top of page

Dialoguing with Machines

Elastic, entangled forms sprout like vegetation from the base of the image frame. Groupings of blues, purples, and teals push against the yellow ground as color and composition meld in this vibrant abstraction. Peculiar as the imagery may be, the novel methods for its production may overshadow its aesthetic qualities. It is the result of a decades long collaboration between the artist Harold Cohen and an artificial intelligence referred to as AARON.

Harold Cohen and AARON, 040502, 2004. Source.

Through extensive research and development, Cohen crafted a system that could create original imagery and over the course of decades adapted and refined AARON to suit his audience and ideas. While AARON faced mixed responses from viewers, that was more a result of its novelty than its inherent qualities. As one of the first in a new generation of artists working with technology, Cohen and AARON exemplified the generative nature of these collaborations and forged a path for future artists in the field of art and technology.

A Restless Painter

Harold Cohen began his career as a painter and rapidly became proficient and well respected. He produced dynamic abstractions that dialogued with abstract expressionism, received numerous awards for his work, and even represented Great Britain in the 33rd Venice Biennale (McCorduck 1991, 3). His work, Vigil Completed, almost vibrates with a reckless sort of energy. The painting features rugged line-work that ripples across a textured surface. Its aesthetic appears unapologetically emphatic and modern. But although Cohen’s paintings lacked hesitation, his drive as a painter began to falter as his momentum shifted.

Harold Cohen, Vigil Completed, 1966. Source.

Cohen later looked back on his work and acknowledged his accomplishments, but as he described, “I wish that somebody else had done them; I didn’t really feel they had anything to do with me” (McCorduck 1991, 20). Quite simply, at some point Cohen’s interest in his relationship to painting faded and he began to feel drawn to new areas of study. This transition took time, but eventually led him to develop AARON, who in turn helped Cohen to reinvent his relationship with painting.

Observations on Image Construction

Cohen’s practice as an artist truly began to radiate outward from painting as he researched and considered the psychology behind image construction. As an avid thinker, he was drawn to abstract ideas and their application. This methodology led to a revolution in his artistic practice, based in the fundamental question, “What is the minimum condition under which a set of marks functions as an image?" (Ostrow 2009, 123). While many spend their efforts debating what qualifies as art, Cohen cut to the essence of the question and used his own experiences and observations as the basis for his research.

He observed his children as they drew scribbles and doodles and he watched them negotiating boundaries and developing meaning through the process (Cohen 2018, 46). Cohen recognized the importance of studying early human development as a way of understanding basic human tendencies. Eventually, his observations and research would come to shape both his understanding of the viewer’s relationship to art and how best to design AARON (Ostrow 2009, 123).

Application through Technology

In the same way that Cohen felt drawn to research the psychology behind children’s drawings, he also felt a pull towards the use of modern machines. His curiosity about computing in part arose because, as he described, there “were more interesting things happening outside my studio than inside it” (Cohen 2018, 42). Once he had his foot in the door and gained access to computing technology, he got to work and soon began developing AARON.

During the decades he spent working with AARON, he developed numerous strongly developed opinions, especially about creativity and intelligence. In particular, he stated that, “Machine intelligence is not the same sort of thing as human intelligence” (Cohen, 2002, 60). Indeed, we have every reason to believe that human and machine intelligence function inherently differently, despite the fact that humans design and train machines.

While there were overlaps in his study of the human mind and AARON, there was a point where these fields of research came into stark contrast and Cohen “was obliged to devise strategies for an increasingly alien entity” (Cohen 2002, 62). Cohen believed that AARON’s intelligence resided in its ability to learn from its experiences in a way that caused it to develop “an original visual inventory that reflects its evolving "intelligence”” (Ostrow 2009, 122). But Cohen always acknowledged that AARON functioned in its own way, which created space for feedback between him and the machine.

Creativity arose from the discourse between human and machine and thus did not belong to one or the other (Sundararajan 2021, 413). Cohen experienced bursts of energy that arose from his dialogues with AARON and they formed a unique partnership that strengthened Cohen as an artist and established him as a pivotal figure in the movement of artists working with technology.

Dialogue with AARON

Cohen learned to make sense of the nature of his discourse with AARON through a philosophy introduced to him by Louise Sundararajan. The idea arose from the work of Charles Sanders Peirce who believed that thinking arose from dialogue that followed a “triadic structure of self-other-self” (Sundararajan 2021, 413). AARON constituted the “other” in this paradigm. With that structure in mind, the shifts in AARON’s design make a great deal more sense.

