Half a Century Before Midjourney, There Was AARON

In 2022, New York’s Museum of Modern Art (MoMA) exhibited and later acquired “Unsupervised – Machine Hallucinations – MoMA” (2022), a massive screen-based digital artwork by the Turkish-American artist Refik Anadol that filled the museum’s midtown lobby with undulating waves of color and sound. Its immersive visuals, generated by a machine-learning model trained by a data set of images of objects from the museum’s 200-year-old collection, along with its modernist living room-like environs, was at once heralded and vilified in the art press. Nevertheless, its acquisition was an institutional power move that captured the cultural zeitgeist by positioning bleeding-edge artificial intelligence within the art historical canon. 

One year prior, the Whitney Museum of American Art quietly began collecting AARON, a prototypical version of artificial intelligence software by the late London-born painter-cum-programmer Harold Cohen that evolved continuously from designs made in approximately the late 1960s to his death in 2016. Drawn entirely from the museum’s collection — it is the only institution actively collecting multiple versions of the software — Harold Cohen: AARON complicates, and clarifies, the historical narrative around artificial intelligence and art. The exhibition was organized by longtime Curator of Digital Art Christiane Paul, a colleague of mine when I served as the museum’s director of digital media. It introduces new audiences to Cohen’s work, which was shown widely earlier in his career. When viewed within the cultural context of the present, Cohen’s practice — which includes a series of self-designed plotters and painting machines run by the AARON software, along with a trove of wildly vibrant drawings and paintings produced by both — proves prescient. 

This show makes an extraordinary effort to demonstrate Cohen’s breadth of process, positing that computer code itself might be considered an art object that, in this case, begets other objects. Several plotting machines draw versions of AARON in the galleries in real-time. Equally impressively, conservators recreated the technical environment (a process known as emulation) to run one of the earlier versions of the software (“AARON KCAT,” 2001), which is projected on the gallery walls, revealing how Cohen’s works were drawn over time. Surrounded by the physical drawings and paintings they produced, the machines and projections lend dimension to our understanding of how AARON functions, not to mention pleasure in the experience of looking. 

Cohen’s background as a painter clearly informed his work as an engineer:  His diagrams and notebooks in particular, which often contain small sketches and are situated throughout the galleries, signify his concern for figuration. How could AARON “understand” the human form well enough to produce a responsive gesture, for example, or to situate a figure in relation to its surroundings? AARON constitutes what is known as “symbolic” AI — a much simpler, rules-based system that the artist devised himself, writing code freehand. This early version of artificial intelligence is based on individual knowledge — not the massive mashups of data, or Large Language Models (LLMs) that current versions of artificial intelligence software train on. The resulting drawings and paintings, with their Fauvist palette and wide-ranging level of abstraction, demonstrate the software’s evolution: Cohen wrote around 60 versions of AARON in his lifetime. His prescience as a programmer and an inventor is, in my estimation, as relevant as — and perhaps even more exciting than — the artworks produced by AARON, as joyous as they are. 

Cohen’s work, shown in this context, invites an institutional metacommentary, as well as a social one: Lack of transparency on the part of the companies that develop AI platforms such as ChatGPT, DALL-E2 and Midjourney is one of the core sources of public anxiety surrounding modern-day artificial intelligence. In terms of the arts, AI poses an exceptional threat to closely held ideals around artistic originality, as evidenced by any number of recent copyright lawsuits against major AI companies. 

As museums wade into this highly fraught fray on increasingly grand scales, they create the conditions for a unique form of public dialogue about AI, one situated within the work of art itself. Old questions about how digitally based artworks might be displayed and interpreted arise anew, posing opportunities to revisit history. The Institute of Contemporary Arts London, for example, hosted a symposium at the end of February to commemorate its groundbreaking 1968 exhibition Cybernetic Serendipity — and push museological conventions that account for a new era of AI. 

The ultimate existential concern surrounding artificial intelligence, one that permeates so many industries, is that AI is developing more rapidly than human capability — that it will somehow supplant humanity itself. Indeed, not a day goes by when we aren’t confronted with hysteria-steeped media speculation around the subject. As an exhibition that effectively demystifies technology, Harold Cohen: AARON offers us the space to contemplate the still-symbiotic relationship between the human mind and the machine. It says, “We could be fine, after all.” 

Harold Cohen: AARON continues at the Whitney Museum of American Art (99 Gansevoort Street, West Village, Manhattan) through May 19. The exhibition was organized by Christiane Paul, Curator of Digital Art, with David Lisbon, Curatorial Assistant.