Christiane Paul on Curating Cohen’s AARON

Christiane Paul's curation of the Whitney's Harold Cohen: AARON informs a conversation with Peter Bauman (Monk Antony), where they discuss Cohen's persistent impact to this day while reflecting on the nature of digital expression.
About the Author
Harold Cohen, AARON KCAT, 2001. Screenshot. Artificial intelligence software. Dimensions variable. Whitney Museum of American Art, New York; purchase, with funds from the Digital Art Committee 2023. © Harold Cohen Trust

Christiane Paul on Curating Cohen’s AARON

Christiane Paul's curation of the Whitney's Harold Cohen: AARON informs a conversation with Peter Bauman (Monk Antony), where they discuss Cohen's persistent impact to this day while reflecting on the nature of digital expression.

Peter Bauman: You’re curating Harold Cohen: AARON at the Whitney Museum of American Art in February 2024. How does this exhibition speak to the peculiar challenges we face with AI as a society today? In other words, why this exhibition now?

Christiane Paul: The exhibition definitely can speak to all of the peculiar challenges we're facing with AI as a society today. But AI's role in art and culture obviously has become a really pressing topic since the mainstream arrival of software tools such as DALL-E, Midjourney and Stable Diffusion from 2021 onwards.

Artistic practice involving AI has definitely gained new momentum so revisiting Harold Cohen's AARON is, I believe, crucial at this moment in time since it can teach us so much about AI art practice.

First of all, AARON really illustrates the long history of AI art, which hadn't just arrived with text-to-image tools. AARON underscores that artists have been engaging with artificial intelligence for decades, basically since its existence. The other aspect is that AARON shows us a very different approach to art-making with AI than the current tools do at this moment. AARON obviously is early; it employs symbolic AI. It's a kind of expert, procedural system that operates on the basis of rules and hasn't been trained on previously existing images. It therefore doesn't entail any of the standardization, averaging or optimization that is used in the current models that have been trained on massive data sets of existing images.

So AARON was always coded as an art-making program that has external knowledge of the world and internal knowledge of how to visually represent that. It's a very different proposition or an approach to AI that I think is interesting. But of course there are so many important issues of authorship and agency that AARON raises and that are really crucial to today's discussions, I would say.

Peter Bauman: Those questions of authorship and agency have become even more convoluted since Cohen’s time. What do you think AARON continues to contribute to those discussions today?

Christiane Paul: Harold Cohen, of course, really conceived AARON as a collaborator, as another self. He encoded his vision of art into this system. That is very different from artists working with tools, even if they're doing it in very sophisticated ways, such as Stable Diffusion, DALL-E or Midjourney, because they are more engaged with a dialog of tweaking an existing optimized system. While what Harold is doing is really having control over authorship. The question of who is the author is clearer, I think, in AARON, because Cohen created the tool, while with corporate software, obviously, that authorship isn't there. So there, the engagement lies much, much more in how artists explore the specific technology, how they tweak prompts, how they train and work within that system. While with AARON, the questions are different and there's more control when it comes to the agency of the software and also collaboration with it. I would not say that an artist who is using text-to-image models today collaborates with them in the sense that Cohen collaborated with AARON, where it was a constant back and forth between those entities. So in a nutshell, AARON asks important questions about where agency resides in work with AI and what form it can take.

Harold Cohen, AARON KCAT, 2001. Screenshot. Artificial intelligence software. Dimensions variable. Whitney Museum of American Art, New York; purchase, with funds from the Digital Art Committee 2023. © Harold Cohen Trust

Peter Bauman: What contemporary artists do you see asking those similar questions? Whether a prompt artist or not, somebody that comes to mind is Sougwen Chung, who I had the fortunate opportunity to chat with a few months ago about Cohen’s AARON.

Christiane Paul: Yes, Sougwen Chung is more in the vein of AARON, in the sense of creating the system as a collaborator. I think that's also baked into that work. In terms of text-to-image tools, I was thinking about David Salle, for example, in terms of really training the AI or having a similar approach to the line as he does, or to structuring the canvas and then tweaking that through prompts by poets. So I think that is also a model or an approach to working with AI that is more geared towards a form of collaboration.

Peter Bauman: You collaborated on a book with two other authors, including the wonderful Margot Lovejoy. In that book, Context Providers (2010), you write about the importance of artists as context providers, specifically in the digital age. In what ways do you think Harold Cohen was a context provider?

