesthetic possibilities, blurring the lines between natural and artificial creation.
The exhibition of these pieces within gallery spaces challenges traditional definitions and underscores the evolving landscape of artistic practice. While some may view AI as an autonomous creator, potentially supplanting human creativity, a deeper examination reveals a symbiotic relationship where artists act as a mediators, curators, and collaborators in the creative process.
The agency of the artist in guiding AI tools, the deliberate disruptions of AI-generated imagery, and the exploration on new aesthetic realms all point to a redefinition of artistic practice in the ‘artificial’ age. As we navigate this evolving landscape, it becomes evident that AI art is not merely about the final product but also the dialogue, experimentation, and exploration it fosters between human creators and machine intelligence.
In conclusion, the emergence of AI art challenges us to rethink established paradigms, inviting us to embrace the possibilities of collaborative creation and the ever-expanding boundaries of artistic expression. As we witness the ascent and admission of AI-generated works ad legitimate artworks, we embark on a journey of discovery, creativity, and continuous evolution within the realm of contemporary art.
Bibliography
Barale A., Arte e intelligenza artificiale. BE MY GAN, Jaca Book, Milano, 2020.
Boden M. A., The Creativite Mind: Myths and Mechanisms, Routledge, London 2004.
D’Isa F., La rivoluzione degli algoritmi nel mondo dell’arte in «Il Tascabile», (21.07.2022).
D’Isa F., La rivoluzione algoritmica delle immagini, Luca Sossella Editore, 2024.
Grunert F.P., Craglia M., Gómez E., Thielen-del Pozo J., HumaniTies and Artificial Intelligence, Noema Publishing, European Union, 2022.
Mascellaro V., Arte e intelligenza artificiale, quale relazione? in «ArteIn» n.16, settembre/ottobre 2023.
Sineglossa, We gave magic mushrooms to an a.i., pamphlet for the exhibition The Rights from Future Generations – A Perspective on (A)rt and (I)nnovation, 2023.
Vittoria Mascellaro (Monza, 1996). After I received a degree in Philosophy at Università degli Studi in Milan, I attended a two-year specialisation in Visual Arts and Curatorial Studies in NABA. I am an independent curator and I investigate the relationship between art and artificial intelligence both from an ethical and curatorial point of view. At the moment I am “cultrice della materia” Sociology of Art at Accademia di Belle Arti in Catania and professor of Introduction to AI and Ethics in Artificial Intelligence at ITSAR Milano.