For example, Cohen initially needed to color in AARON’s images for it. But with time he was able to train AARON how to color the images. However, later on he changed his approach and reclaimed the task of coloring for himself. Throughout each of these stages, he negotiated the balance of conversation between him and AARON until he found a steady state that suited the nature of their relationship. (Cohen 2018, 46-49)

While his approach may have varied from time to time, the drive to work with AARON remained consistent since Cohen discovered renewed meaning through their collaboration. As he described, he began to think that his work with this technology “would allow me to get back on course, and to examine what was in my head in a clearer way than I had ever been able to do by painting alone” (Cohen 2018, 43). Working with AARON created momentum; however, it also produced a new kind of resistance in relation to the viewer.

Harold Cohen and AARON, Socrates' Garden, 1984. Source.

The Viewer According to Cohen

While Cohen appeared to possess a relatively independent metric for success in his work, he also invested a great deal of thought and effort into connecting with his viewers. The idea central to his posture was that the artist’s goal ought to be “to reach an audience capable of hearing what he has to say, without compromising what that is in the process” (Cohen 2018, 42). Cohen knew that given the right prodding and searching, he would be able to situate his work where the right audience could observe it, but the translation of meaning lay solely in the domain of the viewer.

Cohen wrote that he thought of artwork as “meaning generators that evoke meaning in the viewer rather than inform the viewer what someone else […] intended to communicate” (Cohen 2018, 44). His observations of children drawing may very well have strengthened this conceptual framework. The artistic intent of a child may be difficult to decipher, but the images themselves possess meaning unique to the artist and to the viewer.

By playing with the boundary of what constitutes an image, Cohen arrived at the thin veil separating his understanding of his work from the viewer’s. But rather than try to penetrate that barrier, he recognized it as an inherent aspect of the exchange from “self” to “other” to “self” again (Sundararajan 2021, 413). He chose to adapt as needed to the viewers’ responses while remaining true to his process and purposes.

AARON and the Viewer

One prominent example of Cohen’s willingness to adapt to the viewer is the “turtle.” While the thinking, planning, and designing of the image resided in AARON’s software, a small robot that resembled a turtle would manually produce the artwork, functioning as a sort of body for AARON. However, the turtle unintentionally captivated the audience and distracted from the drawings themselves (Cohen 2018, 44). Routine processes, such as the robot washing out its cups and brushes, captured the interest of the viewers and effectively detracted from Cohen’s greater efforts with AARON (Cohen 2018, 49).

Over time, Cohen found better ways to produce the images in front of the viewers while directing their focus where he intended. Much later on, Cohen even transitioned AARON to a program that could be accessed and observed online (Cohen 2018, 50). As with any work that relies on modern technological systems, AARON morphed and evolved to match Cohen’s intentions. Likewise, Cohen reformed his methods to facilitate dialogue with his viewers. In many instances, he was met with enthusiasm and admiration. But in others, he was confronted with blank stares or an aura of unimpressed disinterest and even contempt.

Social Engagement

What Cohen attempted to accomplish was thoroughly unprecedented and, consequently, received mixed reactions. As Cohen explained, he could not rely on previously made justifications for his effort or prior examples that proved what computers could do for contemporary artists (Cohen 2018, 43). As a result, the public’s response fractured into two general categories: “un-sceptical believers and unbelieving sceptics” (Cohen 2018, 44). The first group accepted Cohen’s project with enthusiastic vigor, but without genuine understanding. Meanwhile the second group, the skeptics, did not acknowledge AARON for what it was and some of the literature on Cohen’s work reflects this bias (Cohen 2018, 44).

One publication makes the determination that Harold Cohen prioritized development of artificial intelligence over “issues germane to contemporary art” and concluded that “AARON'S images, whatever their formal virtues, demonstrate how far we have to go before art as a human activity is truly threatened” (Ostrow 2009, 123). These blunt and absolutist conclusions fail to engage effectively with Cohen’s vision for AARON and display the unfounded fear that machines somehow pose a pending risk to humanity.

Responses like these are not limited to Cohen’s work and are a fairly common reaction to the intersection between any technological progress and art. But it seems odd that they would have been directed at Cohen since he never expressed the desire to take over the ‘art world’ using machines. In fact, works produced by AARON in 1983 were available for a mere $10 during his show at the Brooklyn Museum (New York Times, 1983).