Christiane Paul: I think Harold definitely was a context provider for artmaking with AI because of all the questions he raised. He's a context provider for a form of automated artistic practice and what that means and what this engagement with systems means. Harold's art was always art about meaning-making in art and art itself. And I think he also provides context for that. He invites us to ask what art is and what defines art. He always made work, inquiring at what point a mark makes meaning and how an artist perceives and represents the world. So he's really getting to the essence of image-making or of painting itself and providing context for that. At the core of AARON is, of course, Cohen's freehand line algorithm that makes all the output look like it has been drawn by hand. For Harold, something could be art only if it had intention. And AARON is all about the intentionality of the line. That is also part of the context that particular work is providing.

Peter Bauman: And speaking of AARON's actual mark-making, I think it's really important that you included these live demonstrations of pen plotters at the Whitney exhibition. Can you elaborate on how they enhance the viewer's understanding of AARON's creative process and its significance?

Christiane Paul: I think what these live pen-plotting demonstrations are adding here is that they nicely demonstrate the output of AI across different materialities. The creations of the AARON software are shown within the exhibition as projections. They're shown as paintings and drawings and as these plotter drawings, created live in the gallery. What that shows us is what it takes for software to generate art, an image, and what it takes to make that image manifest. I think that can vanish a little bit in pure digital imagery and screen-based imagery. We're so used to today's AI just generating something for us on the screen. Having the software drawing in the gallery and being plotted live gives you a different relationship to image-making and to materiality. It also really highlights different qualities when it comes to media. The plotted drawings obviously have a very different feel to them than the paintings on the wall or the projections that you see on the wall. So on the one hand, it underscores the interdependency of the output’s materiality, how software can manifest in very different ways. And on the other hand, it enhances medium specificity when it comes to the different media.

Harold Cohen, AARON KCAT, 2001. Screenshot. Artificial intelligence software. Dimensions variable. Whitney Museum of American Art, New York; purchase, with funds from the Digital Art Committee 2023. © Harold Cohen Trust

Peter Bauman: I love showcasing the breadth of software’s materiality, which could be read as a challenge to our screen-dominated digital consumption. I'm wondering if you see this exhibition as challenging or contextualizing the screen-based text-to-image AI art of today?

Christiane Paul: I would say both. And probably “both and..." On the one hand, it contextualizes text-to-image models. Historically, it shows that there have been different approaches and trajectories to it. I think many people who are coming to image-making with AI only now and whose first experience with it is with these text-to-image models might think that this is the only way AI operates and makes images. And obviously, it's not. There are different approaches to it so there is a contextualization happening. On the other hand, I would also say there is a certain kind of challenge, or maybe even a proposal. You could see AARON as a proposal for a combination of more old-fashioned expert systems with today's statistical text-to-image models. And personally, I think that AI will return a bit more to expert systems. That's already underway. I think the future will be maybe a little bit more of a combination of the two—the rule-based approach with the approaches we're seeing right now.

Peter Bauman: Why do you think there wasn’t more creative investigation of Cohen’s symbolic-logic approach and that we moved towards the networked, big-data AI of today?

Christiane Paul: Well, I think one part of it is simply processing power and what became possible. And I also fully agree that there is enormous potential in it.

AI can certainly achieve more by being trained on these massive data sets than humans trying to encode the knowledge of the world in the system, which is what Harold Cohen tried to do. That is ridiculous on one level but also marvelously beautiful.

So I think it came with technological evolution and different technologies converging. Neural networks have also existed for a long time but through this convergence of neural networks, computing power and big data, we saw a completely new era. But there still is a role for more symbolic logic that can work in tandem with the current approaches.

Harold Cohen, AARON KCAT, 2001. Screenshot. Artificial intelligence software. Dimensions variable. Whitney Museum of American Art, New York; purchase, with funds from the Digital Art Committee 2023. © Harold Cohen Trust

Peter Bauman: This exhibition might even help spark such approaches. You touched earlier on Cohen's conception of AARON as this other self. Other artists I've spoken to have mentioned something similar: William Mapan and, again, Sougwen Chung. I thought that was really what made Cohen’s AARON so contemporary. It recognized so early on the power of computers to not only enhance creativity but to actually amplify our very beings and our artistic beings. Could you share more insight into Cohen's perspective on creativity and this relationship between the artist and the program?

Christiane Paul:
This idea of the other self, of course, means that AARON was encoded with Harold Cohen's sensibilities as an artist and his specific perspective of the world and rendering that world.

AARON is almost a form of self-portraiture but in a very different way. Not in the sense that it's creating a portrait of Harold Cohen but it's a form of self-portraiture of an artist and an artist's sensibility.

I mean, I absolutely agree with you that, on a general level, there's this element of another being. But in this particular case, this idea of the other self also very much means an image or representation of a specific artist's aesthetics and sensibility, which I find very interesting.

Peter Bauman: Cohen would in some way have been familiar with traditional self-portraiture because he started off as a painter, and quite a successful one. He studied at the Slade and was even an art history professor in the 1950s. He then went into plotted computer art in the ‘60s. So he had quite an established and fascinating career well before AARON in 1973. Considering his contributions to systematic painting, computer-plotted art, AI and digital art, how does this exhibition address his legacy and impact on so many intersections of art and technology?

Christiane Paul: As you say, it's important to keep in mind that Harold was a very established painter. He represented the UK in the Venice Biennale in 1966. Tate had an exhibition that same year, Two Decorative Works by Henri Matisse and Harold Cohen. His paintings also were very interested in systems. One of them is actually titled Systems. I think he also got frustrated with painting as a medium and the limitations of that medium. I always want to be careful about making overreaching claims of impact on the intersection of art and technology per se. But what Harold Cohen's contributions certainly do is highlight the complexities of that art-technology intersection. His work makes it very clear that technology is not only a tool but that artists also can conceive that tool. What he created was an artist-written tool. So it's a very active engagement with that technology.

What Cohen's work with AARON really highlights is the inherent characteristics and features of the digital. It's generative, modular, iterative, procedural, potentially real-time and interactive nature. Basically, once again, the medium-specificity of technological art is, I believe, at the core of Harold Cohen's legacy.

Peter Bauman: You just listed the characteristics of digital art, which are part of your iconic definition of digital art as “art that explores digital technologies as a medium by making use of its medium’s key features, such as its real-time, interactive, participatory, generative, and variable characteristics, or by reflecting upon the nature and impact of digital technologies.” How do you define generative art, and do you consider it more of a movement or a method?

Christiane Paul: I always go back to Philip Galanter's definition of generative art. I think he's really the expert on this. I go with him defining generative art as any kind of art practice in which the artist uses a system, set in motion with some degree of autonomy, thereby creating or contributing to a completed work of art. And that, of course, also means that generative art is not specific to the digital. Generative art is not digital art per se. I would say that digital art is a subcategory of generative art, or generative digital art, at least.

I'm not sure if method is the best term, but definitely an approach and movement. For me, movement is always socially defined as a group of artists at a specific moment in time, for example, working together. I would say there have been various generative art movements. You could even see Fluxus as an expression of that. I think then if you look at the Algorists, founded by Roman Verostko and Jean-Pierre Hébert, but that included Manfred Mohr, Frieder Nake, Vera Molnár, Herbert W. Franke and others. I think that was a kind of generative art movement. Then you can see a younger generation today; many of them are also working with NFTs, where it has become a form of movement again. All of these stages were different groups of people in different kinds of social environments. So I want to be careful about a single movement per se, because many of the generative art artists today would not necessarily always see themselves in that company or in that tradition.

Harold Cohen, AARON KCAT, 2001. Screenshot. Artificial intelligence software. Dimensions variable. Whitney Museum of American Art, New York; purchase, with funds from the Digital Art Committee 2023. © Harold Cohen Trust

Peter Bauman: Of course you included the Galanter definition in your A Companion to Digital Art (2016), where he emphasized the importance of generative art’s materiality as well. Galanter goes on to touch on issues of aesthetics, classifying generative approaches based on their visual complexity. As the writer of the book Digital Art (2003), I'd really love to get your opinion on digital aesthetics, because it's something that I think a lot about, in particular how to reconcile traditional notions of aesthetics with the digitally native characteristics you listed earlier. For instance, where do you stand on skeuomorphism versus flatness, or digitally native elements? Do you think there's a place for physical representation in digital art?