Responses to AARON and Cohen fluctuated over time, but there was a particular interaction with a viewer that created a remarkably positive impression on Cohen. During one of his shows while AARON produced artwork before the viewers, Cohen watched as one particular individual lingered for quite a long time before saying, “The Medicis didn’t know what they would get when they commissioned Michelangelo to do something, did they?” (Cohen 2018, 49). After asking this question, he requested to buy the piece that AARON would make next, without knowing how it would turn out. Following this exchange, Cohen was struck with the idea that, “For $25 this man has bought himself a place in history and won a new understanding of his relationship to art” (Cohen 2018, 49).

The patron that day managed to span an invisible gap between past and present understandings of art, viewership, and patronage. This interaction continued to inspire Cohen’s efforts to make AARON accessible online in the hopes that it could continually make art while the viewer, the patron, watched. The point, while lost upon some, was always was to watch AARON and to see it learn and adapt as it produced work, just as Cohen had observed his own children producing art. Distinct from humans in innumerable ways, AARON was the embodiment of Harold’s questions, observations, and conversations. It was never meant to be a competitor, rather a collaborator.

Machines, Mortality, and the Melancholy

While Cohen and AARON’s partnership carried on for decades, it came to a sudden end in 2016. Prior to this, Cohen often liked to joke that he “would be the first artist in history to have a posthumous exhibition of new work” (Cohen 2002, 59). This was not so much a goal in his practice as it was a curious musing that he liked to chew on. AARON could theoretically have continued Cohen’s legacy indefinitely. However, Cohen acknowledged the unfeasibility of that actually happening and believed that AARON’s work without him would be “original only in a very limited sense” (Cohen 2002, 64).

When thinking about AARON pressing on in his stead, he stated, “I have to confess to feeling less like a potential immortal than like Moses looking out over the promised land he will never enter (Cohen 2002, 64). At the root of it all, he could not imagine why someone would take his place in working with AARON when they, “should build up their own other selves” (Sundararajan 2021, 412). His hunch proved itself to be true when Cohen passed away in 2016. In the absence of his direction and drive, AARON became collateral, a final song drifting into silence (Sundararajan 2021, 412). Fragments of Cohen and AARON’s work remain accessible through a website documenting their collaborations, but the following note prefaces the page:

Enquiries about the availability of work for exhibition, reproduction or purchase contact The Harold Cohen Trust at:

Below this, a grid of videos functions as a sort of dynamic gallery. But the effect inspires a haunting question. Where and what is AARON now that Cohen is gone? Odds are that I will never be able to answer that question and in some ways I do not want to. Cohen was too entwined in AARON and perhaps it would be even more tragic for AARON to linger on without him. In a way, AARON saved Cohen and inspired him with new purpose and motion and we can choose where to find the meaning in that, in our own ways, as Cohen would have wanted us to.

I, for one, am content knowing that while it lasted Cohen could see through AARON where he wanted to go next. With a melancholy sense of curiosity and respect, I now hear him say, “at least I have the feeling that, without being able to see more than a very little way down the road, I do have my head pointing in the right direction. That is a feeling every artist needs” (Cohen 2018, 50). I must agree; that is a feeling I need too. But now it is his legacy that guides me onward.

As with the use of any new technology, AARON confronted numerous challenges and biases and Cohen met them with a willingness to adapt counterbalanced with a sense of resolve. He modeled a lifelong collaboration with AI and his example demonstrates the potentially generative nature of these partnerships. AARON renewed his love for painting and inspired him to reconsider and redefine the boundaries between artist and viewer.



“Aaron the Computer,” The New York Times. (New York, N.Y: New York Times Company, 1983),

Late Edition (East Coast) edition.

Cohen, “A Million Millennial Medicis,” in Explorations in Art and Technology, Second ed., ed. Linda Candy, Ernest Edmonds, and Fabrizio Poltronieri (London: Springer-Verlag, 2018).

Cohen, Harold. “A Self-Defining Game for One Player: On the Nature of Creativity and the Possibility of Creative Computer Programs.” Leonardo (Oxford) 35, no. 1 (2002): 59–64.

McCorduck, Pamela. Aaron’s Code : Meta-Art, Artificial Intelligence, and the Work of Harold Cohen (New York: W.H. Freeman, 1991).

Ostrow, Saul. “HAROLD COHEN.” Art in America. (New York: Brant Publications, Incorporated, 2009).

Sundararajan, Louise. “Harold Cohen and AARON: Collaborations in the Last Six Years (2010–2016) of a Creative Life.” Leonardo (Oxford) (2021): 412–417.

22 views0 comments


bottom of page