Christiane Paul: I don't have a stance on that, per se, and I don't stand in a specific place when it comes to that discussion. For me, the question always is what conceptually makes sense and whether skeuomorphism is—I don't want to say justified—but is the best expression for something. I think where it often fails and falls flat is in this urge in virtual worlds to present skeuomorphic representations of the physical for no good reason.

The world you're operating in gives you so much freedom to play with completely different constructs.

So I think there have been a lot of misguided approaches to it. But then again, skeuomorphism can also be very interesting and, conceptually, more than justified. I'm thinking of Claudia Hart's The Ruins, for example, which was a skeuomorphic representation of her Bitforms exhibition. She super nicely played with that, because ultimately all of the elements in the work have been digitally created. She is getting to the porous boundaries between the physical and virtual by creating a skeuomorphic representation of that show. So in that case, I would say this is a very interesting approach to it. I'm also thinking about works such as Clement Valla's Surface Proxy, which really plays with those boundaries. So once again, the question always is: “What makes sense conceptually when it comes to skeuomorphism?”

Peter Bauman: You bring up the importance of concept but another thing that I think a lot about with the digital generative art that I see is a tendency to rely on the visual and visual aesthetics, often over concept. This appears to challenge plastic practices in the contemporary art world, which seems to have moved past this aesthetic pursuit for beauty in a lot of ways. Postmodern thought has tended to not think of beauty in terms of aesthetics like the digital realm continues to. Do you think there’s an over-reliance on visual aesthetics in general within the digital space?

Christiane Paul: I see what you mean. And there is certainly a specific segment of visual practice in the digital that can fall into the trap of visual seduction. I would also point out that I don't think that was necessarily the case in the early 2000s. You had super-engaged conceptual, hardcore-conceptual approaches to digital art that did not necessarily have this seductive visual output. And there's nothing wrong with that, of course, with this form of visual seduction. Digital art generative proceduralism also lends itself to that. But I think there are many artists who are also deeply conceptual in their approach. Maybe that comes in waves because, once again, I would say that that was more pronounced twenty years ago. Right now, we're certainly in a phase where we see more of that visual seduction. But it’s not mutually exclusive. Conceptual work can also be a harder sell. I know that many artworks I deeply admired in the earlier days for their conceptual radicality, are about as difficult to sell as conceptual art with cardboard boxes. So it's not that easy [laughs].

Harold Cohen, AARON KCAT, 2001. Screenshot. Artificial intelligence software. Dimensions variable. Whitney Museum of American Art, New York; purchase, with funds from the Digital Art Committee 2023. © Harold Cohen Trust

Peter Bauman: [Laughs] right, engaging with the viewer on a sensory level to some extent matters. This foundational incongruence in terms of what to pursue may contribute to a sense of skepticism within the mainstream art world toward digital art. But as a digital art curator at a major museum, do you even distinguish between those two worlds as many often do?

Christiane Paul: Not really. And I actually don't think that the skepticism towards digital art easily maps onto the mainstream art world, whether that is the traditional art fairs or museums. But I also deeply believe one needs to be careful about that because that skepticism doesn't necessarily easily map onto those categories. I mean that the skepticism of digital art is much broader than that. And there has been a lot of acceptance of those practices within the mainstream art world. So it depends. It's very difficult to draw the lines between these two worlds or classify them.

Peter Bauman: You, more than maybe any other person, have advocated for digital art’s acceptance over the last two decades. Are you more optimistic today than when you wrote Digital Art in 2003?

Christiane Paul: I would say, “Yes and no.” I know that's always a frustrating answer. We've definitely seen more acceptance of digital art over the years in the quote-unquote mainstream art world. I'm thinking just of recent types of exhibitions: Jacolby Satterwhite at the Met or, whatever you think of the work, Refik Anadol’s Unsupervised at MoMA. Stephanie Dinkins did a night at the Guggenheim on January 25, 2024. So we see these artists, who have been very influential and seminal in the medium, in really mainstream art institutions and getting quite a bit of exposure. I think that's certainly a positive sign. Overall, I would say I'm always a little bit skeptical because we have seen these waves before. For instance, 2000 was a big one when, at that moment, the Whitney did BitStreams and Data Dynamics and suddenly all of the institutions were commissioning net art. And then it all collapsed. I'm not saying that that will happen. I think this is a more sustained engagement but there are also always ups and downs.

What I have not seen change at all, ultimately, is a deeper understanding of the aesthetics of the digital medium and its language—what it really requires and what the needs of this medium are—when it comes to institutional infrastructure. And I think that's a much, much harder battle in which we haven't made that much progress.

Peter Bauman: How do we gain this deeper understanding of the essence of the digital realm that you were just describing? How would you encourage the readers of this piece to engage in a deeper way with digital media?

Christiane Paul: I would say it's always a good start to educate yourself and start following the art in a more consistent way. Show that it has validity because everybody's engagement with it also is a sign that people are interested in it. At this moment in time, there are a lot of organizations devoted to it. I mean, if you look even at what is on view in galleries in New York City or any given city, there might be something. Pay attention. Take a look. When it comes to even reading the New York Times, look at the digital articles rather than just jumping into the tried-and-true. It's also interesting that even at this moment in time, and that's right to my point, many of the reviews of digital art could be buried in the technology or design section and not in the art section.

I would encourage everyone to pay attention and educate themselves about the work.

Look at what digital art organizations are doing, from Rhizome to others. Support them, even if it's a small donation. And for the people who have more monetary bandwidth and can really make a larger contribution, look into the acquisition committees that museums might have. The Whitney has a Digital Art Acquisition Committee. Help build collections and contribute that way.

Because I think that's also one of the major challenges—that digital art hasn't been collected enough. It's getting collected more and more. But most institutions do not have enough digital art to really tell a story of the medium and a story of its evolution.

Harold Cohen, AARON KCAT, 2001. Screenshot. Artificial intelligence software. Dimensions variable. Whitney Museum of American Art, New York; purchase, with funds from the Digital Art Committee 2023. © Harold Cohen Trust

Peter Bauman: That is incredibly good advice and something I will certainly take to heart. I think providing the scaffolding for participants in this space to fully engage with the medium is incredibly important. Related to that, I wrote about a Framework for appreciating generative work holistically by investigating artistic intent. Can you walk us through your inner monologue when you’re assessing the success or failure of a work? How does this differ from static and plastic compositions compared to the digitally native?

Christiane Paul: I hate to disappoint you because I do not have an inner monologue and there are no steps I walk through at all. It's very much a gut feeling and it does not differ at all from me looking at a painting or at a digital artwork because the question for me always is whether a work expresses itself in conceptually and materially sophisticated ways.

Does it make me think about the world or make me perceive the world in new ways? Does it ask interesting, challenging questions? So it's, as always in art, I think, this fusion of concept and of its expression in its materiality.

By materiality, I could also mean software art. I know that's a very meta level. But what that requires, of course, when you dig down a little bit, and this is, I believe, what you're talking about, is an understanding of the medium itself—whether that's the medium of painting or whether that's the medium of digital art. And I think one has to understand that language; one has to be able to see, for example, emergence at work and how it manifests and how it plays out in a work, how an artist sets up this visual or conceptual framework in which something unfolds.

So that is definitely what one needs to pay attention to. Also, if it's interactive, really see response as a medium itself. So the meta-level assessment I'm describing here would not be possible without an understanding of the language of the medium. And I have internalized that so much that I don't need to have it in a monologue.

Peter Bauman: It's automatic at this point.

Christiane Paul: It's like, “Oh, wow, this is great." And then I can rationalize it and break it down into: This works and this is done in a terrific way.

Peter Bauman: So you have the initial reaction and then you break it down further later. Does the piece need to justify that breakdown in the first place? If the piece doesn't capture you, do you bother with the second, third and fourth steps?Christiane Paul: I still do. I would still ask myself, “Why didn't it capture me? Or what's the issue here? Why doesn't it blow me away? Why don't I respond the way I maybe would like to respond to it?”


Christiane Paul is a curator, author and professor who has written extensively about new media arts and given international lectures on art and technology. She received the Thoma Foundation's 2016 Arts Writing Award in Digital Art. Her most recent books include A Companion to Digital Art (Wiley Blackwell, 2016), Digital Art (Thames and Hudson, 3rd revised edition, 2015), Context Providers - Conditions of Meaning in Media Arts (Intellect, 2011; Chinese edition, 2012), co-edited with Margot Lovejoy and Victoria Vesna, and New Media in the White Cube and Beyond (UC Press, 2008).

Peter Bauman (Monk Antony) is Le Random's Editor-in-Chief.